as a PDF
Transcription
as a PDF
Volume 29, No. 1, 2008 Edited by / Édité par Craig Wilson Edwards School of Business University of Saskatchewan FINANCE Proceedings of the annual conference of the Administrative Sciences Association of Canada Finance Division Actes de congrès annuel de la section finance de l’association des sciences administratives du Canada Halifax, Nova Scotia ACKNOWLEDGEMENTS REMERCIEMENTS I would like to thank everyone who helped put this program together. In particular, I appreciate the authors who submitted very high quality articles this year, the volunteers who agreed to chair sessions and discuss papers, and especially the reviewers who provided very detailed reports. Special thanks go to Sean Cleary (Division Chair) and Iraj Fooladi (Conference Chair) for their helpful advice. Je voudrais remercier chacun qui a aidé à assembler ce programme. Particulièrement, j'apprécie les auteurs qui ont soumis des articles très de haute qualité cette année, les volontaires qui ont accepté de présider des séances et discuter de papiers, et surtout les critiques qui ont fourni des rapports très détaillés. Merci spécial est pour Sean Cleary (président de division) et Iraj Fooladi (président de conférence) pour leur conseil utile. Craig Wilson University of Saskatchewan Craig Wilson Université de Saskatchewan ii REVIEWERS / ÉVALUATEURS Sebouh Aintablian Abdelouahid Assaïdi Nizar Atrissi Salma Ben Amor Hamdi Ben Nasr Richard Bozec Mohammed Charmouh Hai Feng Chen Diego Cueto Nabil El-Meslmani Iraj Fooladi Hatem Ghouma Omrane Guedhami Haibo Jiang Paul Kalyta Sana Mohsni Jérémy Morvan Rick Nason Wissam Nawfal Alex Ng Maria Pacurar Anne-Lise Renard-Ronsse Arturo Rubalcava Jacques Saint-Pierre Anis Samet William Sodjahin Chengye Sun In Sunwoo Oumar Sy Samir Trabelsi Rossitsa Yalamova Jun Yang Ashraf Zaman Lebanese American University CREG, CNAM Paris Université Saint-Joseph (Beirut) ÉSG, UQÀM Université Laval University of Ottawa Université de Lille 2 (France) University of Manitoba Concordia University Concordia University Dalhousie University HEC Montréal Memorial University Concordia University University of Ottawa Concordia University Université de Bretagne Occidentale Dalhousie University Concordia University University of Northern British Columbia Dalhousie University Université de Valenciennes University of Regina Université Laval HEC Montréal Université Laval Carleton University CGI Group Inc Dalhousie University Brock University University of Lethbridge Acadia University Saint Mary’s University iii TABLE OF CONTENTS / TABLE DES MATIÈRES Corporate Goverance, Ownership Structure and Bank Performance: Evidence from the Middle East and North Africa (MENA) ……………………………………………………………………………………1 Sebouh Aintablian (Lebanese American University), Wajih Al Boustany (American University of Beirut) Reassessing Canadian Hedging Practices: A Survey Study ……………………………………………15 Ashraf Al Zaman, Karen Lightstone (Saint Mary's University) La rationalité mimétique dans la formation des recommandations des analystes financiers: Etude de terrain ……………………………………………………………………………………………………….32 Abdelouahid Assaïdi (CREG, CNAM Paris) Repurchases and Post-SEO Underperformance …………………………………………………………49 Nilanjan Basu, Haibo Jiang, Parianen Veeren, Neiliane Williams (Concordia University) Reverse Split Announcements, Effective Dates and Survival ……………………………………………62 Marie-Claude Beaulieu, William R. Sodjahin (Laval University) Modèle prévisionnel de la défaillance financière des PME Québécoises emprunteuses …………………81 Salma Ben Amor, Nabil Khoury, Marco Savor (ÉSG, UQÀM) Why do Foreign Firms Issue a Specific ADR? …………………………………………………………98 Narjess Boubakri, Jean-Claude Cosset, Anis Samet (HEC Montreal) Earnings Management and Bond Costs and Ratings ……………………………………………………115 Narjess Boubakri, Hatem Ghouma (HEC Montreal) Securitization buy out: Le cas français …………………………………………………………………130 Christian Cadiou, Nathalie Cotillard, Jérémy Morvan (Université de Bretagne Occidentale) Creation de valeur financière et stratégique lors des fusions et acquisitions ……………………………148 Mohammed Charmouh (Université de Lille 2) Real and Nominal Duration: A Multi-Dimension Hedging Strategy …………………………………168 Iraj Fooladi (Dalhousie University) An Empirical Comparison: Two Special Cases of CEV Option Pricing Model and Black-Scholes Model on S&P Canada 60 Index Call Options …………………………………………………………………182 Haibo Jiang (Concordia University) Making it Personal: The Impact of CEO Retirement Plan on Firm Risk ………………………………195 Paul Kalyta (University of Ottawa) Capital Costs for Domestic and Cross-Listed Canadian Firms …………………………………………211 Lawrence Kryzanowski, Sana Mohsni (Concordia University) iv Do Family Firms use More or Less Debt? ………………………………………………………………228 Imen Latrous (Université du Québec à Chicoutimi), Samir Trabelsi (Brock University) Reconstructing the Historical Performance of Merged Ecommerce Mutual Funds ……………………245 David Rankin (University of Toronto), Jun Yang (Acadia University), Eric Wang (Athabasca University) Sélection de portefeuilles et prédictibilité des rendements via la durée de l’avantage concurrentiel……255 Jacques Saint-Pierre, Chawki Mouelhi (Université Laval) Life Science Venture Capitalists: An Exploratory Study ………………………………………………272 Marianne Schovsbo, Alex Ng (University of Northern British Columbia) Market Valuation of Corporate Charitable Donations …………………………………………………289 In Sunwoo (Concordia University) Does Diversification Explain Market Anomalies? ……………………………………………………305 Oumar Sy (Dalhousie University), Vihang Errunza (McGill University) v Sebouh Aintablian School of Business Lebanese American University ASAC 2008 Halifax, Nova Scotia Wajih Al Boustany Institute of Financial Economics American University of Beirut CORPORATE GOVERNANCE, OWNERSHIP STRUCTURE AND BANK PERFORMANCE: EVIDENCE FROM THE MIDDLE EAST AND NORTH AFRICA (MENA)1 We study the relationship between ownership structure and performance for a sample of 89 commercial, investment, specialized, and Islamic banks in the Middle East and North Africa (MENA) region for the period 1998-2003.) Our results for commercial banks show that the performance of banks is weakened with board/management separation and concentrated ownership, and enhanced with government ownership. I. Introduction The field of corporate governance has grown in recent years into a dynamic area of study due to the increased role of corporations and their impact on the economies in which they operate. The interest in this subject was earlier limited to developed nations; however, with the globalization trend, developing and emerging markets are now incorporating governance restructuring in their reform programs.With the increasing demand for investment funds in developed and developing nations, and the decreasing need for barriers to the free flow of capital, policy makers have now recognized the role corporate governance may play in attracting capital. They also have come to the conclusion that weak corporate governance systems will definitely hinder the efficient allocation of resources, weaken opportunities of competition and ultimately slow down investment and economic development. Ever since the East Asian crisis in mid 90’s and the more recent events concerning the Enron scandal, both public and private sectors have been increasingly involved in finding mechanisms through which to boost investors’ confidence in the firms and consequently the countries in which they operate. In this respect, assessing the corporate governance framework comes as a natural step in that direction. Jim Wolfensohn, the former president of the World Bank, is quoted saying that “the proper governance of companies will become as crucial to the world economy as the proper governance of countries.” 2 1 Partial funding for this paper was provided by the University Research Council at the Lebanese American University. Research assistantship was provided by Paola Boghossian at the School of Business, Lebanese American University. 2 Wolfensohn, J. D. (1999). A Battle for Corporate Honesty, the Economist: the World in 1999 at 38. Financial Times. 1 Corporate governance could be briefly explained as the set of rules, laws and frameworks that govern how fiduciary authority is exercised. Assigning proper controls on the relationships between the different stakeholders is essential in the set up of a proper check-and-balance system and the promotion of well defined lines of responsibility. Nowadays, investors around the globe are demanding improved measures to safeguard their investments and its returns to go along with ever growing flows of capital across borders caused by movements towards globalization. Hence, it is the policy makers’ job to recognize the importance of sound corporate governance as a tool in both attracting these capitals and the creation of appropriate atmosphere for the development of systems for the efficient allocation of resources. The recent crashes in many Gulf Cooperation Council (GCC) stock markets that spread as far as North African countries has shed the spotlight on the need for reform in the corporate as well as the financial sector in the MENA region. Given the fact that the region is heavily endowed by ever increasing capital due to oil revenues, reforms are needed in the banking sector in order to promote better governance strategies and to boost investor confidence. The current paper aims to test empirically the relationship between ownership structure and performance for a sample of MENA banks. The contribution of this study is two fold: First, it aims to fill the gap in banking literature by focusing on the banking sector in the MENA region. Second, it aims to gain an understanding of a significant determinant, ownership structure, which is associated with bank performance. The paper is divided into six sections. Section II summarizes different corporate governance systems and models that are implemented worldwide and reviews the literature. Section III discusses the state of corporate governance in the MENA region with particular focus on the financial sector. Section IV describes the data and the methodology used in the current study. Section V includes an analysis of the empirical results. Section VI concludes with a summary of the main findings. II. Corporate Governance and Bank Performance: Review of Literature The main aspects of corporate governance revolve around the “principal-agent” problem or the agency problem. This problem arises when the owner of capital is different from the person who controls and manages it. There is no single definition of corporate governance that is used today. Rather, different organizations as well as researchers have made considerable efforts to present concise definitions of the term. A slightly more general definition is the one put forward by Shleifer and Vishny (1997) stating that “we can define corporate governance as the set of methods to ensure that investors (suppliers of finance, shareholders, or creditors) get a return on their money.”3 There are three main stakeholders in a certain firm: shareholders, managers and creditors. Problems stem from the fact that in most cases, each of these stakeholders may have different interests in the firm. Agency theory suggests that managers acting as agents for owners have tendencies to pursue strategies that meet their own goals rather than those of the owners (Jensen and Meckling, 1976). From here comes the need for certain control mechanisms that can efficiently manage these interests, so as one is not exercised at the expense of another. Stiglitz (1985) states that the most important mechanism to ensure proper respect for interests is through the concentration with which the financial claims of the firm are held. In other words, if equity of a certain firm is concentrated, this will give shareholders enough incentives to pursue their interests more closely and consequently keep a closer eye on management through increased investment in information acquisition and monitoring. Stiglitz (1985) concludes that large shareholdings give investors the increased ability to control 3 Shleifer, A., and Vishny, R. W. (1997). A Survey of Corporate Governance. Journal of Finance, 52(2), 737783. 2 through the power of voting or representation on the board of directors. In addition, if debt is concentrated in the hands of a few creditors this will give the latter enough incentive to monitor management decisions. The banking system in the world remains a vital sector in both developed and developing countries. It plays a critical role in the transmission of monetary policy and the channeling of savings to firms and households. Hence, it is important to study whether banks are affected by corporate governance mechanisms and whether the agency problem has any severe implications on the performance of these banks. The financial sector worldwide is highly regulated. This has to some extent substituted the weaker mechanisms of corporate control of this sector, a weakness that arises from the oligopolistic nature of the banking sector in certain countries. However, regulatory intervention is seen to be more costly than healthy market mechanisms. Regulators usually are motivated to push for regulations which target reducing the probabilities of failure rather than increasing shareholder wealth. The bulk of research on bank governance is interested in how corporate governance mechanisms vary in different legal and regulatory environments. Research has predominantly concentrated around two different aspects. First, corporate governance in financial firms is contrasted with that of nonfinancial firms. Second, researchers compare financial firms across countries to explain how different legal and regulatory environments affect governance mechanisms. Hannan and Mavinga (1980) studied differences in input expenditure between banks that were classified as ownercontrolled (largest shareholder held more than 25% stake in the bank) versus manager-controlled banks (largest shareholder held less than 10% stake in the bank). Their results show that banks which were classified as manager-controlled operating in noncompetitive markets spent more on items that are likely to be preferred by managers such as salaries and wages, equipment and expenses related to the premises of the banks than did owner-controlled banks.Glassman and Rhoades (1980) examined over 1400 US banks in 1975 and 1976. They studied whether profit rates, costs and growth rates were related to the ownership structure of the bank. Banks were classified as owner-controlled if 5% of owners held more than 60% of equity stake, otherwise they were classified as manager-controlled. The results showed that owner-controlled banks had higher profit rates and that large banks in general exhibited behavior consistent with being under control of managers.Smirlock and Marshall (1983) studied the effects of bank size on the expense-preference behavior of managers. They found that as bank size increases, ownership becomes more and more dispersed since capital demands of large banks surpass individual investors’ abilities. This creates a problem for owners as it becomes more difficult for them to monitor managers. This is reflected in higher expense-preference behavior by managers. Akella and Greenbaum (1988) studied the difference between two forms of institutions: mutual versus stock savings and loans. Mutual savings and loans are legally owned by depositors but they do not have any ownership rights or responsibilities, unlike stock savings and loans where some responsibilities are held by owners and thus incentive and power to monitor managers. They found that mutual savings and loans tend to expand their deposit and loans portfolios beyond profit maximizing levels.Saunders et al. (1990) found that banks with higher degree of management ownership (lower managerial slack) are riskier, since these managers benefit from the incentives provided by fixed rate deposit insurance.Allen and Cebenoyan (1991) examined the relationship between ownership structure in banks and acquisitions. They found that banks with entrenched management tend to be heavily involved with acquisitions. This is explained by managers’ preference for lower risk levels as well as the increased perquisites managers have to gain by increasing the size of the corporation. Murto (1994) presents evidence that Finnish banks in the late 1980s got into trouble by expanding their respective market shares too fast, which is a reflection of managers’ preference for size at the expense of profits. Gorton and Rosen (1995) studied the decline of banking in the 1980s in the US. They inferred that moral hazard problems did not play a major role in this decline. They set the problem at the level of corporate controls, whereby managers ultimately hold the final lending decisions and consequently the amount of risk that could be taken. They found that managers have a tendency to take excessive risks when the industry is unhealthy and their stake in ownership is too large to make outside monitoring very costly but at a point where their ownership is not so large as to align their interests with those of shareholders. However, if the industry is healthy, 3 entrenched managers tend to behave too conservatively. Gorton and Schmid (1999) studied ownership structure in Austrian cooperative banks. They found that firm performance declines as the number of cooperative members increases, corresponding to a greater separation of ownership and control. The decline in firm performance is attributed to an increase in efficiency wages. Altunbas, Evans, and Molyneux (2001) found that there is little evidence to suggest that privately owned banks are more efficient than mutual or public sector banks. Using data on German banks from the years 1989-1996, they concluded that banks of the three forms were able to benefit from economies of scale associated with increased size. Public sector and mutual banks seem to have slight cost and profit advantages as compared to privately owned banks, which are caused by lower cost of funds. La Porta, Lopez-deSilane and Shleifer (2002) studied government ownership of banks in the world. Using data on banks in 1995, they found that government ownership of banks does not lead to subsequent growth and development, which is consistent with the political view of government ownership of banks whereby the resource allocation process is politicized leading to reduced efficiency. Anderson and Campbell (2004) studied governance activity of over 100 TSE-listed Japanese banks for a 12-year period during the banking crisis in the early 1990s. By examining top executive turnover at Japanese banks and how it is affected by bank performance, they did not find a relation between bank performance and nonroutine turnover of bank presidents in the pre-crisis period of 1985–1990, suggesting that Japanese bank executives were insulated from disciplinary dismissal because of poor relative performance prior to the banking crisis. In contrast, they found a significantly negative relation between non-routine presidential turnover and bank performance measures such as stock returns or profitability in the crisis years of 1991–1996. Crespi et al (2004) examined the governance of Spanish banks and found a negative relationship between performance and governance intervention for banks, but the results change for each form of ownership and each type of intervention. Internal-control mechanisms work for independent commercial banks, but savings banks show weaker internal mechanisms of control and the only significant relationship between performance and governance intervention that appears is for mergers. Levine (1999) suggests that in time of crisis or economic stagnation, the presence of foreign banks with internationally diversified asset portfolios have stabilizing influence. Barth, Caprio, and Levine (2001) found that the likelihood of a major banking crisis is positively associated with limitations on foreign bank entry and ownership. In another study, Barth et al (2004) found that regulations that encourage and facilitate the private monitoring of banks tend to boost bank performance, reduce non-performing loans and enhance bank stability. III. Corporate Governance in the MENA Region The issue of corporate governance is still taking its first steps in the MENA region. Although there is recognizably more interest in the matter, the region still lags behind others, especially if compared with the more industrialized countries in aspects pertaining to the establishment of the proper frameworks for corporate governance. In this respect, the partnerships with the European Union that many regional countries have undertaken are expected to speed up the process of recognizing problems that lie within the corporate governance frameworks. A. The State of Financial Development in the MENA Region Creane et al (2004) study the financial sector development in the MENA region. Table 1 presents several indices relating to the state of financial development in the region. The first index “Financial Development Index” is the weighted average of several indices such as monetary sector and policy, financial openness, and institutional environment. As observed in Table 1, some countries in the region have faired well in all of the indicators. However, the region as a whole barely makes it into the average category. These indicators also show that the banking sector has managed higher scores than the non-bank financial sector. This is more indicative of the severe underdevelopment of the non-bank financial sector rather than relative development of the banking sector. 4 Saidi (2004) summarizes different key issues to be considered as essential building blocks in diagnosing the problem of corporate governance in the region. The first point is the lack of data or clear image on the state of governance in the MENA region. Saidi (2004) mentions that only six countries (Algeria, Jordan, Kuwait, Morocco, Tunisia and UAE) in the region have presented an ROSC (Report On The Observance Of Standards And Codes) which is a report detailing the standards of accounting and auditing. In addition only four countries (Algeria, Kuwait, Tunisia and UAE) have accomplished a Financial Sector Assessment Program which is jointly prepared with the International Monetary Fund and the World Bank to assess the financial sector’s strengths and weaknesses. This indicates that the countries with studies on the status of their respective financial sectors are very limited. Table 1 Financial Development in the MENA Region4 Financial Development Index Banking Sector Non-Bank Financial Sector Regulation and Supervision Bahrain 7.7 7.3 5 9.3 Lebanon Jordan Kuwait United Arab Emirates Saudi Arabia Pakistan 7 6.9 6.8 6.6 6.4 6 8.7 7.1 7.4 7.9 7.8 5.8 3.3 6.3 5 5 3.3 6.3 7.7 8.7 8 6.7 8 7.7 Oman Qatar Tunisia Morocco Egypt 5.9 5.7 5.6 5.5 5.4 6.1 6.8 7.7 5.6 6 5 0.7 4.7 4.7 6.3 8.3 6.7 5.3 7.3 5.3 Sudan Dj ibouti Yemen, Republic of Mauritania Algeria 4.7 4.1 3.9 3.5 32 5.7 3.8 4.1 3.8 2.5 0.7 1.3 0.7 0.7 3 3.7 5 3.3 3 3.5 Iran, Islamic Rep. Syrian Arab Republic Libya 2.5 1.1 1 1.9 1.9 1.3 3.3 0.7 0.7 4.7 0 2 5 5.5 3.3 5.7 Average Scale: Very low: below 2.5, Low: 2.5-5.0, Medium: 5.0-6.0, High: 6.0-7.5, Very high: above 7.5. 4 Creane et al.(2004), Financial Sector Development in the Middle East and North Africa 5 Saidi (2004) also surveys 298 companies and their CEOs in Lebanon on different issues relating to corporate governance. Figure 1 displays the priorities for reform as concluded from the survey. Accordingly, enforcing laws, rules and procedures is seen as the most pressing problem in facing Lebanese businesses. B. Corporate Governance in Banks of the MENA Region Similar to the state of corporate governance of non-financial firms in the MENA region financial firms lag behind their counterparts in the developed countries. Also the issue of corporate governance in banking has not attracted enough attention. There are still some misconceptions within bank management that consider improving governance as a costly process requiring time and effort without any financial return. Figure 1 Corporate Governance reform Priorities by Topic5 The Union of Arab Banks (UAB) (2006) has surveyed 5 out of 18 banks operating in the Sultanate of Oman representing 93% of the total assets of the Omani banking sector. The survey focused on issues related to capital adequacy, risk management, asset liability management, transparency and corporate governance. The survey concludes that Oman is improving its regulatory framework in order to comply with the requirements of the Basel Committee’s Core Principles for Effective Banking Supervision. The survey also finds that accounting and provisioning practices in Omani banks are in compliance with Basel II requirements. However, risk management systems are still not up to the minimum standards needed. In 2002, the banking law in Oman was amended to regulate the functions of external auditors. The newly amended law makes it the board of governors’ duty to ensure the qualifications of external auditors. It enforces the importance of periodical rotation, the timeliness of reporting and the need to ensure the complete compliance with the regulations and standards of the CBO. The central Bank of Jordan has put forward a handbook in an effort to explain and promote better governance practices within Jordanian banks6. This handbook emphasizes two aspects of 5 Saidi, N. (2004, June 3-5, 2004). Corporate Governance in MENA Countries: Improving Transparency and Disclosure. Paper presented at the The Second Middle East and North Africa Regional Corporate Governance Forum, Beirut. 6 Bank Directors' Handbook of Corporate Governance. Central Bank of Jordan, 2006. 6 corporate governance: Internal and External corporate governance. Internal corporate governance deals with the relationship between the various groups in direct control of the Bank: Shareholders, Board of directors and management as well as other shareholders. Good corporate governance is one where managers are held accountable by the board of directors, and where directors are held accountable by shareholders. External corporate governance involves the establishment of laws and regulation, capital market infrastructure, and accounting standards .The Central Bank of Jordan concludes that both appropriate internal and external environments are needed to ensure a good corporate governance framework. These environments should be based on four guiding principles: fairness, transparency, accountability and responsibility. Al-Muharrami, Matthews, and Khabari (2006) investigated the market structure of Arab GCC banking industry during the period 1993-2002 and found that Kuwait, Saudi Arabia and UAE have moderately concentrated banking sector, and operate under perfect competition. On the other hand, Qatar, Bahrain, and Oman are highly concentrated markets and operate under conditions of monopolistic competition. IV. Data and Methodology The current study aims at studying the relationship between ownership structure and performance of banks in the MENA region. The initial sample includes 89 banks in four categories: commercial, investment, specialized, and Islamic banks. We define the variables that determine the ownership structure of the banks, and then test whether these variables are significantly related to bank performance. The data used are obtained from the Arab Banks and Financial Institutions Directory 19982004 published by the Union of Arab Banks. It covers a six year span for the period 1998-2003. The variables are: • Natural Logarithm of Net profits of banks: used as a proxy of performance. • Natural Logarithm of Shareholders' Equity (US Dollars) and Loan Advances (US Dollars) included as control variables. • Five dummy variables that differentiate the ownership structure of banks. The data on the continuous variables (Net profits, Shareholders’ Equity and Loan Advances) is the average for the period sampled 1998-2003. The five dummy variables are the concentrated versus non-concentrated ownership variable, government ownership variable, foreign/local ownership variable, ownership/board separation variable and board/management separation variable. a. Concentrated versus non-concentrated ownership. Banks with concentrated ownership are coded as 1 while banks with non-concentrated ownership are coded as 0. Banks with a single owner with 50 percent or more of total shares are considered as ownership concentrated. It should be noted that banks in MENA region, have less dispersed ownership than banks in other regions with few banks actively involved in dispersed public ownership. The first hypothesis to be tested follows: H1: Banks with concentrated ownership perform better than banks with dispersed ownership 7 Table 2 Characteristics of banks in the region7 Arab1 Foreign (Non-Arab)2 Local3 Foreign4 Algeria Bahrain Egypt Jordan Kuwait Lebanon Lybia Morocco Oman Quatar Saudi Arabia Sudan Tunisia UAE 16 35 43 19 11 55 17 14 11 10 15 32 22 40 0 31 3 3 0 15 0 1 5 5 0 1 2 12 0 4 20 10 5 29 7 2 7 7 6 11 7 12 0 4 6 6 1 13 0 2 0 0 0 4 2 2 Yemen 15 2 7 0 1. Number of Arab banks in the respective country 2. Number of Non-Arab banks in the respective country 3. Number of Local banks. 4. Number of Foreign banks. . b. Type of ownership. The government ownership dummy is coded 1 when a bank is owned by a governmental institution, and foreign/local ownership dummy is coded as 1 if the bank is owned by foreign investors by more than 50 percent of its outstanding shares. The next two hypotheses follow: H2: Privately owned banks perform better than government owned banks H3: Foreign banks perform better than domestic banks c. Ownership/board separation. This dummy is used to describe whether owners choose to be directly represented within the board of directors. Board of directors, with more than half of its members being owners of the bank or direct representatives of holding institution, is considered to have no ownership/board separation and coded as 1. This variable is important in assessing to which level bank owners play an active role in guiding the general policies of the bank operations. H4: Banks with no ownership/board separation perform better than those with ownership/board separation . d. Board/management separation. This dummy measures whether there is a separation between board membership and management. Banks with half of the top level management 7 Source: Union of Arab Banks Directory 1998-2003 8 composed of board of directors are considered to have no board management separation and coded as 1. This variable tests to which extent stockholders control management activities. H5: Banks with board/management separation perform better than those with no board/management separation. The following regression equation is formed to test for the above hypotheses: Ln PERFORMi = β0 + β1GOV + β2 FOREIGN + β3 CONCENTRATED + β4OWNER/BOARD + β5BOARD/MNGT + β6ln EQUITYi + β7ln LOANADVi + ε V. Empirical Results The multivariate analysis is comprised of weighted least square regressions. This is an efficient method that makes good use of small data sets. It also shares the ability to provide different types of easily interpretable statistical intervals for estimation, prediction, calibration and optimization. In addition, the main advantage that weighted least squares enjoy over other methods is the ability to handle regression situations in which the data points are of varying quality. The dataset available is a cross-sectional one with 89 banks represented. White heteroskedasticity-consistent standard errors and covariance is used to correct for heteroskedasticity in the regressions. In addition, the natural logarithm of continuous variables is used given that these variables are not normally distributed. The results of Table 3 show that all variables are insignificant except for board/management separation and the two control variables: LN Shareholders' Equity and LN Loan Advances. We next exclude 34 bank observations in three categories (investment, specialized, and Islamic banks) due to inconsistent and missing information on the relevant variables tested. In order to get “cleaner” results, we run a least squares regression for commercial banks only (n=56). Table 4 summarizes the results for commercial banks. The coefficient of the dummy variable for board/management separation is positive and significant at 5% level. This indicates that the performance of the banks is enhanced when there is no board/management separation. The result is consistent with the argument that moral hazard problems are reduced when board members have an active role in management and that no separation will decrease managerial slack thus increasing profitability and efficiency. The concentrated ownership variable is negatively related to performance implying that concentrated ownership is constraining management ability to invest in higher risk/higher return assets. Increased monitoring of management through concentrated ownership happens at the expense of reduced profits. The coefficient of the government ownership variable is positive and significant implying that government ownership of commercial banks has a positive effect on net profits, which is a plausible result that contrasts the standard view that government ownership leads to inefficient and wasteful behavior. Finally, ownership/board separation coefficient and the foreign ownership coefficient are both insignificant. This implies that the performance of MENA banks cannot be explained by these variables. Hence, any inference based on these variables becomes inconclusive. 9 Table 3 Least Square Regression for All Banks (n = 89) White Heteroskedasticity-Consistent Standard Errors & Covariance Least Squares Regression Variable Coefficient t-Statistic -6.0416 (***) Prob. Constant -2.1687 Government Ownership 0.1728 0.8664 0.3889 Foreign/Local Ownership Concentrated Ownership -0.0647 -0.2056 -0.3127 -1.2867 0.7553 0.2019 Ownership/Board Separation -0.0575 -0.3198 0.7499 Board/Management Separation 0.2761 (*) 0.0736 LN Shareholders' Equity (USD) LN Loan Advances (USD) 0.8403 0.1406 R-squared Adjusted R-squared 0.8662 0.8546 F-statistic 74.8841 Prob(F-statistic) 0.0000 1.8125 10.3339(***) 2.0581(**) 0.0000 0.0000 0.0428 (***) Significant at 1% level (**) Significant at 5% level (*) Significant at 10% level Table 4 Least Squares Regression for Commercial Banks (n = 56) White Heteroskedasticity-Consistent Standard Errors & Covariance Least Squares Regression Variable Coefficient t-Statistic (***) Prob. Constant Government Ownership -2.3856 0.8809 -4.1642 2.8024(***) 0.0001 0.0073 Foreign/Local Ownership 0.2193 0.7629 0.4492 (**) 0.0211 Concentrated Ownership -0.5367 -2.3845 Ownership/Board Separation Board/Management Separation -0.0495 0.4667 -0.2013 2.1657(**) 0.8413 0.0353 LN Shareholders' Equity (USD) 0.6053 3.8888(***) 0.0003 LN Loan Advances (USD) 0.3603 (**) 0.0271 R-squared Adjusted R-squared 0.8996 0.8850 F-statistic 61.4688 Prob(F-statistic) 0.0000 (***) Significant at 1% level (**) Significant at 5% level (*) Significant at 10% level 10 2.2793 VI. Summary The current paper examines empirically the relationship between ownership structure and performance for a sample of MENA banks over a six year period 1998-2003.The initial sample included 89 banks in four categories: commercial, investment, specialized, and Islamic banks. We run a least square regression with the natural logarithm of net profits of banks as a proxy of bank performance in addition to five dummy variables for the ownership structure of banks: concentrated versus non-concentrated ownership, government ownership, foreign/local ownership, ownership/board separation and board/management separation .The natural logarithm of shareholders' equity and loan advances are included as control variables. After finding insignificant results for the whole sample, we focus on the “clean” sample of commercial banks (n=56). The coefficient of the dummy variable for board/management separation is positive and significant at 5% level. This indicates that the performance of the banks is enhanced when there is no board/management separation. The result is consistent with the argument that moral hazard problems are reduced when board members have an active role in management and that no separation will decrease managerial slack thus increasing profitability and efficiency. The concentrated ownership variable is negatively related to performance implying that concentrated ownership is constraining management ability to invest in higher risk/higher return assets. Increased monitoring of management through concentrated ownership happens at the expense of reduced profits. The coefficient of the Government ownership variable is positive and significant implying that government ownership of commercial banks has a positive effect on net profits, which is a plausible result that contrasts the standard view that government ownership leads to inefficient and wasteful behavior. Finally, ownership/board separation coefficient and the foreign ownership coefficient are both insignificant. This implies that the performance of MENA banks cannot be explained by these variables. Hence, any inference based on these variables becomes inconclusive. Our results show that the performance of banks is weakened with board/management separation, and concentrated ownership and enhanced with government ownership. References Adwan, C. D. (2003). The Task Force on Corporate Governance in Lebanon: Focus Group on Corporate Governance in Lebanon. Paper presented at the Global Corporate Governance Forum. Akella, S., and Greenbaum, S. (1988). Savings and Loan Ownership Structure and ExpensePreference. Journal of Banking and Finance, 12, 419-437. Al-Muharrami,S., Matthews, K., and Khabari,Y.(2006), Market Structure and Competitive Conditions in the Arab GCC Banking System, Journal of Banking and Finance 30, 3487-3501. Allen, L., and Cebenoyan, A. S. (1991). Bank Acquisitions and Ownership Structure: Theory and Evidence. Journal of Banking and Finance, 15(2), 425-448. 11 Altunbas, Y., Evans, L., and Molyneux, P. (2001). Bank Ownership and Efficiency. Journal of Money, Credit and Banking, 33(4), 926-954. Ang, J. S., Cole, R. A., and Lin, J. W. (2000). Agency Costs and Ownership structure. Journal of Finance, 55(1), 81-106. Arab Banks and Financial Institutions Directory. (2003-2004). Union of Arab Banks. Bank Directors' Handbook of Corporate Governance. Central Bank of Jordan. Barth,J.R., Caprio Jr., and Levine, R., (2001).Banking Systems around the Globe: Do Regulations and Ownership affect Performance and Stability? In: Mishkin,F.S. (Ed.), Prudential Supervision: What Works and What Doesn’t . University of Chicago Press. Barth,J.R., Caprio Jr., and Levine, R., (2004).Bank Regulation and Supervision: What Works Best? Journal of Financial Intermediation 13,205-248. Beatty, R., Santomero, A., and Smirlock, M. (1987). Bank Merger Premiums:Analysis and Evidence (Vol. 3). New York: Salomon Brothers Center for the Study of Financial Institutions. Berle, A., and Means, G. (1932). The Modern Corporation and Private Property. New York: Macmillan. Blair, M. (1995). Ownership and Control: Rethinking Corporate Governance for the Twenty-First Century. Washington DC: Brookings Institution. Coase, R. H. (1937). The Nature of the Firm. Economica, 4, 386-405. Corporate Governance for Morocco, Egypt, Lebanon and Jordan. (2003, September 7, 2003). Paper presented at the Middle East and North African Corporate Governance Regional Working Group, Cairo, Egypt. Corporate Governance Survey in the Arab Banking Sector: The Case of the Sultanate of Oman. (2006). Union of Arab Banks. Creane, S., Goyal, R., Mobarak, A. M., and Sab, R. (2004). Financial Sector Development in the Middle East and North Africa. International Monetary Fund Working Paper WP/04/201. Diamon, D. W. (1984). Financial Intermediation and Delegated Monitoring. Review of Economic Studies(51), 393-414. Edwards, F. R. (1977). Managerial Objectives in Regulated Industries: Expense Preference Behavior in Banking. Journal of Political Economy, 85, 147-162. Enhancing Corporate Governance for Banking Organisations. (2006). Bank for International Settlements. Gedajlovic, E. R., and Shapiro, D. M. (1998). Management and Ownership Effects: Evidence from Five Countries. Strategic Management Journal, 19(6), 533-553. Glassman, C. A., and Rhoades, S. A. (1980). Owner vs. Manager Control Effects on Bank Performance. The Review of Economic Statistics, 62(2), 263-270. Gorton, G., and Rosen, R. (1995). Corporate Control, Portfolio Choice and the Decline of Banking. Journal of Finance, 50(5), 1377-1420. 12 Gorton, G., and Schmid, F. (1999). Corporate Governance, Ownership Dispersion and Efficiency: Empirical Evidence From Austrian Cooperative Banking. Journal of Corporate Finance, 5, 119-140. Hannan, T. H., and Mavinga, F. (1980). Expense Preference and Managerial Control: The Case of the Banking Firm. Bell Journal of Economics, 11, 671-682. Hart, O. (1995). Corporate Governance: Some Theory and Implications. The Economic Journal, 105(430), 678-689. Hart, O. (1995). Firms, Contracts, and Financial Structure. Oxford: Oxford University Press,. James, C. M. (1984). An Analysis of the Effect of State Acquisition Laws on Managerial Efficiency: The Case of Banking. Journal of Political Economy, 95, 355-370. Jensen, M.C. and Meckling W.H. (1976).Theory of the firm: Managerial Behavior, Agency Costs and Ownership Structure, Journal of Financial Economics 3, 305-360. Kole, S., and Lehn, K. (1997). Deregulation, the Evolution of Corporate Governance Structure, and Survival. The American Economic Review, 87(2, Papers and Proceedings of the Hundred and Fourth Annual Meeting of the American Economic Association), 421-425. La Porta, R., Lopez-de-Silane, F., and Shleifer, A. (2002). Government Ownership of Banks. Journal of Finance, 57(1), 265-301. La Porta, R., Lopez-de-Silane, F., Shleifer, A., and Vishny, R. W. (2000). Investor Protection and Corporate Governance. Journal of Financial Economics, 58(1-2), 3-27. La Porta, R., Lopez-de-Silanes, F., and Shleifer, A. (1999). Corporate Ownership Around the World. Journal of Finance, 54(2), 471-517. La Porta, R., Lopez-De-Silanes, F., Shleifer, A., and Vishny, R. W. (1997). Legal Determinants of External Finance. Journal of Finance, 52(3, Papers and Proceedings Fifty-Seventh Annual Meeting, American Finance Association, New Orleans, Louisiana January 4-6, 1997), 11311150. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., and Vishny, R. W. (1998). Law and Finance. Journal of Political Economy, 106(6), 1113-1155. Levine, R., (1999).Foreign Bank Entry and Capital Control Liberalization: Effects on Growth and Stability. University of Minnesota, mimeo. Li, J. (1994). Ownership Structure and Board Composition: A Multi-Country Test of Agency Theory Predictions. Managerial and Decision Economics, 15(4, Special Issue: Aspects of Corporate Governance), 359-368. Macey, J. R. (1997). Institutional Investors and Corporate Monitoring: A Demand-Side Perspective. Managerial and Decision Economics, 18(7/8, The Use of Finance and Economics in Securities Regulation and Corporation Law), 601-610. Maher, M., and Andersson, T. (1999, November 4-5). “Corporate Governance: Effects on Firm Performance and Economic Growth. Paper presented at the Convergence and Diversity in Corporate Governance Regimes and Capital Markets, Tilburg University. Mathiesen. (2002). What is Corporate Governance? Retrieved March 27, 2006, from www.encycogov.com 13 Morck, R., Shleifer, A., and Vishny, R. W. (1989). Alternative Mechanisms for Corporate Control. American Economic Review, 79, 842-852. Murto, R. (1994). Finnish Banking Crisis: Can we Blame Management? Finnish Economic Papers(7), 55-68. Pedersen, T., and Thomsen, S. (1997). European Patterns of Corporate Ownership: A Twelve-Country Study. Journal of International Business Studies, 28(4), 759-778. Prowse, S. (1997). The Corporate Governance System in Banking: What do we Know? BNL Quarterly Review, Special Issue, March 1997. Qintai, C. (2004). State Shareholders Should Become an Active Force in Promoting and Establishing Effective Corporate Governance. Paper presented at the Shanghai OECD Corporate Governance Forum, Shanghai, China. Saidi, N. (2004, June 3-5, 2004). Corporate Governance in MENA Countries: Improving Transparency and Disclosure. Paper presented at the The Second Middle East and North Africa Regional Corporate Governance Forum, Beirut. Saunders, A., Strock, E., and Travlos, N. G. (1990). Ownership Structure, Deregulation, and Bank Risk-Taking. Journal of Finance, 58(2), 643-654. Schranz, M. S. (1993). Takeovers Improve Bank Performance: Evidence from the Banking Industry. Journal of Political Economy, 101, 299-326. Shleifer, A., and Vishny, R. W. (1997). A Survey of Corporate Governance. Journal of Finance, 52(2), 737-783. Smirlock, M., and Marshall, W. (1983). Monopoly Power and Expense Preference Behavior: Theory and Empirical Evidence to the Contrary. Bell Journal of Economics, 14, 166-178. Stiglitz, J. (1985). Credit Markets and the Control of Capital. Journal of Money, Credit and Banking(17), 133-152. Stone, A., Hurly, K., and Khumani, R. (1998). Business Environment and Corporate Governance: Strengthening Incentives for Private Sector Performance. The World Bank/IMF Annual Meetings, 1998 Background Papers. Tirole, J. (2001). Corporate Governance. Econometrica, 69(1), 1-35. Tricker, R. I. (1995). International Corporate Governance: Text, Readings and Cases: Prentice Hall. Wolfensohn, J. D. (1999). A Battle for Corporate Honesty, the Economist: the World in 1999 at 38. Financial Times. 14 ASAC 2008 Halifax, Nova Scotia Ashraf Al Zaman Karen Lightstone Sobey School of Business Saint Mary's University REASSESSING CANADIAN HEDGING PRACTICES: A SURVEY STUDY Using a unique set of Canadian data, we analyze the hedging behavior of non-financial Canadian firms. We find that firms are selectively hedging risks and have the ability to naturally hedge. Internal controls are in place however the effectiveness is diminished due to a lack of reassessment. Accounting regulations are posing problems for firms wanting to or using derivatives. We also find that in some industries, experience is more important than a higher level of education. Finally, we find that derivative instrument use is dependent on the decentralization or delegation of risk management decision-making. Introduction The importance of risk management, as well as the number of risk management products and their use, has increased substantially (Bodnar et al., 1998). In a more recent triennial study, the Bank of International Settlements (2005) reports that the notional amount of over-the-counter (OTC) derivatives increased from $99.7 trillion in June 2001, to $220 trillion in June 2004. Given these developments, understanding the need for hedging for corporations, the strategies followed, the assessment of the efficacy of the strategies followed, and the importance of regulation in this area is important. Firms face different types of risks including exchange rate risk, interest rate risk and risk related to obtaining raw material. According to Modigliani and Miller (1958), in a marketplace without friction, hedging to minimize various types of risks should not create any additional value for the firm. Therefore, with market friction, empirical and theoretical analyses have shown the value created by hedging (see Muller & Verschoor, 2006). In exploring this issue, academics have addressed not only the issues directly related to cash flows, but also additional corporate concerns in the context of hedging such as external financing, financial distress, agency conflicts and taxes.1 A limited number of studies, such as Borokhovich et al. (2004) and Géczy et al. (2007), have started to address the issue of corporate governance and hedging. But the internal control of hedging is not well studied and understood. Whereas the importance of hedging is well emphasized, concerns have also been expressed with regards to potential welfare loss due to hedging in studies such as Tufano (1998). The major emphasis of studies on corporate hedging has been on the use of derivative by nonfinancial firms and the determinants of corporate hedging policies. 2 Many of the theoretical predictions on hedging have been assessed, but the results are at best mixed. In a recent survey, on foreign exchange 1 For financial distress and hedging consult Smith & Stulz (1985) and Stulz (1996). Agency cost and corporate hedging is considered by Chidambaran et al. (2001). For analysis of reduction of income taxes consult Graham et al. (2002) and Graham & Clifford (1999). 2 Stulz (1990), Belk and Glaum (1990), Tufano (1996), Bodnar et al. (1996), Belk & Edelshain (1997), Bodnar et al. (1998), Fatemi & Glaum (2000), and Graham & Harvey (2001) present discussions on derivative use by 15 risk exposure, Muller and Verschoor (2006) document the mixed results. Existing empirical analysis often finds risk exposure to be non-significant, which leads to questioning the necessity of hedging. Bartram and Bodnar (2007) claim, using exchange rate exposure, that the observed inconsistencies are mainly due to endogeneity of operative and financial hedging at the firm level. Firms can minimize their risk exposure not only through derivative use, but also through structuring and locating firm operations in such a way that the net exposure is minimized. These intra-firm activities are not captured through assessments of available market data. So accurate measurement of risk exposure is important for success in empirical research. This point is further supported by Crabb (2002), where he shows, using 276 US multinational firms, that the active hedging strategies of firms mitigate exchange rate risk exposure, but this is not captured in much of the empirical work. In addition, Bartram (2007) makes a similar claim using a clinical study of a German multinational company. From existing empirical research, it is clear a substantial number of firms are active in hedging. When firm value creation or reduction through hedging is still being questioned, corporations are active in hedging, but are the companies able to assess their total exposure in order to manage their risk effectively? Loderer and Pichler (2000) investigate this question using data on risk management practices of Swiss industrial corporations. They find corporations are unable to quantify their currency risk. This indicates possible ineffective hedging at the corporate level. Given the strong theoretical predictions on the importance and use of hedging, the mixed results of empirical work, the limitation in measuring risk and the potential limitation on the part of corporate executives in assessing the overall risk exposure, it is a challenging job to generate robust results on corporate hedging. But given the dynamic nature of the hedging process, due to changes in the industry and regulations, having the opportunity to get first-hand feedback from corporate executives is valuable. Executive surveys are valuable tools in this process. In this study, we present the results from our survey conducted with a group of Canadian non-financial firms.3 We investigate the types of risks firms are exposed to, their use of derivative tools, attitude toward risk exposure, hedging strategy, organizational structure with regards to hedging decisions and reporting, and the individuals involved in this process. Through our study, we make several contributions to the literature. First, we provide a country perspective on Canada. Some of our analysis is similar to other studies on Canada such as Jalilvand (1999). But our study updates existing findings with regards to derivative use, and complements the existing findings by reporting on the role of corporate control and management with respect to hedging. Given the dynamic nature of the hedging environment, new studies create value by providing information on most recent practices. Methodologically our study is close to Fatemi and Glaum (2000), but we introduce additional elements such as corporate control over risk management. Secondly, our study provides a way of comparing recent Canadian hedging practices with hedging practices in other countries. Loderer and Pichler (2000), Marshall (2000), and Jalilvand, Switzer and Tang (2000) document substantial variations in hedging practices across countries. Our contribution allows us to learn about hedging practices of Canadian non-financial firms. Thirdly, we provide a unique analysis of the relationship between level of centralization or internal control over decision-making, industry, hedging and hedging instruments, which is missing in the literature, especially in the Canadian context. This allows us to relate Canadian practices with the practices in the USA reported in Bodnar et al. (1998). Finally, we provide some insight into the limitations of doing survey research with executives involved in hedging. This may provide some future guidance to survey researchers in designing effective surveys. corporations. Fok et al. (1997), Géczy et al. (1997), Jalilvand (1999), El-Masry & Abdel-Salam (2007), and Géczy et al. (2007) discuss issues related to the determinants of corporate hedging policies. 3 For an example of derivative use of a financial institution consult Hogan & Rossi (1997). 16 The remainder of the paper is organized as follows: In the following section we describe the methodology of this study. In the results section, first we present our findings on the basis of specific issues in corporate hedging; then we complement our findings with a multivariate study. In the final section we conclude and make recommendations for future research. Methodology For our study all companies with audited annual financial statements for 2005 were identified through the System for Electronic Document Analysis and Retrieval (www.SEDAR.com), a public company database. This resulted in a population of 2,136 companies. The population was reduced by removing scholarship plans, derivative product companies, financial services companies, and other financial companies. An additional criterion was that the companies be listed on the Toronto Stock Exchange. This yielded a population of 945 non-financial firms. In May 2007, a mailing list was compiled using the list of non-financial companies and the current mailing addresses listed on SEDAR of the executives in charge of risk management. The questionnaires were mailed out and a similar questionnaire was made available on the Internet. The survey responses were collected anonymously. Our initial mail-out resulted in 29 companies responding – 27 completed the paper questionnaire and 2 submitted Internet surveys. We also received 80 return-to-sender envelopes. A second follow-up letter was sent to the remaining 865 companies in July 2007 reminding them about the survey. A further 11 electronic questionnaires were received as well as an additional 66 return-to-sender envelopes. The questionnaire contained 39 detailed questions. A third attempt to increase the response rate was made in November 2007 which included a letter directing participants to the Internet version of the survey as well as a $2.00 Second Cup Café card as an incentive. We received an additional 16 responses and 20 more return-to-sender envelopes. This resulted in a sample population of 56 respondents from 779 Canadian public companies. Our response rate is low (7.2%), but in corporate finance surveys low response rates have been a problem for a long time. Trahan and Gitman (1995) had a response rate of 12% and Graham and Campbell (2001) had a response rate around 9%. In a recent paper, Géczy et al. (2007) also discuss this issue to certain extent. This issue is addressed in more detail in the Appendix. Results Our results are presented in the following manner. The aggregate responses to our questionnaire are presented first. This provides a descriptive analysis of the respondents. In some instances, we find a statistically significant relationship between the industry of the respondent and one of the other variables.4 Where this occurs, we include it in the discussion. At the end, the results of our statistical analysis are presented along with a discussion. Description of the population As discussed above, our sample population consists of 56 firms spanning all industry sectors as defined by the major industry categories in SEDAR. Some industries are better represented than others with 28.6% in mining, oil and gas; 21.4% in manufacturing; and 12.5% in wholesale/retail. The “Other” category for industry represents 16.1% of the respondents and consists primarily of real estate investment and multi-industrial companies. 4 It is standard practice to use linear regression to determine basic relationships among variables. We recognize the limitations of this methodology in the current context and have also run some logit models to take into account our quality dependent variables later in the paper. 17 Our respondents are predominantly members of the executive committee (73.2%) or are department heads (19.6%) of their respective companies. They are responsible for risk management issues. Sixty-one percent have a chartered accountant designation, 14.3% are certified management accountants and 21.4% have an accounting designation as well as a master’s of business administration. The respondents’ age is quite wide ranging with 14.3% 30-39 years old; 46.4% 40-49; 30.4% 50-59; and 8.9% 60 years and older. One quarter of the respondents has been a chief financial officer for 1-3 years; 21.4% have been in the position for 4-6 years; 7.1% for 7-9 years; and most (37.5%) have worked in the position for more than 9 years. Our analysis reveals a significant relationship between education and industry (t=2.002, p<.05). Breaking education into two categories: university degree or more and no university degree, also results in a significant relationship with industry (t=2.141, p<.05). Respondents in the utility industry, 62.5% of mining, oil and gas, and 41.7% of the manufacturing industry do not have a university degree. When we combine education, industry and length of time in the chief financial officer position, we find that many (55.6%) of the mining, oil and gas respondents with no university degree have been in the position for more than 9 years. These results suggest experience is more important than educational background in some industries. Risk management strategies Respondents were asked about the importance of four main categories of risk. They stated that financial risk is the most important (mean of 3.98 on a scale of 1 to 5, with 1 representing not important). This is followed by industry risk (mean of 3.79), operational risk (3.66) and lastly general environmental risk (mean of 3.32). Not all types of risk are controllable and in many cases managers are not concerned about controlling all risks. Respondents were then asked what they were doing to address the four main risks. The options provided were doing nothing, observing, observing and quantifying, and actively setting limits with respect to acceptable risk levels. Observing and quantifying or actively setting limits occurs in 69.9% of the respondents with respect to financial risk; 44.6% for operational risk; 37.5% for industry risk; and 14.3% for general environmental risk. This is contrary to Loderer and Pichler’s (2000) findings in Swiss firms. With respect to financial risks, Table 1 presents the results regarding the degree of risk acceptable to the company. The responses are on a scale of 1 (risk averse) to 7 (risk seeking) with 4 representing a risk neutral position. Table 1: Acceptable levels of financial risk Type of financial risk Exchange rate Interest rate Raw material price Credit Mean 3.09 3.11 3.14 3.41 Risk neutral 17.9% 37.5% 39.3% 41.1% As denoted by the low mean, Table 1 indicates that for all types of financial risk, managers are seeking to reduce risk. However, a reasonable percentage of respondents said they were risk neutral, which might suggest a lack of activity regarding taking measures to reduce the particular risk. Regression analysis indicates that interest rate risk and industry were significantly related (t=-2.259, p<.015). 18 The development of a risk management system to assist managers in observing, quantifying and controlling risk exposure is crucial to the organization. However, when asked about the current stage of their risk management system, 71.4% of respondents state the system is either in the planning stage or had some parts in place and operating. Only 28.6% of respondents state the system is fully developed and operational. There is no noticeable advancement in the stage of risk management systems from Jalilvand, Switzer and Tang’s 2000 study. The level of difficulty in implementing a risk management system seems to depend partly on the industry. We find a significant relationship between implementation and industry (t=-2.122, p<.038). Most respondents indicate they are having some problems implementing a risk management system. However, 37.5% of mining, oil and gas company respondents and 33.3% of manufacturing companies state the implementation of a risk management system poses no problems. Risk management goals Respondents were given a list of eight goals for risk management and asked to rate the importance of each on a scale of 1 (not important) to 5 (very important). The top three goals were the protection of the existence of the enterprise (mean of 3.96); reduction in the volatility of cash flows (mean of 3.91); and increase in the market value of the enterprise (mean of 3.73). Interestingly, there was a significant relationship between industry type and the goal of reducing the volatility of earnings (t=2.2124, p<.039). This goal was the fourth most important out of the eight listed (mean of 3.38). Respondents did not rank the goals of reducing income taxes (mean of 2.61 and 28.6% of respondents selecting not important) or controlling the behaviors of employees or subsidiary companies (mean of 2.8 and 21.4% of respondents selecting not important) very high. The lack of importance over reducing income taxes is predictable given the income tax structure in Canada. Canadian income taxes are assessed on the cash basis of accounting whereas the income statement uses accrual accounting. U.S. income tax regulations utilize the income statement for assessing taxes and therefore studies have found this to be a more important goal of risk management (see Graham & Smith, 1999). Jalilvand (1999) shows a weak relationship between taxes and hedging which is contrary to our findings. Strategic and organizational structure of Treasury management Treasury management was defined in the survey as being financially responsible for short term liquidity. Respondents were asked whether their Treasury operated as a service, cost or profit centre. Half the respondents selected service center and 35.7% selected cost centre. One respondent did not answer the question and the remaining 12.5% selected profit centre. Jalilvand, Switzer and Tang (2000) found a higher percentage of firms operating their treasury departments as service centres. The majority of respondents (71.4%) said that all decisions were made by the company head office or by appropriate predetermined guidelines, a high degree of centralization. For the execution of hedging measures, 73.2% said that this function was highly centralized. The effectiveness of a risk management system also depends considerably on the efficiency of the information systems used. When asked about the type of software system used in the Treasury management, most responded (82.1%) that spread-sheet programs are extensively used. This indicates that over a period of time technology has not advanced (see Jalilvand, Switzer & Tang, 2000). The use of a variety of programs which are not integrated was found to be significantly related to industry (t=-2.283, p<.05) as well as future plans to invest in new information systems designed for risk management (t=2.210, p<.05). Derivatives 19 Respondents were asked to indicate the frequency with which various types of derivatives were used. On a scale of 1 (very frequently) to 4 (never) Table 2 presents the mean scores for each of the instruments in order of frequency of use. Four companies (7.1%) from three different industries responded “never” for all types of derivatives indicating they do not use them. Table 2: Types of derivatives by frequency of use Type Mean response Forward foreign contracts 2.61 Forward rate agreements 3.02 Interest rate swaps 3.14 Over-the-counter foreign currency options 3.25 Currency swaps 3.29 Goods/Commodity derivatives 3.30 Caps/floors 3.38 Structured derivatives 3.66 Exchange traded foreign currency options 3.73 Interest rate futures 3.84 There is a significant relationship between industry and type of derivative used. Both over-thecounter foreign currency options and goods/commodity derivatives produced a statistically significant regression result with t=2.189 (p<.05) and t=2.772 (p<.05), respectively. When asked about the primary purpose for using derivatives 82.1% stated they were used for hedging exposed risk positions. It is interesting to note that only one respondent stated they were used extensively for speculating in the market. All respondents, regardless of whether they used derivatives or not, were asked about the impact on their company of various situations. The question presented a scale of 1 (no effect) to 5 (great effect) and Table 3 provides the mean responses to each situation. Table 3: Impact on the organization Impact Mean response Missing information about derivatives within our enterprise 3.09 Inaccurate knowledge of the risk positions of our enterprise 3.61 Price (value) of our derivatives 3.11 Evaluation of the risks of derivatives 3.02 Transaction costs 2.70 Liquidity risk 3.20 Business failure risk 3.34 Fiscal problems 3.07 Disclosure obligations 3.52 Public perception 3.30 Canadian accounting regulations 3.68 International accounting regulations (IAS/US GAAP) 2.89 Canadian accounting regulations have the greatest impact on the firm according to the respondents. This is not surprising given that the accounting standards regarding derivatives have recently changed and are continuing to change as Canada prepares to adopt International Accounting Standards. 20 The second greatest effect on companies is an inaccurate level of knowledge of the risk positions within the enterprise. Management of exchange rate risks The next series of questions focused only on exchange rate risks. Respondents were asked about their approach for measuring exchange rate risks with respect to accounting, transaction and economic exposures. For each exposure respondents could indicate whether they did nothing, observed the exposure or set targets for controlling the exposure. Accounting exposure was defined as the effect of exchange rates on the equity/profit reported in the balance sheet. This exposure was actively controlled by (16.1%) of respondents. This indicates that the translation of the balance sheet for reporting purposes is being managed, likely by a forward contract. The majority of respondents observe the impact (57%) and 25% do nothing. Transaction exposure was defined as the effect of exchange rates on the Canadian dollar receivables and payables compared to other currencies receivables and payables. Target-oriented control was indicated by 12.5%, with 58.9% observing and 26.8% taking no measures. Economic exposure was defined as the effect of exchange rates on the competitive ability and thus on the fluctuation in expected cash flows. Only 8.9% selected active control, 62.5% observe and 26.8% do nothing. (In all cases 2% did not respond.) The results for transaction and economic exposure may be partly explained by the hedging strategies taken by the responding firms. The majority of respondents (51.8%) balance foreign receipts and commitments, and 42.9% said they match expected future foreign currency revenues and expenses. Sixteen percent also stated they use the actual equity positions of their foreign subsidiaries. Respondents were able to select any or all of the hedging options. There was also a significant relationship between the ability to match expected future foreign currency revenues and expenses and industry (t=-2.461, p<.05). The results indicate there are natural hedging opportunities available to many organizations. This conclusion is consistent with the responses to the risk management strategies indicated in Table 4. One explanation for 39.3% of the respondents not taking any protection would be that they have the ability to naturally hedge with an offsetting transaction (This question was not specifically asked). Table 4: Exchange rate management strategies Impact Frequency No response 3.6% Generally no protection against foreign currency exposure 39.3% Immediate protection against foreign currency exposure 14.3% Follow guidelines for partial protection against foreign currency exposure 28.6% Selected security for foreign currency exposures based on expected or 10.7% predicted exchange rate fluctuations Profit-oriented approach to foreign currency exposures depending on 3.6% expected or predicted exchange rate fluctuations Respondents were asked if they engage in any forecasting or evaluating of future exchange rate fluctuations. Nearly 45% said they did. Those engaging in forecasting were then asked how important various methods were on a scale of 1 (not important) to 5 (very important). The most important method of forecasting utilizes free information (mean of 3.88) followed by fundamental analyses (based on macroeconomic variables; mean of 3.16) and personal expectations (without formal analysis; mean of 3.04). Paid for forecasts were the least important (mean of 1.96).Regression analysis indicated a significant relationship between the use of fundamental analyses for forecasting future foreign exchange rates and industry (t=-2.203, p<.05). 21 With respect to foreign exchange rates, respondents were asked whether they agreed, strongly agreed, disagreed or strongly disagreed to a series of statements. The results are presented below grouped by agreement, disagreement. (Non-responses have been removed).5 Table 5: Reaction to statements about foreign exchange rates Statement In “good times,” i.e. in periods with relative high gains, we protect against unexpected exchange rate fluctuations less than we would otherwise Smoothing income against expected gains is an important target in order to reduce tax payments What our competitors are doing regarding risk management plays a large role in our risk management policies Due to the international level of our activities exchange rate risk is somewhat reduced The most reliable source of information for risk management decisions is the financial markets which we consider to be efficient Our predictions of exchange rate movements have been very accurate therefore we have been able to obtain high yields in the passed years through our selective hedging strategy in comparison to a full hedging strategy The correlation of exchange rate risk to other enterprise risk is not influencing our hedging decision We regularly measure the success of our risk management policies % agree 27.8 % disagree 72.2 13.2 86.8 13.2 86.8 39.6 60.4 70.4 29.6 26.0 74.0 62.7 37.3 59.3 40.7 Management of interest rate risk Interest rate risk is the focus of the next set of questions. Questions similar to exchange rate risk were asked of respondents. They were asked whether changes in interest rates were monitored or actively controlled. The majority of respondents (76.8%) stated they either did nothing about interest rate changes or monitored them. Some monitor and quantify interest rate changes (21.4%) and only one respondent actively controls them through set targets. The policy with respect to interest rate fluctuations is significantly related to industry (t=2.1, p<.05). The respondents who monitor interest rate risk changes were asked to indicate how important various types of unexpected interest rate fluctuations were on a scale of 1 (represents no risk for us) to 5 (represents a substantial risk for us). Table 6 provides the mean responses. The responses suggest that long-term debt may predominantly be at variable rates rather than at fixed rates of interest. The lower scores for a reduction in interest rates are consistent with this view because the question was asking about risk and a variable rate of interest being reduced would not present much of a risk. The majority (41.1%) of respondents state they have no security for managing interest rate risk in the organization. Taking partial or minimum security was selected by 21.4% and taking selective security depending on expectations/predictions of future interest rates was selected by 19.6% of respondents. 5 Our findings are consistent with Bodnar et al. (1996, 1998), Bodnar et al. (1995), Jalilvand, Switzer & Tang (2000). 22 Table 6: Risk of unexpected interest rate fluctuations Increase in interest with respect to forthcoming capital funding needs A reduction in interest rates regarding capital funding in place A reduction in interest rates regarding future cash flows Increase in interest rates regarding cash flows in place Long-term effects of interest rate fluctuations on operational cash flow Mean response 3.18 2.28 2.31 3.15 3.36 Respondents were asked to select strongly agree, agree, disagree or strongly disagree to a series of statements regarding the management of interest rate risks. The responses have been divided between those that agreed and those that disagreed to each statement and are presented in the following table. (Non-responses have been removed.) Table 7: Interest rate risk management statements Statement % agree The management of interest rate risks in our enterprise is not as developed as our 47.1 management of exchange rate risks The management of interest rate risks and the management of exchange rate risks 39.2 are, in most cases, considered separate functions and are the responsibility of the department this function falls under Our predictions of interest rate movements have been very accurate, therefore we 48.9 have been able to obtain high yields in the passed years through our selective hedging strategy in comparison to a full hedging strategy We regularly measure the success of our interest rate strategies 51.0 % disagree 52.9 60.8 51.1 49.0 These results have changed over the years as prior research suggests that interest rate risks were not as developed as exchange rate risks. However, this statement is significantly related to industry suggesting that the development is stronger in some industries (t=2.038, p<.05). Prior research also shows that the management of interest rate and exchange rates tended to be the responsibility of different departments. This finding is not supported by our results. Internal controls The final section of questions related to internal controls in the organization. Little information about internal controls over derivative use has been included in prior risk management studies. The majority of studies contain two main questions, which are whether there is a documented risk management policy and how often derivative positions are reported to the Board of Directors. (Ginting & Helliar, 2003; Mallin et al, 2001; De Ceuster et al, 2000; Jalilvand et al, 2000; Bodnar & Gebhardt, 1999; Bodnar et al, 1998 and 1996.) Lam (1997) presents seven broad sections that should form the foundation of any internal control system for risk management. They are laid out as lessons in his article, “Firm wide Risk Management: An Integrated Approach to Risk Management and Internal Control.” The internal control questions asked in this survey were formulated based on Lam’s lessons as well as a document developed by Deloitte & Touche LLP as part of the Committee of Sponsoring Organizations of the Treadway Commission (COSO) initiative in the U.S.A. The report is part of a three-volume set on internal control commencing with an integrated framework developed by COSO in 1992. They in turn requested Deloitte & Touche LLP to develop an information tool that would help companies implement the framework. 23 The first question asked about overall general controls such as the existence of an independent board of directors and an audit committee. As expected, all but one company has an independent board of directors and all have an audit committee. Other questions specifically related to specific risk management. Our findings suggest that segregation of duties within Treasury management – in particular the planning, handling and evaluating of derivative positions and detailed risk management policies and processes – i.e. written guidelines determining the level of risk, management can take are not widely practiced. With respect to derivative trading respondents were asked to identify which controls were in place in their enterprise. The following table provides the control and the percentage of respondents stating that it was in place in their organization. Table 8: Percentage of respondents having derivative trading controls Authorized signature required to initiate a trade Confirmation of trades required from trading institution in writing Trading orders and receipt of trade confirmation are handled by two separate departments All trades must be identified as either a hedge or a trade Hedges must be documented as such and linked to a particular identified exposure Monthly reconciliation of trade accounts to cash balance Reconciliation reviewed and authorized Regular, at least monthly, quantification of the potential loss position of all derivatives (i.e. using a value-at-risk approach, etc.) Separate review, at least monthly, of quantification of net trading position 80.4% 80.4% 33.9% 46.4% 66.1% 69.6% 66.1% 53.6% 41.1% Trading orders not being handled by separate departments is consistent with the majority of respondents stating the lack of segregation of duties within Treasury management. Identifying derivative transactions as hedges or trades is potentially related to the new Canadian accounting regulations. The low percentage of respondents having this control may lead to greater volatility in future income statements as the fair value of the derivatives are adjusted to current market conditions at each reporting period. Regular quantification of the potential loss position is significantly related to industry (t=-2.234, p<.05). This suggests that some industries are far more vigilant about the impact of derivatives on their organization. One of Lam’s (1997) lessons is to keep an eye on the cash. This is crucial in any organization and often receives a good deal of attention. In a risk management system, managers have the ability to trade in many different currencies and to trade in derivatives to manage the foreign exchange rate fluctuations in each currency. Keeping an eye on the cash is particularly difficult which many corporations suffering from large losses have found out. With this in mind, one might have expected a much higher percentage of firms selecting the monthly reconciliation of trade accounts to cash balance. The results also suggest that companies tend to have documentation type controls in place with less emphasis on following up as suggested by the lower percentages on reviewing controls. Respondents were asked about the attitude within the company toward risk management by selecting whether certain statements were true or not. Table 9 presents the percentage of respondents who selected “yes” to each statement. The “no” and non-responses have been eliminated. (In all cases, only 1 or 2 respondents did not answer the question.) The low percentage of respondents (55.6%) feeling that they have regular meetings to communicate lessons learned supports the previous finding that companies tend to have stronger documentation controls and weaker follow-up or investigation style controls. In relation to the strong responses to some of the attitudes, the lower percentage feeling appropriately rewarded for meeting risk management goals warrants further exploration. 24 Table 9: Percentage of positive responses to risk management attitudes Do You feel that the top level of management believes in the guidelines established for risk management? Do you feel you have an avenue to discuss your risk management ideas/suggestions with top management? Do you feel you are appropriately rewarded when you meet the risk management goals? When a loss position is expected or occurs from a particular risk exposure, is there an avenue to communicate the results to top management? Are there regular meetings to communicate lessons learned? Do you feel there are adequate training and development programs to assist you in your responsibilities? 92.7 98.2 75.9 98.2 55.6 74.1 Measuring performance, whether system generated or human, is an integral part of most corporations. Respondents were asked about the frequency of a continual recognition process for all risks the organization is exposed to. The majority of respondents (60.7%) stated that it occurred less than once a month and 17.9% selected monthly. (10.7% selected not applicable or did not answer the question). Measuring the risks that the company is exposed occurs monthly in 37.5% of respondents’ companies and less than once a month in 46.4%. Regular written reports to management regarding the results of Treasury management appraisals and the current risk positions occurs either monthly or less than once a month in 84% of the respondents’ firms. The compensation and incentive process is implemented in an organization by measuring performance. Human behavior is influenced by the type of compensation and incentives awarded. Empirical research has shown that a common method of evaluating derivative managers is by the impact on the bottom line (Bodnar et al, 1998). Respondents were asked how the compensation and incentive process takes place in their organization with respect to the employees responsible for risk management. The choices listed can be grouped into income or profit related measures and the ability to manage risks. Of the respondents who answered the question, 55.3% reward based on some form of income or profit and 44.7% reward employees responsible for risk management based on their ability to manage risk. Logit Analysis In this part of analysis, we present the potential relationships among some of the variables described above. Given the rich data set, it is possible to run many different regression analyses. However, we focus on a few specific issues: types of risks, types of hedging instruments used, centralized versus decentralized decision-making for risk management, and specific instruments used in different industries. Given our qualitative variables, we find it useful to use logit model for our analysis. Our sample is reduced to 54 when we consider only those observations that indicate hedging instrument use. This is the base sample that we will be using for the remaining of the analysis. In addition, all the derivative financial instruments are collapsed into four broad categories (as is done in the literature): Currency instruments, Interest rate instruments, Commodity instruments, and Other. Since we observe that only one respondent responded to the Other category, we focus on the first three categories.6 For each instrument the sample is split into two groups: frequent users and non-frequent users. Through the first logit analysis, we assess the relationship between a specific instrument, industry, and the centralized control of decision-making related to risk management. We collapse the industries into three broad categories: Manufacturing, Mining, oil and gas, and Other. We then introduce 6 We analyzed the Other category but did not find any significant result. 25 two dummy variables for the first two categories. The Other category works as the base category. The variable Control is an ordinal variable. It assumes three values: 1, 2, and 3, depending on the degree of centralization of decision-making with respect to risk management. In this logit model, the dependent variable is the derivative user type: frequent users (1) versus non-frequent users (0). The result of the maximum likelihood estimation is presented in Table 10. Given that the goodness of fit is of secondary importance in logit analysis, we focus on the economic content of the signs of the parameters and their statistical significance (using Wald test). Table 10: Use of hedging instruments across industries and centralized decision making on hedging Currency instruments 3.947 (1.905)*** Interest rate instruments -0.146 (1.553) Commodity instruments -4.715 (2.445)** Control -1.563 (0.648)*** -1.070 (0.555) 0.774 (0.823) Manufacturing 1.423 (0.811)** 0.085 (0.706) 1.975 (0.971)*** Mining, oil and gas 0.000 (0.703) -0.153 (0.679) Intercept 1.669 (0.949)** Note: *** and ** represent levels of statistical significance at less than 5% and 8%, respectively. For currency instruments, we find Manufacturing industries are expected to use them more frequently. But the statistical significance of the Intercept term shows that the Other industry category firms are also likely to be frequent users of the currency instruments. Given the large export sector of Canada the significance of the parameters and their positive signs makes intuitive sense. It is also clear that frequent users of the instruments are likely to enjoy some level of decision-making privileges. This is observed through the negative sign of the parameter for Control. Interestingly enough, frequent use of Interest rate instruments and instruments categorized as Other, not shown in Table 10, are not related to industry or level of control in decision-making. This result may be driven by the type of risks these instruments allow users to hedge against. Some intuitive explanation may be provided; first, since over exposure to interest rate risk can increase the cost of capital for the firm most non-financial firms may organize their debt structure in such a way that they are not prone to much interest rate risk. A second reason may be the level of risk exposure. Even though, some firms may be exposed to this risk, it is of secondary importance from the perspective of risk management. Or, it may be the case that some of these firms manage these risks less frequently. Finally, coefficients for Commodity instrument use are positive and significant both for Manufacturers and Mining, oil, and gas companies. They are more likely to be frequent users of the instruments. Manufactures would likely hedge against the risk of raw materials and transportation costs, and Mining, oil and gas companies would be interested in securing stable cash flow for their shareholders. The second logit analysis is aimed at gaining specific insight into the purpose of hedging. Also in this model, we want to reassess the role of centralization of decision-making so we again use the Control variable. We consider four broad categories of risks: General environmental risk, Industry risk, Enterprise risk and Financial risk. In order to isolate an individual risk, we introduce a dummy variable that assumes a value of 1 if the respondents respond by selecting 4 or 5 out of a scale of 1 to 5 (1 indicating the risk 26 being not important and 5 indicating the risk being very important). So we have two categories of respondents: respondents for whom a specific risk is substantially important and respondents for whom it is not important. The frequency of the use of a specific instrument is the dependent variable and risk type and control are independent variables. Please note that the risk dummy represents the risk under consideration. The parameter value of the logit regression is presented in Table 11. Table 11: Types of risk exposure, hedging instruments used and centralized decision making Currency instruments General environmental risk Intercept Risk dummy Control Industry risk Intercept Risk dummy Control Enterprise risk Intercept Risk dummy Control Financial risk Intercept Risk dummy Control Interest rate instrument Commodity instruments 4.262 (1.886)*** -0.255 (0.583) -1.526 (0.657)*** -0.608 (1.578) 0.788 (0.571) -0.095 (0.564) -4.804 (2.606)** 1.727 (0.858)*** 0.818 (0.876) 2.589 (1.835) 1.467 (0.663)*** -1.319 (0.638)*** 0.107 (1.624) -0.233 (0.589) -0.153 (0.559) -3.741 (2.523) 0.411 (0.776) 0.726 (0.843) 3.903 (1.866)*** 0.169 (0.582) -1.474 (0.650)*** -0.151 (1.561) 0.024 (0.561) -0.117 (0.554) -3.214 (2.372) -0.032 (0.709) 0.639 (0.820) 4.007 (1.942)*** 1.411 (0.752)** -1.856 (0.748)*** -0.773 (1.606) 1.264 (0.727)* -0.237 (0.573) -3.452 (2.328) 0.452 (0.865) 0.596 (0.814) Note: ***, ** and * represent levels of statistical significance at less than 5%, 8% and 10% , respectively. In Table 11, the top section presents the estimated parameter value of the regression where we assess the use of Currency instruments by corporations significantly concerned about General environmental risk. We notice that the parameter for the Risk dummy is not significant, which indicates that general environmental risk does not have any bearing on Exchange rate risk. This is intuitive given the use of the instrument; this instrument is used to hedge currency risk not the other type of risks such as political risks and legal risks. The significance of the intercept term indicates that frequent use of this instrument is important for other types of risks. Interest rate instruments are not important for this type of risk. But the significance observed for the Commodity instrument may be ascribed to the high correlation between political risks and commodity prices, especially oil prices. Firms uncertain about the impact of the global political environment may want to hedge their raw material price risk as well as risk of higher cost of logistics. Note that the signs of the parameters for the Control variable are negative and significant for all 27 types of risks. Since the interpretations of these are the same as provided earlier, we will not elaborate further. In international finance, it is always emphasized that operational risk arises due to currency fluctuation. Multinational firms or exporting and importing firms may be negatively affected by unfavorable movements in exchange rate changes. For example, unfavorable exchange rate movements may cause the firm to lose its competitive edge. This concern is captured through the Industry risk dummy in our analysis. So rightfully, we find exchange rate instrument use to be important for the firms concerned about Industry risk. None of the other instruments are used frequently by any of the firms to hedge this kind of risk. Enterprise Risks, such as accident risk, are not well hedged by any of the instrument types under consideration. Some of these kinds of risks are better managed by insurance services. Whereas, this risk coefficient for the dummy is not significant, the Intercept term is statistically significant. This indicates that concern for other kinds of risk may motivate frequent use of Currency instruments. Finally, we find that for financial risk concerns, frequent use of currency instruments is important. In addition, we note that to a certain extent in managing financial risk, corporations may also use Interest rate instrument (parameter significant at 10% level). Intuitively this risk type is naturally one of the most important for all companies given its immediate impact on corporate cash flow. For this type of risk, Commodity instruments are not as effective a risk management tool. Conclusion In this study, we reassessed the hedging practices of Canadian firms. We find that Canadian firms are selectively hedging their exposed risk positions. This suggests that natural hedging opportunities exist. The instruments being used depend on the industry the firm is operating in. We find that manufacturing companies are frequent users of currency instruments and mining, oil, and gas companies are frequent users of commodity instruments. Currency instruments continue to be the most popular risk management instrument to manage industry risk and financial risk. General environmental risks, due to risks such as global political instability, are managed using Commodity instruments. We also uniquely link the day-today decision-making for risk management. Whether the decision for risk management will be made centrally or whether it will be delegated across the organization depends on the type of instrument being used. With exchange rate instruments the control is limited .i.e. the responsibilities are delegated. Accounting regulations are creating problems with using derivatives. Also, internal controls are not as effective because there is no reassessment mechanism in place. On methodological grounds, we note that mail surveys and web-based surveys complement each other. Incentive design continues to be a challenge in motivating high level executives to participate in corporate surveys. So we need to develop newer incentive mechanisms for data generation. In addition, although low response rates are a problem with this type of financial research, employing a proper econometric methodology and with consideration given to existing empirical findings value may be created. Appendix Low response rate has been a persistent obstacle in survey research. It is not only an issue in corporate finance research, but it is also an important issue in other areas of social science (Dey, 1997). Specifically, surveys involving high level corporate executives obtain response rates ranging from 10% to 30%. Surveys involving more detailed questionnaires tend have even lower response rates. Sometimes we 28 also find that trade organization sponsored surveys, such as the Certified Financial Analysts’ Institute, have higher response rates due the commitment of the members to the organizational cause. Our survey, which included 39 detailed questions, was directed at top executives involved in the hedging process. We suspect that our low response rate is due to the length of the questionnaire. In addition, although anonymity was promised, it may be the case that some executives did not want to reveal their practices out of privacy concerns. Providing anonymity however, comes at a price. We are not able to link the responses to a particular company, and, therefore, unable to do further analysis based on publicly available financial information. Aside from the limitation of a low response rate, we explored the reasonableness of our conclusion through documented findings and econometric analysis. Most of the findings that we obtained are similar to the findings of surveys having higher response rates. So we can conclude that our sample is not dissimilar to other samples, and we may use econometric analysis to gain new insights. With regards to the survey methodology, we would like to share some of our experienced. As has been noted in the main text, empirical analysis on hedging is difficult to conduct due to some endogeneity issues (Bartram & Bodnar, 2007). The available market data my also be limited in revealing hedging needs and effectiveness. Therefore, survey data allows us to obtain information on current practices. We conducted our survey through two medium: mail and the Internet. All the respondents were allowed to complete the survey in either medium. Half the respondents used each medium, which indicates that the two mediums are complementary. For designers of new surveys this insight can be valuable. We also realize a limitation with respect to our survey incentive. There was no material incentive for the executives to participate in either the first or second round. The executives were given the incentive of having the opportunity to have a copy of our report. But in the third round of the survey, we provided a coffee card as an incentive for filling out the form. Even though, we received 16 additional responses (on top of 40 responses from the first two rounds), we would not confidently say that it is due to the coffee card. Given the commitments of top executives, it is hard to provide any material incentive. In this case, one of the key incentives may be generated by the trade organizations. With all of its limitations and potentials, survey research should be pursued to collect unique useful data to complement other sources of data. Low response rates should not deter us from evaluating the responses; without evaluating the responses, we would be deprived of the knowledge that those responses may confer. If proper theoretical guidance is followed and effective econometric methodologies are employed, it is possible to assess findings in light of an existing knowledge base. References Bank for International Settlements (2005). Triennial Central Bank Survey: Foreign exchange and derivatives market activity in 2004. Basel, Switzerland: Bank of International Settlements. Bartram, S.M., & Bodnar, G.M. (2007). The exchange rate exposure puzzle. Managerial Finance, 33(9), 642-666. Bartram, S.M. (2008). What lies beneath: Foreign exchange rate exposure, hedging and cash flows. Journal of Banking and Finance, in press. Belk, P.A., & Glaum, M. (1990). The management of foreign exchange risk in UK multinationals: an empirical investigation. Accounting and Business Research, 2(81), 3-13. Berkman, H., Bradbury, M.E., & Magan, S. (1997). An international comparison of derivative use. Financial Management, 26(4), 69-73. Bodnar, G.M., Hayt, G., & Marston, R.C. (1995). 1994 Wharton survey of derivatives usage by US nonfinancial firms. Financial Management, 24(2), 104-114. 29 Bodnar, G.M., Hayt, G.S., & Marston, R.C. (1996). 1995Wharton survey of derivatives usage by US nonfinancial firms. Financial Management, 25(4), 113-133. Bodnar, G.M., Hayt, G., & Marston, R.C. (1998). 1998 Wharton survey of financial risk management by US non-financial firms. Financial Management, 27(4), 70-91. Bodnar, G.M, & Gebhardt, G. (1999). Derivatives usage in risk management by U.S and German nonfinancial firms: A comparative survey. Journal of International Financial Management and Accounting, 10(3), 153-187. Bodnar, G.M., Jong, A.D, & Macrae, V. (2003). The impact of institutional differences on derivatives usage: A comparative study of US and Dutch firms. European Financial Management, 9(3), 271-297. Borokhovich, K., Brunarski, K., Crutchley, C., & Simkins, B. (2004). Board composition and corporate use of interest rate derivatives. The Journal of Financial Research, 27(2), 199-216. Chidambaran, N.K., Fernando, C.F., & Spindt, P.A. (2001). Credit enhancement through financial engineering: Freeport McMoRan’s gold-denominated depository shares. Journal of Financial Economics, 60, 487-528. Crabb, P.R. (2002). Multinational Corporations and Hedging Exchange Rate Exposure. International Review of Economics and Finance, 11(3), 299-314. De Ceuster, M.J.K., Durinck, E., Laveren, E., & Lodewyckx, J. (2000) A survey into the use of derivatives by large non-financial firms operating in Belgium. European Financial Management, 6(3), 301-318. Deloitte & Touche LLP (1996). Internal control issues in derivatives usage: An information tool for considering the COSO internal Control - integrated framework in derivatives applications. New Jersey: American Institute of Certified Public Accountants. Dey, E.L. (1997). Working With Low Survey Response Rates: The efficiency of Weighting Adjustment. Research in Higher Education, 38(2), 215-227. El-Masry, A., & Omneya, A. (2007). Exchange rate exposure: do size and foreign operations matter?. Managerial Finance, 33(9), 741-765. Fok, R.C.W., Carroll, C., & Chiou, M.C. (1997). Determinants of corporate hedging and derivatives: a revisit. Journal of Economics and Business, 49, 569-85. Géczy, C., Minton, B.A., & Schrand, C. (1997). Why firms use currency derivatives. Journal of Finance, 52(4), 1323-1354. Géczy, C., Minton, B.A., & Schrand, C. (2007). Taking a view: Corporate Speculation, Governance, and Compensation. Journal of Finance, 62(6), 2405-2443. Ginting, Y.I., & Helliar, C.V. (2003). Financial Risk Management and Derivatives usage in Indonesian non-financial firms. University of Dundee: Discussion paper series. Graham, J., & Smith, C.W., Jr. (1999). Tax incentives to hedge. The Journal of Finance, 54(6), 22412262. Graham, J., & Harvey, C. (2001). The Theory and Practice of Corporate Finance: Evidence from the Field. Journal of Financial Economics, 60, 187-243. Graham, J., & Rogers, D. (2002). Do firms hedge in response to tax incentives?. The Journal of Finance, 57(2), 815-839. Hogan, A., & Rossi, C. (1997). The Characteristics of Derivatives Users: An Analysis of the Thrift Industry. Journal of Economics and Business, 49, 587-598. Jalilvand, A. (1999). Why firms use derivatives: evidence from Canada. Canadian Journal of Administrative Sciences, 16(3), 213-28. Jalilvand, A., Switzer, J., & Tang, C. (2000). A global perspective on the use of derivatives for corporate risk management decisions. Managerial Finance, 26(3), 29-39. Lam, J. C. (1997) Firmwide risk management: An integrated approach to risk management and internal control. In R. J. Schwartz & J. S. Clifford Jr. (Eds.), Derivatives Handbook: Risk Management and Contro (pp. 406-429). New York, NY: John Wiley & Sons Inc. Loderer C., & Pichler K. (2000). Firms, do you know your currency risk exposure? Survey results. Journal of Empirical Finance, 7, 317-344. 30 Mallin, C., Ow-Yong, K., and Reynolds, M. (2001). Derivatives usage in UK non-financial listed companies. The European Journal of Finance, 7, 63-91. Marshall, A.P. (2000). Foreign exchange risk management in UK, USA and Asia Pacific multinational companies. Journal of Multinational Financial Management, 10(2), 185-211. Modigliani, F., & Miller, M. (1958). The cost of capital, corporation finance and the theory of investment. American Economic Review, June, 261-297. Muller, A., Verschoor, W.F.C. (2006). Foreign exchange risk exposure: Survey and suggestions. Journal of Multinational Financial Management, 16 (4), 385-410. Smith, C.W., Jr., & Stulz, R.M. (1985). The determinants of firms’ hedging policies. Journal of Financial and Quantitative Analysis, 20(4), 391-405. Stulz, R.M. (1990). Managerial discretion and optimal hedging policy. Journal of Financial Economics, 26(1), 3-27. Stulz, R.M. (1996). Rethinking risk management. Journal of Applied Corporate Finance, 9(3), 8-24. Trahan, E. A., & Gitman, L. J. (1995). Bridging the Theory-Practice Gap in Corporate Finance: A Survey of Chief Financial Officers. Quarterly Review of Economics and Finance, 35(1), 73-87. Tufano, P. (1996). Who manages risk? An empirical examination of the risk management practices in the gold mining industry. Journal of Finance, 51(4), 1097-137. Tufano, P. (1998). Agency costs of corporate risk management. Financial Management, 27(1), 67-77. 31 ASAC 24-27 mai 2008 Halifax, Nouvelle-Ecosse ASSAÏDI Abdelouahid (étudiant) CREG, CNAM Paris LA RATIONALITE MIMETIQUE DANS LA FORMATION DES RECOMMANDATIONS DES ANALYSTES FINANCIERS : ETUDE DE TERRAIN Résumé Notre étude s’intéresse au rôle joué par la « rationalité mimétique » dans les recommandations des analystes financières. La notion de mimétisme à laquelle on s’intéresse correspond aux situations où les analystes délaissent leurs propres croyances et basent leurs décisions sur les actions collectives même si ces dernières contredisent leurs propres prédictions. L’étude de terrain nous a permis de mieux cerner la fonction d’analyste et de comprendre leurs réactions sur le marché financier. Introduction Les analystes financiers constituent un lien indispensable entre l’entreprise et les investisseurs. Leur rôle ne se réduit pas exclusivement à l’énonciation des recommandations. La tâche de l’analyste se résume à faire des estimations des perspectives des firmes dont il en charge le suivi (structure financière, bilan, performances générales, qualité de management, stratégie de croissance annoncée, etc.). Pour formuler des prévisions et former des recommandations, l’analyste financier a recours à de nombreuses et diverses sources d’informations telles que : l’information comptable, les rencontres privées avec les dirigeants des entreprises, les assemblées générales, les réunions entre analystes au sein de la SFAF, les études diverses, la presse spécialisée et générale, etc. Ses recommandations prennent la forme de conseil, traduits dans des notes et rapports qu’ils rédigent régulièrement. L’histoire nous enseigne que l’environnement de l’analyste est un environnement mouvant. Autrement dit, sur un marché financier, l’analyste financier agit dans un univers incertain qui rend délicat ses prévisions et recommandations. Les multiples sources informationnelles et les comportements des acteurs contribuent à la construction de l’opinion de l’analyste. Comment peut-il ainsi prendre ses repères ? Si l’hypothèse de base de la finance moderne est sans doute celle de l’efficience des marchés financiers où les prix reflètent toute l’information disponible, la littérature a cependant amassé, durant ces deux dernières décennies, un nombre substantiel d’observations d’anomalies apparentes par rapport à cette hypothèse (DeBondt, Thaler, 1989). D’autre part, le marché financier contrairement à la vision orthodoxe subit de temps à autre des « chocs » dus à l’interaction des intervenants se traduisant le plus souvent par des comportements mimétiques. Ces comportements font probablement partie de la nature humaine et caractérisent les instincts les plus basiques de chacun comme le note Dupuy (1979), « L’Autre est un modèle, et le Sujet ne songe qu’à l’imiter… » « … Toute socialisation procède d’un apprentissage, qui n’est jamais autre chose qu’une imitation. ». L’étude des phénomènes mimétiques présente un intérêt réel dans la mesure où ils proposent des réponses à certaines anomalies décelées sur le marché financier. Le mimétisme peut être le résultat 32 d’agissements irrationnels ou à l’inverse rationnels des analystes et peut se manifester sous différentes formes. Dans ce présent travail, nous nous intéressons au mimétisme rationnel, c'est-à-dire que l’organisation du marché ou la répartition de l’information implique qu’il est rationnel pour les analystes de copier d’autres analystes en ne tenant pas compte de leur information privée (Artus, 1995). La notion de mimétisme sur le marché financier à laquelle on s’intéresse correspond aux situations où les individus délaissent leurs propres croyances et basent leurs décisions sur les actions collectives même si ces dernières contredisent leurs propres prédictions. Chacun des agents a une idée personnelle de son activité, de ses capacités d'analyse, de décision et de son aptitude à recueillir la « meilleure » information. Mais les questions qu'ils se posent en permanence sont : pour ma recommandation, dois-je prendre une décision autonome et par conséquent risquer de rester en retrait ? Ou dois-je coopérer en me calquant sur les idées des autres ? Dans les deux cas de figure, ils se ressassent finalement la question, à savoir s’ils auront tort ou raison (dans un futur proche) sur le marché face à la présence des autres agents. Autrement dit, notre problématique peut se formuler en ces termes simples à savoir : Quel rôle joue la rationalité mimétique dans la formation des recommandations des analystes financiers sur le marché financier ? La première section décrit les caractéristiques de la profession d’analyste financier et son rôle sur le marché. Elle décrit également l’approche rationnelle du mimétisme. La seconde section retrace la méthodologie employée lors de ce travail et enfin la troisième section décrit les résultats. Quel est le rôle de l’analyste financier ? La prise de conscience du rôle grandissant et de l’influence des analystes financiers sur les marchés financiers s’est généralisée avec l’essor et la dérégulation des marchés financiers et l’internationalisation des entreprises (Bayle et Schwartz, 2005). Ce sont ces mutations importantes de l’analyse financière qui ont poussé le législateur à définir de manière précise le métier d’analyste. C’est ainsi que la loi de sécurité financière (loi du 1er août 2003) qui a certes renforcé les contrôles et les organes qui en sont chargés, mais surtout donne pour la première fois une définition juridique de la profession d’analyste financier : « exerce une activité d’analyse financière toute personne qui, à titre de profession habituelle, produit et diffuse des études sur les personnes morales faisant appel public à l’épargne, en vue de formuler et de diffuser une opinion sur l’évolution prévisible desdites personnes morales et, le cas échéant, sur le prix des instruments financiers qu’elles émettent ». Dans leur fonction première, ils agissent comme conseillers en placement pour la clientèle que représentent les investisseurs en bourse. Cela suppose qu’ils procèdent à une évaluation régulière des stratégies de développement suivies par les dirigeants et qu’ils en tirent des conséquences sur l’intérêt ou non de détenir les actions concernées en portefeuille. On comprend que cette fonction nécessite une grande liberté de jugement et une certaine forme d’indépendance par rapport à l’institution financière dont ils font partie. Leur rôle a été tel durant les années 1990 qu’elles furent même qualifiées « d’ère des analystes », bien qu’après l’éclatement de la bulle Internet et les scandales américains, ils aient été mis sur la sellette. On leur a reproché en effet d’avoir alimenté la bulle spéculative, de ne pas être suffisamment indépendants des établissements financiers qui les emploient (conflits d’intérêts avec les activités de banques d’affaires) et de n’avoir pas anticipé la détérioration de la situation financière des 33 grands groupes internationaux comme Enron, Vivendi Universal, Worldcom..., d’avoir suivi aveuglément quelques leaders d’opinions et donc de n’avoir pas été assez rationnel dans un contexte euphorique. Le rôle principal des analystes financiers consiste à fournir des prévisions et des recommandations d’achat ou de vente de titre au marché et, en particulier aux gérants de portefeuille (investisseurs). Ils collectent ainsi des informations auprès de différentes sources pour les interpréter, les retraiter et les restituer de manière synthétique et intelligible afin de permettre aux gérants de prendre des décisions optimales et générer ainsi des profits. Autrement dit, le travail des analystes financiers consiste à recueillir des informations précises dans l’objectif de formuler des prévisions et des recommandations en direction des investisseurs moyennant une commission. Ils émettent en ce sens un signal privé plus précis que celui contenu dans les prix. De nombreux articles ont en effet confirmé l’utilité des prévisions et recommandations pour les investisseurs (Stickel, 1991). Park et Pincus (2000) montrent que les prévisions des analystes financiers affectent directement les anticipations des investisseurs et leur comportement à l’annonce des résultats comptables. En bref, les relations entre les analystes financiers et les autres opérateurs du marché ou encore entre analystes financiers eux-mêmes sont si étroites qu’elles engendrent des influences réciproques. A l’évidence, il convient de souligner que ces influences n’enlèvent rien au caractère rationnel des opérateurs sur le marché financier. Solliciter les conseils d’un analyste financier se justifie souvent par une insuffisance d’information nécessaire à une prise de décision « rationnelle » sur le marché financier. Cette sollicitation résulte également de l’existence de croyances multiples et hétérogènes. Cela veut dire qu’en réalité tous les intervenants sur le marché financier ne sont pas au même niveau d’information (Van Loye et Fontowicz, 2004) contrairement au cadre Walrasien où toute l’information existante est disponible et sans coût, et donc que tous les intervenants se situent sur un même pied d’égalité. Pour Mandelbrot (2005) l’information peut être source de risque lorsque celle-ci est inconnue ou inatteignable, lorsqu’elle est cachée ou présentée de manière inexacte ou encore lorsqu’elle est mal interprétée. L’histoire financière nous enseigne à ce propos l’ampleur de ruines financières lorsqu’il y a peu ou trop d’informations. Les évolutions de l’environnement économique et financier amènent les firmes financières de plus en plus nombreuses à faire du traitement d’information et de l’analyse financière leur spécialité, et pour ce faire employer des analystes financiers en leur donnant pour mission de dissiper l’opacité des marchés financiers. En résumé, l’analyste recueille dans un premier temps des informations économiques et financières générales et spécifiques des sociétés suivies, procède ensuite à leur analyse et enfin établit les prévisions afin d’élaborer l’objectif de cours et ses recommandations. Autrement dit et comme le note à juste raison Pierret (1986), l’analyste est perçu comme un « homme de synthèse ». L’approche rationnelle du mimétisme Que signifie la notion de mimétisme sur le marché financier… ? Il est vrai que l’étude des marchés financiers attire le regard surtout des courbes, des formules mathématiques et la performance des différents titres, au dépend bien souvent de l’analyse du comportement humain, qui joue pourtant un rôle non des moindre. Bien que l’on ne puisse nier l’essor sans précédent de l’ingénierie financière et de l’outil informatique, l’homme demeure toutefois le 34 principal organe qui se révèle par les décisions prises et l’impact que ses décisions ont sur les cours. C’est donc à l’homme que revient cette habileté à régner sur le marché le plus raisonnablement possible. L’individu est par définition un être qui a besoin d’autrui pour prendre ses repères et exister en quelque sorte. Par conséquent, il bâti ses gestes, ses comportements sur ceux d’autrui (très souvent, c’est la personne la plus influente qui mène le groupe). Mais est-ce vraiment rationnel ? Pas vraiment si l’on considère que la prise de décision doit se faire de façon autonome et sans influences. La notion de mimétisme en finance n’est pas simple à saisir dès lors que l’on s’attache à la théorie classique traditionnelle l’homo oeconomicus, selon laquelle chaque intervenant cherche à maximiser son utilité de façon individuelle et isolée. L’histoire économique et financière nous enseigne à cet effet, que chaque krach ou crise financière grave trouve sa source dans l’emballement psychologique des opérateurs provoquant des phénomènes de contagion et de mimétisme. La diversité des acteurs sur le marché financier implique logiquement une interaction entre eux, et les influences collectives sont telles, que les prises de décision ne relèvent pas uniquement du sort individuel mais dépendent fortement de l’entourage donc des autres opérateurs. Vu la complexité du système financier et l’interprétation des mouvements erratiques des prix constatés, le mimétisme apparaît comme rationnel, et l’une des règles intangibles permettant de comprendre les motivations décisionnelles des intervenants sur les marchés financiers dans un univers incertain. Et celle de rationalité mimétique ? Sur un marché financier, le jeu consiste à anticiper de manière à être le plus proche possible du prix fixé par le marché. Or, si l’on croit dans les vertus du marché, ce prix correspond à la meilleure estimation que l’on puisse faire de la valeur fondamentale, compte tenu des informations disponibles. Puisque sur un marché coexistent aussi bien des agents informés que des agents non informés, Orléan (1986), note que les agents non informés imitent de façon rationnelle, car grâce à l’imitation, les agents non informés accèdent effectivement à l’évaluation juste de l’actif. Autrement dit, même en ayant presque pas d’information sur la valeur fondamentale future d’un actif, les opérateurs ignorants pourraient tirer profits de leur investissement. Cela reste vrai que si toutefois les agents informés sont en nombre importants, et ce pour ne pas créer des « bulles » source de crise lorsqu’elles éclatent. Artus (1995), définit ainsi trois types de mimétisme sur le marché financier : « Le mimétisme résulte de ce qu’il y a de nombreux individus mal ou pas informés qui se rencontrent aléatoirement et se copient ». En effet, tous les intervenants sur le marché financier ne maîtrisent pas forcément les données existantes du marché et encore moins les prévisions futures, l’interaction pousse logiquement au comportement grégaire. « Le mimétisme résulte de la structure concurrentielle du marché ». Comme l’on sait, chaque intervenant est tenu à satisfaire de manière optimale ses besoins et donc de s’accaparer des plus grandes parts de marché faute de quoi le marché l’évince. Chacun exprime alors le désir de rester en course et pour se faire, se rattacher aux autres et faire comme eux. Une trop grande complexité rapproche les hommes dans leur analyse et dans leurs actes. L’imitation s’impose alors d’elle-même et apparaît comme une protection. - « le mimétisme est aussi rationnel : l’organisation du marché ou la répartition de l’information impliquent qu’il est rationnel pour certains investisseurs de copier d’autres en oubliant leur information privée. » Sur le marché financier, le comportement de chaque analyste et investisseur est directement interprétable par les investisseurs suivants et se trouve être un reflet de sa base décisionnelle. On assiste dès lors à un comportement mimétique lorsqu’un investisseur est amené à reconsidérer son information privée pour suivre délibérément celle des personnes qui le précèdent. Un phénomène de 35 cascade d’informations peut alors apparaître sur le marché dans le cas où les analystes et investisseurs agissent tous dans le même sens, indépendamment de leurs informations privées et du reste du monde. Banerjee (1992) propose un modèle mimétique séquentiel. Les acteurs dans ce modèle interviennent les uns après les autres, les derniers pouvant systématiquement observer les prises de positions des premiers. Banerjee montre que sous certaines conditions, il peut être optimal pour certains intervenants d’abandonner leurs propres sources d’informations pour suivre aveuglément le comportement des autres agents. Dans les modèles de Bikhchandani, Hirshleifer et Welch (1992), les agents acquièrent des informations utiles par l’observation des décisions précédentes des autres agents, et peuvent ignorer complètement leurs propres signaux sans pour autant dévier de la rationalité et de l’optimalité. La réputation et les modes de rémunérations peuvent inciter à des comportements mimétiques rationnels. En effet, l’évaluation des performances des analystes et gestionnaires est souvent basée sur la performance relative. Dans le cas des gestionnaires de fonds, Maug et Naik (1996) développent un modèle dans lequel ils montrent que ces derniers ont tendance à calquer leur portefeuille sur celui du benchmark par peur de réaliser une performance inférieure à celle du benchmark et donc d’avoir une rétribution moins importante. Welch (2000) montre que le consensus prévalant et les deux révisions les plus récentes influencent les recommandations des analystes. Les révisions des autres analystes exercent une influence plus forte si elles sont récentes et si elles prédisent convenablement les rendements des actions ex-post. Comportement mimétique des analystes financiers D’après Bickchandani et Sharma (2001), un individu est dit mimétique, si, sans connaître la décision des autres investisseurs, aurait procédé à l’investissement, mais n’y procède pas lorsqu’il constate que les autres investisseurs ont décidé de ne pas réaliser cet investissement. On parlera donc de comportement mimétique des analystes financiers lorsque ceux-ci décident d’ignorer leurs propres informations et signaux pour suivre les décisions observées des autres analystes. D’autres auteurs tels que Grinblatt, Titman et Wermers (1995), et Nofsinger et Sias (1999) proposent une définition plus large du mimétisme, comme étant le fait qu’un groupe d’investisseurs transige sur le même titre, dans une même direction (achat ou vente), pendant une période donnée du temps. Cependant, la corrélation dans le comportement des investisseurs ne signifie pas pour autant que ceux-ci s’influencent mutuellement. Cette corrélation peut être observée si les investisseurs sont indépendamment influencés par des facteurs et/ou des informations communes. Hirshleifer et Teoh (2001) font la distinction entre les deux types de comportements ; ils désignent ainsi par « mimétisme » (herd behavior) les convergences de comportements et par « cascades informationnelles » les situations où l’individu choisi son action en se basant sur l’observation des autres indépendamment de son propre signal informationnel. Ainsi, les analystes financiers vont s’engager dans des comportements mimétiques ou des cascades informationnelles dans leur choix des titres mais encore et surtout dans leur démarche de recommandation. Si le mimétisme peut s’expliquer par différentes raisons, et que si, pour beaucoup, il résulte de comportements irrationnels et irresponsables, pour d’autres, le mimétisme est tout à fait rationnel. L’étude du marché financier constitue un riche enseignement des comportements des individus au sein d’un groupe. Les agissements des uns ne laissent souvent pas indifférents les autres en quête de performances. Pour y parvenir, ces derniers calquent généralement leur position sur celle des premiers, ils les imitent. Le mimétisme est une caractéristique irréfutable des marchés financiers, même si beaucoup se défendent d’une telle attitude. Ce qui en soi est fort compréhensible, puisque leur faire remarquer qu’ils ont un comportement mimétique, reviendrait à dire que leur personnalité est insuffisamment 36 affirmée et qu’ils font l’objet de manipulations. Mais la question du mimétisme relève avant tout du domaine de la psychologie. C’est ce phénomène psychologique alternatif qui explique en partie les anomalies du marché financier. Ceci ne veut pas dire que les professionnels du marché ne saisissent pas l’évolution de celui-ci, mais tout simplement qu’ils ont des aptitudes limitées et par conséquent semblent suivre un certain modèle de comportement chaque fois qu’il leur manque une indication claire. Il est nul doute que le facteur psychologique est un paramètre d’une importance cruciale dans l’étude du marché financier notamment chez les analystes financiers. De nombreux chercheurs s’intéressent particulièrement à ce biais cognitif et à son impact sur les marchés financiers. En effet, plus l’analyste financier ou l’investisseur réussit de transactions (par chance ou par calcul), plus celui-ci aura tendance à reproduire l’opération toujours à une échelle plus élevée. Cet optimisme semblera alors justifié aux yeux de nombreux autres intervenants qui par mimétisme observerons une conduite identique. C(est dans ce cadre qu’Olsen (1996) établit une relation entre l’optimisme des analystes financiers et le phénomène de mimétisme. L’existence de biais cognitifs récurrents (surconfiance en soi, surréaction aux annonces) ou encore l’usage par les agents de modèles mentaux (Shiller, 1989, 2000) influencent les analyses réalisées et les conseils donnés par les analystes. Ce caractère d’optimisme observé dans cette population de professionnels traduit la réputation des uns et des autres et donc le pouvoir à un moment donné d’influencer le marché, c'est-à-dire d’être suivi par la masse, ne dure effectivement pas une éternité. Nos entretiens nous révèlent que les performances continues des uns et des autres sur le marché sont plutôt rare à quelques exceptions près. En bref, les relations entre les analystes financiers et autre acteurs du marché sont si étroites qu’elles engendrent des influences réciproques. A l’évidence, il convient de souligner que ces influences n’enlèvent rien au caractère rationnel des opérateurs sur le marché financier. A ce propos, les analystes financiers que nous avons interrogés se disent être « les agents les plus rationnels du marché » au sens économique. Les comportements mimétiques relèvent d’un manque (manque de confiance en soi, manque de charisme, manque d’informations, de connaissances, etc.). Un analyste financier nous affirmait à propos de la notion de rationalité mimétique : « Je crois qu’il y a du rationnel, mais ce rationnel, il est influencé par ce qui a autour. C’est pas possible de penser tout seul. Vous êtes dans du relatif et donc vous êtes influencé par ce qui se passe autour de vous ». Les analystes financiers ont un rôle non négligeable sur le marché financier. Leurs décisions ne laissent rarement le marché insensible, elles ont un impact direct sur son orientation. Très souvent, les techniques et méthodes bien que sophistiquées ne leurs permettent pas toujours d’obtenir une valorisation optimale. Dans un contexte aussi complexe que le marché financier, gravitent plusieurs paramètres parfois contradictoires. Chaque opérateur est en effet doté de capacités computationnelles différentes. Ainsi, les objectifs, les horizons, les attentes, etc. divergents selon les individus. Or, ceuxci sont amenés à prendre des décisions dans un milieu commun à tous. Comment agir en fonction de sa seule et unique croyance ? Le marché étant le reflet des opinions de tous, par conséquent son orientation dépendra de l’opinion moyenne. Aller à son encontre relève de l’irrationnel. Etant soumis à diverses pressions, ils n’ont de choix que se conformer aux croyances collectives. Ce mimétisme est justifié, puisque agir autrement entraîne un manque à gagner significatif avec des conséquences consécutives, c’est à dire des pertes de profits pour l’employeur, une rentabilité négative pour les clients mais également affaibli la notoriété de l’analyste. 37 Méthodologie Conscient de l’existence d’ambiguïtés et de difficultés dans la compréhension du sujet par les analystes financiers vue sa dimension théorique, nous avons pris soin de les rencontrer afin d’éclaircir certaines notions et d’expliciter notre démarche. Nous avons voulu dépasser la dimension théorique et apprécier de plus près leur travail. Etude de terrain Cette démarche a pour objectif d’enrichir cette recherche par un travail exploratoire de terrain et lier la dimension théorique et empirique. Pour ce faire, nous avons dans une première étape consulté la revue de littérature relative à notre thème de recherche pour mieux en cerner les contours et mieux comprendre la conduite de l’analyste financier. L’aboutissement de la revue de littérature nous a permis de nous familiariser avec les différents concepts pour ensuite construire un guide d’entretien destiné à cette population. C’est pourquoi nos entretiens ont été menés à l’aide d’une suite de questions ouvertes. L’objectif est de mieux cerner la réalité du métier d’analyste financier en France et celle du fonctionnement du marché financier. Mener des entretiens de face à face nous a semblé la meilleure solution, parce qu’une telle démarche présente de multiples avantages pour recueillir des informations de qualité. Cela permet d’orienter l’interlocuteur lorsque ce dernier présente des difficultés de compréhension, de demander l’approfondissement ou l’éclaircissement de certains points qui nous auraient paru flous, de justifier nos questions lorsque celles-ci étaient mal comprises, et d’avoir une certaine spontanéité dans les réponses du répondant. La collecte des données par entretiens exploratoires (données qualitatives) nous a amené vers une démarche inductive c'est-à-dire réfléchir aux hypothèses à partir de la revue de littérature, des retranscriptions des entretiens et de leur analyse. Pour ce faire, une série de quinze entretiens a été réalisée pour la population des analystes financiers. Ils travaillent tous sur le marché parisien des actions. Ce travail exploratoire a été une étape cruciale puisque nous avons pu mieux saisir certaines réalités dans le fonctionnement du marché financier et de mieux connaître le travail quotidien des analystes financiers et surtout de légitimer notre étude. Le questionnaire Après le recueil des données qualitatives et leur retranscription, la seconde étape de notre travail exploratoire a consisté à utiliser ces données pour construire un questionnaire (toutes les questions sont des questions fermées), les différents items utilisés ressortent des affirmations des interviewés (étape précédente). Ces questionnaires ont d’abord été testés auprès d’analystes financiers (5 exactement) que nous avons rencontrés dans le souci d’apporter des rectifications si toutefois une question, ou un item semblait flou ou incompris pour le répondant. Ces questionnaires ont été ensuite envoyés à un échantillon beaucoup plus important d’analystes financiers sell-side membres de la SFAF (au nombre de 539) afin d’obtenir des données quantitatives significatives. L’exploitation des questionnaires a été menée à l’aide du logiciel SPSS. 38 Analyse exploratoire du questionnaire Les questionnaires ont été analysés selon la démarche suivante : Figure 1 : Etapes de l’analyse exploratoire Analyse exploratoire Fiabilité cohérence Analyse factorielle Cette étude vise d’abord à établir la dimensionnalité du questionnaire. Elle cherche aussi à vérifier la cohérence interne et leur fiabilité. L’enquête a été menée auprès d’un échantillon de 88 analystes financiers sell-side exerçant dans différentes institutions, sociétés de bourse, d’investissement et banques. On compte au sein de la population ayant participé à l’enquête 48 hommes (soit environ 55%) et 40 femmes (soit environ 45%). Soixante six analystes financiers ayant répondu à l’enquête travaillent dans des sociétés de bourse (soit 75%). Vingt et un travaillent au sein d’une banque (soit environ 24%) et un seul dans une société d’investissement (soit environ 1%). Elaboration du questionnaire A la lumière de cette étude par questionnaire, nous avons produit un ensemble d’items provenant de l’étude de terrain (lorsque nous nous sommes entretenus avec les analystes financiers). Nous interrogeons les analystes à propos de leur profession de manière générale avant d’évoquer la notion de mimétisme, leur demandant leurs réactions et comportements éventuels notamment dans des contextes d’interaction ou encore d’incertitude. Chaque question est composée de plusieurs items. Les réponses sont étalées sur une échelle de Likert à 6 points. Chacun doit noter chaque item sur une de échelle de 1 à 6 (1 correspondant à ‘pas du tout d’accord’ et 6 à ‘tout à fait d’accord’). Matrice des corrélations et AFCP Le traitement des données est effectué par une analyse factorielle en composantes principales (AFCP). La fonction première de l’analyse en composantes principales en est une réduction des données. Autrement dit, c’est une approche qui vise à réduire un grand nombre d’informations sur un sujet donné à un petit nombre d’éléments plus facilement interprétables (Stafford & Bodson, 2006). Pour ce faire, nous avons donc pour chaque question étudié les interrelations entre les différentes variables, puis regrouper ces variables dans des composantes ou groupes limités et établit ensuite entre ces groupes de variables une hiérarchie basée sur la valeur explicative de chacun des facteurs. 39 Le principe de résolution va être de trouver successivement un premier facteur résumant le mieux l’information contenue dans la matrice initiale, puis un second, indépendamment du premier, résumant le mieux l’information résiduelle et ainsi de suite. On obtient ainsi un certain nombre d’axes factoriels, la somme des valeurs propres associées à ces axes est égale à la variance totale. Certaines variables sont corrélées à deux ou même à plusieurs axes avec des coefficients de corrélation différents. Pour résoudre ce problème et mieux interpréter les facteurs, nous avons effectué une rotation Varimax dans l’espace factoriel de façon à augmenter la valeur des coefficients de corrélation de certaines variables en les rapprochant de l’un des axes. L’alpha de cronbach L’alpha de cronbach est une estimation de la variance du score total de l’échelle due à tous les facteurs communs propres aux items de l’échelle testée (Igalens et Roussel, 1998). Lorsque l’alpha se rapproche de 1, l’ensemble des items à une bonne cohérence interne, à l’inverse lorsque l’alpha se rapproche de 0, la cohérence interne est faible. Pour Evrard et al. (2003), l’alpha de cronbach permet seulement de savoir si les indicateurs utilisés, censés être équivalents sont cohérents entre eux. La fiabilité est une condition nécessaire à la validité. Lorsque le facteur est composé d’un seul item, le logiciel ne calcul pas l’alpha de cronbach Résultats et discussion Pour chacun des tableaux, nous avons calculé le test de Kaiser-Meyer-Olkin qui représente la corrélation partielle entre l’ensemble des variables (items). Lorsque celui- est inférieur à 0,7, on considère qu’il est de mauvaise qualité, ce qui signifie une faible corrélation entre l’ensemble des variables. Nous avons ensuite calculé l’alpha de cronbach pour tester la cohérence entre les variables. Nous avons donc calculer l’alpha pour chaque composante obtenue après la rotation orthogonale varimax. les sources d’informations utilisées par les analystes pour établir leur jugement Les analystes financiers utilisent plusieurs sources d’information pour établir un jugement. Ces sources d’informations sont de nature différentes. On pose la question à un échantillon d’analystes, chacun doit noter chaque item sur une de échelle de Likert de 1 à 6 (1 correspondant à ‘pas du tout important’ et 6 à ‘tout à fait important’). 40 Variance en % Composantes et variables 1 Composante 1 : les informations comptables 1. Rapports de gestion 2. Bilan 3. Compte de résultats 4. Annexes comptables 5. Tableaux des flux de trésorerie L’alpha de cronbach Composante 2 : les informations continues et les réunions d’informations 1. Des banques de données (World Equities, Data Stream, Chahine, etc.) 2. Des informations en continu (Reuters, Bloomberg, AFP, etc.) 3. Les réunions organisées par la SFAF 4. Les réunions organisées par les entreprises cotées L’alpha de cronbach Composante 3 : information générale et consensus 1. La presse économique générale (Echos, Tribune, Agifi, Le Monde, etc.) 2. Le consensus des analystes financiers L’alpha de cronbach Composante 4 : Les informations sectorielles et environnementales 1. Les études des secteurs 2. Les rapports du développement durable 3. Des banques de données (World Equities, Data Stream, Chahine, etc.) L’alpha de cronbach Composante 5 : les informations directes 1. Les rencontres avec la direction de l’entreprise Composante 6 : les informations des rapports des autres analystes 1. Les études de certains de vos confrères Total Coefficients 2 0,868 0,771 0,725 0,883 0,706 Réelle* 3 Interne 4 22,2 31,3 0,867 0,523 0,809 0,801 0,642 15,6 22 0,738 0,707 9,8 13,8 0,824 0,547 0,607 0,717 8,6 12,1 0,506 0,359 0,739 8 11,1 0,878 6,4 70,9 9 100 L’indice KMO (Kaiser-Meyer-Olkin): 0,629 Le tableau ci-dessus dresse le bilan final de l’analyse en composantes principales des sources d’information utilisées par les analystes financiers pour établir leur jugement. Dans la colonne 1, nous avons les composantes et les variables, la composante 1 regroupe les variables qui sont les plus importantes pour établir un jugement ; dans la colonne 2, nous avons les coefficients placés par ordre de grandeur ; dans la colonne 3 est présentée la variance expliquée par le modèle factoriel. Ainsi la composante 1 représente 22,2% de la variance, la composante 2 représente 15,6% de la variance, la composante 3 représente 9,8%, la composante 4 représente 8,6%, la composante 5 représente 8% et enfin la composante 6 représente 6,4% de la variance. Le modèle lui-même explique à 70,9% des sources d’informations utilisées par les analystes financiers pour établir un jugement : cela signifie que 29,1% des sources d’information reste inexpliquée par les variables choisies ; dans ce cas, il aurait fallu ajouter de nouvelles variables, ce qui rend toutefois plus difficile la collecte des données sur le terrain. Dans la dernière colonne, nous avons la variance interne. L’analyse des composantes principales des sources d’informations utilisées pour établir un jugement tend à montrer que parmi la multitude des sources d’informations à la disposition des analystes financiers, ces derniers ont tendance à privilégier les informations comptables pour établir leur jugement. L’indice KMO est de 0.629. Il est inférieur à 0,7. Il est de mauvaise qualité. Les corrélations partielles entre les variables sont faibles. Le test de Bartlett sur les variables montre que le modèle factoriel est approprié ( test de Bartlett significatif). L’alpha de cronbach est bon (0,867), il indique une bonne cohérence interne entre les variables de la première composante. En effet, ces résultats entérinent les résultats de l’étude exploratoire selon lesquels, les informations comptables constituent la base, la matière première de l’analyste. Ce qui n’empêche nullement ces derniers d’utiliser d’autres sources d’informations pour affiner leur jugement, telles que les réunions organisées par la SFAF, ou encore les informations obtenues par les entreprises sans oublier que l’analyste est constamment à jour en consultant les informations diverses 41 (presse, TV, radio, Internet…). On notera cependant que les variables « études de certains de vos confrères » de la composante 5 et 6 sont des sources nécessaires mais non déterminantes pour établir un jugement. La rationalité de l’analyste On pose la question à un échantillon d’analystes à propos de sa rationalité lors de ses recommandations. Chacun doit noter chaque item sur une de échelle de Likert de 1 à 6 (1 correspondant à ‘pas du tout d’accord’ et 6 à ‘tout à fait d’accord’). Variance en % Composantes et variables 1 Composante 1 : informations collectives 1. Lorsqu’il bénéficie de relations dans le milieu (réseau social) 2. Lorsqu’il se fie aux informations obtenues par les collègues du même secteur 3. Lorsqu’il consulte le consensus 4. Lorsqu’il consulte les notes diffusées par les analystes des autres brokers L’alpha de cronbach Composante 2 : informations individuelles 1. Lorsqu’il détient des informations personnelles 2. Lorsqu’il privilégie sa seule capacité d’analyse 3. Lorsqu’il se fie à son intuition L’alpha de cronbach Composante 3 : informations fondamentales 1. Lorsqu’il utilise des outils et méthodes dans ses valorisations 2. Lorsqu’il détient les informations publiées par l’entreprise L’alpha de cronbach Total Coefficients 2 Réelle* 3 Interne 4 26,037 42,199 0,608 0,793 0,606 0,859 0,715 0,790 0,586 0.715 23,899 38,734 0,574 0,582 0,862 11,764 19,066 0,609 61,700 100 L’indice KMO (Kaiser-Meyer-Olkin): 0,603 L’analyse des composantes principales de la rationalité des analystes financiers lors de leurs recommandations montre une approche de la définition de la rationalité mitigé, puisque au regard du tableau ci-dessus, la rationalité prend une signification différente selon les analystes interrogés. En effet, la première composante qui représente 26,037% de la variance signifie que l’analyste financier est rationnel lors de ses recommandations lorsqu’il détient ou bénéficie des informations collectives (réseau, collègues, consensus), avec un alpha de 0,715, donc une bonne cohérence de la mesure. A l’inverse, la seconde composante montre que l’analyste est rationnel lorsqu’il détient ou privilégie des informations personnelles (intuition, capacité d’analyse personnelle). Cette composante représente 23,899% de la variance, avec un alpha faible (0,574). La troisième composante relève que l’analyste est rationnel lorsqu’il utilise ou privilégie les informations fondamentales, l’alpha est acceptable (0,609). Ces trois composantes décrivent une rationalité s’expliquant de diverses façons. Ces résultats corroborent ceux de l’étude exploratoire selon lesquels, les analystes se disent les acteurs les plus rationnels du marché, et donc à chacun sa façon d’être rationnel. Ce qui semble caractériser le fonctionnement du marché financier et les attitudes des analystes financiers. Ici, l’analyste semble ne pas rester indifférent aux comportements des autres acteurs sur le marché. Selon les analystes financiers, être rationnel, c’est intégrer les positions des autres pour optimiser leurs chances de rentabilité et de survie. 42 Influences dans la modification des recommandations de l’analyste Variance en % Composantes et variables 1 Composante 1 : prise en compte du consensus et de l’environnement relationnel 1. Un changement d’attitude de la part de vos confrères (consensus) 2. Une directive émanant de votre supériorité hiérarchique 3. Une crainte de compromettre vos relations avec les dirigeants d’entreprise 4. Une crainte de compromettre vos relations avec vos clients L’alpha de cronbach Composante 2 : prise en compte des convictions personnelles et des rumeurs 1. Une rumeur sur le marché financier 2. Une information nouvelle 3. Une conviction personnelle de dernière minute L’alpha de cronbach Composante 3 : prise en compte du marché 1. Une crainte de sanction du marché Total Coefficients 2 0,744 0,912 Réelle* 3 Interne 4 42,764 55,897 0,937 0,948 0,917 0,726 0,652 0,863 20,600 26,926 0,626 0,931 13,140 76,505 17,175 100 L’indice KMO (Kaiser-Meyer-Olkin): 0,720 L’analyse des composantes principales des éléments pouvant conduire l’analyste à modifier sa recommandation au moment de son émission tend à montrer que ce dernier semble sensible au consensus, aux directives de sa hiérarchie, aux relations entretenues avec les entreprises et celles entretenues avec les clients. Ceci conforte nos analyses précédentes selon lesquelles, l’analyste doit tenir compte des différents acteurs du marché et surtout ceux qui y ont un certain intérêt. Les intérêts en jeu de chacun étant souvent élevés, l’analyste doit prendre soin de ne léser personne avant de faire valoir son opinion finale. L’alpha de cronbach (0,917) indique une bonne cohérence interne entre les variables. L’analyste prend en compte non seulement les réactions du marché (sanctions), mais aussi ses convictions personnelles et les toutes nouvelles informations pouvant influencer le marché. L’analyste a parfaitement conscience qu’il fait partie d’un ensemble et que par conséquent il ne peut y avoir d’emprise individuelle sur le marché qui obéit à une majorité d’opinion. D’ou l’intérêt de connaître leur position au préalable avant un quelconque engagement. Le comportement mimétique de l’analyste financier Variance en % Composantes et variables 1 Composante 1 : contraintes informationnelles et environnementales 1. Un manque d’informations précises à propos d’une valeur, d’une entreprise, d’un secteur, d’un marché, etc. 2. Un manque d’expérience 3. Une crainte de sanctions du marché si vous allez à son encontre 4. Une pression exercée par votre employeur en quête de résultats 5. Une pression exercée par le (s) vendeur (s) 6. Une pression exercée par votre employeur en quête de résultats L’alpha de cronbach Composante 2 : contexte euphorique 1. L’opportunisme (réaliser un gain sur le court terme) 2. Un effet de mode (situation euphorique selon le contexte) 3. Une attention particulièrement portée sur les tendances de l’analyse technique L’alpha de cronbach Composante 3 : manque de recul et d’analyse 1. Un manque de recul et d’analyse face à une situation inédite Total Coefficients 2 Réelle* 3 Interne 4 34,943 55,519 0,6 15 0,533 0,691 0,748 0,897 0,890 0,846 0,751 0,758 18,648 29,628 0,553 0,589 0,794 L’indice KMO (Kaiser-Meyer-Olkin): 0,750 43 9,348 62,939 14,852 100 L’analyse des composantes principales des causes qui pourraient conduire l’analyste à adopter un comportement mimétique montre que la contrainte informationnelle et environnementale explique le mimétisme des analystes. L’analyste adopte un comportement mimétique lors d’un déficit informationnel, à l’image du mimétisme informationnel d’Orléan (1999) ou encore par un manque d’expérience. Cette composante explique 34,943% de la variance. L’alpha est satisfaisant (0,846), c’est à dire une bonne cohérence interne entre les variables de la composante. L’indice KMO de 0,750 indique une corrélation partielle entre les variables. De plus le test de Bartlett est significatif. La mutation de l’environnement de l’analyse financière bouleverse la fonction même d’analyste financier. Celui-ci, doit désormais faire face et rendre compte à des acteurs différents du marché financier. Les enjeux sont tels que ces derniers restent très sensibles aux résultats de son analyse, ce qui le contraint à l’adoption d’un comportement mimétique. C’est ce que relève les quatre derniers facteurs de la première composante : « une crainte de sanctions du marché si vous allez à son encontre », « une pression exercée par votre employeur en quête de résultats », « une pression exercée par le (s) vendeur (s) », « une pression exercée par votre employeur en quête de résultats) ». Ces résultats corroborent ceux de l’étude exploratoire et ceux de l’ACP de la question précédente. La seconde composante montre des analystes adoptant un comportement mimétique par opportunisme ou parce qu’il existe un effet de mode. Ici, l’analyste prête davantage attention à son environnement plutôt qu’à ses fondamentaux. Il suit la tendance du marché financier. Ce genre de comportement s’explique par une euphorie existante à un moment donné et qui influence les positions par l’appât du gain. Le comportement mimétique s’explique dans la troisième composante par un manque de recul face à une situation inédite. Elle explique 9,348% de la variance. Elle s’avère donc nécessaire mais pas déterminante dans le comportement mimétique. l’interaction humaine et la prise de décision de l’analyste financier Variance en % Composantes et variables 1 Composante 1 : apport informationnel 1. Mieux apprécier et connaître une valeur, une entreprise, un secteur, etc. 2. Apporter des solutions, des justifications aux questions que vous vous posez 3. Renforcer les convictions vis-à-vis d’un secteur ou d’une valeur 4. Faire émerger une idée cruciale qui vous paraissait secondaire L’alpha de cronbach Composante 2 : formation de l’opinionn 1. Corriger les prévisions 2. Se forger une opinion L’alpha de cronbach Composante 3 : connaissance 1. Connaître la position des confrères sur les valeurs suivies Composante 4 : compréhension 1. Mieux comprendre les orientations du marché à un moment donné Total Coefficients 2 Réelle* 3 Interne 4 26,658 41,582 0,687 0,631 0,679 0,656 0,615 0,798 0,809 14,650 22,852 0,747 0,881 12,618 19,682 0,961 10,183 64,108 15,884 100 L’indice KMO (Kaiser-Meyer-Olkin): 0,608 L’interaction humaine sur le marché financier joue un rôle non négligeable dans la prise de décision de l’analyste financier. L’A.C.P montre que l’interaction permet à l’analyste de dissiper ses incertitudes. Cela lui permet entre autre de « mieux comprendre et connaître une valeur, une entreprise, un secteur, etc. », « renforcer les convictions vis-à-vis d’un secteur ou d’une valeur » ou encore « faire émerger une idée cruciale qui paraissait secondaire ». Cette composante explique 26,658% de la variance. L’alpha (0,615) indique un cohérence entre les variables de la composante fiable et un indice KMO de 0,608. Toutefois le test de Bartlett est significatif. Les compétences hétérogènes impliquent une influence sur chacun des acteurs du marché financier et une certaine valeur ajoutée. 44 La seconde composante composée de trois facteurs relève les effets bénéfiques de l’interaction. Avec les interférences, ils sont ainsi en mesure de « corriger leurs prévisions », « se forger une opinion » ou encore « donner une meilleure assurance lors des recommandations » et donc pouvoir améliorer leurs performances sur le marché financier. L’attitude adoptée par l’analyste financier lorsque le contexte boursier est incertain Variance en % Composantes et variables 1 Composante 1 : influence des autres analystes et du consensus 1. Vous prenez vos décisions de recommandation allant dans le sens de celles des autres analystes lorsque vous ne possédez pas d’informations 2. Vous vous alignez sur le consensus 3. Vous prenez une décision allant dans le sens de l’analyste employé par un « gros » broker de renom 4. Vous prenez une décision allant dans le sens de l’analyste leader d’opinion L’alpha de cronbach Composante 2 : mimétisme opportuniste 1. Vous renforcez votre information propre observant le comportement et l’action des autres analystes 2. Vous adoptez un comportement de suiveur lors de situations euphoriques L’alpha de cronbach Composante 3 : aucune influence : action individuelle 1. Vous agissez selon votre seule conviction sans tenir compte des opérations des autres intervenants Total Coefficients 2 Réelle* 3 Interne 4 37,745 52,348 0,657 0,834 0,847 0,845 0,804 0,727 19,324 26,80 0,969 0,500 0,960 15,034 20,850 72,103 100 L’indice KMO (Kaiser-Meyer-Olkin): 0,720 Cette question rend compte d’une situation extrêmement délicate pour l’analyste. Il s’agit de connaître son comportement dans un contexte boursier incertain et lorsque les avis des confrères divergent. Les résultats montrent que l’analyste n’agit pas avec ses seules convictions sans tenir compte des autres intervenants, mais qu’au contraire, dans un environnement opaque, il est fortement influencé par le comportement des autres analystes. sa décision va donc dans le sens de la majorité des autres analystes. De tels résultats confirment l’étude exploratoire et la notion de mimétisme informationnelle d’Orléan. En effet, lorsque l’analyste financier ne possède pas d’informations, il préfère suivre le comportement des autres analystes qu’il pense mieux informés. Une telle attitude est en soi rationnelle, la décision de suivre des acteurs supposés mieux informés n’enlève rien à la situation de départ de l’analyste suiveur. Cette composante représente 37,745% de la variance et un alpha de 0,804. La seconde composante montre un analyste adoptant un comportement opportuniste, c'est-à-dire un analyste opérant selon le contexte de l’environnement. Cette composante représente 19,324% de la variance, avec une faible cohérence interne (alpha de 0,5) Enfin la troisième composante souligne que dans un contexte boursier incertain l’analyste ne subit aucune influence et préfère agir selon sa seule conviction sans tenir compte des opérations des autres intervenants. La composante explique que 15% de la variance. 45 La signification de la rationalité mimétique pour les analystes financiers Variance en % Composantes et variables 1 Composante 1 : rationalité mimétique 1. Prendre des décisions de recommandation allant dans le sens de celles des autres analystes lorsque vous ne possédez pas d’informations claires 2. Délaisser sa propre information pour adopter celle des autres analystes supposés mieux informés 3. Aller dans le même sens que la majorité censée représenter l’orientation du marché 4. Adopter les décisions de recommandation d’un autre analyste lorsque vos connaissances sont limitées L’alpha de cronbach Composante 2 : rationalité normative 1. Adopter les décisions de l’analyste classé supposé plus performant 2. Adopter les décisions de l’analyste employée par un broker de renom L’alpha de cronbach Total Coefficients 2 Réelle* 3 Interne 4 35,611 64,391 0,779 0,657 0,718 0,814 0,759 0,831 0,880 19,694 35,610 0,780 55,304 100 L’indice KMO (Kaiser-Meyer-Olkin): 0,642 Les analystes prêtent une attention particulière à l’entourage et ce type d’attitude est pour eux une attitude rationnelle. En effet, la première composante explique 35, 611% de la variance. L’alpha (0,759) montre une certaine cohérence entre les variables de la composante. L’indice KMO se révèle être faible, mais le test de Bartlett est significatif. Il s’agit de la rationalité mimétique telle que nous l’avons définie. L’analyste est prêt à abandonner ses propres informations et convictions, lorsque celui-ci pense que les autres possèdent de meilleurs éléments desquels il pourrait tirer profit. La seconde composante explique 19,694% de la variance semble aller dans le même sens que la première, à la différence qu’ici l’analyste semble influencé par la réputation des autres analystes. La notoriété des analystes leaders d’opinions laisse rarement indifférents les autres en quête justement de cette reconnaissance. Conclusion Cette étude nous a permis de mieux cerner le rôle de la rationalité mimétique dans la formation des recommandations des analystes financiers. Elle montre que le processus décisionnel ne s’exerce pas dans un lieu isolé mais plutôt dans un lieu où prédomine une multitude d’agents aux exigences hétéroclites. De ce fait, les opinions des uns influencent et/ou peuvent être influencées par celles des autres d’où des comportements mimétiques. L’information constitue la matière première dans le travail des analystes financiers. Celle-ci étant abondante et multiple, la tâche essentielle est sa sélection puis son traitement pour prodiguer les conseils et décisions en face avec l’évolution du marché boursier. Dans cette tâche de collecte et de sélection, certains analystes privilégieront les contacts directs avec les dirigeants d’entreprise cotées, ou les réunions SFAF pour une information alors que d’autres calqueront leur opinion sur le consensus, c’est à dire l’opinion majoritaire qui prédomine à un moment donné. La versatilité du marché conduit à des comportements mimétiques rationnels, les risques encourus pouvant être fatals pour la carrière de l’analyste. C’est finalement avec l’interdépendance des opérateurs sur le marché financier, que la rationalité économique dégénère en rationalité mimétique 46 Références 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. Abou A., Sassenou N., (2001). “La prime de risque anticipée sur le marché des actions : mesure et analyse sur le panel d’experts et de titres du SBF 250”, document de travail Caisse des dépôts et consignations. Artus P., (1995). Anomalies sur les Marchés financiers Paris Editions Economica. Banerjee A., (1992), “A simple Model of Herd Behavior,” Quartely Journal of Economics, 107, 797-818. Bayle E. et M. Schwartz, (2005). “A quoi servent les analystes financiers,” Revue d’économie financière, 81, , 211-235. Bikhchandani S., Hirshleifer D., Welch I., (1992). “A Theory of Fads, Fashions, Customs and Cultural Change as Informational Cascades,” Journal of Political Economy, 100, 992-1096. Bikhchandani S., Sharma S., « Herd Behavior in Financial Markets : A review. » IMF Working Paper WP/00/48. www.imf.org/external/pubs/ft/wp/2000/wp0048.pdf De Bondt W., R. 7 Thaler, (1989). “Anomalies: A mean-reverting walk down Wall Street,” Journal of Economic Perspectives, 3, , 189-202. Dupuy J P., Dumouchel P., (1979). L’enfer des choses, René Girard et la logique de l’économie, Paris, Seuil. Evrard Y., Pras B., Roux E., (2003), “Market, études et recherches en marketing”, Dunod Grinblatt, M., Titman S., Wermers R., (1995). “Momentum Investment Strategies, Portfolio Performance, and Herding: A study of Mutual Fund Behavior,” American Economic Review, 85(5), 1088-1105. Hamon J., Préface in Bruno-Laurent Moschetto, (1998). Mimétisme et Marchés financiers, Paris Economica. Hirshleifer D., Teoh S.H, (2001). “Herd Behavior and Cascading in Capital Markets : A Review and Synthesis,” Dice Center Working Paper, No. 2001-20. Igalens J., Roussel P. (1998), “Méthodes de Recherche en Gestion des Ressources Humaines”, Economica Keynes J.M., (1936). Théorie Générale de l’emploie, de l’intérêt et de la monnaie, Paris, Payot. Mandelbrot B., (2005).Une approche fractale des marchés, risquer, perdre et gagner, Paris, Odile Jacob. Maug E., Naik N., (1996). “Herding and delegated portfolio management: The impact of relative performance evaluation on asset allocation,” London Business School. Nofsinger J.R., Sias R.W., (1999). “Herding and feedback trading by institutional and individual investors,” Journal of Finance, 54, 2263-2295. Olsen R.A., (1996). “Implications of Herding Behavior for earnings Estimation, Risk Assessment, and Stock returns,” Financial Analysts Journal, 37-41. Park C.W., Pincus M., (2000). “Market Reactions to Changes in Analyst Consensus Recommendations Following Quarterly Earnings Announcements,” Papier de recherche, Hong Kong University of Science and Technology, University of Iowa. Pierret J.P., (1986). “L’analyste financier et l’accès à l’information”, Analyse Financière, 1215. Samuelson P.A, (1974). “Challenge to judgment,” Journal of Portfolio Management, 1, 1719. Shiller R.J, (1989). “Market Volatility,” Cambridge : MIT Press. Shiller R.J., (2000). Irrational Exuberance, Princeton University Press. Stafford J., Bodson P. (2006), “L’analyse multivarieé avec SPSS”, Presse de l’Université du Québec Stickel S.E., (1991). “Common Stock Returns surrounding Earnings Forecast Revisions : More Puzzling Evidence” The Accounting Review, 66, 402-416. Van Loye G., L. Fontowicz, (2004). “La crédibilité des producteurs d’information le cas des analystes financiers,” Revue Française de Gestion, 30(151), 81-95. 47 27. Welch I., (2000). “Herding among security analysts,” Journal of Financial Economics, 58, 369-396. 48 Nilanjan Basu ASAC 2008 Haibo Jiang (student) Halifax, Nova Scotia Parianen Veeren (student) Neiliane Williams (student) John Molson School of Business Concordia University REPURCHASES AND POST-SEO UNDERPERFORMANCE We investigate firms that issue stock and then buy them back. Self-tender offers that follow SEOs appear to signal undervaluation whereas open market repurchases do not. The strength of this signal is related to the underperformance that follows the SEO. The results suggest that selftender offers can be used by issuing firms to reverse the post-issue underperformance of stock. Introduction Why do firms repurchase stock? The obvious reason is to return cash to the shareholders. As such, repurchases are potentially superior to dividends since they may result in a lower tax for shareholders. However, as noted in BusinessWeek1, many firms repurchase their shares around the same time that they issue equity. Billett and Xue (2007) provide one possible explanation for this apparently paradoxical action. They show that companies planning an equity issuance can repurchase part of their shares prior to the issuance in order to signal higher firm value. This allows them to obtain a more favorable pricing for their equity issue. This potential explanation, however, leaves the repurchase announcements made soon after the completion of an equity issue unaddressed. In this paper we examine the related scenario where a firm buys back some of the shares that were previously issued in a seasoned equity offering (SEO). Prior research2 informs us that SEOs are often associated with stock price underperformance. We find that self-tender repurchases provide one way 1 2 http://www.businessweek.com/magazine/content/06_04/b3968099.htm See for example Loughran and Ritter (1995). 49 that a firm can, at least partially, reverse such price declines. The abnormal returns earned at the announcement of such repurchases are positive and higher than those earned by repurchases that are not preceded by share issuance. On the other hand, we find that in the case of open market repurchases, the abnormal return earned upon the announcement of the repurchase is lower if it is preceded by an SEO. We hypothesize that self-tender offers represent a commitment on part of the repurchasing firm and are thus a credible signal of undervaluation. In contrast, open market repurchases are not commitments and could conceivably be perceived by the market as further attempts to time the market. In order to explore the link between the SEO and the repurchase in more detail, we test the relationship between the underperformance following the SEO and the abnormal returns earned at the time of the repurchase. We find that, at the time of the repurchase, the extent of preceding underperformance is a significant predictor of the abnormal returns. On the other hand, firms that conduct repurchases that do not have a foregoing SEO earn smaller returns at the time of the repurchase. Moreover, at the time of the repurchase, the corresponding abnormal returns are not significantly related to prior stock performance. Overall, our results indicate that repurchases which follow SEOs could be driven by the desire of issuing firms to reverse the underperformance initiated by the SEOs. This does not imply that the returns earned by repurchasing firms, in general, can be explained exclusively by the behaviour of this sub-sample of firms that issued equity prior to repurchase. After all, only a small fraction of repurchasing firms have preceding SEOs. Nevertheless, these firms may contribute disproportionately to the overall positive returns that are associated with the announcement of repurchases. The remainder of our paper is organized as follows. The next section provides a brief summary of prior work in this area. The remaining sections review the characteristics of our sample, describe our methodology, and report our results and conclusions. Prior work and hypotheses The motivation for stock buybacks has been studied extensively in earlier research. Although the direct impact of repurchases is to return cash to shareholders, there are other potential reasons why a firm would repurchase stock. Dittmar (2000), for example, notes many such possibilities. They include, in addition to returning excess cash to shareholders, the signaling of undervaluation, altering the leverage ratio, fending off takeovers, and countering the dilution effects of stock options as alternative explanations. However, as stated by Louis and White (2007), “the most commonly cited explanation for repurchases in the academic literature is signaling”. In keeping with the popularity of this hypothesis, several researchers, such as Dann (1981), Vermaelen (1981), Comment and Jarrell (1991), and Lie and McConnell (1998) have found evidence consistent with the signaling hypothesis. Similar to repurchases, equity issues have also been extensively studied by researchers. One of the more controversial findings with respect to equity issuance pertains to the post-issue underperformance associated with both initial public offerings (IPOs) and SEOs. In particular, Spiess and Affleck-Graves (1995) and Loughran and Ritter (1995) find that firms which undertake SEOs experience a prolonged period of underperformance. However, as pointed out in Brav, Geczy, and Gompers (2000), the underperformance appears to be concentrated in small firms with low book-to-market ratios. We aim to enhance our understanding of these two critical issues (i.e. SEOs and repurchases) by examining a sample of firms that issue shares and then buy them back. The opposite effects of the two actions on capital structure and cash would tend to rule out the possibility that both actions were taken 50 solely in order to adjust leverage or pay back excess cash to shareholders, thus, enabling us to focus on the signaling effects of repurchase. Billett and Xue (2007) report in this context that repurchases could be an effective way for firms to signal value and therefore, raise their prices before an SEO. Here we address the complementary problem that confronts the issuing firm in the form of a post-issuance underperformance. We hypothesize that a self-tender offer could be executed in order to counteract the decline in prices that followed the SEO. As a result, we are able to arrive at the following testable hypotheses. Hypothesis 1: To the extent that self-tender offers are positive signals, we would expect them to have a higher impact in case of firms that have a greater need to credibly signal their value. As reported by Loughran and Ritter (1995) and Spiess and Affleck-Graves (1995), firms that issue seasoned equity tend to underperform during the post-SEO period. In particular, firms that issue equity for reasons other than market timing will feel the need to separate themselves from firms that do time the market for their SEOs. One way for them to do so would be via a self-tender repurchase. As a result, we would expect that: Firms that issue stock and then repurchase their shares through a self-tender offer will experience higher returns at the time of the repurchase announcement than comparable firms that do not issue stock prior to the announcement of their self-tender offer. Hypothesis 2: Although, as outlined by McNally (1999), open market repurchases could serve as signals, they may also be related to opportunistic stock buybacks as explained by Ikenberry and Vermaelen (1996). Likewise, Grullon and Michaely (2004), find no evidence of improved operating performance following an open market repurchase. Further, if such open market repurchase announcements are preceded by an SEO, they could be interpreted by investors as an effort by management to time the market. In contrast, open market repurchases that are not preceded by an SEO are relatively more likely to be a positive signal. Therefore we expect that: Firms that issue stock and then announce open market repurchases will experience lower announcement returns at the time of the repurchase than comparable firms that do not issue stock but announce open market repurchases. Hypothesis 3: To the extent that the self-tender repurchase is triggered by the underperformance that follows the SEO, we expect that a larger underperformance would result in a greater recovery of the stock price upon announcement of the self-tender offer. Underperformance following the SEO is a significant predictor of the announcement returns at the time of the self-tender offer. Data We start by identifying all self-tender and open market repurchases conducted by U.S. firms between 1989 and 2007. This sample of repurchasing firms is obtained from SDC Platinum. We remove all financial and utility firms from this sample and then match the remaining firms with a corresponding sample of all SEOs that are available on SDC Platinum between the years 1989 and 2007. We require that the stock issuance take place no more than four years prior to the announcement of the repurchase. Our primary sample consists of 417 buyback announcements of which 25 are self-tender offers and the remaining 392 are open market repurchases. Each of the 417 announcements is preceded by an SEO. 51 For each firm in our sample, we then identify a firm of comparable size from the same 2-digit SIC industry. We require that the matching firm announced the same type of stock buyback (open market or self-tender) 30 days before or after the repurchase announcement made by the sample firm. Further, we require that there is no issue of shares during the four years preceding the repurchase announcement made by the matching firm. The details of the sample selection process are outlined in Table 1. Table 1: Sample selection Panel A: Selection of sample firms and number of observations Stock repurchases of U.S. listed firms from SDC Platinum; Transaction types: Open Market and Self Tender; Period: Jan. 1989 to Oct. 2007 4984 Seasoned Equity Offers (SEO) of U.S. Listed firms from the SDC Platinum; Period: Jan. 1989 to Oct. 2007 6176 SEOs after excluding Financial and Utility firms 5787 Repurchases with available accounting data at Repurchase year 4233 Repurchase firms with a preceding SEO within 4 years 702 Repurchase firms with a preceding SEO within 4 years 702 Only self-tender repurchases 30 Eliminate 1 duplicate observation and 2 multiple SEOs (by keeping the latest SEO) 27 SEO followed by self-tender repurchases; available historical stock prices from CRSP 25 Repurchase firms with a preceding SEO within 4 years 702 Only open market repurchase 672 After eliminating duplicate observations and multiple SEOs (by keeping the latest SEO) SEO followed by open market repurchases; available historical stock prices from CRSP 52 423 392 Panel B: Selection of matched repurchase-only firms and number of observations Self-tender repurchase firms without preceding SEO 296 Self-tender repurchase firms without preceding SEO; announcement date is within 30 days of sample repurchase announcement 73 Industry matched firms (based on 2-digit SIC codes. If a 2-digit match is not possible we use 1-digit) 56 Closest match based on total assets 25 Open market repurchase firms without preceding SEO 3137 Open market repurchase firms without preceding SEO; announcement date is within 30 days of sample repurchase announcement 1850 Industry matched firms (based on 2-digit SIC codes. If a 2-digit match is not possible we use 1-digit) 1425 Closest match based on total assets 392 We calculate the cumulative abnormal returns (CAR) for our sample and matched firms using data from the Center for Research in Security Prices (CRSP) and the Fama-French factors from the website of Kenneth R. French3. For the calculation of the CARs we use the Fama-French four-factor model. However, the results are robust to the use of the Fama-French three factor model or the market model. Supplementary data on firm characteristics are collected from Compustat. We describe our sample in Table 2 below. We depict the calendar time clustering of our sample in panel A. Our sample does not appear to have any significant clustering in calendar time, either in terms of issuance or in terms of the repurchase announcement. Panel B describes the characteristics of the firms. The value of total assets reported by the median firm conducting self-tender offers in our sample is $356 million while the median firm conducting open market repurchases has assets worth $270 million. The average self-tender offer takes place about two and a half years after the SEO while the average open market repurchase is announced in about two years from the SEO. The firms in both groups have moderate leverage with a median debt to total assets ratio of about 40% and 20% for firms announcing self-tender and open market repurchases respectively. Overall, there are no indications that our sample selection is biased towards a certain time period or towards specific firm characteristics. 3 We thank Kenneth R. French for making this data publicly available via his website at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ 53 Table 2: Descriptive statistics Panel A: Distribution of the sample across years Sample characteristics are reported in this table. Panel A provides the clustering of our sample in calendar time. Panel B reports mean firm characteristics with medians in parentheses. Year Number of selftender offer announcements Number of open market repurchase announcements SEOs linked to subsequent selftender offer SEOs linked to subsequent open market repurchases 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Total 0 0 0 0 2 2 0 2 3 3 2 2 1 3 0 4 1 25 0 0 0 2 26 18 36 46 61 43 36 28 30 8 20 20 18 392 0 2 1 1 1 2 6 0 1 5 0 1 0 3 1 1 0 25 3 3 15 32 32 52 49 39 30 47 26 15 14 15 13 7 0 392 Panel B: Firm characteristics SEO followed by self-tender offer SEO followed by open market repurchase Size 1245.1724 (356.4120) 2868.7756 (270.5755) Market to book ratio 1.6746 (1.1732) 2.4073 (1.8365) Days between SEO and repurchase 908 (874) 750 (699) Debt to total assets ratio 0.2640 (0.4243) 0.1904 (0.2033) 25 392 Number of observations 54 Results Our primary analysis utilizes the standard event study methodology. We define abnormal returns as actual returns earned by a stock minus the expected returns for the stock. Expected returns are predicted using a Fama-French four factor model4. The stock and factor return data for the estimation of the factor betas are collected over an estimation window that stretches from 302 trading days before the event to 50 trading days before the event. The univariate tests for the level of abnormal returns are reported in Table 3. Consistent with prior research, we find that announcements of repurchases in general result in positive abnormal returns. Once again, and consistent with prior research, we find that the returns are usually higher for self-tender offers than they are for open market repurchases. Table 3: Abnormal returns earned upon announcement of repurchase We measure abnormal returns using the Fama-French four factor model. The factor betas are estimated over an estimation window of (-302, -50) and the abnormal returns earned at the time of the event over different event windows are tabulated below. The first column reports the event windows. The third column (Sample firms) refers to our sample of firms that issued stock and then bought them back. The second column (Matched non-SEO firms) refers to the matched firms that repurchased shares without a preceding SEO. The last column reports the differences in the two samples. The mean values are reported with median values in parentheses. Significance is tested using a paired t-test for means and a paired Wilcoxon test for medians. Significance at the 10% level is marked with *, at the 5% level with **, and at the 1% level with ***. Panel A: Dutch-Auction and Fixed Price Self-Tender Repurchase Matched non-SEO firms Sample firms Difference (-5,+5) 0.028976 (0.025527) 0.081373 (0.068837)** -0.05240 (-0.03103) (-2,+2) 0.026486 (0.014349)* 0.118010 (0.090781)*** -0.09152 (-0.04238)** (-1,+1) 0.045199 (0.036732 )*** 0.111570 (0.073928)*** -0.06637 (-0.02811)** (+0,+1) 0.038519 (0.015234)** 0.107404 (0.068284)*** -0.06888 (-0.02722)** 4 Fama and French (1993). Our results are qualitatively unchanged when we use the standard market model or the four-factor model that includes the momentum factor. 55 Panel B: Open market repurchases Matched non-SEO firms Sample firms Difference (-5,+5) 0.019915 (0.010125)** 0.00235 (-0.00008) 0.017764 (0.011068) (-2,+2) 0.038592 (0.023237)*** 0.012735 (0.015524)* 0.025857 (0.009663)** (-1,+1) 0.047694 (0.025902)*** 0.018113 (0.015494)*** 0.029581 (0.012944)*** (+0,+1) 0.047300 (0.022700)*** 0.026564 (0.016748)*** 0.020735 (0.006157)** In order to compare our sample firms with an appropriate benchmark, we match each firm in our sample to a corresponding firm that announced a stock buyback but did not issue equity in the previous four years. From the potential matching firms, we selected the firm that is in the same industry and from the potential matches in the same industry we pick the one with the closest size to the sample firm as measured by total assets. In order to further ensure comparability, we match sample firms that make selftender offers to other firms that make self-tender offers and sample firms that announce open market repurchases to other firms that announce open market repurchases. In Panel A of Table 3, we compare the abnormal returns earned by two groups of firms that announced a self-tender offer. The first group (labeled “Matched non-SEO firms”) consists of firms that repurchased shares in a self-tender offer without a preceding SEO. In comparison to these firms, our sample of firms that issue equity and then buy them back via self-tender offers show abnormal returns that are 5-9% higher. In contrast, firms that announce open market repurchases following an SEO exhibit lower abnormal returns than comparable firms that announce open market repurchases without a preceding SEO. Overall, the results in Panel A are consistent with the hypothesis that self-tender offers are signals and that they are at least partially effective at reversing the post-issue underperformance of SEOs. In Panel B of Table 3 we provide a similar comparison for open market repurchases. Unlike selftender offers, open market repurchases that are preceded by an SEO actually earn lesser abnormal returns than do open market repurchases that are not preceded by an SEO. Clearly, if the open market repurchases were meant to signal value and thus partially offset the post issue underperformance, they are not as good a choice as self-tender offers. Alternatively, it is possible that open market repurchases along with equity issuances are part of a larger effort to time the market. In either case, the self-tender offers appear to be the superior choice for a firm that wishes to signal undervaluation and offset its post-issue underperformance. In order to better address the link between the self-tender offer and the SEO we examine the post issue underperformance of our sample of firms compared to the benchmark of a sample of firms matched by industry, size and the market-to-book ratio. The results are reported in Table 4. Underperformance is 56 measured by the ratio of the final wealth, calculated via an equally weighted buy and hold strategy of the sample firms (or matched firms that bought back their shares in a self-tender but did not issue shares) to the final wealth obtained from the benchmark firms derived from a similar equally-weighted buy and hold strategy. Table 4: Post-issue underperformance Sample firms that issue and then self-tender are matched on the basis of industry, size and market-to-book ratio to a firm not in the sample. Underperformance for each of the 25 pairs of firms is defined as the difference in their performance measured as a wealth relative. Panel A measures the underperformance from the day after the SEO up to 5 trading days before the announcement of the self-tender offer. Panel B measures the underperformance from 5 trading days before the repurchase through to one year before the repurchase. Significance at the 10% level is marked with *, at the 5% level with **, and at the 1% level with ***. Panel A: The day after the SEO to 5 trading days before the repurchase Sample firms Matched non-SEO firms Difference Annualized -0.20931 (-0.25037)** 0.191455 (0.029067) -0.40076 (-0.19340)** Cumulative -0.28302 (-0.40114)** 1.199445 (0.072195) -1.48246 (-0.35001) Panel B: One year till 5 trading days before the repurchase Sample firms Matched non-SEO firms Difference Annualized -0.26016 (-0.28599)** 0.080357 (0.038274) -0.34051 (-0.25584)** Cumulative -0.26175 (-0.28467)*** 0.077114 (0.038167) -0.33886 (-0.25455)** As shown in Table 4, firms that issue stock and subsequently repurchase their shares through a self-tender offer have fairly large annualized post issue underperformance of over 20%. In contrast, the matched firms that bought back their shares without having previously issued stock in an SEO typically outperformed their benchmark. In order to address the sensitivity of our results, we redo the analysis using alternate performance benchmarks. Instead of using a sample of matched firms we also try using unadjusted returns and market adjusted returns. Our results still hold after considering these alternate specifications. 57 If our conjecture is true and the post-issue underperformance is, in fact, an important determinant of the returns earned by the self-tender offer, we should expect to see a cross-sectional relationship between the extent of the underperformance and the abnormal returns at the time of the tender offer. Table 5: Underperformance as a predictor of self-tender abnormal returns We predict the abnormal return associated with the self-tender offer as a function of the prior underperformance following the SEO. For sample firms, underperformance is measured from the day after the SEO until 5 trading days before the announcement of the repurchase. For the matched firms, we define the underperformance based on the same calendar time period as the corresponding sample firm. The dependent variable in each case is the abnormal return earned over the (-2, +2) event window surrounding the self-tender announcement. Logsize refers to the log of total assets. Loggap refers to the log of the number of days between the SEO and the self-tender offer. M/Bassets refers to the market value of assets divided by the book value of assets. Shares% refers to the percent of shares that are to be repurchased. R&D refers to the R&D expenditure scaled by sales. Underperform refers to the annual postissue underperformance. SEO_ST is a dummy variable that takes on a value of 1 if the firm is part of our sample of firms that issued stock and then bought them back. It takes on a value of 0 if the firm is part of the matched sample that repurchased stock but did not have an SEO preceding the repurchase. SEO_ST*Underperform is the product of Underperform and SEO_ST. Columns (1) and (2) pertain to the sample firms. Columns (3) and (4) refer to the matched firms. Column (5) refers to the pooled regression of sample and matched firms. The t-statistics are reported in parentheses. Significance at the 10% level is marked with *, at the 5% level with **, and at the 1% level with ***. Intercept (1) (2) (3) (4) (5) 0.091 (2.89)*** 0.0177 (1.31) 0.0513 (0.81) 0.0112 (1.10) 0.0939 (1.28) -0.0097 (-0.87) -0.1289 (-1.91)* 0.866 (1.61) -0.0277 (-1.29) -0.0875 (-1.08) -0.0184 (-0.77) 0.0016 (1.35) -0.0174 (-0.98) -0.0757 (-1.00) 0.0459 (2.63)** -0.0507 (-1.49) -0.001 (-1.93)* 0.0055 (0.08) 0.0504 (2.74)** -0.0166 (-0.92) 0.0001 (0.18) -0.0029 (-0.21) 0.04319 (1.42) 0.0929 (2.54)** -0.1424 (-2.12)** 25 25 25 25 50 0.10 0.20 0.20 0.24 0.20 Logsize Loggap M/B assets Shares% R&D Underperform SEO_ST SEO_ST* Underperform # of observations 2 Adjusted R 58 We run regressions of the form: CARi = β 0 + β1 Logsizei + β 2 Loggapi + β 3 M / Bassetsi + β 4 Shares% i + β 5 R & Di + β 6Underperformi + β 7 SEO _ STi + β 8 SEO _ ST *Underperformi + ε i The dependent variable in each case is the abnormal return earned over the (-2, +2) event window surrounding the self-tender announcement. Logsize refers to the log of total assets. Loggap refers to the log of the number of days between the SEO and the self-tender offer. M/Bassets refers to the market value of assets divided by the book value of assets. Shares% refers to the percent of shares that are to be repurchased. R&D refers to the R&D expenditure scaled by sales. Underperform refers to the annual postissue underperformance. SEO_ST is a dummy variable that takes on a value of 1 if the firm is part of our sample of firms that issued stock and then bought them back. It takes on a value of 0 if the firm is part of the matched sample that repurchased stock but did not have an SEO preceding the repurchase. SEO_ST*Underperform is the product of Underperform and SEO_ST. In columns (1) and (2) of table 5, we consider the relationship between post-issue underperformance and abnormal returns upon announcement of the repurchase for our sample of firms announcing self-tender offers. The coefficient estimates for β6 are negative and sometimes significant. The results weakly support the hypothesized negative relationship between the post-issue underperformance and the abnormal returns earned upon the announcement of the self-tender offer. In other words, the abnormal returns at the time of the repurchase announcements are higher if the firm’s stock has gone down sharply following the SEO. In contrast, the abnormal returns are positively associated with prior stock performance for firms that announce a self-tender offer but have not issued stock. In column (5) we test whether the nature of the relationship between the underperformance and the CAR differs for the sample and matched firms. The term SEO_ST*Underperform captures the difference between the issuing sample and the non-issuing matched sample. It is negative and significant, suggesting that the relationship between prior stock underperformance and abnormal returns at the time of announcement of self-tender offers is specific to issuing firms. Conclusion To our knowledge, this is the first study to examine firms that repurchase shares soon after an SEO. Billett and Xue (2007) consider the complementary situation where firms repurchase their stock prior to an SEO. They conclude that the positive signal sent by the repurchase of stock may enable issuers to obtain a more favorable pricing for their SEO. We find that even after the SEO, repurchases can have a useful function. They can help firms counteract the post issue underperformance of their share price. Our contributions from this study lie in several areas. We provide evidence that self-tender repurchases can act as powerful signals for firms that have issued stock. Although prior research by Comment and Jarrell (1991) and Lie and McConnell (1998) have found evidence as to the signaling power of self-tender offers, we provide further insights into the motivation for such signals. The results have a clear and direct implication for current management practice. Post-SEO underperformance of a firm’s stock price can impair a firm’s ability to effectively raise capital. Based on our evidence it appear that firms that have strong underlying performance can signal that to the market through self-tender offers. 59 Our results have implications for other research as well. Past research on repurchases has often found signaling to be a powerful explanation for the repurchasing behavior of firms. Yet, as pointed out by Louis and White (2007), “in spite of the popularity of the signaling proposition, there is little empirical evidence on whether managers intentionally use repurchases to signal their private information.” We are still not sure why and when managers signal. However, our results suggest that post-issue underperformance could be one reason. We do not suggest that post-issue underperformance is the only or even the primary driver of self-tender offers or of the positive returns earned upon announcement of a self-tender offer. However, our results indicate that it would be fruitful to explore the motivations behind signaling in greater detail. 60 References Billett, M., & Xue, H. (2007). Share Repurchases and the Need for External Finance. Journal of Applied Corporate Finance, 19(3), 42 – 55. Brav, A., Geczy, C., & Gompers, P. (2000). Is the abnormal return following equity issuances anomalous? Journal of Financial Economics, 56, 209 – 249. Comment, R., & Jarrell, G. (1991). The relative signalling power of Dutch-auction and fixed price selftender offers and open-market share purchases. Journal of Finance, 46, 1243 – 1271. Dann, L. (1981). Common stock repurchases: an analysis of returns to bondholders and stockholders. Journal of Financial Economics, 9, 113 – 138. Dittmar, A. (2000). Why do firms repurchase stock? Journal of Business 73(3), 331 – 355. Fama, E., & French, K. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, 3 – 56. Grullon, G., Michaely, R. (2004). The information content of share repurchase programs. Journal of Finance, 59, 651 – 680. Henry, D. (2006). The dirty little secret about buybacks. BusinessWeek, Issue 3968: Retrieved from http://www.businessweek.com/magazine/content/06_04/b3968099.htm. Ikenberry, D., & Vermaelen, T. (1996). The option to repurchase stock. Financial Management 25, 9 – 24. Lie, E., & McConnell, J. (1998). Earnings signals in fixed-price and Dutch auction self-tender offers. Journal of Financial Economics, 49, 161 - 186. Loughran, T., & Ritter, J. (1995). The new issues puzzle. Journal of Finance, 50(1), 23 - 51. Louis, H., & White, H. (2007). Do managers intentionally use repurchase tender offers to signal private information? Evidence from firm financial reporting behavior. Journal of Financial Economics, 85, 205 – 233. McNally, W. (1999). Open market stock repurchase signaling. Financial Management, 28, 55 - 67. Spiess, K., & Affleck-Graves, J. (1995). The long-run performance following seasoned equity issues. Journal of Financial Economics, 38, 243 - 267. Vermaelen, T. (1981). Common stock repurchases and market signaling. Journal of Financial Economics, 9, 139 – 183. 61 ASAC 2008 Halifax, Canada Marie-Claude Beaulieu Department of Finance and Insurance (FSA) Laval University William R. Sodjahin (PhD Candidate) Department of Finance and Insurance (FSA) Laval University REVERSE SPLIT ANNOUNCEMENTS, EFFECTIVE DATES AND SURVIVAL1 We examine separately the determinants of reverse split decision, the reverse split factor choice and the timing of the reverse split realization. The results reveal that while an important run down in stock price and poor growth opportunities lead to a reverse split decision, the choice of the reverse split size is determined by the size of the firm, the preannouncement price level and the liquidity. We also observe a strong age and size effect in setting the reverse split realization delay. Last, we find that both reverse split factor and reverse split execution delay can predict the reverse splitting firms’ survival. I. Introduction This paper investigates the factors influencing the probability of a reverse split decision and the determinants of the reverse split factor. It further characterizes the reverse split realization delay and its determinants. We seek to understand how this delay varies across firms and what the determinants of this variation are. Our paper is the first one to examine separately the reverse split decision, the choice of the reverse split factor and the reverse split realization delay and the influence of managers’ and directors’ incentives on these different steps of reverse split policy. The paper also investigates whether the reverse split factors or execution delays can predict the survival of the reverse splitting firms. Finally, the paper investigates how managers’ and directors’ incentives can affect the probability of reverse splitting firms’ survival. These investigations are important since they will give more insight into the reverse split process and will help better understand the motivations behind the reverse split. Our results will offer guidance on using the reverse split size, the reverse split execution delay to foresee firms’ survival. The reverse split is nothing more than “substitution of one new share for a certain number of outstanding shares” (Han, 1995). Compared to the stock split, little research has been done regarding the motivation for reverse split and information content of the reverse split factor and the reverse split execution delay. Most of the previous literature has documented negative abnormal returns around reverse split dates2. Empirical evidence shows that reverse splits are viewed as a negative signal about future performance, since the only way for reverse splitting firms to improve their stock price is through artificial means (Spudeck and Moyer, 1985). A rational manager 1 M.-C. Beaulieu, RBC Chair in Financial Innovations, gratefully acknowledges funding from Social Sciences and Humanities Research Council of Canada (SSHRC), Fonds Québecois pour la Recherche sur la Société et la Culture (FQRSC) and Institut de Finance Mathématique de Montréal (IFM2). W. Sodjahin is grateful for financial support from Institut de Finance Mathématique de Montréal (IFM2). 2 See Woodridge and Chambers (1983), Lamoureux and Poon (1987), Han (1995) and Desai and Jain (1997) among others. 62 expecting earnings and price growth would not incur the reverse split costs. Moreover, managers could undertake reverse splits to maintain listing on the current exchange. Indeed, Peterson and Peterson (1992) distinguished discretionary versus nondiscretionary reverse splits. Note however that some empirical studies have also revealed some value-improving results. Reverse splits increase liquidity (Han, 1995), increase the ability of the stock to be purchased on margin (Masse, Hanrahan, and Kushner, 1997), improve the image of the stock and make it more attractive for institutional investors3 (West and Brouilette, 1970) and decrease risk (Peterson and Peterson, 1992). Masse et al. (1997) have observed a positive market reaction to Canadian reverse split. But to date, there is no study investigating factors that determine the probability of reverse split decision, the information content of the reverse split factor, the timing of reverse split realization and the extend to which reverse split factors or the reverse split execution delay can provide information about the likelihood of a firm survival. These different results raise the following general questions: What are the real underlying motivations for engaging in reverse splits? How can one distinguish between “good” and “bad” motivations? More specifically, this study seeks to answer the following questions: What factors, continued poor price and earning performance or growth opportunities, influence the reverse split decision? Similarly to stock splits, do managers’ and directors’ incentives have any role to play (Beaulieu and Sodjahin, 2007)? Is there any information content in reverse split factors? Furthermore, what factors determine the reverse split realization delay? Do reverse split factor and reverse split execution delay help discriminate between “good” and “bad” motivation for reverse split decision? Do reverse split factor, reverse split realization delay help foresee the survival of reverse splitting firms? How do managers’ and directors’ incentives affect the probability of reverse splitting firms’ survival? This study contributes to the literature by providing insights on the factors that are associated with reverse split decisions, reverse split factor choice and reverse split execution delay and the predictability of the reverse splitting firms’ survival by the reverse split size and execution delay. We estimate a probit regression of reverse split decision and analyze especially the influences of past performances, prospect indicators and managerial and directors incentives on this decision. Second, we examine whether there an incremental management’s private information4, using the Inverse Mills Ratio5 (see Heckman, 1979; Nayak and Prabhala, 2001), revealed through the reverse split factors and reverse split execution delay. Third, we estimate robust least squares models of the reverse split factors and the realization delay to find different factors that can determine these decisions. We finally estimate probit models of reverse splitting firms’ survival to investigate the role reverse split factors or reverse split execution delay in the probability of survival. We find, consistent with the signaling hypothesis, that the reverse split decision is essentially motivated by unusual run-downs in stock price and poor growth opportunities. We find respectively negative and positive conditional incremental information in the size of the reverse split factor and in reverse split realization delay. More precisely we observe that the choice of the reverse split factor is determined by the size of the firm, the preannouncement price level and liquidity while reverse split execution is affected by the age and the size of the firm. Finally, results indicate that the likelihood to survive 12 months after the reverse realization is positively related to the reverse split size chosen but negatively associated to how long firms wait to execute the announced reverse split. We also find that when firms are not doing well, as it is often the case when a reverse split occurs, granting loans to exercise stock options could be burdensome and directors’ ownership could make them overbearing and disastrous (see Holmstrom, 2005; Song and Thakor, 2007) for 3 4 In fact, “penny stocks” are in disfavor with institutional investors. Beside the reverse split decision itself. 5 The predicted values of the reverse split decision are used to generate the Inverse Mills Ratio which is the ratio of the probability density function over the cumulative distribution function of a distribution. 63 splitting firms and affect negatively the probability of survival. The remainder of this paper is organized as follows. Section 2 develops the research hypotheses in the context of the prior literature. Section 3 describes the sample, data sources, variable specification, and testing methodology. Section 4 provides descriptive statistics for the sample firms and discusses the results of empirical tests. Section 5 concludes. II- Theory and hypotheses This section develops the research hypotheses in the context of prior literature. The central questions that we address are whether there is distinct information that can be drawn from the reverse split decision itself, the reverse split factor and execution delay and whether reverse split factor and reverse split realization delay can signal the survival of the reverse splitting firms. Four hypotheses are tested. The first hypothesis links the reverse split decision, on the one hand, to past performance and growth opportunities, and on the other hand, to managerial and directors’ incentives. Hypotheses 2 and 3 present respectively a relationship of the reverse split factor and of the reverse split realization delay with past performance and growth opportunities and managerial and directors’ incentives. Hypothesis 4 deals with the effect of split factor and reverse split execution delay and their impact on reverse splitting firms’ survival. In light of many studies that documented average negative abnormal returns (Han, 1995 and Desai and Jain, 1997 among others) reverse splits convey unfavorable private information about the current value of firms. Thus, if reverse splits really signal negative information then the information about poor past performance and weak growth opportunities will have strong positive impact on the reverse split decision contrary to the stock split case (Beaulieu and Sodjahin, 2007). Consequently, we examine the following testable hypothesis: Hypothesis 1a: Firms with poor past market performance and weak growth opportunities are more likely to reverse split. Hypothesis 1b states that if there are managers’ and directors’ incentives, firms will be more concerned by their image and will then undertake a reverse split to get their stock out of the penny stock zone. We therefore test the following hypothesis: Hypothesis 1b: Managers’ and directors’ incentives affect positively the reverse split decision. When there is little run down in preannouncement price and better growth opportunities, firms will logically prefer smaller size of the reverse split. In fact, the stock price will normally increase without a strong artificial intervention. The information content of stock split factor have been tackled in the literature (McNichols and Dravid, 1990; Beaulieu and Sodjahin, 2007 among others) but to date, there is no direct evidence concerning the reverse split factor. As derived by Beaulieu and Sodjahin (2007) the optimal split factor f * is: ϑ f * = ⎡⎣G (m, X 0 ) Z ρ ⎤⎦ , where 0 < ρ < 1 represents the elasticity of the estimated intrinsic value with respect to Z that represents the information about past market performance and future prospect indicators which is 1 > 0 . G(.,.) is a function of the number 1− ρ of shares outstanding before the split, m and the pre-split demand in the firm’s sector, X 0 . Indeed, X t is the state variable that captures market conditions (that is the aggregate demand in the essential for investors to estimate stock prices and ϑ = 64 firm’s sector). Under the risk neutral measure, X t follows the following dynamics: dX t = ( r − δ ) X t dt + σ X t dwˆ t ˆ t is a standard where δ > 0 , r is the risk free interest rate, σ the volatility of the process and w Brownian motion. We get in a reverse split framework when 0 < f * < 1. The reverse split size in this context is 1− f * . * Since ∂f > 0 , low Z (poor past market performance and weak growth opportunities) imply large ∂Z 1− f * . Thus, we test the following hypothesis: Hypothesis 2a: Firms with poor past market performance and weak growth opportunities are more likely to choose larger reverse split magnitude. Conditioning on the preannouncement price and growth opportunities, firms with managers’ and directors’ incentives will choose higher reverse split size to preserve investors’ perception and the “image” of the company in general. This lead to the following hypothesis: Hypothesis 2b: Managers’ and directors’ incentives are positively related to the reverse split magnitude. Reverse split announcement signals that price will keep decreasing (see. Han, 1995 among others) and if that is the case, the manager will not wait long to bring the stock price out of the “penny stock” range to a more attractive trading range. The following hypothesis is then examined: Hypothesis 3a: Firms with poor past market performance and weak growth opportunities will realize the announced reverse split earlier. We also examine the subsequent hypothesis: Hypothesis 3b: Managers’ and directors’ incentives are negatively related to the reverse split realization delay. Firms with managers’ and directors’ incentives will, again for image reasons, quicken the realization of the reverse split to bring up their stock price as soon as possible into a more attractive trading range. Finally, our last hypothesis states that the size of the reverse split factor and the reverse split execution delay can predict the survival of reverse splitting firms: Hypothesis 4: The probability of survival of reverse splitting firms is positively related to the reverse split factor magnitude and to managers’ and directors’ incentives, but negatively associated with the reverse split execution delay. In fact, firms with larger reverse split factors are more likely to survive since their post reverse split price is high enough to meet listing requirements. Moreover, firms with better governance are less likely to face delisting (see Charitou et al, 2007). Last, firms that wait long to get their price out of the “penny stock” range are less likely to survive. Past market performance is measured by the variable Run-down computed as the buy-and-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. It is the same variable called RUNUP in the stock split literature (see. e.g. Desai and Jain, 1997 and Nayak and Prabhala, 2001). 65 Growth opportunities are captured by the variable Market-to-book ratio since this ratio has commonly been used as a proxy for the growth opportunities of a firm (see. e.g. Daniel and Titman, 1999 and D’Mello and Ferris, 2000). Managerial and directors’ incentives include the following variables: o Dirsubstock: equals 1 if directors are subject to stock ownership requirements (otherwise it is equal to zero); o Dirownership: equals 1 if directors’ and officers’ ownership as % of shares outstanding is >5% and <=30% (otherwise it is equal to zero). The alternative choices for this variable in ISS are: officers and directors ownership as % of shares outstanding is <1% or >30% (1) and >=1% and <=5% (2). We choose Dirownership since in reality, very few firms are owned in a proportion higher than 30%. o Stockplan: takes the value of 1 when the company managers are remunerated with options and 0 otherwise. o Loansoption: takes the value of 1 when the company provides loans to executives for exercising options and 0 otherwise. o Incentives Global takes the value of 1 when Dirsubstock, Dirownership, Stockplan and Loansoption take simultaneously the value of 1 and 0 otherwise. Given that specific incentives might in some cases affect the decisions surrounding the split (the decision to split, the choice of the split factor and the delay within which the split becomes effective) in different ways, Incentives Global allow us to determine the overall effect. The reverse splitting firms’ survival is measured by Survival that takes the value of 1 when the ending dates in CRSP data base (variable ENDDT) occur after 12 months from the reverse split effective dates and 0 otherwise. III- Univariate Analysis This section presents the data, summary statistics of our sample and an initial exploration of the relation between reverse split factor and firm’s characteristics on the one hand and timing of reverse split realization and firm’s characteristics on the other. III- a) Data description and summary statistics The sample is collected from four sources: Center for Research in Security Prices (CRSP) daily master files (for reverse split effective date, reverse split factor, etc.), Bloomberg (for the reverse split announcement date), Compustat (for growth opportunities and other accounting data) and from Institutional Shareholder Services (ISS) for managerial and directors’ incentives data. Because of the availability of ISS data, our sample covers the period from January 2003 to December 2005. Our primary sample is composed of 145 reverse splits whose announcement dates information was retrieved from Bloomberg. Table I: Summary Statistics Our initial sample contains 145 reverse splits that occurred on NASDAQ, NYSE and AMEX between January 2003 and December 2005. Panel A reports reverse splits by factors and year. Panel B provides descriptive statistics for reverse split delay. We trim the top 1% of observation from the reverse split execution delay to limit the impact of outliers. The descriptive statistics of the trimmed sample are presented in parentheses. Panel C reports the p-value of tests on the split delay across the split delay tercile. The split execution delay is equal to the number of days from announcement date to effective date. As reported in CRSP, the split factor is FACPR = [ S (t ) S (t ')] − 1 , where s (t ) is the number of shares outstanding after the split, and s (t ' ) is the number of shares outstanding before the split. Also reported are p-values of mean comparisons (with unequal variance) and Man-Whitney (Wilcoxon rank-sum) comparisons between different groups. The symbols ***, **,* indicate significance at respectively a 1% level, a 5% and a 10% level. 66 Panel A: Reverse stock splits by factor and year Reverse CRSP FACPR Split Size 43 for 49 -0.122 11 for 15 -0.267 1 for 2 -0.500 2 for 5 -0.600 1 for 3 -0.667 1 for 4 -0.750 1 for 5 -0.800 1 for 6 -0.833 1 for 7 -0.857 1 for 8 -0.875 1 for 9 -0.889 1 for 10 -0.900 1 for 12 -0.917 1 for 13 -0.923 1 for 15 -0.933 1 for 20 -0.950 1 for 25 -0.960 1 for 30 -0.967 1 for 50 -0.980 Total Panel B: Timing of split realization tercile Mean (in Days) 1 (short delay) 0.98 (0.98) 2 3.07 (3.07) 3 (long delay) 34.74 (28.69) Total 12.28 (10.24) Year Median (in Days) 1 (1) 3 (3) 14.5 (14) 3 (3) 2003 0 0 1 1 6 10 12 4 4 3 1 12 0 0 2 2 0 2 0 60 2004 0 0 3 0 5 5 5 2 0 1 0 7 0 0 1 2 1 0 0 32 2005 1 1 3 2 2 11 13 2 3 1 0 9 1 1 0 2 0 0 1 53 Total 1 1 7 3 13 26 30 8 7 5 1 28 1 1 3 6 1 2 1 145 Std. Dev. Max Min 0.13 (0.13) 0.52 (0.52) 57.50 (40.73) 35.63 (25.83) 1 (1) 4 (4) 307 (203) 307 (203) 0 (0) 2 (2) 5 (5) 0 (0) Number of Observations 58 (58) 41(41) 46 (45) 145 (144) Panel C: Tests of equality of means and medians (p-values) Mean 0.000*** 0.000*** 1 versus 3 1 versus 2 Median 0.000*** 0.000*** Table 1 reports summary statistics for the sample of reverse stock splits analyzed in this paper. Panel A report information related to the distribution of split factors per year. Let s (t ) be the number of shares outstanding after the split, and s (t ' ) the number of shares outstanding before the split. As reported in CRSP the split factor is s (t ) − s ( t ' ) (1) FACPR = s (t ' ) Notice that, contrarily to splits, FACPR is negative for reverse splits. Reverse split factors range from -0.122 (43 for 49) to -0.980 (1 for 50). The most common reverse split factors are respectively -0.800 (1 for 5), -0.900 (1 for 10) and -0.750 (1 for 4). The highest number of reverse splits occurred in 2003 (41%) followed by 52 (37%) in 2005 and 32 (22%) in 2004. Number of reverse splits Number of reverse splits each month 2005Oct 2005Jul 2005Apr 2005Jan 2004Oct 2004Jul 2004Apr 2004Jan 2003Oct 2003Jul 2003Apr 2003Jan 14 12 10 8 6 4 2 0 Month Figure 1: This Figure provides an overview of the number of reverse stock split through months and years. The sample includes 145 reverse stock splits announced between January 1, 2003 and December 31, 2005. 67 Figure 1 provides an overview of the number of reverse splits through months and years. The number of reverse splits fluctuates highly throughout the year. The highest occurrences of reverse splits in 2003, 2004 and 2005 are respectively observed in June, October and December. These results reveal that reverse split announcements are not concentrated during a specific period of the year. Panel B reports an average elapsed time between the announcement date and the effective date of 12.28 days (10.24 for the trimmed sample) compared to 36.89 days in our previous study on stock splits (see Beaulieu and Sodjahin, 2007). We trim the top 1% observations from the reverse split realization delay to limit the impact of outliers. The timing of reverse split realization exhibits a high variability of 35.63 (25.83 for the trimmed sample) but a low median time of 3 days. This high variability makes the reverse execution delay important to study. Notice that the standard deviation of the reverse split execution delay for tercile three is far higher (57.50 and 40.73 for the trimmed sample) followed by tercile two (0.52) and tercile one (0.13). Panel C shows that the difference between the short and the long split execution delay is significant for both the trimmed and non trimmed samples. III- b) Reverse split factor and firm’s characteristics Panel A of Table II tests on a univariate basis the relationship between reverse split factors and firms characteristics. We observe that firms which choose larger reverse split magnitude (1-for-5 and more) have significantly smaller size and lower price level and are more liquid. Even though the median run-down of firms that choose 1-for-5 factors and more is significantly more important, the average run-down is not significantly different from zero. There are no statistically significant differences between the larger split magnitude (1-for-5 and more) and the smaller one (less than 1for-5) for growth opportunities. Table II: Reverse split factor and firm characteristics Panel A presents univariate relations between reverse split factor delay and firms’ characteristics.. Panel B tests the effects of directors’ ownership, directors and officers’ ownership (5%-30%), stock option plan and loan for option exercise on reverse split delay. The reverse split execution delay is equal to the number of days from announcement date to effective date. Log(age) is the natural logarithm of the age of the firms at the announcement date (days from listing date to split announcement date). P_Price and Log(size) are respectively the price level and the natural logarithm of the market value on trading date [-5], where [0] represents the split announcement date. Rundown is the buy-and-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. Market-to-book is monthly market-to-book ratio prior the reverse split announcement. Volume is computed as the ratio of the average number of shares traded in the month prior the split announcement to the total quarterly number of outstanding shares before the reverse split announcement. Also reported are p-values of mean comparisons (with unequal variance) and Man-Whitney (Wilcoxon rank-sum) comparisons between different groups. The symbols ***, **,* indicate significance at respectively a 1% level, a 5% and a 10% level. Panel A: Reverse split factor and firm characteristics 1 (<1 for 5) Firm’s attributes Log(Age) P_Price Log (Size) Past performance Run-down Growth opportunities Market-to-Book Split factor Split factor Liquidity Volume 2 (1 for 5) Mean Median Mean Median Mean Median Diff. mean (median) 1 versus 2&3 p-Value 3.45 6.01 7.50 3.41 2.24 7.28 3.31 0.84 6.98 3.31 0.64 7.01 3.44 0.85 6.35 3.41 0.96 6.14 0.375 (0.458) 0.001*** (0.000***) 0.000*** (0.000***) 0.20 -0.05 0.70 -0.39 -0.43 -0.50 0.288 (0.000***) 3.18 1.63 3.77 1.86 1.42 0.65 0.549 (0.101) -0.66 -0.75 -0.80 -0.80 -0.90 -0.90 0.000*** (0.000***) 0.13 0.07 0.29 0.16 0.21 0.08 0.026** (0.058*) 68 3 (>1 for 5) Panel B: Reverse split factor and managerial and directors’ incentives Split factor Directors ownership Mean -0.654 -0.799 -0.802 -0.790 -0.782 -0.796 -0.813 -0.786 -0.800 -0.798 Yes No Yes No Yes No Yes No Yes No Directors and officers ownership (5%-30%) Stock option plan Loan for option exercise Incentives Global Diff. mean p-value 0.215 (0.076*) 0.542 (0.964) 0.629 (0.363) 0.167 (0.752) 0.812 (0.280) Median -0.625 -0.800 -0.800 -0.800 -0.800 -0.800 -0.800 -0.800 -0.792 -0.833 Results in Panel B of Table II show that firms don’t choose significantly different reverse split factors when there are manager’s and directors’ incentives or more precisely when directors and officers own stock. III- c) Timing of reverse split realization and firm’s characteristics Table III deals with the relationship between the reverse split execution delay and firm characteristics. Panel A tests on a univariate basis the relation between reverse split execution delays and firms characteristics. We observe that firms which realize the announced reverse split earlier (tercile one) have significantly smaller size and preannouncement price than firms that do so later. The mean differences between terciles one and three are statistically significant at the 1% level. There are no statistically significant differences between terciles one and three for other characteristics such as past performance, growth opportunities, reverse split factor and liquidity. Table III: Timing of reverse split realization and firm characteristics Panel A presents univariate relations between reverse split execution delay and firms’ characteristics. Panel B tests the effects of directors’ ownership, directors and officers’ ownership (5%-30%), stock option plan and loan for option exercise on reverse split delay. The reverse split execution delay is equal to the number of days from announcement date to effective date. Log(age) is the natural logarithm of the age of the firms at the announcement date (days from listing date to split announcement date). P_Price and Log(size) are respectively the price level and the natural logarithm of the market value on trading date [-5], where [0] represents the split announcement date. Rundown is the buy-and-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. Market-to-book is monthly market-to-book ratio prior the reverse split announcement. Volume is computed as the ratio of the average number of shares traded in the month prior the split announcement to the total quarterly number of outstanding shares before the reverse split announcement. Also reported are p-values of mean comparisons (with unequal variance) and Man-Whitney (Wilcoxon rank-sum) comparisons between different groups. The symbols ***, **,* indicate significance at respectively a 1% level, a 5% and a 10% level. Panel A: Timing of split realization and firm characteristics 1 (short) Firm’s attributes Log(Age) P_Price Log (Size) Past performance Run-down Growth opportunities Market-to-Book Split factor Split factor Liquidity Volume 2 Mean Median Mean Median Mean Median Diff. mean (median) 1 versus 3 p-Value 3.41 1.53 6.61 3.40 0.63 6.36 3.32 1.28 6.54 3.29 0.62 6.39 3.52 5.36 7.47 3.55 1.31 7.22 0.096* (0.063*) 0.021** (0.007***) 0.000*** (0.000***) -0.05 -0.39 0.37 -0.18 -0.21 -0.42 0.501 (0.851) 2.20 1.10 1.91 1.02 3.34 1.55 0.540 (0.136) -0.81 -0.83 -0.81 -0.80 -0.76 -0.80 0.131 (0.230) 0.18 0.08 0.24 0.10 0.18 0.07 0.992 (0.761) 69 3 (long) Panel B: Timing of split realization and managerial and directors’ incentives Directors ownership Directors and officers ownership (5%-30%) Stock option plan Loan for option exercise Incentives Global Yes No Yes No Yes No Yes No Yes Split execution delay Mean Median 1.59 1.73 0.52 0.48 0.50 0.48 0.58 0.48 0.66 0.48 0.53 0.48 0.58 0.48 0.53 0.48 0.60 0.48 No 0.49 0.48 Diff. mean (median) p-value 0.087* (0.008***) 0.418 (0.266) 0.436 (0.479) 0.683 (0.646) 0.242 (0.601) Panel B shows that when there are manager’s and directors’ incentives or more specifically when directors and officers own stock, firms reverse split execution delay is not significantly different from the delay in firms with no such incentives. Even though the difference in medians for directors stock ownership requirements is significant at the 1% level, the difference in means is only significant at 10%. IV- Multivariate analysis In this section we present and discuss tests of the research hypotheses developed above. The results are presented for the reverse split decision, the reverse split factor choice, the delay of reverse splits execution and the survival model. Table IV deals with reverse split decision, Table V presents conditional reverse split effects. Table VI presents the results for the reverse split factor choice while Table VII deals with the delay in reverse split execution. Table VIII reports the influence of reverse split magnitude, execution delay and incentives on the probability of survival of reverse splitting firms. We distinguish between the choice of the reverse split factor and the reverse split decision itself for a better understanding of explanatory factors relative to each decision. a) Reverse split decision The main question we address in this section is: Why do firms decide to undertake reverse splits? To answer this question we specify an empirical model for RSPLi , the latent variable ruling a reverse split decision which takes the value of 1 if firm i announced a reverse split during the calendar year and 0 otherwise. We also define Rundowni the past market performance, MB the market-to-book, Incentives the managers’ and directors’ incentives, C the control variables, and the error term ui which is assumed to be normally distributed and independent of Rundown, MB , Incentives, and C . The empirical model is RSPLi = γ 1 + γ 2 Rundowni + γ 3 MBi + γ 4 Incentivesi + γ 5Ci + ui (2) Under the assumption of ui normality, the reverse split decision is a standard probit model. The control variables C include preannouncement stock price, size, age, trading volume and earnings changes before and around reverse split announcement. We include preannouncement price as firms with lower stock prices are more likely to announce reverse splits, if reverse splits are really intended to get stock out of the “penny stock range”. We include size since similar to stock split (McNichols and Dravid, 1990 and Nayak and Prabhala, 2001), conditional on price, small firms could be more likely to reverse split. The firm size also is used as a proxy for the level of information asymmetry (see, e.g. Grossman and Stiglitz, 1976; Zeghal, 1984 and Freeman, 1987). 70 We also include the age of the firm at the reverse split announcement date (days from listing date to reverse split announcement date), since this variable can capture prudence motives (see, e.g. Del Guercio, 1996 and Gompers and Metrick, 2001). We expect age to be negatively related to reverse split decision, reverse split factor and positively related to the reverse split execution delay. We include trading volume since Han (1995), among others, suggests that reverse splits are associated with changes in trading volume. We finally consider earnings changes before and around reverse split announcement dates because some studies (see. e.g. Robinson, 2007) observe that the market reaction is positively related to the earnings performance before and after the reverse split. Table IV: Probit models of the reverse split decision This table presents the results of robust probit regressions explaining the reverse split decision. The dependent variable takes the value of 1 if the firm announced a reverse split during the calendar year and 0 otherwise. Log(age) is the natural logarithm of the number of days from listing date to reverse split announcement date. Rundown is the buy-and-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. Market-to-book is monthly market-to-book ratio prior to the reverse split announcement. Volume is computed as the ratio of the average number of shares traded in the month prior the reverse split announcement to the total quarterly number of outstanding shares before the reverse split announcement. Earnings Changes=1 if ΔEPS (Earnings Per Share) > 0 around reverse split announcement and Earnings Changes=1 if ΔEPS (Earnings Per Share) > 0 before reverse split announcement. Dirsubstock: equals 1 if directors are subject to stock ownership requirements and 0 otherwise. Dirownership equals 1 if directors and officers ownership as % of shares outstanding is >5% and <=30% (otherwise it is equal to zero). Stockplan takes the value of 1 when the company’s directors are remunerated with options and 0 otherwise. Loansoption takes the value of 1 when the company provides loans to executives for exercising options and 0 otherwise. Incentives Global takes the value of 0 when Dirsubstock and Dirownership and Stockplan and Loansoption take simultaneously the value of 0 and 1 otherwise. Heteroscedasticityconsistent p-values are provided in parentheses. ***, **,* indicate significance at respectively a 1% , a 5% and a 10% level. (1) Firm attributes Log(Age) P_Price Log(Size) Past Performance Rundown Growth opportunities Market-to-book Liquidity Volume Earnings changes =1 if ΔEPS>0 around split =1 if ΔEPS>0 before reverse split Managerial and directors’ incentives Dirsubstock Dirownership Stockplan Loansoption Incentives Global Intercept (2) (3) (4) 0.032 (0.903) -0.009 (0.580) -0.230 (0.029)** 0.100 (0.710) -0.022 (0.232) -0.188 (0.080)* 0.006 (0.981) -0.017 (0.396) -0.216 (0.046)** 0.068 (0.795) -0.012 (0.480) -0.211 (0.047)** -0.149 (0.001)*** -0.140 (0.002)*** -0.148 (0.001)*** -0.147 (0.002)*** -0.0001 (0.000)*** -0.0002 (0.000)*** -0.0001 (0.000)*** -0.0001(0.000)*** 0.041 (0.818) 0.052 (0.770) 0.049 (0.781) 0.044 (0.805) 0.118 (0.469) 0.130 (0.423) 0.527 (0.392) -0.134 (0.418) 0.112 (0.661) -0.017 (0.923) -0.172 (0.281) 1.411 (0.228) 0.754 (0.525) 1.461(0.218) 1.266 (0.284) 13.19% 12.85% 13.65% 13.46% 41.71 (0.000)*** 312 38.27 (0.000)*** 300 44.16 (0.000)*** 312 41.02 (0.000)*** 312 Diagnostics Pseudo R2 Wald χ test (all coeff.=0) No. of observations 2 Table IV presents probit regression results. In the first model, the explanatory variables are firm attributes, preannouncement performance, growth opportunity and liquidity. Model 2 extends Model 1 by introducing earnings changes while models 3 and 4 add managers’ and directors’ incentives variables. Results show that the coefficient for firm size is negative and significant at the 5% level through all specifications except for Model 2 (10% level). This suggests that smaller firms are more likely to engage in reverse splits compared to larger ones, probably because small firms are more likely to be in the “penny stock” range. Opposite the stock split case (see. e.g. Nayak and 71 Prabhala, 2001), we find that the coefficients for Rundown6 and Market-to-book are negative and strongly significant throughout our four models. This suggests first that firms whose stocks have experienced unusual run-downs are more likely to make reverse split decisions and moreover that past market performance influences more the reverse split decision than operating performance. Second, poor growth opportunities play a significant role in the reverse split decision. These results are consistent with the signaling hypothesis. Moreover, we find no evidence of the impact of liquidity, earnings changes and managers’ and directors’ incentives on the reverse split decision. These results are in full accordance with Hypothesis 1a but not with Hypothesis 1b. Reverse split decision doesn’t seem to be determined by managers’ and directors’ incentives contrary to the stock split decision (Beaulieu and Sodjahin, 2007). Before studying the determinants of the reverse split factor and the execution delay, let us examine if there is any incremental information in the reverse split factor and the realization of the timing above the reverse split decision itself7. Let X irs be the market preannouncement information set for firm i . As in Nayak and Prabhala (2001), X irs includes preannouncement price, size, rundown and trading volume. Let ψ irs represent the announcing firm private information. Firm i announces a reverse split (RS) if RSPLi = X irsθ +ψ irs > 0 (3) If there is respectively negative and positive incremental information in the reverse split magnitude and in the timing of reverse split realization, we should find that both reverse split factors8 and reverse split execution delay are positively related to the unrevealed information of the reverse split decision. Consequently both β f and β d should be positive in the following regression E ( RSFACTi | RS ) = γ f + β f E (ψ irs | X irs ' θ + ψ irs > 0) (4) E ( Delayi | RS ) = γ d + β d E (ψ irs | X irs ' θ + ψ irs > 0) (5) The previous equations represent respectively the conditional reverse split factor and the conditional reverse split execution delay, given the vector of characteristics X irs associated with the reverse splitting firm. Under the assumption that ψ irs is normally distributed and defining λˆrs as the Inverse Mills Ratio (IMR) for the reverse split announcement, we can write equations (4) and (5) as E ( RSFACTi | RS ) = γ f + β f λˆrsi E ( Delay | RS ) = γ d + β d λˆ i (6) (7) rsi Defining φ (.) as the density probability function, and Φ (.) as the cumulative probability function, we find λˆrs which is estimated using a probit estimates for the vector of parameters, θ , that is λˆrsi = φ ( X irs ' θ ) . Φ ( X irs ' θ ) (8) 6 We also investigated the influence of the operating performance measured by ROA (Return on Asset). Even though the ROA coefficient (-0.002 in model 1) has the same sign as Rundown, it is not statistically significant (p-value 0.120). 7 Whether or not a reverse split is announced to the market. 8 Recall that, contrarily to split factors, reverse split factors are negative given that the number of shares in the firm diminishes after the reverse split. 72 We empirically estimate equation (8) using a probit model since ψ irs is assumed to be normally distributed. We also estimate equations (6) and (7) using least squares with adjusted standard errors (see Heckman, 1979). Results are presented in Table V. Table V: Conditional incremental reverse split effects This table presents the results of conditional incremental information in reverse split size and in reverse split realization delay. Panel A reports the robust probit reverse split decision with explanatory variables reduced to Nayak and Prabhala (2001) market preannouncement information set. The dependent variable takes the value of 1 is 1 if the firm announced a reverse split during the calendar year and 0 otherwise. Panel B and C represent least squares regressions respectively for reverse split magnitude (absolute value of reverse split factor) and reverse split realization delay with the estimated Inverse Mills Ratio (IMR) as explanatory variable. P_Price and Log(size) are respectively the price and the natural logarithm of the market value on trading date [-5], where [0] represents the reverse split announcement date. Rundown is the buy-and-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. Volume is computed as the ratio of the average number of shares traded in the month prior the reverse split announcement to the total quarterly number of outstanding shares before the reverse split announcement. Heteroscedasticityconsistent p-values are provided in parentheses. ***, **,* indicate significance at respectively a 1% , a 5% and a 10% level. Panel A: Probit model for reverse splits Dependent variable: REVSPL Market’s prevent information: P_Price Log(Size) Run-down Volume Constant Diagnostics -0.009 (0.589) -0.234 (0.025)** -0.151 (0.001)*** 0.0.33 (0.852) 1.553 (0.028)** Pseudo R2 13.08% 2 Wald χ test (all coeff.=0) No. of observations 23.07 (0.000)*** 314 Panel B: Second stage regression with reverse split factor Dependent variable: |RSFACT| λrs (Inverse Mills Ratio) Constant Diagnostics -0.215 (0.003)*** 0.991 (0.000)*** R2 F (all coeff.=0) No. of observations 28% 9.39 (0.003)*** 120 Panel C: Second stage regression with reverse split realization delay Dependent variable: Log(Delay) λrs (Inverse Mills Ratio) Constant Diagnostics 0.489 (0.041)** 0.121 (0.529) R2 F (all coeff.=0) No. of observations 6.16% 4.26 (0.041)** 120 Panel A of Table V reports the probit reverse split decision with explanatory variables reduced to Nayak and Prabhala (2001) market preannouncement information set. Only coefficients of Log(size) and Rundown are significant as in Table VI. Panels B and C present adjusted least squares standard error regressions for the reverse split magnitude (absolute value of the reverse split factor)9 and 9 Indeed, given that the split factor is negative in the case of a reverse split, it implies that the larger the reverse split magnitude, the more negative the reverse split factor will be. 73 execution delay. Results reveal significant coefficients of the estimated Inverse Mills Ratio (IMR) negative in reverse split magnitude and positive in execution delay equations. Our results suggest that whether or not a reverse split is announced and conditioning on the market preannouncement information set X irs (including preannouncement price, size, rundown and trading volume), there is negative incremental information in the reverse split magnitude and positive incremental information in the reverse split realization delay. These results reveal that some publicly unobserved characteristics that increase the likelihood of a reverse split contribute to further reduce the reverse split size and increase the reverse split execution delay. If we interpret the unobserved characteristics as private information on the firm, then our results suggest that the firm private information is consistent with smaller reverse split size and longer reverse split execution delay. b) Reverse split size We seek to answer the following question: What determines the choice of the reverse split size? To answer this question we estimate least squares with standard errors adjusted models of reverse split factor. Results are reported in Table V. Table VI: OLS models of reverse split factor This table presents the results of standard errors adjusted least squares regressions explaining the reverse split factors. The dependent variable is the reverse split magnitude measured by the absolute value of the reverse split factor RSFACT = [ S (t ) S (t ')] − 1 , where s (t ) is the number of shares outstanding after reverse the split, and s(t ' ) is the number of shares outstanding before the reverse split. Log(age) is the natural logarithm of the number of days from listing date to reverse split announcement date. Rundown is the buy-and-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. Market-to-book is monthly market-to-book ratio prior to the reverse split announcement. Volume is computed as the ratio of the average number of shares traded in the month prior the reverse split announcement to the total quarterly number of outstanding shares before the reverse split announcement. Earnings Changes=1 if ΔEPS (Earnings Per Share) > 0 around reverse split announcement and Earnings Changes=1 if ΔEPS (Earnings Per Share) > 0 before reverse split announcement. Dirsubstock: equals 1 if directors are subject to stock ownership requirements and 0 otherwise. Dirownership equals 1 if directors and officers ownership as % of shares outstanding is >5% and <=30% (otherwise it is equal to zero). Stockplan takes the value of 1 when the company’s directors are remunerated with options and 0 otherwise. Loansoption takes the value of 1 when the company provides loans to executives for exercising options and 0 otherwise. Incentives Global takes the value of 0 when Dirsubstock and Dirownership and Stockplan and Loansoption take simultaneously the value of 0 and 1 otherwise. Heteroscedasticity-consistent p-values are provided in parentheses. ***, **,* indicate significance at respectively a 1% , a 5% and a 10% level. (1) Firm attributes Log(Age) P_Price Log(Size) Past Performance Rundown Growth opportunities Market-to-book Liquidity Volume Earnings changes =1 if ΔEPS>0 around split =1 if ΔEPS>0 before reverse split Managerial and directors’ incentives Dirsubstock Dirownership Stockplan Loansoption Incentives Global Intercept (2) (3) (4) 0.062 (0.038)** -0.009 (0.000)*** -0.035 (0.002)*** 0.055 (0.083)* -0.006 (0.000)*** -0.040 (0.001)*** 0.064 (0.981)** -0.010 (0.000)*** -0.035 (0.002)*** 0.062 (0.037)** -0.009 (0.000)*** -0.035 (0.002)*** -0.005 (0.353) -0.005 (0.369) -0.005 (0.271) -0.005 (0.365) 0.0005 (0.688) 0.001 (0.374) 0.001 (0.289) 0.0006 (0.677) 0.059 (0.003)*** 0.053 (0.008)*** 0.062 (0.001)*** 0.059 (0.003)*** 0.017 (0.338) 0.019 (0.273) 0.111 (0.013)** -0.017 (0.311) 0.023 (0.324) -0.002 (0.885) -0.005 (0.783) 0.843 (0.000)*** 0.329 (0.000)*** 0.844 (0.000)*** 0.841 (0.000)*** 55.33% 14.92 (0.000)*** 118 42.99% 18.59 (0.000)*** 113 57.32% 24.10 (0.000)*** 118 55.36% 12.62 (0.000)*** 118 Diagnostics R2 F (all coeff.=0) No. of observations Model 1 of Table VI focuses on the impact of a reverse splitting firm attributes (pre-announcement 74 price, age, and size) past market performance, growth opportunities and liquidity on the choice of the reverse split magnitude. Model 2 adds earnings changes while models 3 and 4 introduce managers’ and directors’ incentives variables. Pre-announcement price and firm size exhibit negative and strongly significant coefficients in all specifications. These results suggest that low preannouncement prices lead to large reverse split magnitude. It further reveals that smaller firms are more likely to choose larger reverse split magnitudes. Moreover the age coefficient is positive and significant at the 5% level in all specifications except in Model 2 (10% level). This suggests that younger firms are more likely to prefer small reverse split magnitude than the older ones, probably by prudence. We also find a strong and positive impact of preannouncement trading volume on the magnitude of the reverse split in all specifications. Indeed, the low volume preannouncement firms will likely be preoccupied with getting out of the “penny stock” range and attract institutional investor. These firms will avoid large reverse split magnitude to keep small investors holding their stock. Finally, we observe that controlling for price level and prospect indicators, firms will choose higher reverse split magnitude to preserve investors’ perception and the “image” of the company in general, when directors are subject to stock ownership requirements. However, we find no evidence of a global impact of managers’ and directors’ incentives, past performance and growth opportunities on the choice of the reverse split magnitude. c) Timing of reverse split realization The main question we address in this section is: What determines the choice of the split execution delay? We investigate the effect of firm attributes, past market performance, growth opportunities and managers’ and directors’ incentives on the split realization delay. We define Log ( Delay ) as the natural logarithm of days elapsing from announcement date to effective date for firm i . Rundowni is the past performance, MB the market-to-book, Incentives is the managers' and directors’ incentives, C is the control variables including firm’s attributes, and the error term ωi is assumed to be normally distributed and independent of Rundown, MB and Incentives. We estimate the following empirical model Log ( Delay )i = δ1 + δ 2 Rundowni + δ 3 MBi + δ 4 Incentivesi + δ 5Ci + ωi (9) Results of adjusted standard error least squares from equation (9) are presented in Table VII. Table VII: OLS models of reverse split realization delay This table presents the results of standard errors adjusted least squares regressions explaining the timing of reverse split realization. The dependant variable is the log(days from announcement date to effective date). The reverse split magnitude is measured by the absolute value of the reverse split factor RSFACT = [ S (t ) S (t ')] − 1 , where s (t ) is the number of shares outstanding after the reverse split, and s (t ' ) is the number of shares outstanding before the reverse split. Log(age) is the natural logarithm of the number of days from listing date to reverse split announcement date. Rundown is the buyand-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. Market-to-book is monthly market-to-book ratio prior to the reverse split announcement. Volume is computed as the ratio of the average number of shares traded in the month prior the reverse split announcement to the total quarterly number of outstanding shares before the reverse split announcement. Earnings Changes=1 if ΔEPS (Earnings Per Share) > 0 around reverse split announcement and Earnings Changes=1 if ΔEPS (Earnings Per Share) > 0 before reverse split announcement. Dirsubstock: equals 1 if directors are subject to stock ownership requirements and 0 otherwise. Dirownership equals 1 if directors and officers ownership as % of shares outstanding is >5% and <=30% (otherwise it is equal to zero). Stockplan takes the value of 1 when the company’s directors are remunerated with options and 0 otherwise. Loansoption takes the value of 1 when the company provides loans to executives for exercising options and 0 otherwise. Incentives Global takes the value of 0 when Dirsubstock and Dirownership and Stockplan and Loansoption take simultaneously the value of 0 and 1 otherwise. Heteroscedasticity-consistent p-values are provided in parentheses. ***, **,* indicate significance at respectively a 1% , a 5% and a 10% level. 75 (1) Reverse split factor |RSFACT| Firm attributes Log(Age) P_Price Log(Size) Past Performance Run-down Growth opportunities Market-to-book Liquidity Volume Earnings changes =1 if ΔEPS>0 around split =1 if ΔEPS>0 before reverse split Managerial and directors’ incentives Dirsubstock Dirownership Stockplan Loansoption Incentives Global Intercept (2) (3) (4) 0.960 (0.095)* 0.822 (0.175) 0.062 (0.009)*** 0.009 (0.470) 0.279 (0.000)*** 0.444 (0.018)** 0.012 (0.431) 0.262(0.001)*** 0.505 (0.004)*** -0.007 (0.463) 0.228 (0.002)*** 0.493 (0.003)*** 0.002 (0.819) 0.232 (0.001)*** -0.046 (0.067)* -0.044 (0.099)* -0.050 (0.081)* -0.053 (0.039)** 0.0003 (0.980) 0.0006 (0.964) 0.004 (0.742) 0.0003 (0.980) -0.019 (0.891) -0.026 (0.851) 0.062 (0.744) 0.026 (0.849) 0.025 (0.824) 0.054 (0.600) 0.805 (0.002)*** -0.059 (0.612) 0.159 (0.312) 0.085 (0.510) 0.119 (0.229) -3.660 (0.000)*** -3.482(0.000)*** -2.772 (0.001)*** -2.795 (0.000)*** 28.50% 6.15 (0.000)*** 118 27.85% 4.80 (0.000)*** 113 30.88% 12.26 (0.000)*** 118 27.71% 5.77 (0.000)*** 118 Diagnostics R2 F (all coeff.=0) No. of observations The first two models add the reverse split magnitude to the previously described explanatory variables. The reverse split execution delay is not significantly determined by the magnitude of the chosen factor. Results from the four models reveal strong and positive age and size effects on the timing of the reverse split realization. This indicates that smaller or younger firms are more likely to realize the announced reverse split early. Indeed, these firms are more subject to information asymmetry (see, e.g. Zeghal, 1984 and Freeman, 1987) and a long waiting period can be negatively interpreted by institutional investors that these firms want to attract. Moreover, small firms are more affected by small prices and will not wait long to bring the stock price out of the “penny stock” range to a more attractive trading range. Contrary to our expectations, we find that when directors are subject to stock ownership requirements, the split execution delay is strongly and positively affected, probably because directors’ oversight is greater and then sets back the execution of the reverse split. This means that the reverse split execution delay is longer when directors are subject to ownership requirements. Nevertheless, managers’ and directors’ incentives taken globally do not affect the timing of reverse split realization contrarily to the timing of split realization (Beaulieu and Sodjahin, 2007). Past performance, growth opportunities and liquidity do not affect significantly the reverse split execution delay either. d) Reverse split factor, execution delay and probability of survival This section investigates whether the reverse split magnitudes or the reverse split execution delay can predict the survival of the reverse splitting firms. The influence of managers’ and directors’ incentives on the probability of survival is also examined. We define Survival to take the value of 1 when the ending dates10 in the CRSP data base (variable ENDDT) occur within 12 months from the reverse split effective dates and 0 otherwise. |RSFACT| is the reverse split magnitude, Log(Delay) is the natural logarithm of days from announcement date to effective date, Incentives is the 10 This date corresponds to the end of the stock database. 76 managers’ and directors’ incentives, C is the control variables including firm’s attributes but also profitability and leverage since reasonably more profitable and low leverage firms are more likely to survive. To study the link between the likelihood of survival and reverse split factor and execution delay, we use a probit regression model: Survivali = η1 + η2 RSFACT i + η3 Log ( Delay )i + η4 Incentives + η5Ci + υi (10) The error term υ is assumed to be normally distributed and independent of the explanatory variables. The empirical results are presented in Table VIII below. Table VIII: Probit models of reverse splitting firms’ survival This table presents the results of robust probit regressions explaining the probability of reverse splitting firms’ survival. The dependant variable Survival that takes the value 1 when the ending dates in CRSP data base (variable ENDDT) occur after 12 months from the reverse split effective dates and 0 otherwise. The reverse split magnitude is measured by the absolute value of the reverse split factor RSFACT = [ S (t ) S (t ')] − 1 , where s (t ) is the number of shares outstanding after the reverse split, and s(t ' ) is the number of shares outstanding before the reverse split. Log(Delay) is the natural logarithm of days from announcement date to effective date. Log(age) is the natural logarithm of the number of days from listing date to reverse split announcement date. Rundown is the buy-and-hold return from one year before the reverse split announcement through five trading days prior to the reverse split announcement. Market-tobook is monthly market-to-book ratio prior to the reverse split announcement. Volume is computed as the ratio of the average number of shares traded in the month prior the reverse split announcement to the total quarterly number of outstanding shares before the reverse split announcement. ROA is the preannouncement return on assets. Total debt/total Assets is defined as the sum of Long-Term Debt and Debt in Current Liabilities, divided by Total Assets. Dirsubstock: equals 1 if directors are subject to stock ownership requirements and 0 otherwise. Dirownership equals 1 if directors and officers ownership as % of shares outstanding is >5% and <=30% (otherwise it is equal to zero). Stockplan takes the value of 1 when the company’s directors are remunerated with options and 0 otherwise. Loansoption takes the value of 1 when the company provides loans to executives for exercising options and 0 otherwise. Incentives Global takes the value of 0 when Dirsubstock and Dirownership and Stockplan and Loansoption take simultaneously the value of 0 and 1 otherwise. Heteroscedasticity-consistent p-values are provided in parentheses. ***, **,* indicate significance at respectively a 1% , a 5% and a 10% level. Reverse split factor |RSFACT| Realization delay Log(Delay) Firm attributes Log(Age) P_Price Log(Size) Past Performance Run-down Growth opportunities Market-to-book Liquidity Volume Profitability ROA Leverage Total debt/total Assets Managerial and directors’ incentives Dirsubstock Dirownership Stockplan Loansoption Incentives Global Intercept Model 1 Model 2 Model 3 Model 4 4.524 (0.007)*** 6.284 (0.000)*** 7.252 (0.000)*** 6.488 (0.000)*** -0.883 (0.001)*** -1.010 (0.001)*** -0.992 (0.001)*** -1.022 (0.001)*** -0.476 (0.329) 0.016 (0.595) 0.503 (0.015)** - 0.745 (0.073)* - 0.612 (0.198) - 0.511 (0.285) 0.618(0.003)*** 0.850 (0.000)*** 0.735 (0.000)*** 0.008 (0.001)*** 0.007 (0.005)*** 0.005 (0.026)** -0.005 (0.189) -0.008 (0.032)** -0.007 (0.063)* -0.050 (0.553) 0.034 (0.258) 1.945 (0.093)* 2.779 (0.103) -1.462 (0.035)** -0.041 (0.907) 0.056 (0.899) -0.906 (0.016)** -0.806 (0.022)** -4.346 (0.074)* -4.911(0.079)* 21.04% 23.88 (0.002)*** 118 28.33% 26.68 (0.000)*** 121 -6.873 (0.035)** -5.838 (0.057)** Diagnostics Pseudo R2 2 Wald χ test (all coeff.=0) No. of observations 28.54% 41.14 (0.000)*** 123 26.28% 30.44 (0.000)*** 123 In full accordance with Hypothesis 4, Table VIII reveals a strong and positive impact of the reverse split magnitude and of reverse split execution delay in the four models presented. These results 77 indicate that the likelihood to survive 12 months after the reverse split execution is positively related to the reverse split magnitude chosen. Indeed, firms with larger reverse split magnitude are more likely to survive since their post reverse split price is high enough to attract institutional investors and to meet the exchange listing requirements. Moreover, the probability to survive 12 months after the reverse split realization is negatively related to how long firms wait to execute the reverse split. Intuitively, firms that wait long to get their price out of the “penny stock” range will be less likely to survive even though the reverse execution delay may have positive incremental information as we found previously. Also, the results in Table VIII logically suggest that large firms and more profitable firms are more likely to survive 12 months after their reverse split. In fact, firm size and ROA exhibit strong positive and significant coefficients in all specifications. Furthermore, managers’ and directors’ incentives, contrary to our expectations (Hypothesis 4), negatively affect the survival of reverse splitting firms. This is confirmed by the negative and statistically significant coefficient of Overall Incentives. More specifically, results indicate that firms that grant loans to exercise stock options will be less likely to survive. Indeed, granting loans to exercise stock options could be burdensome especially for reverse splitting firms which have weak growth opportunities (see table IV). Moreover, there is a negative effect of directors stock ownership requirements (see variable Dirsubstock) on the reverse splitting firms’ survival. In fact, directors’ ownership could make them overbearing, particularly when the firm is not doing well, and as noted by Holmstrom (2005): “Powerful boards can be disastrous for a company…” and also Song and Thakor (2007): “greater board oversight can produce the opposite of its intended effect”. This might also explain why longer delays in the reverse split realization leads to a lower probability of survival. It might result from management having to deal with an overbearing board. V- Conclusion In this paper, we investigate the factors that separately determine the reverse split decision, the choice of reverse split factor and the reverse split execution delay. We also examine whether reverse split factors and reverse split execution delays can help predict the reverse splitting firms’ survival and how managers’ and directors’ incentives can affect the likelihood of a firm to survive a reverse split. The results reveal that the reverse split decision is essentially motivated by unusual run-downs in stock price and poor growth opportunities. These results are consistent with the signaling hypothesis advocated in the literature (see Spudeck and Moyer, 1985 among others). We observe that beside the fact that whether a reverse split is announced or not and based on the market preannouncement information set including preannouncement price, size, rundown and trading volume, there is negative incremental information in reverse split factors and positive incremental information in reverse split realization delays. Moreover, the choice of the reverse split factor is determine by the size of the firm and the preannouncement price level. In fact, low preannouncement prices lead to large reverse split magnitude. Also large firms are more likely to prefer small reverse split magnitudes. Furthermore, younger firms are more likely to prefer a small reverse split magnitude compare to the older ones, probably by prudence. Low preannouncement trading volume firms will be more preoccupied with getting out of the “penny stock” range and avoid large reverse split size so not to discourage small investors from holding their stock. Controlling for price level and prospect indicators, firms are more preoccupied by investor’s perception and their image in general and will choose higher reverse split size when directors are subject to stock ownership requirements. Our empirical results also reveal that smaller firms or younger firms are more likely to realize the announced reverse split earlier. Contrary to our expectations, we find that when directors are subject to stock ownership requirements, the split execution delay is longer (strong and positive impact), probably because directors’ oversight is greater and then sets back the execution of the announced reverse split. 78 Lastly, results of this study indicate that the likelihood to survive 12 months after the reverse execution is positively related to the chosen reverse split magnitude. Indeed, firms with larger reverse split sizes are more likely to survive given that their post reverse split price is high enough to attract institutional investors and to meet listing requirements. Moreover, the probability to survive 12 months after the reverse execution is negatively related to how long firms wait to make the reverse split effective. Intuitively, firms that wait a long time to get their price out of the “penny stock” range will be less likely to survive. We also find that for reverse splitting firms, granting loans to exercise stock options could be burdensome and directors’ ownership could make them overbearing and lead to disastrous consequences (Song and Thakor, 2007) which reduce the probability of survival. References Beaulieu, M-C., and W.R. Sodjahin, 2007, “To wait or not to wait: When do announced splits be made effective?,” Working paper, Université Laval. Charitou, A., Louca, C., and N., Vafeas, 2007, “Boards, ownership structure, and involuntary delisting from the New York Stock Exchange,” Journal of Accounting and Public Policy 26, 2, 249262. Daniel, K., and S. Titman, 1999, “Market efficiency in an irrational world,” Financial Analysts Journal 55, 28-40. Del Guercio, D., 1996, “The Distorting Effect of the Prudent-Man Laws on Institutional Equity Investment,” Journal of Financial Economics, 40, 31-62. Desai, H. and Jain, P.C., 1997, “Long-run common stock returns following stock splits and reverse splits,” Journal of Business 70,3, 409-433. D’Mello, R., and S. P. Ferris, 2000, “The information effects of analyst activity at the announcement of new equity issues,” Financial Management 29, 78-95. Freeman, R., 1987, “The association between accounting earnings and security returns for large and small firms,” Journal of Accounting and Economics 9, 195-228. Greene, William, 2003, “Econometric Analysis,” Fifth Edition. Prentice Hall. Gompers, P., and A. Metrick, 2001, “Institutional Investors and Equity Prices,” Quarterly Journal of Economics, 116, 229-259. Han, K.,1995, “The effects of reverse splits on the liquidity of the stock,” Journal of Financial and Quantitative Analysis, 30, 159–169. Heckman, J., 1979, “Sample selection bias as a specification error,” Econometrica 47, 153-161. Holmstrom, Bengt, 2005, “Pay without performance and the managerial power hypothesis: a comment,” Journal of Corporation Law 30, 703-713. Ikenberry, D., G., Rankine, and E. K., Stice, 1996, “What do stock splits really signal?,” The Journal of Financial and Quantitative Analysis 31,3, 357-375. 79 Lamoureux, C.G. and Poon, P., 1987, “The market reaction to stock splits,” The journal of finance 42, 5, 1347-1370. Masse, I., Hanrahan, J. R., and Kushner, J., 1997, “The effect of Canadian stock splits, stock dividends, and reverse splits on the value of the firm,” Quarterly Journal of Business and Economics, 36, 51–63. McNichols,M. and A., Dravid, 1990, “Stock dividends, stock splits and signaling,” Journal of Finance 45, 857-880. Nayak, S. and Prabhala, N.R., 2001, “Disentangling the dividend information in splits: a decomposition using conditional event-study methods,” The Review of Financial Studies 14, 4, 1083-1116. Peterson, D. R., and Peterson, P. P., 1992, “A further understanding of stock distributions: The case of reverse stock splits,” Journal of Financial Research, 15, 189–205. Robinson, D., 2007, “The information content of reverse stock splits,” Advances in Accounting 23, 1-314. Song, Fenghua, and Anjan Thakor, 2007, “Intrinsically-motivated CEOs, overbearing boards, and diversity in corporate governance,” Working paper, Washington University in St. Louis. Spudeck, R. E., and Moyer, R. C., 1985, “Reverse splits and shareholder wealth: The impact of commissions,” Financial Management, 14, 52–56. West, R. R., and Brouilette, A. B., 1970, “Reverse stock splits…harbinger of bad times or valid management technique,” Financial Executive, 38, 12–17. Woolridge, R., and Chambers, D., 1983, “Reverse splits and shareholder wealth,” Financial Management, 12, 5–15. Zeghal, D.,1984, “Firm size and the information content of financial statements,” The Journal of Financial and Quantitative Analysis 19,3, 299-310. 80 ASAC 2008 Halifax, Nouvelle-Écosse Salma Ben Amor Étudiante, Ph.D Administration ESG, UQAM Nabil Khoury Marco Savor MODÈLE PRÉVISIONNEL DE LA DÉFAILLANCE FINANCIÈRE DES PME QUÉBÉCOISES EMPRUNTEUSES La présente recherche se démarque des études antérieures portant sur le risque de la défaillance financière en se concentrant sur les PME québécoises non cotées. Notre objectif est de vérifier le degré d’adéquation des modèles existants avec ce type de contexte et de proposer, le cas échéant, une méthodologie d’analyse. Introduction Parvenir à détecter aussi tôt que possible la défaillance financière est une problématique qui retient l’attention du monde scientifique depuis plusieurs années. La lourdeur des conséquences de cet événement non seulement sur toute l’entreprise, mais aussi sur les bailleurs de fonds rappelle vivement la nécessité de le prévoir. Dans ce contexte, depuis les années soixante, de nombreuses études empiriques se sont focalisées sur la prévision de la défaillance financière des entreprises. Ces études se fondent sur l’analyse économique et financière d’entreprises défaillantes et d’entreprises non défaillantes afin de déterminer les variables, principalement les ratios financiers, qui distinguent au mieux ces deux groupes. À cet égard, la définition de la défaillance prend toute son importance puisqu’elle est à la base du choix des ratios financiers qui la caractérisent et de l’élaboration des modèles qui la prédisent. Certains auteurs identifient la défaillance avec l’amorce d’une procédure judiciaire. Cependant, d’autres chercheurs considèrent comme « défaillante » toute entreprise qui a connu un défaut de paiement. Wruck (1990), par exemple, définit la défaillance financière comme étant la situation où les cash-flows sont insuffisants pour couvrir les obligations courantes. Dans la même veine, selon John (1993), une entreprise est dite « défaillante » si les actifs liquides ne sont pas suffisants pour honorer les engagements financiers immédiatement exigibles. Somme toute, on peut définir la défaillance financière comme un manque de liquidité ou une insolvabilité temporaire qui peut être résolue par une restructuration de l’actif ou du passif. L’objet commun des modèles de prévision de la défaillance financière est de tenter de classer une entreprise quelconque dans l’un des deux groupes suivants : les entreprises défaillantes et les entreprises non défaillantes. Pour se faire, une variété de techniques statistiques ont été utilisées par les chercheurs. L’approche unidimensionnelle de Beaver (1966) semble être le point de départ de plusieurs études empiriques publiés. Il s’agit d’une classification dichotomique fondée sur un ratio unique. L’objectif de Beaver (1966) est de classer les entreprises sur la base du ratio le plus discriminant. Bien que cette méthode fournit un indicateur à la fois simple et efficace, le manque de robustesse lié à l’unicité du ratio utilisé explique l’orientation des recherches vers les analyses multidimensionnelles permettant une description plus riche de la situation de l’entreprise. Dans ce cadre, Altman (1968), Edmister (1972), Altman, 81 Haldeman et Narayanan(1977), Altman et Lavallée (1980) ont eu recours à l’analyse discriminante multidimensionnelle. Cette méthode aboutit à la construction d’une fonction appelée score, qui est une combinaison linéaire des variables explicatives retenues. À titre d’exemple, Altman et Lavalée(1980) utilisent cinq ratios financiers dans leur analyse des entreprises canadiennes pour construire leur score, alors que Altman, Haldeman et Narayanan(1977) ont construit leur modèle en combinant linéairement sept ratios pour prévoir la défaillance financière d’un échantillon d’entreprises américaines. La classification des entreprises en défaillantes et non défaillantes se fait alors sur la base de la valeur calculée de ce score. Bien que cette méthode ait connu une grande popularité, son application requiert des conditions statistiques strictes à savoir la normalité des variables comptables utilisées et l’homogénéité de la matrice variance- covariance entre les deux groupes d’entreprises. À cet égard, R. Eisenbeis (1977) montre que la capacité prédictive de ces modèles peut être sévèrement affectée par le caractère non normal des variables. Partant de ce constat, certains auteurs ont préféré recourir à d’autres méthodes qui supposent une distribution différente des variables comptables. Ohlson (1980) a appliqué la régression logistique à la prévision de la défaillance financière. Dans cette méthode, la variable endogène est une variable binaire qui prend la valeur 0 ou 1, selon que l’entreprise est défaillante ou non. Elle présente l’avantage qu’elle permet de combiner plusieurs variables indépendantes sans que l’hypothèse de normalité soit une condition nécessaire. À partir des années 90, d’autres études ont eu recours à des méthodes statistiques non paramétriques comme les réseaux de neurones artificiels et les algorithmiques génétiques. La première application des réseaux de neurones à l’estimation du risque de défaillance financière a été réalisée par Bell et Alii (1990). L’utilisation de cette technique s’est ensuite intensifiée avec les travaux de Tam (1991), Tam et Kiang (1992) et Altman et al (1994). Au terme d’une étude critique de ces modèles, on peut constater qu’ils ont été développés principalement sur la base de données issues de grandes entreprises. Ils font abstraction des petites et moyennes entreprises bien que celles-ci soient plus concernées par ce risque. De plus, très peu de ces études ont traité spécifiquement le cas des PME québécoises. Dans ce cadre, Veronnault et Legault (1991) ont développé un modèle de prévision appelée CA score sur la base d’un échantillon de 173 PME québécoises dont la moitié étaient des entreprises saines et l’autre moitié des faillies. Leur modèle qui s’appuie sur trois ratios financiers a réussi à classer correctement 85,71% des entreprises une année avant la faillite. Cependant, ce modèle est applicable exclusivement aux PME manufacturières québécoises. Par ailleurs, les particularités financières et organisationnelles des PME québécoises doivent nous convaincre de l’importance d’aller plus loin dans cette voie de recherche et de développer des modèles qui cadrent avec ce contexte particulier. Ces entreprises sont au cœur du développement économique de la région. Elles sont responsables de prés de la moitié du produit intérieur brut québécois et emploient plus de 50% de la main d’œuvre québécoise. Ainsi, les gouvernements, tant canadien que québécois accordent une attention particulière aux intérêts de ces entreprises, surtout en matière de financement. Tel que discuté dans une étude antérieure1, 15% de financement par capital de risque reçu par les PME québécoises en 2004 provenait de sources gouvernementales alors que ce pourcentage était seulement 9% pour l’ensemble des PME canadiennes. L’enquête présentée ici qui s’inscrit dans ce cadre de recherche. Elle se démarque des études antérieures par la nature des données utilisées. L’objet de notre étude est donc d’évaluer le risque de défaillance financière des PME québécoises, de vérifier le degré d’adéquation des modèles existants avec ce type de contexte et de proposer, le cas échéant, une méthodologie 1 N. Khoury, M. Savor et R. Toffoli 2006, « La couverture des risques financiers par les PME québécoises », Revue Canadienne des sciences de l’administration 23(1), 67-80. 82 d’analyse qui permet d’anticiper le plus adéquatement possible la santé financière future des PME québécoises. La suite de cet article s’articule comme suit : dans un premier temps, nous présentons une brève description des particularités des PME québécoises. Nous exposons ensuite la méthodologie et décrivons l’échantillon des PME qui fait l’objet de l’étude. L’analyse des résultats empiriques et leur comparaison avec ceux des études canadiennes sur le sujet feront l’objet de la partie suivante. L’étude s’achève par une conclusion qui souligne ses principales contributions à la littérature. Méthodologie et description des données Définition de l’échantillon Grâce à la collaboration de la Fédération des Caisses Desjardins, nous avons pu obtenir une base de données qui comprend 56768 demandes de financement reçues en 2005, 2006 et 2007 de la part de PME Québécoises. De ce nombre, 23166 PME ont réussi à obtenir un financement en 2005, 29895 autres PME l’ont obtenu en 2006 et 3707 PME l’ont obtenu en 2007. Le nombre total des PME défaillantes dans la base de données s’élève à 1063 (1,87% de l’ensemble des données). Selon la Fédération des Caisses Desjardins, est considérée comme défaillante toute entreprise qui manque à ses engagements en vertu de son contrat d’endettement pour une période de trois mois consécutifs. À partir de la base de données disponible, nous avons pu retenir pour l’analyse 165 entreprises défaillantes et 6709 entreprises non défaillantes dont les données comptables sont disponibles en 2005, ainsi que 470 entreprises défaillantes et 18192 entreprises non défaillantes ayant des données comptables disponibles en 2004. Ces entreprises réalisent un montant total des ventes qui varie entre – 6192$ et 20 millions$ en 2005, et entre – 14000$ et 20 millions $ en 2004. Pour construire l’échantillon final, nous avons procédé à un appariement en faisant correspondre à chaque entreprise défaillante une entreprise non défaillante de même taille. Après l’élimination de certains états financiers contenant des données aberrantes, la structure finale de l’échantillon se compose d’un sous groupe de 303 PME dont les données comptables sont disponibles en 2005 et d’un autre sous groupe de 330 dont les données comptables sont disponibles en 2004. Pour les besoins de l’analyse et afin de se conformer à la composition et à l’étendue de la base des données, nous avons considéré qu’une PME est défaillante si la cessation de paiements dans le cadre de son emprunt intervient en 2006. Selon cette définition, 141 PME de l’échantillon de 2005 et 165 PME de l’échantillon de 2004 peuvent être considérées comme étant en défaut de paiement en 2006. Afin de mettre ces chiffres en perspective, soulignons que le nombre total des PME en faillite au Québec s’élevait à 1837 entreprises en 20062 (soit 0,38% de la population). Cependant, il importe de souligner que ce nombre correspond aux PME qui ont atteint la faillite totale selon la définition de Statistique Canada. Cette définition ne correspond toutefois pas à la définition attribuée aux PME défaillantes par la Fédération Des Caisses Desjardins ce qui rend inappropriée la comparaison entre cette statistique et celle issue de notre base de données. Les données comptables utilisées dans cette étude concernent deux années avant la date de défaillance à savoir les années 2004 et 2005. Le tableau 1 résume la composition de notre échantillon. 2 Source : Statistique Canada, CANSIM, tableau 177-0006. 83 Tableau 1 : nombre d’observations par année d’étude 2005 (année (-1)) 2004 (année (-2)) entreprises défaillantes 141 165 entreprises non défaillantes 162 165 nombre total d’observations 303 330 Méthodologie L’étude effectuée sur notre échantillon se fera en trois étapes. Nous procédons, dans un premier temps à une analyse à l’aide des modèles théoriques de prévision de la défaillance financière les plus connus dans la littérature afin d’établir un étalon de comparaison pour nos résultats subséquents. À cette fin, nous procèderons à une application du modèle d’Altman (1968) ainsi que de celui de Veronnault et Legault (1991) sur notre échantillon pour les deux années d’étude. Les résultats de ces deux modèles, bien connus des praticiens, serviront ensuite de base de comparaison pour le modèle que nous développons. Il est important de noter que l’application de ces deux méthodes analytiques ne représente en aucune façon une vérification empirique de leur validité. En effet, une telle vérification aurait nécessité le développement de variables et de paramètres propres à notre échantillon. Notre objectif étant limité à l’obtention d’un point de comparaison pour l’utilité pratique de nos résultats originaux, nous nous sommes limité à l’application des variables et des coefficients de ces deux modèles à titre de référence. Dans un deuxième temps, nous procédons au développement d’un modèle qui va nous permettre d’avoir une information plus pertinente sur le degré de vulnérabilité des PME québécoises au risque de défaillance. Notre objectif est d’essayer de trouver le meilleur outil de prévision de la défaillance financière qui permet de traduire le plus convenablement possible les particularités des PME québécoises. À cette fin, nous avons choisi d’appliquer la régression logistique puisqu’elle n’exige pas la normalité des variables en plus de sa plus grande robustesse telle que démontrée par Kira et al (1997) et par Lennox (1999). Les variables explicatives du modèle que nous développons sont des ratios financiers qui traduisent la liquidité, la solvabilité et la profitabilité des entreprises de notre échantillon. Il s’agit là des trois dimensions considérées par les analystes financiers comme les plus représentatives de la santé financière des entreprises. À partir des données comptables disponibles, nous avons pu ainsi retenir 19 ratios financiers sélectionnés parmi ceux qui ont servi dans les travaux empiriques antérieurs. Une seconde sélection est ensuite effectuée sur les ratios ainsi calculés afin de ne conserver que ceux qui s’avèrent les plus significatifs, c'est-à-dire ceux qui discriminent le mieux entre les entreprises défaillantes et non défaillantes. Pour se faire, un test « t » de différence de moyennes est appliqué. L’application de la régression logistique doit être précédée par une analyse de la corrélation entre les variables explicatives étant donné que cette technique nécessite que les variables indépendantes soient non corrélées. Sur la base du résultat du test de corrélation, nous construisons plusieurs modèles à partir d’une combinaison linéaire des ratios non corrèles qui peuvent servir à une prévision plus précise de la défaillance financière des PME québécoises. Le choix du meilleur modèle sera basé sur sa capacité prédictive, son niveau de précision dans la classification et le degré de signification des coefficients de la régression. Ces modèles sont estimés en se basant sur un échantillon d’estimation alors que leur capacité prédictive est testée sur un échantillon de contrôle sélectionné d’une manière aléatoire. 84 À la troisième et dernière étape, nous avons essayé de détecter l’effet de l’industrie sur l’analyse de la défaillance financière des entreprises faisant partie de secteurs d’activités bien particuliers. Plus précisément, nous avons réestimé le modèle retenu à l’étape précédente en utilisant les données comptables d’entreprises appartenant aux deux secteurs d’activité déclarant le plus grand pourcentage de défaillance à savoir le secteur des services et le secteur du commerce en détail. Cette étape nous permet d’analyser le degré de signification des ratios financiers utilisés dans le contexte d’un secteur bien particulier. Résultats empiriques Applicabilité des modèles théoriques dans le contexte des PME québécoises Étude d’Altman (1968). En utilisant l’analyse discriminante linéaire, Altman (1968) a développé un modèle de prévision de la défaillance financière à partir d’une combinaison linéaire de cinq ratios financiers. Appliqué sur un échantillon de 66 entreprises américaines dont la valeur de l’actif varie entre 0,7 million $ et 25,9 million $, ce modèle a réussi à classer correctement 95% des firmes faillies une année avant la défaillance. L’équation du modèle est la suivante : Z = 0.012 X1 + 0.014 X2 + 0.033 X3 + 0.006 X4 + 0.999 X5 X1 : Fonds de roulement/ actif total X2 : Bénéfice non repartis / actif total X3 : Bénéfice avant impôts et intérêts / actif total X4 : Avoir des actionnaires/ passif total X5 : Ventes / actif total Z : Score discriminant L’application du modèle est simple. Pour prévoir l’état futur d’une entreprise, il suffit de calculer les ratios X1 – X5 et de déterminer ensuite la valeur du Z correspondante. Pour distinguer les deux catégories de firmes, Altman a utilisé un point de coupure moyen de 2,675. En appliquant ce modèle à notre échantillon, nous obtenons les résultats de classification présentés aux tableaux suivants, rappelons encore une fois que l’objet de cette classification n’est pas de vérifié la validité du modèle analytique d’Altman avec nos données mais de fournir tout simplement un point de référence pour mettre en perspective l’utilité pratique des résultats de notre modèle. Tableau 2 : Application du modèle d’Altman (1968) sur les données comptables de 2005 de notre échantillon CLASSIFICATION DU MODÈ D’ALTMAN Défaut Non défaut Défaut 84 57 Non défaut 109 52 CLASSIFICATION D’ORIGINE Défaut % DE BONS CLASSEMENT Non défaut 141 59,57 162 32,3 45,03 85 Tableau 3 : Application du modèle d’Altman (1968) sur les données comptables de 2004 de notre échantillon CLASSIFICATION DU MODÈ D’ALTMAN Défaut Non défaut Défaut 107 58 Non défaut 108 57 CLASSIFICATION D’ORIGINE Défaut % DE BONS CLASSEMENT Non défaut 165 64,85 165 34,54 49,69 Afin d’évaluer l’efficacité de la méthode utilisée, nous procédons au calcul du taux de bons classements qu’elle procure. Le taux de bons classements des entreprises non défaillantes correspond au pourcentage des entreprises non défaillantes correctement classées dans l’ensemble des entreprises non défaillantes considérées. Le taux de bons classements des entreprises défaillantes correspond au pourcentage des entreprises défaillantes correctement classées dans l’ensemble des entreprises défaillantes considérées. Une mesure globale est également envisageable : la moyenne des taux de bons classements pondérés par l’effectif respectif de chaque sous-échantillon. Comme les deux tableaux ci –dessus le montre, les résultats obtenus ne soutiennent pas la performance empirique du modèle d’Altman (1968) lorsqu’on l’applique aux données de PME québécoises. En effet, le modèle n’a réussi à classer correctement que 45,03% des entreprises une année avant la défaillance et 49,69% des firmes deux années avant la défaillance. Comme on le sait, Altman (1968) a développé son équation discriminante en utilisant un échantillon d’entreprises américaines opérant principalement dans le secteur manufacturier. Ceci pourrait expliquer la faiblesse des résultats obtenus compte tenu de la divergence existante entre l’environnement économique et organisationnel des entreprises américaines et celui des PME québécoises. L’utilisation d’un modèle de prévision qui se rapproche davantage de notre contexte d’étude pourrait améliorer les résultats de classification. Étude de Veronnault et Legault (1991). Veronnault et Legault (1991) se sont penchés sur un échantillon de petites entreprises québécoises manufacturières afin de développer leur modèle de prévision de la défaillance financière. L'équation algébrique de leur modèle est la suivante: CA score = 4,5913* R1 + 4,5080* R2 + 0,3936*R3 - 2,7616 R1 : Avoir des actionnaires / actif total R2 : Bénéfice avant impôts et intérêts/ actif total R3 : Ventes / actif total Pour les fins d’analyses, Veronnault et Legault (1991) considèrent qu’un score négatif ou proche de -0,3 est un signal d’alarme indiquant l’apparence de graves difficultés financieres. En utilisant -0,3 comme point de coupure, les taux de bons classements des entreprises de notre échantillon, calculés selon la formule pré-citée, sont présentés dans les tableaux suivants : 86 Tableau 4 : Application du modèle de Veronnault et Legault (1991) sur les données comptables de 2005 de notre échantillon CLASSIFICATION DU MODÈ CLASSIFICATION DE VERONNAULT ET D’ORIGINE LEGAULT Défaut Non défaut Défaut Non défaut Défaut 36 105 Non défaut 48 113 141 % DE BONS CLASSEMENT 25,53 162 70,19 49,34 Tableau 5 : Application du modèle de Veronnault et Legault (1991) sur les données comptables de 2004 de notre échantillon CLASSIFICATION DU MODÈ CLASSIFICATION DE VERONNAULT ET D’ORIGINE LEGAULT Défaut Non défaut Défaut Non défaut Défaut 87 78 Non défaut 48 117 165 % DE BONS CLASSEMENT 52,72 165 70,91 61,82 Comme les tableaux 6 et 7 le montre, il y a une amélioration dans les pourcentages de bons classements une année et deux années avant la défaillance, et ce surtout pour les entreprises non défaillantes. Par contre, le pourcentage de bons classements des entreprises défaillantes s’est détérioré ce qui est non souhaitable étant donné les coûts énormes qui en découlent. Bien que ce modèle soit plus proche de notre environnement d’étude, les résultats obtenus semblent être moins performants que ceux du modèle d’Altman(1968). La détection précoce de la défaillance financière : un modèle suggéré Résultats de l’estimation pour une année avant la défaillance. Tel qu’indiqué précédemment, les variables explicatives retenues pour notre modèle sont au nombre de 19, soit 10 ratios financiers de liquidité de l’entreprise, 5 ratios de solvabilité et 4 ratios mesurant la profitabilité. La première étape du développement de notre modèle consiste à effectuer un test de différence de moyenne afin d’identifier les ratios qui présentent un pouvoir discriminant significatif entre les deux groupes d’entreprises. D’après les résultats de ce test, nous retenons les ratios énumérés au tableau 8, jugés les plus significatifs pour l’analyse (seuil de signification de 5%) : 87 Tableau 6 : Ratios financiers retenus pour l’analyse d’une période avant la défaillance Ratios Mesure Avoir des actionnaires/ passif total Liquidité Fonds de roulement/ actif total Liquidité Fonds de roulement/ passif total Liquidité Actifs à court terme / actif total Liquidité Passif à court terme/ actif total Liquidité Benefice avant impôt et intérêt / frais financiers Solvabilité Actif à court terme/ passif à court terme Liquidité Benefice avant impôt et intérêts/ actif total profitabilité (Actifs à court terme- stock)/ passif à court terme Liquidité Benefice non repartis/ actif total Profitabilité L’étape suivante consiste à effecteur un test de corrélation en raison que la régression logistique nécessite que les ratios retenus soient non corrélés. En se basant sur les résultats de ce test, nous avons défini quatre modèles prédictifs à partir d’une combinaison linéaire des ratios non corrélés du tableau 8, sur lesquels nous avons appliqué la régression logistique. L’estimation de ces modèles a été effectuée sur un échantillon composé de 152 PME (71 firmes défaillantes et 81 firmes non défaillantes) alors que la qualité de la prévision est testé sur un échantillon de contrôle composé de 151 PME ( 70 firmes défaillantes et 81 firmes non défaillantes). Les résultats de cette estimation sont présentés au tableau suivant : 88 Tableau 7 : Résultats de l’estimation des modèles développés pour une période avant la défaillance COEFFICIENTS ESTIMÉS VARIABLES Modèle 1 Constante 0,229 (0,55) -1,086 (0,018) -0,652 (0,397) Avoir des actionnaires/ passif total Fonds de roulement/ actif total Modèle 2 0,086 (0,735) -1,341 (0,01) -0,549 (0,43) Modèle 3 -0,387 (0,311) Fonds de roulement/ passif total Actifs à court terme / actif total -0,355 (0,65) Passif à court terme/ actif total Benefice avant impôt et intérêt / frais financiers -0,12 (0,005) -0,139 (0,003) 0,065 (0,0009) 0,063 (0,002) 0,139 (0,209) Actif à court terme/ passif à court terme Benefice avant impôt et intérêts/ actif total (Actifs à court terme- stock)/ passif à court terme -1,242 (0,131) 1,705 (0,029) -0,145 (0,002) 0,005 (0,9) 0,066 (0,0008) Modèle 4 -0,174 (0,624) -1,013 (0,335) -0,698 (0,391) 0,058 (0,93) -0,117 (0,005) 0,0612 (0,0012) Benefice non repartis/ actif total Test de la capacité prédictive Log vraisemblance (Contraint) L0 Log vraisemblance (Non Contraint) L1 Indice du ratio de vraisemblance (%) 3 Goodness of Fit test ( βi = 0 ∀ i ) -102,247 -97,735 -100,7 -102,24 -72,326 -69,52 -76,431 -76,68 29,26 22,17 24,1 25 21,508 11,03 8,45 12,04 Niveau de précision (%) Firmes “ Non défaut” 82,28 82,05 79,75 79,75 Firmes “Défaut” 71,01 71,88 67,16 69,57 Total 77,03 77,46 73,97 75 Avant d’interpréter les résultats obtenus, il convient de définir les concepts clés : L0 est la valeur maximale de la fonction log de vraisemblance du modèle contraint, c’est à dire, quand tous les paramètres, à l’exception de la constante, sont mis à zéro. L1 est la valeur maximale de la même fonction pour le modèle non contraint composé de tous les paramètres en plus de la constante. Plus la différence entre ces deux valeurs est grande, plus l’amélioration apportée par le modèle non contraint par rapport au modèle contraint est significative. L’indice du ratio de vraisemblance4 est une mesure du pouvoir prédictif proposé par Mc Fadden (1974) pour les modèles logistiques. Son interprétation est semblable à celle du coefficient de détermination R2 pour une régression linéaire. Plus cet indice est proche de 1, plus adéquate est la capacité 4 L’indice du ratio de vraisemblance = 1 – L1/L0 89 prédictive du modèle. Les résultats présentés au tableau 9 montrent que les modèles 1 et 2 présentent le niveau de précision de classification le plus élevé avec un pourcentage de 77,03% et 77,46% respectivement. Il faut noter que le degré de signification des coefficients intervient dans le choix du meilleur modèle. Ce degré de signification est mesuré par le test de « Goodness-ofFit ». Son hypothèse nulle implique des coefficients non significatifs. Les résultats de ce test indiquent que cette hypothèse est rejetée pour le modèle 1 alors qu’elle n’est pas rejetée pour le modèle 25. Le modèle 1 est donc considéré comme le meilleur modèle de prévision de la défaillance financière. Les variables significatives obtenues à partir de ce modèle mesurent la liquidité (Avoir des actionnaires/ passif total), la solvabilité (Bénéfice avant impôts et intérêts/ frais financiers) et la profitabilité de l’entreprise (Bénéfice avant impôts et intérêts/ actif total). Ceci implique que la réduction des profits de l’entreprise pourrait entraîner graduellement la détérioration de sa performance. Si les fonds disponibles pour les opérations courantes sont insuffisants, l’entreprise se trouve dans l’obligation de recourir à l’endettement. L’incapacité de remboursement de ces dettes augmente certainement la probabilité de défaillance financière. Résultats de l’estimation pour deux années avant la défaillance. Nous retenons dans cette partie les mêmes ratios explicatifs utilisés précédemment. En appliquant le test de comparaison de moyenne, on obtient les ratios les plus significatifs énumérés au tableau 10. Tableau 8 : Les ratios financiers retenus pour l’analyse de deux périodes avant la défaillance Ratios Mesure Fonds de roulement/ ventes Liquidité Avoir des actionnaires/ passif total Liquidité Fonds de roulement/ actif total Liquidité Fonds de roulement/ passif total Liquidité Passif à court terme/ actif total Liquidité Benefice non repartis/ actif total Profitabilité Actif à court terme/ passif à court terme Liquidité (Actifs à court terme- stock)/ passif à court terme Liquidité Avoir des actionnaires / actif total solvabilité De nouveau, nous avons défini quatre modèles à partir de combinaisons linéaires différentes des ratios du tableau 10 sur lesquels nous avons appliqué la régression logistique. L’estimation de ces modèles a été effectuée sur un échantillon d’estimation composé de 165 PME (83 défaillantes et 82 non défaillantes) et la validation du modèle retenu est testée sur un échantillon de contrôle composé de 165 autres PME (82 défaillantes et 83 non défaillantes). Les résultats de l’estimation sont présentés au tableau 11. Les valeurs calculées pour ce test sont comparées à une valeur critique de : χ2 = 12,59 90 Tableau 9 : Résultats de l’estimation des modèles pour l’analyse de deux périodes avant la défaillance VARIABLES Constante Fonds de roulement/ ventes Avoir des actionnaires/ passif total Fonds de roulement/ actif total Fonds de roulement/ passif total COEFFICIENTS ESTIMÉS Modèle 1 0,056 (0,752) -0,301 (0,369) -0,209 (0,244) -0,372 (0,51) -0,164 (0,693) Passif à court terme/ actif total Modèle 2 0,045 (0,86) -0,417 (0,218) -0,222 (0,228) Modèle 3 0,172 (0,405) -0,056 (0,614) Modèle 4 0,076 (0,657) -0,311 (0,27) -0,021 (0,75) -0,186 (0,56) 0,202 (0,514) Benefice non repartis/ actif total Actif à court terme/ passif à court terme (Actifs à court terme- stock)/ passif à court terme -0,132 (0,585) -0,89 (0,019) -0,088 (0,48) -0,085 (0,75) Avoir des actionnaires / actif total -1,242 (0,006) Test de la capacité prédictive Log vraisemblance (Contraint) L0 -114,36 -114,36 -114,36 -114,36 Log vraisemblance (Non Contraint) L1 -108,29 -108,67 -105,65 -103,84 Indice du ratio de vraisemblance (%) 5,31 4,97 7,61 9,20 Goodness of Fit test ( βi = 0 ∀ i ) 12,02 18,55 10,52 11,89 Niveau de précision (%) Firmes “ Non défaut” 56,10 59,76 67,07 70,73 Firmes “Défaut” 68,67 69,88 66,27 63,66 Total 62,42 64,85 66,67 67,27 D’après les résultats obtenus, le modèle 4 semble être le meilleur modèle de prévision de la défaillance financière des PME deux périodes avant la défaillance étant donné qu’il présente le taux de bons classement le plus élevé ( 67,27%). Le test de « Goodness-of-Fit » indique que les coefficients sont significatifs6 et l’indice du ratio de vraisemblance présente la valeur la plus élevée indiquant que ce modèle est susceptible de mieux prévoir la situation future des PME québécoises deux périodes avant la défaillance. Dans ce modèle, seul le ratio « avoir des actionnaires/ actif total » est significatif. Ce ratio mesure bien la solvabilité de l’entreprise. Il s’en suit que l’état de l’insolvabilité est le facteur déterminant de la dégradation de la santé financière des PME québécoises, les conduisant ainsi à la défaillance financière deux années plus tard. Il est à noter que le taux de bons classements une année avant la défaillance est plus grand que celui 6 Les valeurs calculées pour ce test sont comparées à une valeur critique de : χ2 = 11,07 91 deux années avant la défaillance. Ceci nous permet de conclure que plus l’horizon de prévision est court, plus précise sera la prévision de la défaillance. Choix de point de coupure optimal. Les résultats de classification présentés ci-dessus sont obtenus en adoptant un seuil appelé point de coupure qui permet d’isoler les entreprises en situation financière saines de celles en difficultés financières. L’observation des études théoriques antérieures utilisant la régression logistique dans leur modèle de prévision montre qu’elles utilisent un point de coupure de 0,5 pour distinguer les entreprises défaillantes de celles non défaillantes. Ceci implique qu’une entreprise qui affiche une probabilité de défaut supérieure à 50% sera classée dans le groupe « en défaillance ». On suppose ainsi implicitement que la fonction de perte est symétrique entre les deux types d’erreur de classification. À toutes fins pratiques, la qualité de l’indicateur de risque ainsi élaboré est jugée au regard des erreurs de classement, à savoir le pourcentage d’entreprises défaillantes considérées comme saines – erreur de type I – et, inversement, le pourcentage d’entreprises non défaillantes considérées comme risquées – erreur de type II. Pour obtenir un point de coupure optimal, il est nécessaire de connaître les coûts de ces erreurs ainsi que les probabilités à priori de la survie et de la défaillance ce qui est difficile à obtenir. Quelques modèles (Ohlson 1980) proposent une alternative qui consiste à considérer plusieurs points de coupure dans la détermination des pourcentages de bonne classification. Le point optimal dans ce cas est celui qui minimise la somme des erreurs de types I et II. Il faut noter que les coûts de l’erreur de type I sont plus significatifs que ceux de l’erreur de type II. Pour un bailleur de fonds, accorder un prêt à une entreprise qui va probablement faire défaut (erreur de type I) est plus coûteux qu’un refus de dossier de crédit à une autre firme qui pourrait être en situation financière saine (erreur de type II). Dans le but de déterminer le point qui minimise les erreurs de classification, nous allons expérimenter avec plusieurs points de coupure différents. Les résultats de classification ainsi que les erreurs de type I et II qui en découlent sont présentés au tableau suivant : Tableau 10 : résultats de l’application de plusieurs points de coupure sur le modèle retenu ANNÉE (-1) ANNÉE (-2) Point de coupure 0,1 % de bons classements 60,14 Erreur type I 0 Erreur type II 74,68 % de bons classements 52,94 Erreur type I 0 Erreur type II 96,34 0,2 59,46 7,25 69,62 53,94 0 92,68 0,3 69,59 10,14 48,10 53,33 4,82 89,02 0,4 79,73 13,04 26,58 61,82 13,25 63,41 0,5 77,03 28,99 17,72 67,27 36,14 29,27 0,6 73,65 42,03 12,66 64,24 62,65 8,54 0,7 70,95 56,52 5,06 56,97 81,93 3,66 0,8 65,54 69,57 3,08 52,12 92,77 2,44 0,9 62,16 79,71 1,27 51,52 95,18 1,22 Comme le tableau 12 le montre, le taux de bons classements varie entre 59,46% et 79,73% pour une année avant la défaillance alors qu’il se situe entre 51,52% et 67,27% pour deux années avant la défaillance. Étant donné qu’on cherche le point de coupure qui maximise le taux 92 de bons classements tout en minimisant la somme des deux types d’erreurs, les point de coupure optimal seront 0,4 et 0,5 pour une année et deux années avant la défaillance respectivement. Ces deux points déterminent le taux de bons classements le plus élevé (79,73% et 67,27% respectivement) minimisant ainsi les erreurs de types I et II (20,2 % et 32,73% respectivement). Validation des modèles suggérés. Pour tester l’efficacité des deux modèles retenus dans la détection de la défaillance financière des PME québécoises, nous appliquons les deux points de coupure optimaux ainsi définis sur nos échantillons de contrôle. Les tableaux suivants présentent les résultats de ces analyses. Tableau 11 : Résultats de l’application du modèle 1 sur l’échantillon de contrôle (une année avant la défaillance) CLASSIFICATION DU MODÈLE SUGGÉRÉ Défaut Non défaut Défaut 55 15 Non défaut 26 55 % DE CLASSIFICATION DE CLASSIFICATION DE L’ÉCHANTILLON L’ÉCHANTILLON DE DE CONTRÔLE QUI CONTRÔLE CORRESPOND AU MODÈLE Défaut Non défaut Défaut Non défaut 70 78,57 81 67,90 72,84 Tableau 12 : Résultats de l’application du modèle 4 sur l’échantillon de contrôle (deux années avant la défaillance) CLASSIFICATION DU MODÈLE SUGGÉRÉ Défaut Non défaut Défaut 36 46 Non défaut 14 69 % DE CLASSIFICATION DE CLASSIFICATION DE L’ÉCHANTILLON L’ÉCHANTILLON DE DE CONTRÔLEQUI CONTRÔLE CORRESPOND AU MODÈLE Défaut Non défaut Défaut Non défaut 82 43,9 83 83,13 63,63 Tableau 13: Tableau récapitulatif Point de coupure % de bons classements 93 Erreur type I Erreur type II 0,4 0,5 Année (-1) Année (-2) 72,84 63,63 21,43 56,09 32,09 16,86 D’après les résultats obtenus, les taux de bons classements des entreprises de l’échantillon de contrôle pour les deux années d’études se situent approximativement au même niveau que ceux trouvés pour l’échantillon d’estimation témoignant l’efficacité de nos modèles logistiques suggérés. Il convient de souligner que plus l’analyse se rapporte à des données proches de l’événement de défaillance, plus grande est l’exactitude de classification et de prévision. Mise en évidence de l’effet de l’industrie. La méthode de la prévision de la défaillance financière basée sur les ratios financiers cherche essentiellement à déterminer les variables qui différencient au mieux les entreprises défaillantes des entreprises non défaillantes. L’appartenance d’une entreprise donnée à un secteur d’activité bien particulier exerce certainement une influence sur le degré de signification des ratios du modèle de prévision. Nous mettons l’accent dans cette partie sur les secteurs d’activités les plus touchés par la défaillance financière des PME québécoises à savoir le secteur des services et le secteur du commerce en détail. L’idée est de réestimer les modèles de prévision retenus précédemment en se limitant aux données financières des PME appartenant à ces deux secteurs d’activité. Les résultats obtenus pour les deux années d’études sont présentés aux tableaux suivants : Tableau 14: la détection de l’effet industrie une année avant la défaillance Coefficients estimés Services Constante Avoir des actionnaires/ passif total Fonds de roulement/ actif total Actifs à court terme / actif total Bénéfice avant impôt et intérêt / frais financiers Bénéfice avant impôt et intérêts/ actif total -0,129 (0,76) -0,322 (0,205) -1,204 (0,132) -0,861 (0,33) -0,024 (0,32) 0,07 (0,0009) Commerce en détail 3,093 (0,01) -1,157 (0,41) 0,308 (0,88) -3,038 (0,086) -0,437 (0,035) 0,042 (0,30) Niveau de précision % Firmes « non défaut » 84,62 73,91 Firmes « défaut » 73,77 89,29 Total 79,37 82,35 94 Tableau 15: la détection de l’effet industrie deux années avant la défaillance Coefficients estimés Services Constante Fonds de roulement/ ventes Avoir des actionnaires/ passif total Fonds de roulement/ passif total Avoir des actionnaires / actif total Commerce en détail 0,189 (0,436) -2,616 (0,069) -0,028 (0,656) 0,945 (0,155) -2,514 (0,0002) 0,568 (0,06) 0,279 (0,738) -2,0001 (0,12) -0,812 (0,377) 1,245 (0,421) Niveau de précision % Firmes « non défaut » 75,38 55,81 Firmes « défaut » 58,73 75,61 Total 67,19 65,48 Comme les tableaux 16 et 17 le montrent, les ratios les plus discriminants différent d’un secteur à l’autre et d’une période à l’autre. Pour une année avant la défaillance, le ratio « Bénéfice avant impôt et intérêt/ actif total » est significatif pour le secteur des services alors qu’il ne l’est pas pour le secteur du commerce en détail. Inversement, le ratio « Bénéfice avant impôt et intérêt/ frais financiers » discrimine entre les deux groupes d’entreprises pour le secteur commerce en détail alors qu’il n’a pas d’effet discriminant pour le secteur des services. On pourrait ainsi déduire que le manque de profits pour les PME opérant dans le secteur des services est le premier facteur responsable de la dégradation de la santé financière de ces entreprises alors que l’état de l’insolvabilité est le facteur déterminant de la défaillance des entreprises opérant dans le secteur du commerce en détail. Les résultats pour deux années avant la défaillance montrent que l’état de l’insolvabilité est l’élément dominant pour le secteur du service alors que tous les ratios sont non significatifs pour le secteur du commerce en détail. Conclusion L’objet de la présente étude est de développer un modèle de prévision de la défaillance financière des PME québécoises non cotés en bourse. L’étude révèle que les modèles théoriques de la prévision de la défaillance financière, connus dans la littérature, ne reflètent pas réellement les spécificités des PME québécoises en matière de financement dans le sens qu’ils ne permettent pas d’anticiper correctement les difficultés économiques et financières qu’une PME est susceptible de rencontrer. Partant de ce constat, nous nous sommes appliqués à développer un modèle qui traduit efficacement les particularités des PME québécoises. Les résultats empiriques démontrent clairement l’efficacité de nos modèles logistiques suggérés à prédire la défaillance des PME un an et deux ans avant l’évènement. En effet, les taux de bons classements de nos 95 modèles sont nettement supérieurs à ceux des modèles théoriques existants, et ce surtout pour les entreprises défaillantes. De plus, les résultats de l’application de nos modèles sur des échantillons de contrôle confirment bien leur robustesse. L’étude indique également que plus l’horizon de prévision est court, plus grande est l’exactitude de la prévision. L’analyse des résultats de l’étude sectorielle fait ressortir que le pouvoir discriminant des ratios financiers retenus pourrait différer d’un secteur à l’autre et d’une période à l’autre. Autrement dit, un ratio de liquidité pourrait jouer un rôle important dans la discrimination entre les deux groupes d’entreprises pour un secteur d’activité bien déterminé alors qu’il ne présente aucun effet dans un autre secteur. En somme, étant donné que l’accès au capital est d’importance primordiale pour les PME québécoises pour soutenir leur croissance, la prévision de leur défaillance financière est une nécessité majeure. L’étude révèle que la régression logistique appliquée sur des ratios financiers s’avère un moyen efficace et robuste de prévision. Il est intéressant aussi d’effectuer une étude sectorielle en mettant l’accent sur les secteurs d’activités les plus concernés par la défaillance afin d’identifier la nature des variables significatives dans chaque cadre d’analyse. Références Altman, E. I., (1968). Financial Ratios, Discriminant Analysis and the Prediction of the Corporate Bankruptcy. Journal of Finance, 23, 589 – 609. Altman, E. I., R. G. Haldeman, and P. Narayanan, (1977). ZETA Analysis: A New Model to Identify Bankruptcy Risk of Corporations. Journal of Banking and Finance, 1, 29 – 54. Altman, E. I. and M. Y. Lavallee, (1980). Business Failure Classification in Canada Journal of Business Administration.12, 147 -162. Baldwin, J., Bian, L., Dupuy, R. and G. Gellatly, (2000). Taux d’échec des nouvelles entreprises canadiennes: Nouvelles perspectives sur les entrées et les sorites. Statistique Canada. Banque Nationale, (1988). Un modèle de prévision de la faillite : le CA Score . Service du crédit aux entreprises, Information crédit, 29. Beaver, W. H., (1966). Financial Ratios as Predictors of Failure. Journal of Accounting Research, 4, 71 – 111. Edmister, R., (1972). An Empirical Test of Financial Ratio Analysis for Small Business Failure Prediction . Journal of Financial and Quantitative Analysis, 7, 1477. Eisenbeis R., Avery R.B., (1972). .Discriminant Analysis and Classification Procedures: Theory andApplications., working paper. Fahey, R., P. Paradis, (2003). Pour une contribution optimale à l’essor des PME québécoises . Fédération Canadienne de l’Entreprise Indépendante. 96 Industrie Canada, (2002). Les lacunes dans le financement des PME : Cadre d’analyse. Khoury, N., M. Savor et R. Toffoli, (2006). La couverture des risques financiers par les PME québécoises , Revue canadienne des sciences de l’administration 23(1), 67-80. Lecavalier, C., (2006), Tendances nationales et régionales des faillites d’entreprises, 1980 à 2005. Division de l’analyse microéconomique, Statistique Canada. Machauer, A. and M. Weber. (1998). Bank behavior based on internal credit ratings of borrowers. Journal of Banking and Finance, 22, 1355-1383. Ohlson, J. A., (1980). Financial Ratios and the Probabilistic Prediction of Bankruptcy. Journal of Accounting Research, 18, 109 – 131. Paradis, P.E., (2004). Québec, regard sur la PME, Fédération Canadienne de l’Entreprise Indépendante. Tam K.Y. (1991). Neural network models and the prediction of bank bankruptcy. Omega, 19, 429-445. Tam K.Y., Kiang M.Y. (1992). Managerial application of neural networks: the case of bank failure predictions, Management science, 38, 926-947. Wruck, K. (1990). Financial Distress, Reorganisation and organizational Efficiency. Journal of Financial Economics, 27, October, 419- 444. Zavgren, C. V., (1985), Assessing the Vulnerability to Failure of American Industrial Firms. A logistic Analysis, Journal of Business Finance and Accounting, 19-45. 97 ASAC 2008 Halifax, Nova Scotia Narjess Boubakri Jean-Claude Cosset Anis Samet (Ph.D. Student) HEC Montreal WHY DO FOREIGN FIRMS ISSUE A SPECIFIC ADR? 1 We study the determinants of a firm’s decision to issue one of the four available ADR programs (Level I, Level II, Level III, and Rule 144A). We find that the firm's attributes and the firm's home-country institutional variables condition this choice. We also examine the issuing activity and the determinants of the ADR choice before and after the enactment of the Sarbanes-Oxley (SOX) Act. Following this structural change, we provide evidence of a reallocation between ADR programs. Compared to the preSOX period, firms from emerging markets, and those from countries with weak legal protection of minority shareholders, are more likely after SOX to choose Rule 144A and Level III, respectively. 1. Introduction Cross-listings on U.S. markets have become a major phenomenon over the past two decades. These cross-listings can be achieved via a direct listing, New York Registered shares, global registered shares, or American Depositary Receipts (ADRs). Firms that cross-list under ADR programs come from a wide array of developed and developing countries, while those under direct cross-listing are mostly Canadian.2 Focusing on ADRs thus allows us to bring to light the impact of home-country variables on the cross-listing decision. Any firm that cross-lists via an ADR has basically four options to choose from: Level I, Level II, Level III, and Rule 144A programs, all of which have distinct attributes. For instance, Level III and Rule 144A offer an access to U.S. primary capital markets (i.e., raising capital), whereas Levels I and II allow an access to U.S. secondary markets only. Moreover, the governance and disclosure requirements vary across the four ADR programs, and are more restrictive in Levels II and III (listed programs) than in Level I and Rule 144A (unlisted programs). After the enactment of the Sarbanes-Oxley Act (SOX hereafter) in 2002, these governance and disclosure requirements have become more stringent and costly for listed firms, U.S. and foreign alike. Finally, Rule 144A allows foreign firms to target only U.S. private institutional investors, while Levels I, II, and III give access to public as well as private U.S. investors. 1 We would like to thank Jean-Yves Duclos, Art Durnev, Laurent Fresard, Andrew Karolyi, Iwan Meier, Sergiy Rakhmayil, and Sergei Sarkissian for helpful comments. We are indebted to Chloé Jacob and Andreea Strachinescu for excellent research assistance. 2 Karolyi (2006) reports that in 2003 the United Kingdom, Australia, and Japan accounted for 17%, 10%, and 6% of the ADR listings in U.S., while, South Africa, Mexico, and Brazil, were respectively home to 6%, 5%, and 4% of the firms issuing ADRs. In the same year, ADRs accounted for 73.2% of the U.S. cross-listings. 98 The main objectives of this paper are twofold: first, we analyze the choice of a specific ADR among all four options. Second, we examine whether, and to what extent, the enactment of SOX affects such a choice. In the first part of the analysis, we specifically examine the choice of a particular ADR based on firm-level variables and home-country institutional variables. To date, no previous study on cross-listing has distinguished between the four different ADRs, and thus we consequently consider all four options on an individual basis. The results of our empirical investigation of the choice of an ADR program show that capitalraising Level III attracts large firms, firms with high pre-tax income, those with high growth opportunities, privatized firms, and firms from weak investor protection environments. This latter result is consistent with the bonding hypothesis. We likewise find that firms from weak investor protection environments are attracted by Rule 144A programs. Finally, we document that firms with high ultimate control rights and excess control rights, and those from emerging markets are less likely to choose Level II and more likely to choose Level I.3 In the second part of the paper, we examine whether, and to what extent, the introduction of SOX in 2002 had an impact on the choice of a particular ADR. SOX represents a structural change in the regulatory and legal environment surrounding ADR listings, particularly Level II and Level III programs as it introduces more stringent corporate governance and disclosure requirements for the firms that list on the major U.S. exchanges. After we control for SOX in our multinomial logit estimation, we find that foreign firms are indeed less likely to choose Level II ADRs after SOX. We also find that after SOX, foreign firms are more attracted by Rule 144A programs which allow them to circumvent the new stringent SOX rules and tap the U.S. primary market. This latter result is consistent with Zingales’ (2006) evidence. Furthermore, a closer look at the distribution of firms across the four ADR programs after SOX shows that there is indeed an inter-program reallocation that cannot be explained by a change in firms’ characteristics. More precisely, after SOX, firms are more attracted by capital-raising programs, either Level III or Rule 144A, and are more reluctant to issue Level II. Thus, the possibility to raise fresh capital on U.S. markets seems to drive cross-listing after SOX. By choosing Level III programs, firms subject themselves to more stringent rules but also benefit from the access to U.S. capital resources through public offerings, which is consistent with more bonding and, more generally, enhanced bonding benefits after the implementation of SOX. For those foreign firms that want to avoid such restrictive listings but still raise capital, 144A private placements allow them to do so, as this unlisted program exempts them from governance and disclosure requirements and from compliance to U.S. GAAP (Zingales, 2006). Finally, the results of a re-estimation of our multivariate model over the pre- and post-SOX periods suggest that some attributes have a larger impact on the ADR choice decision in the postSOX period than in the pre-SOX. For instance, being an emerging market firm heightens the probability of choosing Rule 144A. Similarly, coming from a country with weak legal protection of minority shareholders increases the likelihood of cross-listing under Level III in the post-SOX period as compared to the pre-SOX period. This latter result is consistent with more bonding and the enhanced bonding benefits after the implementation of SOX. 3 To calculate the ultimate cash flow rights and ultimate control rights, we follow La Porta et al. (1999), Claessens et al. (2000), and Faccio and Lang (2002). 99 2. Development of hypotheses We conjecture that the choice of a specific ADR depends upon variables related to the firms’ attributes (e.g., size, profitability, growth opportunities, leverage, turnover volume, and country of origin), its corporate governance (privatization, ownership structure, and SOX), and homecountry institutional attributes (accounting standards and investor protection). More precisely, we derive the four hypotheses presented in the following paragraphs. 2.1. Firm attributes and ADR Programs Larger and more profitable firms are more likely to choose Level II and Level III because these two ADR programs require that (1) firms pay large continuing fees and (2) meet size and earnings requirements to cross-list. Firms with high turnover volume (relatively to their local market turnover volume) are more likely to opt for Level II and Level III to enhance their liquidity and circumvent their local market financial constraints. Hence, H1: Larger, firms with higher relative turnover volume, and higher earning firms are more likely to choose a listed ADR (Level II or Level III). Firms with higher growth opportunities generally need additional equity capital. More indebted firms are also more likely to issue equity offerings to finance their operations. Given that only Level III and Rule 144A allow capital-raising, we expect that the higher the leverage ratio and the higher the growth opportunities of foreign firms, the more likely they will choose Rule 144A and Level III. In the same vein, privatized firms are more likely to choose Rule 144A or Level III than Level I or Level II since the aim of privatization through ADRs is usually to raise capital for firms, and is typically done through primary issues. Firms from emerging markets are relatively more capital constrained, and have higher needs to raise external capital (Lins et al., 2005). Therefore, these firms are more likely to choose Rule 144A and Level III. They are also less likely to choose Level II since it is costly in terms of compliance and does not allow the raising of new capital. We summarize our expectation in the following hypothesis: H2: Firms with higher growth opportunities, more indebted firms, privatized firms, and firms from emerging markets are more likely to choose Rule 144A or Level III ADRs. Emerging market firms are less likely to choose Level II. The enactment of SOX is likely to have an impact on the issuer's ADR choice. After SOX, we expect firms to be more likely to issue Rule 144A ADRs which allow firms to raise capital on U.S. markets, and require no particular compliance with SEC, U.S. GAAP, or the SOX Act. The implementation of SOX raised the costs of cross listing by imposing compliance to more stringent new rules; it particularly affected the costs related to the choice of Level II and Level III programs alike. Keeping in mind that only Level III allows firms to raise fresh capital on U.S. markets, we expect foreign firms to find it less interesting to list under Level II after SOX. Accordingly, we enunciate the following hypothesis: H3: Firms issuing ADRs after SOX are more likely to choose Rule 144A and less likely to select Level II. 100 According to Doidge et al. (2007a), when controlling shareholders have tighter control (greater voting rights) of the firm, they are more reluctant to list their firms on a U.S. major stock exchange because the costs of the extraction of private benefits of control exceed the benefits of such listings. Doidge et al. (2007a) find evidence for this conjecture. Therefore, we would expect that controlling shareholders who control a large stake (voting rights) in one firm are more reluctant to relinquish their private benefits of control and are thus more likely to choose less restrictive programs. Additionally, when the separation between control and cash flow rights is less pronounced, this means, according to Claessens et al. (2002), that it is less likely that controlling shareholders extract private benefits of control from minority shareholders. Hence, we expect that the tighter the control in a firm and the larger the difference between the control and cash flow rights, the less likely it is that the firm will choose a listed ADR program, (i.e., Levels II or III) as these two levels increase the costs of extracting private benefits. Thus, H4: Firms where the largest controlling shareholder holds greater control rights, and firms with a high separation between control and cash flow rights are less likely to be listed under Level II or Level III, and more likely to select Rule 144A and Level I. 2.2. Home-country institutional attributes and ADR programs In line with the bonding hypothesis introduced by Coffee (1999, 2002) and Stulz (1999), and discussed above, firms from countries with a lower level of investor protection and weak accounting standards are more likely to choose a listed ADR (i.e., Level II and Level III) to protect minority shareholders against “managerial self-dealing” and private benefit extractions (Karolyi, 2006). However, to avoid the stringent compliances and disclosure requirements, especially those related to SOX, foreign firms from countries with poor investor protection and weak accounting standards may be more willing to choose an unlisted ADR program (i.e., Level I and Rule 144A) rather than an exchange-listed one. Based on these two competing arguments, we cannot put forward any directional hypothesis, and we leave this issue to be resolved by empirics. 3. Data and variables The Bank of New York (BNY), Citibank (CB), the Deutsche Bank (DB), and JP Morgan (JPM), are the major depositaries of ADRs, although BNY alone accounts for 64% of the ADR market.4 We downloaded valuable information from these depositaries' websites regarding ADRs, namely the type, the effective issuance date, the market where the ADR is traded, the sponsorship status (whether the ADR is sponsored or not), the underlying share and its country of origin, the Committee on Uniform Securities Identification Procedures (CUSIP) number of the ADR, and the International Securities Identification Number (ISIN) of the underlying share. We obtain the accounting and financial information on the sample firms one year before the ADR issuance date from different sources which we describe in the Appendix. The final sample consists of 647 ADRs and spans the period from 1990 through 2006. We present summary statistics on this sample in Table 1. Panel A of Table 1 indicates that most ADRs, namely 287 (44.4%), are issued by firms from the Asia/Pacific region. European firms follow with 263 (40.6%) ADRs. Panel B of Table 1 shows 4 See “The depositary receipt markets: The year in review” - 2006, Bank of New York. 101 that firms from high income countries dominate the sample with 442 (68.3%) ADRs. The NYSE attracts more programs than NASDAQ5 (146 (22.6%) versus 45 (7%) ADRs, respectively). The distribution of our sample across ADR programs is close to the universe of sponsored ADRs, since, over our study period, Rule 144A accounted for 26.1% of ADRs, Level I for 44.2%, Level II for 17.1%, and Level III for 12.6%. Table 1: Descriptive statistics This table presents descriptive statistics for a sample of 647 ADRs issued between 1990 and 2006. Based on the World Bank country group classifications, Panel A and B respectively provide the distribution of the issuing firms’ home-countries by geographic location and income category. Panel A: Geographic location Type of ADR Geographic location (countries) 144A Level I Level II Level III Total Percentage Asia/Pacific (13) 77 152 25 33 287 44.4% Europe (20) 35 129 63 36 263 40.6% Latin America (7) 14 31 15 16 76 11.7% Middle East/Africa (3) 4 14 3 0 21 3.2% 100.0% Total (43) 130 326 106 85 647 Percentage 20.1% 50.4% 16.4% 13.1% 100.0% Panel B: Income category Type of ADR Income category (countries) 144A Level I Level II Level III Total Percentage High income (23) 65 235 84 58 442 68.3% Upper middle income (9) 31 56 13 9 109 16.8% Lower middle income (9) 10 35 7 12 64 9.9% Low income (2) 24 0 2 6 32 4.9% 100.0% Total 130 326 106 85 647 Percentage 20.1% 50.4% 16.4% 13.1% 100.0% We consider two categories of variables to examine the choice of an ADR program: those applying to the underlying firms (section 4.1) and to the home country's institutions (section 4.2). The Appendix defines these variables and describes their data sources. 4. Empirical analysis In Table 2, we summarize the predicted relations between the explanatory variables and the probability of choosing a given type of ADR. 5 Note that our sample does not include firms listed on the AMEX because of the unavailability of data for these firms. The Bank of New York reports that only four ADRs are traded on the AMEX. 102 Table 2: The determinants of an issuer’s ADR choice This table reports the predicted signs of the variables that we include in our model of ADR choices, namely, Rule 144A, Level I, Level II, and Level III. The variables are defined in the Appendix. Probability of choosing an ADR Variables Label Rule 144A Level I Level II Level III + + SIZE + + INCOME + + ASSETGR + + LEV + + RELTOV Firm + +/+ EMC + + PRIVA + +/+/SOX + + ULOW + + ULOWDIF +/+/+/+/ACRAT Institutional +/+/+/+/SELFDEAL In Section 5.1, we examine whether the explanatory variables differ across the four ADR programs. We then perform a multivariate analysis in Section 5.2 and present sensitivity tests in Section 5.3. 4.1. Univariate analysis Table 3 presents the means of the explanatory variables for the different types of ADRs. Differences in the means of these variables between the three types of ADRs and Level I (the base outcome) are then tested using a two-tailed t-test of means. Table 3 shows that Level II and Level III firms are larger and have a higher pre-tax income than those choosing Level I. This result, significant at the 1% level, is expected since both the NYSE and NASDAQ impose minimum size and earnings requirements for non-U.S. firms that list on U.S. exchanges. Although they are smaller than Level II and Level III, Rule 144A firms are larger, at the 5% level, than Level I firms. Rule 144A and Level III firms have higher asset growth rates than Level I firms. This result supports our prediction presented in Table 2. In fact, as Rule 144A and Level III allow firms to raise new capital on U.S. markets, firms with relatively high growth opportunities will opt for these two programs. Moreover, Rule 144A firms have a higher leverage than Level I firms, a result which is in line with the predicted relation as Rule 144A allows firms to raise capital on U.S. markets. Non-U.S. firms that choose Level III exhibit a high relative turnover ratio compared to Level I firms, a difference that is significant at the 1% level. Such a result is expected for these firms, which are more likely to seek an ADR that allows them to circumvent the narrowness and illiquidity of their home market (i.e., the financing constraints).Privatizing governments are more likely to choose Rule 144A, Level II, and Level III than Level I. Of the first three, privatizing governments are more likely to choose Rule 144A and Level III than Level II. These results are consistent with the predicted relation, presented in Table 2, since Rule 144A and Level III are the two ADR programs that allow governments to divest gradually through subsequent primary share offerings. 103 Table 3: Comparison between ADR programs This table presents the mean of the different variables for the different types of ADRs, namely Rule 144A, Level I, Level II, and Level III. Our sample consists of 647 ADRs issued between 1990 and 2006. The variables are defined in the Appendix. Differences in the means of the variables between the different types of ADRs and Level I (the base outcome) are tested using two-tailed t-test of means. P-values of this test are reported in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Type of ADR 144A Variables Level I Level II Level lII Label N of Obs. Mean N of Obs. Mean N of Obs. Mean N of Obs. SIZE 130 14.08 326 13.56 106 15.26 85 (0.04)** INCOME 130 ASSETGR 130 (0.00)*** 0.24 326 0.24 106 326 22.67 106 (0.91) 130 71 27.97 326 22.66 106 EMC 130 0.01 239 0.02 79 130 SOX 130 0.88 326 0.47 106 0.12 326 0 106 326 0.45 106 41 42.02 157 31.19 58 114 1.41 157 1.95 58 SELFDEAL 130 85 0.34 22.12 1.67 -10.37 286 -2.71 103 -3.47 85 -0.17 326 -0.01 106 -0.08 (0.03)** 0.18 (0.00)*** 85 0.35 (0.09)* 30 42.08 (0.02)** 30 3.79 (0.14) 78 (0.46) (0.00)*** 0.55 (0.16) (0.74) (0.00)*** Institutional 0.04 0.05 (0.00)*** (0.00)*** (0.56) ACRAT 43 (0.04)** (0.00)*** ULOWDIF 0.26 22.64 (0.99) (0.00)*** 0.57 (0.03)** 41 85 (0.00)*** (0.00)*** ULOW 0.02 33.71 (0.04)** (0.70) (0.00)*** PRIVA 24.29 0.67 (0.00)*** 85 (0.47) (0.46) Firm 85 (0.64) (0.01)*** RELTOV 20.34 14.92 (0.00)*** (0.00)*** 33.28 (0.02)** LEV 1.43 Mean -8.81 (0.00)*** 85 -0.19 (0.00)*** A total of 57% of the Rule 144A ADRs are issued after the implementation of SOX. This result is consistent with Zingales’ (2006) evidence. He points out the large increase in the number of 144A registrations by foreign firms after SOX which, by allowing them to avoid U.S. legal liability, helps them tap the U.S. markets via the “back door.” Firms that list under Rule 144A and Level III have larger ultimate control rights compared to those opting for Level I. In contrast, Level II firms have lower ultimate control rights than Level I firms. As argued previously, the tighter the control of the firm, the more likely that controlling shareholders will extract private benefits of control, and hence the less likely that these shareholders will choose a listed program. In this respect, the univariate results for Level III appear somewhat surprising. In an attempt to provide an explanation for these results, we take a closer look at the data and find that the largest shareholder of many firms listed under Level III is the State. To the extent that governments pursue different objectives from private controlling shareholders, we exclude firms with the State as the largest shareholder from our sample. We find that the ultimate control rights of Level I firms are larger than Level II firms and lower 104 than Rule 144A firms. Moreover, we find that level III firms no longer have higher ultimate control rights than Level I firms. Emerging market firms are more likely to choose Rule 144A and less likely to choose Level II than Level I. This is in keeping with evidence in Lins et al. (2005) that firms from emerging markets are capital constrained, and seek access to U.S. markets through a capital-raising issue that is not allowed under Level II. Anti-self dealing (SELFDEAL) represents the difference in the anti-self dealing indexes of the ADR’s home country and the U.S. This difference is generally negative since the U.S has a higher index. We find that firms choosing Rule 144A, Level II, and Level III ADRs present a higher difference in the anti-self dealing index than firms which choose Level I, i.e., these firms originate from countries with poorer investor protection compared to the U.S. Moreover, the difference in the accounting ratings (ACRAT) (between the home country and the U.S.) is higher for Rule 144A and Level III firms than Level I firms, i.e., firms from countries with weaker accounting standards than the U.S. opt for Rule 144A and Level III. 4.2. Multivariate analysis Once the decision of cross-listing via an ADR is taken, the firm’s manager must decide which type of ADR program to choose. His set of choices includes the four different ADR programs, namely, Rule 144A, Level I, Level II, and Level III. In a multinomial logit model, we cannot estimate all the coefficients for all the choices. In other words, the model is unidentified. To remove this indeterminacy,6 we have to assume a base outcome or a base choice for which all the coefficients are set to 0, and then interpret the estimated coefficients as measuring the change relative to the base outcome for the same variable. The choice of the base outcome (here Level I) is arbitrary and does not affect the predicted probabilities (Greene, 2003). The results of the Hausman test suggest that we cannot reject the IIA assumption for all the specifications; we therefore estimate multinomial logit models, correcting for clustering at the country level. Panel A of Table 4 shows that larger firms (SIZE) are more likely to choose Level II and Level III and less likely to select Level I. These results are respectively significant at the 10%, 5%, and 1% levels. Moreover, this panel shows that the higher the firm’s pre-tax income (INCOME), the more likely that it will choose Level II. These results are consistent with the predicted relations. Indeed, to be listed as Level II or Level III, firms have to meet minimum size and earnings requirements. 6 See Greene (2003), page 721. 105 Table 4: Multinomial logit estimations: the choice between the four ADR programs This table reports the multinomial logit estimations of the choice between the four ADRs programs, namely Rule 144A, Level I, Level II, and Level III. This table reports the marginal effects evaluated at the mean of the explanatory variables for the issued ADRs between 1990 and 2006. The variables are defined in the Appendix. The reported results use Level I as the base outcome and are corrected for clustering at the country level. Values between parentheses represent the Pvalues of the t test for the null hypothesis that the coefficient is equal to zero. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Panel Panel A Panel B Panel C Type of ADR 144A SIZE INCOME ASSETGR LEV RELTOV PRIVA 0.0001 -0.0043 0.0004 0.0025 0.3103 (0.99) (0.92) (0.25) (0.01)*** (0.02)** Level I -0.0373 -0.0475 -0.0012 -0.0005 -0.5509 (0.01)*** (0.20) (0.01)*** (0.73) (0.00)*** Level II 0.0168 0.0344 -0.0001 -0.0008 -0.0775 (0.06)* (0.03)** (0.91) (0.56) (0.10)* Level III 0.0203 0.0174 0.0009 -0.0013 0.3181 (0.04)** (0.35) (0.04)** (0.12) (0.01)*** 144A 0.0003 0.0036 -0.0001 -0.0004 -1.0044 0.1681 (0.92) (0.68) (0.44) (0.41) (0.19) (0.60) Level I -0.0233 -0.0794 -0.0007 0.0007 0.1121 -0.7409 (0.21) (0.28) (0.33) (0.63) (0.93) (0.00)*** Level II 0.0123 0.0701 0.0007 -0.0002 0.8408 -0.2340 (0.47) (0.28) (0.33) (0.91) (0.44) (0.00)*** Level III 0.0107 0.0057 0.0001 -0.0002 0.0515 0.8068 (0.09)* (0.48) (0.49) (0.63) (0.69) (0.01)*** 144A -0.0027 0.0106 0.0000 0.0000 -1.4389 0.6452 (0.51) (0.39) (0.85) (0.99) (0.05)** (0.08)* Level I -0.0146 -0.0820 -0.0003 0.0002 1.5318 -0.7405 (0.35) (0.15) (0.68) (0.84) (0.13) (0.00)*** Level II 0.0067 0.0550 0.0004 0.0001 0.5225 -0.1940 (0.57) (0.21) (0.46) (0.94) (0.44) (0.00)*** Level III 0.0105 0.0163 0.0000 -0.0003 -0.6154 0.2893 (0.27) (0.27) (0.95) (0.59) (0.12) (0.37) SOX ULOW ULOWDIF 0.1055 (0.04)** -0.0214 (0.64) -0.0616 (0.03)** -0.0226 (0.61) 0.0152 -0.0006 -0.0048 (0.15) (0.27) (0.21) 0.0911 0.0070 0.0111 (0.31) (0.00)*** (0.19) -0.1194 -0.0061 -0.0001 (0.16) (0.00)*** (0.98) 0.0131 -0.0003 -0.0064 (0.30) (0.55) (0.12) 0.0169 -0.0005 -0.0023 (0.24) (0.35) (0.23) 0.0577 0.0045 0.0059 (0.39) (0.00)*** (0.08)* -0.0960 -0.0037 0.0018 (0.09)* (0.01)*** (0.31) 0.0214 -0.0003 -0.0054 (0.35) (0.62) (0.08)* 106 EMC ACRAT 0.2993 (0.00)*** -0.1561 (0.16) -0.1591 (0.00)*** 0.0158 (0.73) 0.2661 -0.0023 (0.05)** (0.24) -0.1704 0.0074 (0.22) (0.26) -0.1197 -0.0036 (0.09)* (0.57) 0.0239 -0.0015 (0.18) (0.18) 0.2771 (0.02)** -0.2021 (0.10)* -0.1099 (0.02)** 0.0349 (0.18) Correctly Number Pseudo classified obs. SELFDEAL of obs. R2 647 -0.0790 (0.17) 0.0829 (0.46) 0.0559 (0.56) -0.0599 (0.23) 15.48% 57.34% 196 30.29% 69.90% 214 25.72% 70.10% Results in Panel A of Table 4 also suggest that highly-leveraged firms (LEV) are more likely to issue a Rule 144A ADR that allows them to raise new capital. Furthermore, having a high asset growth rate (ASSETGR) increases the probability of selecting Level III and decreases the probability of choosing Level I. More precisely, because firms with high growth in their investment opportunities generally need to raise fresh capital to finance them, they are less likely to choose Level I since it does not offer this possibility. The fact that a firm comes from an emerging market (EMC) decreases the probability of choosing Level II by 0.1591, and increases the probability of choosing Rule 144A by 0.2993. This result stems from the fact that Level II ADRs require that listed firms observe partial compliance with U.S. GAAP and SEC rules, and does not offer the possibility of raising fresh capital. Therefore, choosing Level II is costly for firms from emerging markets and does not allow them to raise capital on U.S. markets. Issuing a privatization ADR (PRIVAT) increases the probability of choosing Level III and Rule 144A by 0.3181 and 0.3103, respectively, and decreases the probability of choosing Level I and Level II by 0.5509 and 0.0775, respectively. These findings bear out our prediction that governments are less likely to choose Level I and Level II to privatize their firms on U.S. markets. The Sarbanes-Oxley dummy variable (SOX) in Panel A of Table 4 shows that a firm that issues an ADR after April 24, 2002 is more likely to issue Rule 144A and less likely to choose Level II, which is in line with our predicted relation. Panel B and C show that (RELTOV) is significant for Level I firms. In addition, having a higher (ULOW) and (ULOWDIF) decreases the probability of choosing Level II ADRs. In general, the results of the multinomial logit models corroborate the evidence from the univariate analysis as well as the predicted relations between the explanatory variables and the probability of choosing a given type of ADR. 5. Does SOX affect ADR issuance? A number of recent studies focused on the costs and benefits associated with SOX compliance. For instance, Engel et al. (2007) report that 94% of the respondents of a March 2005 survey of 217 public companies by the Financial Executives Institute believe that the costs of SOX compliance exceed its benefits. In what follows, we examine whether and to what extent SOX affected the ADR issuance activity. Specifically, we investigate the following issues: (1) Did SOX lead to a reallocation across ADR programs? (2) Did the issuing firms’ characteristics change around the implementation of SOX? (3) Did the determinants of the likelihood of choosing one type of ADR in the period before SOX differ from the period after? Since our sample spans the period 1990 through 2006 and includes the four ADR programs, it provides us with a unique opportunity to examine these issues. 5.1. Is there any reallocation between the different ADR programs before and after SOX? We draw from the universe of ADRs those that were issued between 1998 and 2001 (the pre-SOX period), and between 2003 and 2006 (the post-SOX period), and we compare the percentage of each ADR 107 program from these two periods. We find that the share of capital-raising programs (i.e., Rule 144A and Level III) increases in the post-SOX compared to the pre-SOX period. Indeed, in the post-SOX period, 30.4% of all ADRs are issued as Rule 144A as compared to 12.7% in the pre-SOX period, and Level III programs attract 15.7% of the total ADRs in the post-SOX period compared to 10.3% in the pre-SOX. We also find that the proportion of Level II ADRs decreased in the post-SOX period, dropping from 27.1% to 14.4%. Likewise, firms issue relatively fewer Level I ADRs in the post-SOX period, decreasing from 41.9% to 39.5%. This evidence suggests that SOX induced a reallocation among ADR programs, as firms tend to issue more Rule 144A and Level III and fewer Level II ADRs in the post-SOX than in the pre-SOX period. Zingales (2006) also documents a significant increase in the number of 144A registrations after SOX. As for Level II, it became more costly after SOX as this program requires virtually the same compliances as Level III without the access to new capital, hence explaining the reluctance of firms to seek Level II ADRs, and therefore opting for Level III in the post-SOX period. This result is in line with what we discussed previously. In the following paragraph, we examine whether the Level II result is an artifact and is simply due to a change in firms’ characteristics that leads to this reallocation. 5.2. Are the characteristics of issuing firms different before and after SOX? We rely on our sample firms to discuss this issue. The unreported results show that firms issuing Rule 144A in the post-SOX period have lower growth opportunities (ASSETGR) and relative turnover ratio (RELTOV) than before SOX. After SOX, firms from emerging markets issue more Rule 144A programs than before. The absolute value of the difference in accounting rating standards (ACRAT) in the post-SOX period is higher than in the pre-SOX period for those firms issuing Rule 144A. More precisely, firms that opt for Rule 144A after SOX originate from countries that present weaker investor protection. Moreover, Rule 144A firms present a lower wedge between the ultimate control and cash flow rights in the postSOX period. Firms that choose Level I after SOX are smaller, and have lower growth opportunities (ASSETGR), leverage (LEV), and relative turnover ratio (RELTOV). The results for SIZE and LEV are consistent with Doidge et al.’s (2007b) findings. Also, the percentage of emerging market firms selecting Level I is smaller in the post-SOX period. The absolute values of the differences in accounting ratings (ACRAT) and investor protection (SELFDEAL) indices are higher. Finally, the ultimate control rights and the difference between the excess control rights are unexpectedly smaller for firms that choose Level I after SOX. Level II firms have a higher pre-tax income (INCOME), a higher difference in accounting ratings and lower leverage ratios after SOX compared to the period before SOX. These results are statistically significant at the 10% level. Finally, for Level III firms, the only difference is in the firms’ leverage, which is lower in the post-SOX than in the pre-SOX period. Overall, these univariate results suggest that generally firms’ attributes, except for leverage (LEV), do not change between the period before and after SOX across the four ADR programs. Moreover, the decrease in Level II programs after SOX is not due to a change in firms’ characteristics because, firstly, contrary to Level II programs, we assist in an increase in Level III ADRs in the post-SOX period, and secondly, the firms’ attributes, especially those required to be listed on the major U.S. exchanges, generally do not change for Levels II and III between the periods before and after SOX. In the next section, we perform a multivariate analysis that allows us to simultaneously control for all these determinants. 108 5.3. SOX and ADRs: multivariate analysis To examine whether the impact of our explanatory variables on the probability of choosing a given ADR program differs across the pre- and post-SOX periods, we re-estimate the multinomial logit models (reported in Table 4) for each sub-period independently. We separately re-estimate Panels A of Table 4 for each sub-period. The results of these estimations are summarized in Table 5. Table 5 shows that, generally, the estimated marginal effects are in accordance with those that we report in Table 4, and are consistent with the predicted relations in Table 2. We find that the Level I marginal effects for the asset growth variable (ASSETGR) exhibit a significant and different sign. Table 5 shows that while they have the same sign, some marginal effects are higher or lower in the postSOX period compared to the pre-SOX. When we examine, for example, the emerging market country dummy, firms from emerging market countries are more likely to choose Rule 144A in the post-SOX period. More precisely, Table 5 show that being an emerging market firm increases the probability of choosing Rule 144A in the post-SOX period by 0.6869. In the pre-SOX period, this variable is statistically significant only in Panel A, and the increase in the probability of choosing Rule 144A for an emerging market firm is only 0.1191.7 The evidence that firms from emerging markets are more willing to issue Rule 144A ADRs after SOX was enacted is consistent with Lins et al.'s (2005) and Zingales’ (2006) findings that emerging market firms that are faced with more financial constraints than are developed country firms need to raise more external capital on U.S markets, and are able to do so through Rule 144A (private placements) or Level III (public offerings). As Level III becomes costly after SOX, especially for emerging markets firms, these latter become more inclined to issue a Rule 144A ADR (a relatively less costly program), and raise external capital on U.S. markets among Qualified Institutional Buyers. 7 For Table 5, we test whether the estimated marginal effect of the emerging market dummy (EMC) in the post-SOX period is higher than in the pre-SOX period using a Wald test. This test shows that the difference between the estimated marginal effects is statistically significant at the 1% level. 109 Table 5: Structural change: multinomial logit estimations before and after the enactment of Sarbanes-Oxley Act This table reports the multinomial logit estimations of the choice between the four ADRs programs, namely Rule 144A, Level I, Level II, and Level III, before and after SOX. This table reports the marginal effects evaluated at the mean of the explanatory variables. The variables are defined in the Appendix.. Values between parentheses represent the P-values of the t test for the null hypothesis that the coefficient is equal to zero. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Panel Period Pre-SOX Type of ADR 144A SIZE INCOME ASSETGR LEV PRIVA 0.0035 -0.0864 0.0004 -0.0005 0.1094 0.1191 (0.60) (0.01)*** (0.18) (0.45) (0.32) (0.04)** Level I -0.0099 -0.0817 -0.0002 0.0007 -0.3907 0.0552 (0.61) (0.45) (0.83) (0.74) (0.00)*** (0.62) 0.0081 0.1009 -0.0008 0.0001 -0.0306 -0.1930 (0.74) (0.15) (0.37) (0.96) (0.78) (0.02)** -0.0017 0.0672 0.0006 -0.0004 0.3118 0.0187 (0.94) (0.20) (0.46) (0.73) (0.03)** (0.86) 144A -0.0087 0.0336 0.0000 0.0072 0.2387 0.6869 (0.67) (0.22) (0.97) (0.00)*** (0.18) (0.00)*** Level I -0.0191 -0.0118 -0.0013 -0.0009 -0.7041 -0.5485 (0.32) (0.73) (0.13) (0.69) (0.00)*** (0.00)*** 0.0113 0.0023 0.0001 -0.0013 -0.0845 -0.1067 Level II Level III Panel A Post-SOX Level II Level III ULOW ULOWDIF EMC (0.00)*** (0.38) (0.72) (0.00)*** (0.00)*** (0.00)*** 0.0164 -0.0242 0.0012 -0.0050 0.5499 -0.0317 (0.24) (0.53) (0.12) (0.01)*** (0.00)*** (0.65) 110 ACRAT SELFDEAL Number of obs. Pseudo R2 196 12.86 244 29.76 6. Conclusion In this paper, we examine the determinants of firms’ decisions to issue one of the four available ADR programs on an individual basis (Level I, Level II, Level III, and Rule 144A). These four options have distinct attributes. On the one hand, only Level III and Rule 144A offer an access to U.S. primary capital markets (i.e., raising capital) through public offerings and private placements, respectively. On the other hand, Level II and Level III (listed programs) are more restrictive in terms of governance and disclosure requirements as compared to Level I and Rule 144A (unlisted programs). Our empirical evidence shows that capital raising Level III programs attract large firms with high pre-tax income, firms with high growth opportunities, privatized firms, and firms from weak investor protection environments, which is consistent with the bonding hypothesis. We likewise find that firms from weak investor protection environments are also attracted to Rule 144A programs. In addition, we find that after SOX foreign firms are more reluctant to issue Level II ADRs as SOX switches the expected costs/benefits associated with these programs. Finally, we document that firms with high ultimate control rights and excess control rights and those from emerging markets are less likely to choose Level II and more likely to choose Level I. We also examine whether the introduction of SOX in 2002 had an impact on the choice of a particular ADR. We find that after SOX, firms are more attracted by capital-raising programs, either Level III or Rule 144A, and are more reluctant to issue Level II. Indeed, this inter-program reallocation shows that raising fresh capital on U.S. markets seems to be an important motive to cross-list after SOX. Our multivariate analysis shows that being an emerging market firm heightens the probability of choosing Rule 144A after SOX compared to the period before. Similarly, coming from a country with weak legal protection of minority shareholders increases the likelihood of cross-listing under Level III in the postSOX period as compared to the pre-SOX. This latter result is consistent with more bonding and the enhanced bonding benefits after the implementation of SOX. Indeed, the corporate governance requirements of SOX strengthen the bonding characteristics of the listed programs (Level II and Level III). 111 Appendix: Variables: definitions and sources Variables SIZE INCOME ASSETGR LEV RELTOV PRIVA SOX ULOW ULOWDIF EMC ACRAT SELFDEAL Definition The natural logarithm of total assets in thousands of U.S. Dollars one year before issuing an ADR The pre-tax income in billions of U.S. Dollars one year before issuing an ADR The annual asset growth of the ADR firm one year before issuing an ADR The leverage ratio, which is equal to total debts divided by total assets one year before issuing an ADR The ratio between the turnover volume of the underlying firm and the turnover volume of its local market one year before issuing an ADR A dummy variable that is 1 when the firm was privatized by issuing ADR, and 0 otherwise A dummy variable for the Sarbanes-Oxley Act which is equal to 1 if one firm issues its ADR after April 24, 2002, and 0 otherwise Sources Worldscope Disclosure, Economatica (for Latin America), Amadeus (for Europe), Orbis, countryspecific company handbooks, firms’ websites, and firms’ financial reports DataStream World Bank, Privatization Barometer website, firms’ websites, and the Bank of New York website Bank of New York, Citibank, Deutsche Bank, and JPMorgan websites, Lexis/Nexis, NYSE website, and Litvak (2007) The percentage of the total ultimate control rights held by the ultimate owner Worldscope Disclosure, Economatica (for Latin America: Brazil, Chile, Colombia, Peru, and of the ADR firm one year before issuing an ADR Venezuela), Amadeus (for Europe), Orbis, countryThe percentage point difference between the ultimate control rights and the specific company handbooks, firms’ websites, and ultimate cash flow rights of the ultimate owner of the ADR firm one year firms’ financial reports before issuing an ADR A dummy variable that is equal to 1 if the firm’s country of origin is an Standard and Poor’s Emerging market Database emerging market, and 0 otherwise (EMDB) The difference in the accounting ratings between the firm’s country of origin La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1998) and the U.S. The difference in the anti-self dealing index between the firm’s country of Djankov, La Porta, Lopes-de-Silanes, and Shleifer origin and the U.S. (2006) 112 Bibliography Bank of New York, The depositary receipts markets, 2005 yearbook. Campbell, J., 1996, Understanding risk and return, Journal of Political Economy 104, 298-345. Citibank, 2005, Depositary receipts: information guide. Claessens, S., Djankov, S., and Lang, L., 2000, The separation of ownership and control in East Asian corporations, Journal of Financial Economics 58, 81-112. Coffee, J., 1999, The future as history: the prospects for global convergence in corporate governance and its implication, Northwestern Law Review 93, 641-707. Coffee, J., 2002, Racing toward the top?: The impact of cross-listings and stock market competition on international corporate governance, Columbia Law Review 93, 1757-1831. Djankov, S., La Porta, R., Lopez-de-Silianes, F., and Shleifer, A., 2006, The law and economics of selfdealing, Journal of Financial Economics, forthcoming. Doidge, C., Karolyi, G. A., Lins, K., Miller, D., and Stulz, R. M., 2007a, Private benefits of control, ownership, and the cross-listing decision, Ohio State University working paper. Doidge, C., Karolyi, G. A., and Stulz, R. M., 2007b, Has New York become less competitive in global markets? Evaluating foreign listing choices over time, Ohio State University working paper. Engel, E., Hayes, R., and Wang, X., 2007, The Sarbanes-Oxley Act and firms’ going-private decisions, Journal of Accounting and Economics, forthcoming. Faccio, M., and Lang, L., 2002, The ultimate ownership of Western European corporations, Journal of Financial Economics 65, 365-395. Greene, W., 2003, Econometric analysis, Prentice Hall. Hausman, J., and McFadden, D., 1984, A specification test for the multinomial logit model, Econometrica 52, 1219-1240. Hensher, D., Rose, J., and Greene, W., 2005, Applied choice analysis: a primer, Cambridge University Press. JPMorgan, February 2006, ADR reference guide. JPMorgan, March/April 2006, The post-SOX challenge for ADRs. Karolyi, G. A., 2006, The world of cross-listings and cross-listings of the world: challenging conventional wisdom, Review of Finance 10, 1-54. La Porta, R., Lopez-de-Silanes, F., and Shleifer, A., 1999, Corporate ownership arround the world, Journal of Finance 54, 471-517. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., and Vishny, R., 1998, Law and finance, Journal of Political Economy 106, 1113-1155. Lins, K., Strickland, D., and Zenner, M., 2005, Do non-U.S. firms issue equity on U.S. exchanges to relax capital constraints?, Journal of Financial and Quantitative Analysis 40, 109-133. Litvak, K., 2007, The effect of the Sarbanes-Oxley Act on non-US companies cross-listed in the US, Journal of Corporate Finance, forthcoming. McFadden, D., 1974, Conditional logit analysis of qualitative choice behavior, in P. Zarembka (ed.), Frontiers of Econometrics, New York: Academic Press, 105-142. 113 Piotroski, J., and Srinivasan, S., 2007, The Sarbanes-Oxley Act and the flow of international listings, University of Chicago working paper. Reese, W., and Weisbach, M., 2002, Protection of minority shareholder interests, cross-listings ine the United States, and subsequent equity offerings, Journal of Financial Economics 66, 65-104. Stulz, R., 1999, Globalization of equity markets and the cost of capital, Journal of Applied Corporate Finance 12, 8-25. Zingales, L., 2006, Is the U.S. capital market losing its competitive edge?, Journal of Economic Perspectives, forthcoming. 114 ASAC 2008 Halifax, Nova Scotia Narjess Boubakri Hatem Ghouma (Ph.D. Student) Department of Finance HEC Montreal EARNINGS MANAGEMENT AND BOND COSTS AND RATINGS We investigate the relationship between managers’ opportunism and the bond costs and ratings. We conjecture that firms where managers are more opportunist bear higher bond costs and have lower bond ratings. Using earnings management to proxy for managers’ opportunism, we find that firms that report small losses/profits have higher costs. Moreover, we find that abnormal accruals increase debt costs and reduce bond ratings. Finally, it appears that bondholders and rating agencies are less tolerant with firms that overstate reported incomes (i.e. with positive abnormal accruals) than with those that understate them. Introduction Recent studies show that lenders do not only rely on the firm’s past profitability and on the issue characteristics in order to infer the expected cash flows (and default probability). In fact, investors also price the firm’s corporate governance structure. This is essentially due to the fact that the firm’s success (and hence its ability to pay back its bondholders) is closely related to the extent of agency conflicts within the firm. Specifically, debtholders face the risk of a potential managers’ opportunistic behaviour. In financial theory, management behaviour can exacerbate the default probability of the firm. The “managerial” firm defined by Berle and Means (1932) is characterized by a separation between ownership and control, and is later defined by Jensen and Meckling (1976) as being “a nexus for a set of contracting relationships”1 among a number of individuals with conflicting objectives. Within this firm, managers are not a perfect agent for shareholders because they may adopt a non value-maximizing behaviour, and engage in self-serving activities such as empire building and perquisite consumption at the expense of shareholders. Moreover, since they invest their human 1 Jensen and Meckling (1976), p. 310. 115 capital in the firm, managers are less willing to engage in risky activities. This, in turn, harms shareholders that seek to have a well-diversified portfolio. Also, because of their limited horizon in the firm, managers have incentives to favour short-run projects rather than projects that ensure a continuity of the firm in the long run. In order to resolve these conflicts, Jensen and Meckling (1976) claim that the firm will support some agency costs that might reduce its value. To reduce information asymmetry, and hence these agency problems, managers are legally required to publish all financial and accounting information that outsiders need in order to assess the true current and future economic situation of the firm. Since these information condition investors’ decision whether to buy or to sell their stocks, managers have incentive to use their discretion in reporting financial information. For example, managers could use their judgment in estimating losses from bad debts, or even in shifting expenditures and gains between periods. SEC Chairman Arthur Levitt says that “earnings reports reflect the desires of management rather than the underlying financial performance of the company.” Manipulating accounting data to be in conformity with the “managers’ desires”, is commonly known as “Earnings Management”. In 1989, Schipper defines Earnings management (EM) as a “purposeful intervention in the external financial reporting process, with the intent of obtaining some private gain.” More recently, Healy and Wahlen (1999) define EM as the fact “when managers use judgment in financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting numbers.” Many factors could motivate the recourse of managers to EM (e.g. to reach analysts’ forecasts, to satisfy contractual obligations, to get promotions or compensations, to reach internal unrealistic plans and budgets,...). Nevertheless, and regardless of the motivation to the managers’ recourse to such practice, both investors and regulatory authorities generally agree on the fact that EM is a misleading activity that hides behind it an opportunistic behavior.2 Thus, for many outsiders, EM means a poor financial and accounting reporting quality that leads to a lack of confidence of investors on how the company is ran. In this paper, we focus on the impact of such an activity (EM) on the attitudes of two types of outsiders: bondholders and rating agencies. These attitudes are reflected respectively on the yields they ask for (for the bondholders) and on the ratings they grant to the firm (for the rating agencies). The majority of studies dealing with EM focused on three main questions: why there is EM, how do managers practice such activity, and what are its consequences. Surprisingly, a large body of this developing literature mainly analyzes the impact of this practice from a shareholders point of view. For example, in relation with the Initial Public Offerings, some studies document that companies use the income-increasing accruals3 2 We should notice that in the EM literature there are two main perspectives : an opportunistic perspective that defines EM as a mean to hide deteriorating performance, and an information perspective that defines it as a way to make accounting reports more informative. In the EM empirical studies, the opportunistic perspective is well documented, while the information perspective has not been tested yet. 3 Accruals are the difference between the earnings and the cash flows of the same period. 116 just before the IPO in order to increase their proceeds (Teoh et al., 1998a and 1998b; DuCharme et al., 2001, 2004; Roosenboom et al., 2003). Some other studies, however, do not confirm these results (Aharony et al., 1993; Kimbro, 2005; Zaluki, Campbell and Goodacre, 2007). Although the remarkable current size and growth of the corporate debt markets (particularly in the USA), a little attention has been devoted to the relation between EM and debt financing. For example, Kieschnick and Urcan (2006) analyze the behavior of firms prior to the issuance of convertible debt. Their results document a possible income increasing discretionary accruals prior to the issuance, especially for firms who use public placement. Furthermore, Chin et al. (2005) find that in Taiwan, earnings management was pervasive in the year of issue of the convertible bonds. In a study relatively closer to ours, Bharath et al. (2004) find that firms with higher earnings management pay higher costs on their bank loans. The authors interpret their finding by the fact that banks are considered as sophisticated investors able to detect income manipulation by firms, and hence to charge higher costs for firms that practice those activities. Whether or not bondholders and rating agencies are sophisticated enough to detect and penalize managers’ opportunism (through earnings management activities) remains an open empirical question that needs to be explored. The present research aims to fill this gap in the literature by trying to extent the examination of the effect of EM on some stakeholders (other than shareholders). Mainly, the paper aims to document to which extent American investors are sophisticated and to shed lights on whether rating agencies assume their role as effective “monitors”. Obviously, an opportunistic EM misleads outsiders’ perception of the true economic performance of the company and thus reduces their utilities. Debtholders, as an important firm’s capital supplier, should worry about the presence of such activities in the firm. Indeed, managers could use their discretion to alter earnings for two main reasons: to find the appropriate (debt) financing for their projects (Sercu, Bauwhede and Willekens, 2006; Kieschnik and Urcan, 2006), and/or to respect their debt contract covenants (DeFond and Jiambalvo, 1994). Expecting such behavior, bondholders will look for solutions to avoid being expropriated. In fact, bondholders decide whether to buy or not the company’s bonds by analyzing its economic performance as well as many other factors. Unfortunately, economic performance is essentially drawn from financial statements that the managers themselves have reported. Bond yields could have a disciplining role in this context, since bondholders could ask for higher yields if they expect an opportunistic behavior through EM. At the same time, rating agencies will downgrade their scores for firms where managers practice EM. Our first testable hypothesis is: H1. Firms with lower earnings management enjoy lower costs of debt and higher bond ratings. Accounting and financial literatures give evidence on the tendency of managers to overstate earnings in years of poor performance and to understate them in years of good performance. Since financial markets are more sensitive to poor performance, and since 117 understating earnings implies a reserve for future years, we expect that bondholders and rating agencies respond asymmetrically to the direction of the earnings manipulations. Thus, our second hypothesis will be: H2. Bondholders and rating agencies are less tolerant with firms that overstate reported incomes (i.e. with positive abnormal accruals) than with those that understate them (i.e. with negative abnormal accruals). Earnings Management Measures In this study, we have retained four proxies for earnings management that reflect different dimensions of the discretion of managers in reporting earnings. These proxies belong to two different families: Income smoothing-based measures and accruals-based measures. Income Smoothing Measures Income smoothing is a particular case of earnings management by which corporate insiders use their discretion to make earnings look more stable over time and soften economic shock that might hit their stability. Smoothing the income increases the attraction of the firm which appears a less risky investment in the eyes of investors (Trueman and Titman, 1988). Depending on the techniques used by managers to make the earnings smoother, we can distinguish between two theories of income smoothing: the information theory and the opportunism theory. If managers use real techniques to smooth earnings (using business decision, example easing credit terms to promote the period’ sales), they could increase the earnings quality and the information communicated to the market. However, if they use artificial techniques (accounting choices based on discretionary accruals), they could opportunistically manipulate income to mislead different stakeholders (Koch, 1981). For this study, we use three main proxies belong to this category of measures. First, we use the standard deviation of Income divided by the standard deviation of Operating Cash Flow (SMOOTH1). If managers are using their discretion to smooth earnings, then one could expect a high standard deviation of operating cash flow compared to the standard deviation of the reported income, resulting in a low value of SMOOTH1. Our second measure is the large loss avoidance (SMOOTH2). It is a dummy variable that equals 1 if the ratio of net income over total asset falls into [-0.01 , 0.01] and 0 otherwise. If the ratio of large loss avoidance falls in this interval, we suspect that earnings were heavily manipulated to avoid larger losses (Degeorge et al., 1999; Burgstahler and Dichev, 1997; Leuz et al., 2003). Finally, our third smoothing measure, SMOOTH3, is the correlation between the changes in operating cash flow and the changes in Abnormal Accruals (measured using performance-matched based on Jones' Model). A high negative correlation implies that managers use their discretion to manipulate earnings (Skinner and Myers, 1999). Accruals Measures 118 Beyond their ability to smooth reported earnings (by reducing its fluctuation), managers have a considerable discretion to manipulate firm’s performance. Scholars have measured this discretion by the magnitude of abnormal accruals (see for example Haw et al., 2003; Teoh et al., 1998a and 1998b among others). We compute the abnormal accruals using the modified Jones’1991 model to which we add the adjustment proposed by Kothari et al. (2005). This adjustment suggests a further ROA-matching to take into account the performance of the firm. Data Sources and Methodology Our Debt data are from The Fixed Investment Securities Database. This database contains information on the type of the issue, the coupon, the yield to maturity, the rating, the maturity, the offering amount as well as many other characteristics (covenants, …). Data on issuer characteristics (performance, risk, size, and leverage) are from Compustat. We analyze all American corporate bond issues during the period between 1995 and 2006. We follow Anderson et al. (2003) and we gather all issues of the same firm during the same year in a portfolio. This lets us to compute for each firm-year the characteristics of the bond portfolio. Firm bond portfolio characteristics (rating, yield to maturity, maturity, offering amount,..) are the weighted-average of the characteristics of the bonds issued during the same year by the same firm. We used the issue size as weightings. The variables used in this study are: COST : RATING : LMAT : ISIZE : COVNT : CONVRT : SMOOTH1 : SMOOTH2 : The weighted-average yield to maturity on the firm’s outstanding traded debt minus the yield to maturity on US treasury bond of similar maturity. We used the issue size as weightings The weighted-average S&P rating scores of the firm’s outstanding traded bonds. Depending on the S&P score, we operate a transformation of this variable as follows: if the S&P score ranges between D and CCC+ then bond rating is equal 1, if it ranges between B- and B+ then bond rating is equal 2, if it ranges between BB- and BB+ then bond rating is equal 3, if it ranges between BBB- and BBB+ then bond rating is equal 4, if it ranges between A- and A+ then bond rating is equal 5, if it ranges between AA- and AA+ then bond rating is equal 6, and if it is AAA then bond rating is set to be 7. The logarithm of the weighted-average years to maturity on the firm’s outstanding bonds. We used the issue size as weightings The logarithm of the weighted-average size (offering amount) of the firm’s outstanding bonds. The percentage of firm's bonds with covenants. It is a weighted-average of a dummy variable that equals 1 if the firm's outstanding bond has a covenant and 0 otherwise. The percentage of firm's convertible bonds. It is a weighted-average of a dummy variable that equals 1 if the bond is convertible, 0 otherwise Standard deviation of Income Before Extra Items (Compustat item #18) Divided by the standard deviation of Operating Cash Flow Large Loss Avoidance. It equals 1 if the ratio of net income over total asset falls into [-0.01, 0.01] and 0 otherwise. 119 SMOOTH3 : The correlation between the changes in operating cash flow and the changes in Abnormal Accruals (measured using performance-matched based on Modified Jones' Model) Abnormal Total Accruals : Abnormal Current Accruals : It is the abnormal long term accruals from the Modified Jones' model (1991) in spirit of Kothari et al. (2005). First we Compute Modified Jones' 1991 abnormal accruals for our sample. Then, we match each firm with another firm-year in the same 2-digit SIC code and having the closest ROA. The Modified Jones-model performance-matched abnormal accrual is the difference between the firm's Modified Jones' model abnormal accrual and the matched firm's Modified Jones-model abnormal accrual for the same year. We process exactly as for the Abnormal Total (Long Term) Accruals with the unique difference is that we exclude the PPE (net property, plant and equipment) when estimating normal accruals. PERFORM : LEVERAGE: RISK : LASSET : Measured by the ROA Leverage as long term debts divided by total assets The standard deviation of the Net Income for the 5 recent years (or the maximum available) Logarithm of total assets To test the relation between earnings management and bond yields and ratings, we use the two following general specifications: (1) COST = f (Earnings management proxy, Issuer Characteristics, Issue Characteristics) (2) RATING = f (Earnings management proxy, Issuer Characteristics, Issue Characteristics) The first model (bond costs) is estimated using the OLS method. The second model (bond ratings) is estimated using an Ordered Probit Model, since the dependent variable is ordinal (S&P ratings are classified in seven ordering categories - see variable description above). Empirical Results Descriptive Statistics Panel A of Table 1 reports descriptive statistics of the variables used in this study. During the sample period, the average (median) of the yield to maturity is about 207.65 bps (163.78). The average (median) rating for our sample during the same period is 5.40 (5) which falls between A- and AA+ (A- and A+) in our transformation scale. Panel B of the same table presents Pearson correlations between the explanatory variables and our two key dependent variables, COST and RATING. Among the five measures of Earnings Management, it seems that SMOOTH1 and SMOOTH3 have no effect either on debt cost or on its rating. However, SMOOTH2 and Abnormal Current Accruals have, as expected, a positive and significant affect on debt costs only, while Abnormal Total Accruals have a negative and significant affect on debt ratings only. 120 Table 1 : Descriptive Statistics and Correlations Panel A:Descriptive Statisitics Panel B: Pearson correlations COST RATING Corr. Sig. N - 0.1442* (0.000)*** 1386 5 Corr. Sig. N 0.1442 (0.000)*** 1386 - 1.819599 .7356276 Corr. Sig. N 0.0450 (0.115) 1228 0.0086 (0.693) 2084 .1732283 .3785067 0 Corr. Sig. N 0.2445 (0.000)*** 1565 -0.0052 (0.790) 2599 2183 -.168238 .6661566 -.2919286 Corr. Sig. N 0.0234 (0.428) 1148 0.0079 (0.732) 1894 Abnormal Total Accruals 2374 -.0166133 1.163239 -.0050626 Corr. Sig. N 0.0258 (0.364) 1238 -0.0568 (0.009)*** 2056 Abnormal Current Accruals 2418 0.031758 1.067584 -0.0058 Corr. Sig. N 0.0554* (0.050)** 1250 -0.0232 (0.288) 2095 PERFORM 2689 .0436945 1.328515 .041327 Corr. Sig. N -0.0199 ( 0.446) 1464 0.0068 (0.739) 2374 LEVERAGE 2766 .3248613 .2471414 .2892902 Corr. Sig. N 0.3451* (0.000)*** 1510 -0.0182 (0.369) 2444 RISK 2729 141.7446 586.0199 38.30792 Corr. Sig. N -0.2025 (0.000)*** 1480 -0.0397 ( 0.051)* 2405 LASSET 2771 7.245956 1.602953 7.170286 Corr. Sig. N -0.4715 (0.000)*** 1510 -0.1506 (0.000)*** 2448 LMAT 3031 2.241498 .5223319 2.279022 Corr. Sig. N -0.1885 (0.000)*** 1642 -0.0635 (0.000)*** 2674 ISIZE 3048 336007 1883613 200000 Corr. Sig. N -0.0761 (0.002)*** 1642 -0.0157 ( 0.417) 2688 COVNT 3048 .6720454 .3838623 1 Corr. Sig. N -0.5145 (0.000)*** 1642 -0.1047 (0.000)*** 2688 CONVRT 3048 .2587329 .4299931 0 Corr. Sig. N -0.1234 (0.000)*** 1642 0.1908 (0.000)*** 2688 COST N 1642 Mean 207.6445 Std. Dev. 159.1381 RATING 2688 5.405878 1.476801 SMOOTH1 2406 .9453357 SMOOTH2 3048 SMOOTH3 121 Median 163.674 Bond Costs and Earnings Management We start with the analysis of the effect of smoothing income measures on corporate bond costs. Table 2 reports results of the regression of our set of explanatory variables on debt costs. In column 2 we report our basic model. All the variables are significant and have their expected signs except for the maturity which is not significant at any conventional statistical level. Particularly, bond spreads are positively affected by the firm’s leverage, and by its risk level. Moreover, bond spreads respond negatively to a higher firm performance, a larger issue and firm size, and to the percentage of bonds with convertible and covenants issued by the firm during the year. The three last columns of the same table reports regressions results when we include one of our three smoothing measures. As we can notice, only SMOOTH2 is significant (at 1% level) is positive and has, as expected, a positive sign. That’s, it appears that bond yields are higher for firms that try to avoid losses by reporting a ratio of income over assets which is very close zero. The two other income smoothing measures are insignificant suggesting that bondholders give less attention to both the correlation between operating cash flows and accruals and to the ratio of standard deviations of income changes and operating cash flow changes. One could explain this result by the fact that SMOOTH2 is simpler and more directly observable for bondholders than the two other measures. Actually, when a firm reports small losses or small profits, bondholders could easily suspect a potential hidden manipulation of earnings that have permitted to managers to reach that earnings’ level and hence avoid larger losses. However, for the SMOOTH1 and SMOOTH3, bondholders need to be more sophisticated to detect and understand such manipulations. Finally, we proxy for management opportunism using two accruals-based measures; Abnormal total Accruals and Abnormal Current Accruals. Abnormal accruals are computed using the performance-matched model (Kothari et al. 2005) based on the modified Jones’ model. Table 3 shows OLS regression results. We report separately results for abnormal total accruals and abnormal current accruals. We start by including in our basic model the unsigned abnormal accruals. At glance, column 3 and column 7 of Table 3 suggest that the unsigned abnormal accruals (both Total and Current) are not significant determinants of bond costs. That means that the magnitude of the abnormal accruals itself doesn’t matter for bondholders. Hence, our second step is to analyze the effect of signed abnormal accruals on costs to see whether the sign (direction of the accruals manipulation) does matter for them. Previous studies suggest that managers can use their discretion to overstate reported earnings during poor performance years and understate reported earnings during good performance years. This is mainly due to the fact that outside investors are very sensitive to the variability of earnings. To see whether bondholders’ perception remains unchanged for firms that overstate earnings and for those that overstate it, we proceed in two ways: first, we simply run our cost model using signed abnormal accruals rather than its absolute values. Second, we split our sample into two sub-samples depending on the sign of the firm’s abnormal accrual. Column 2 and 6 of Table 3 show the results for the whole sample using signed accruals (respectively for Total and Current Abnormal Accruals) as a measure of managerial opportunism. The coefficients of Abnormal Accruals (in both models) are not significant at any conventional statistical level. As for the two sub-samples, columns 4 and 5 of the same table show that abnormal total accruals are significant only for the positive accruals 122 sample (a positive coefficient of 23.98 significant at less than 5% level). That’s, bondholders seem to ask for higher bond costs from firms that overstate their earnings, but they are more tolerant with those that understate earnings. As for Abnormal Current Accruals (columns 8 and 9 of Table 3), there is no significant effect on corporate bond cost. Overall, Bond costs are significantly higher for firms that report small losses or small profits and for those that overstate their income using long term accruals. Table 2: Bond Costs and Income Smoothing measures This table reports OLS Regression results for the following model: COST = a1 PEFORM + a2 LEVERAGE + a3 RISK + a4 LASSET + a5 LMAT + a6.ISIZE + a7.COVNT + a8.CONVRT + a9.Income Smoothing Measure + Year Dummies+ Industry Dummies + ei Dependent variable = COST Basic Model (2) (3) (4) 613.1934 566.3834 616.2243 518.1177 (0.000)*** (0.000)*** (0.000)*** (0.000)*** PERFORM -90.48903 -86.02432 -84.12494 -210.0102 (0.055)* (0.053)* (0.062)* (0.002)** LEVERAGE 100.2361 97.58037 96.53231 114.2041 (0.000)*** (0.000)*** (0.000)*** (0.000)*** Constant RISK LASSET LMAT ISIZE COVNT 0.0393965 0.0405434 0.040561 .041425 (0.000)*** (0.000)*** (0.000)*** (0.000)*** -39.65524 -37.73353 -40.00028 -39.06919 (0.000)*** (0.000)*** (0.000)*** (0.000)*** -3.816442 .1188201 -5.003331 -2.988734 (0.511) (0.981) (0.393) (0.648) -0.00000141 -1.48e-06 -1.38e-06 -1.06e-06 (0.000)*** (0.000)*** (0.000)*** (0.000)*** -102.2153 -87.3329 -101.43 -101.8526 (0.000)*** (0.000)*** (0.000)*** (0.000)*** -199.3893 -183.1057 -196.1392 -202.7507 (0.000)*** (0.000)*** (0.000)*** (0.000)*** Year Dummies Yes Yes Yes Yes Industry Dummies Yes Yes Yes Yes CONVRT -1.487097 SMOOTH1 (0.213) 35.7131 SMOOTH2 (0.002)*** -0.7942628 SMOOTH3 (0.879) N 1461 1249 1461 1174 F 53.25 38.93 47.85 40.45 (0.000)*** (0.000)*** (0.000)*** (0.000)*** 42.05 41.14 42.46 43.61 Sig. Adj. R-Square (%) 123 Table 3 : Bond Costs and Abnormal Accruals measures This table reports OLS Regression results for the following model: COST = a1 PEFORM + a2 LEVERAGE + a3 RISK + a4 LASSET + a5 LMAT + a6.ISIZE + a7.COVNT + a8.CONVRT + a9.Accruals Measure + Year Dummies+ Industry Dummies + ei Dependent variable = COST Constant PERFORM LEVERAGE RISK LASSET LMAT ISIZE COVNT CONVRT Year Dummies Industry Dummies Signed Abnormal Accruals Abnormal Total Accruals (ATA) ATA (UnSinged) Whole Sample ATA Negative Only ATA Positive Only ACA (Singed) Whole Sample ACA (UnSinged) Whole Sample ACA Negative Only ACA Positive Only 533.5892 (0.000)*** -85.13294 (0.100)* 109.1171 (0.000)*** .0443145 (0.000)*** -41.37508 (0.000)*** -3.043231 (0.655) -1.18e-06 (0.000)*** -107.2514 (0.000)*** -197.1486 (0.000)*** Yes Yes 3.764165 (0.447) 534.1587 (0.000)*** -84.44042 (0.101) 109.1034 (0.000)*** .0435025 (0.000)*** -41.2138 (0.000)*** -3.637328 (0.593) -1.18e-06 (0.000)*** -107.7124 (0.000)*** -199.0028 (0.000)*** Yes Yes 644.112 (0.000)*** -106.0791 (0.035)** 87.91049 (0.005)*** .0525562 (0.002)*** -45.47829 (0.000)*** 6.003969 (0.437) -3.87e-07 (0.617) -112.822 (0.000)*** -195.9395 (0.000)*** Yes Yes 484.8831 (0.000)*** -60.99084 (0.416) 142.4236 (0.000)*** .0416447 (0.002)*** -37.65255 (0.000)*** -9.612209 (0.425) -1.63e-06 (0.000)*** -102.5469 (0.000)*** -217.1509 (0.000)*** Yes Yes 625.438 (0.000)*** -86.07373 (0.101) 109.1571 (0.000)*** .044325 (0.000)*** -41.46423 (0.000)*** -3.335486 (0.618) -1.20e-06 (0.000)*** -107.2876 (0.000)*** -197.2673 (0.000)*** Yes Yes -2.738735 (0.840) 620.4142 (0.000)*** -89.80661 (0.095)* 109.1236 (0.000)*** .0436771 (0.000)*** -41.11013 (0.000)*** -3.518731 (0.598) -1.20e-0 (0.000)*** -107.1218 (0.000)*** -197.3362 (0.000)*** Yes Yes 647.534 (0.000)*** -46.12619 (0.266) 109.1553 (0.002)*** .0446312 (0.002)*** -45.65927 (0.000)*** -3.601398 (0742) -1.18e-06 (0.000)*** -105.7841 (0.000)*** -187.133 (0.000)*** Yes Yes 512.0844 (0.000)*** -343.9664 (0.000)*** 95.11097 (0.005)*** .0420148 (0.005)*** -36.97648 (0.000)*** -1.131844 (0.883) -9.73e-07 (0.070)* -102.5479 (0.000)*** -225.4329 (0.000)*** Yes Yes 4.638361 .2909413 23.98019 17.85299 3.116305 10.41084 (0.158) (0.947) (0.029)** (0.205) (0.369) (0.260) 1238 640 598 1250 666 572 UnSigned Abnormal Accruals N F Sig. Adj. R-Square (%) Abnoraml Current Accruals (ACA) ATA (Singed) Whole Sample 1238 1250 48.06 50.24 22.68 43.60 51.94 55.60 33.87 23.38 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** 43.06 43.07 45.75 39.85 43.13 43.20 44.67 43.04 124 Bond Rating and Earnings Management Now, let’s turn to rating agencies. Column 2 of Table 4 reports Ordered Probit Estimations of the rating model without including any proxy for managerial opportunism. Our explanatory variables are significant at less than 10% level with their expected signs, except for the size of the issue and the percentage of the bonds issued with covenants which are not significant. The three last columns of the same table present results when we include separately the income smoothing measures. As we can notice, none of these measures has a significant effect on debt ratings. As we have done with debt cost, we employ the same strategy to analyze the impact of abnormal accruals on bond costs. Table 5 reports results for both Abnormal Total and Current Accruals. Unsigned Abnormal Accruals are insignificant which means that rating agencies don’t (only) rely on the magnitude of abnormal total and current accruals to assess the quality of management reporting. To test whether they rely on the direction of the accruals manipulation (over/understating reported earnings), we regress the ratings on the signed accruals. As we can see from the same table, only abnormal total accruals variable is significant with a negative (as expected) sign. Signed Current accruals (column 6), while it is negative, it is insignificant. Further, we spilt our sample into two sub-samples depending on the sign of the abnormal accruals. Then we run our Probit estimation for each sub-sample. For the abnormal total accruals, it seems that rating agencies are very sensitive to overstating income since the coefficient of our key variable (abnormal total accruals) is negative and highly significant (at less that 1% level) for the positive abnormal accruals sub-sample. As regard the negative abnormal accruals sub-sample, it seems that these agencies are more tolerant since our key variable has no significant statistical effect. Overall, the more the managers overstate their firm’s reported earnings, the worst is the perception of rating agencies, and hence the lower is the rating score for the firm’s bonds. This comforts the role of rating agencies as sophisticated investors who can (indirectly) monitor firm’s behavior. 125 Table 4: Bond Ratings and Income Smoothing measures This table reports Ordred Probit Regression results for the following model: Prob. (Rating=R)= Z(a1 PEFORM + a2 LEVERAGE + a3 RISK + a4 LASSET + a5 LMAT + a6.ISIZE + a7.COVNT + a8.CONVRT + a9.Income Smoothing Measure + Year Dummies+ Industry Dummies + ei ) Dependent variable = RATING Where R is in{1,2,3,4,5,6,7} Basic Model (2) (3) (4) 0.2553364 .2016168 0.2601531 .1870591 (0.05)** (0.136) (0.046)** (0.318) LEVERAGE -0.2180982 -.2805808 -0.2250087 -.3107974 (0.04)** (0.014)** (0.036)** (0.009)*** RISK -0.0000371 -.0000395 -0.0000364 -.0000365 (0.063)* (0.057)* (0.070)* (0.102) -0.1171885 -.1106433 -0.1177833 -.1261967 (0.000)*** (0.000)*** (0.000)*** (0.000)*** -0.0943024 -.1195918 -0.0947962 -.1228112 (0.023)** (0.007)*** (0.022)** (0.008)*** -2.01E-10 -5.29e-10 -1.34e-10 5.38e-10 (0.923) (0.799) (0.949) (0.817) -0.0179948 -.0410517 -0.0168533 -.0213076 (0.788) (0.556) (0.801) (0.774) PERFORM LASSET LMAT ISIZE COVNT CONVRT 0.3488939 .3824678 0.3509905 .3461323 (0.000)*** (0.000)*** (0.000)*** (0.000)*** Year Dummies Yes Yes Yes Yes Industry Dummies Yes Yes Yes Yes (0.320) 0.0671563 SMOOTH2 (0.439) -.0148914 SMOOTH3 (0.686) N Model Chi-2 Sig. Pseudo R-Suqare (%) 2354 2120 2354 1919 313.63 298.98 316.65 272.29 (0.000)*** (0.000)*** (0.000)*** (0.000)*** 3.84 4.06 3.85 4.12 126 Table 5: Bond Ratings and Abnormal Accruals measures This table reports Ordered Probit Regression results for the following model: Prob.(RATING=R) = Z (a1 PEFORM + a2 LEVERAGE + a3 RISK + a4 LASSET + a5 LMAT + a6.ISIZE + a7.COVNT + a8.CONVRT + a9.Accruals Measure + Year Dummies+ Industry Dummies + ei) Where R is in{1,2,3,4,5,6,7} Dependent variable = RATING PERFORM LEVERAGE RISK LASSET LMAT ISIZE COVNT CONVRT Year Dummies Industry Dummies Signed Abnormal Accruals Abnormal Total Accruals (ATA) ATA (Singed) Whole Sample .1633213 (0.217) -.2607456 (0.019)** -.0000441 (0.017)** -.1211458 (0.000)*** -.0807069 (0.073)* 3.74e-10 (0.887) -.0246044 (0.730) .3197405 (0.000)*** Yes Yes -.0551307 (0.001)*** UnSigned Abnormal Accruals N Model Chi-2 Sig. Pseudo R-Suqare (%) 2046 283.66 (0.000)*** 4.05 Abnoraml Current Accruals (ACA) ATA (UnSinged) Whole Sample .1962806 (0.138) -.2512431 (0.025)** -.0000442 (0.016)** -.1195139 (0.000)*** -.0804882 (0.074)* 2.48e-10 (0.926) -.0275793 (0.699) .3241452 (0.000)*** Yes Yes ATA Negative Only .0957721 (0.625) -.3165794 (0.049)** -.0000907 (0.075)* -.1147267 (0.000)*** -.0766382 (0.218) 8.62e-09 (0.171) -.1220156 (0.239) .424733 (0.000)*** Yes Yes ATA Positive Only .2651898 (0.121) -.1976871 (0.211) -.0000214 (0.388) -.1306019 (0.000)*** -.0890514 (0.190) -1.08e-09 (0.718) .0943178 (0.345) .2166758 (0.023)** Yes Yes -.0230773 (0.378) 2046 272.74 (0.000)*** 3.95 .052101 (0.387) 1069 174.49 (0.000)*** 5.24 -.0459472 (0.000)*** 977 148.05 (0.000)*** 3.52 127 ACA (Singed) Whole Sample .1611541 (0.189) -.2473481 (0.026)** -.0000463 (0.009)*** -.1190195 (0.000)*** -.0826941 (0.063)* -1.27e-10 (0.956) -.015088 (0.832) .3415609 (0.000)*** Yes Yes -.0338444 (0.158) 2084 295.50 (0.000)*** 4.14 ACA (UnSinged) Whole Sample .1750449 (0.150) -.2437189 (0.028)** -.0000458 (0.010)** -.1191756 (0.000)*** -.0827272 (0.063)* -1.16e-10 (0.960) -.0135446 (0.849) .3415445 (0.000)*** Yes Yes ACA Negative Only .2446551 (0.083)* -.3405042 (0.036)** -.0000219 (0.778) -.125589 (0.000)*** -.0876284 (0.164) -1.08e-09 (0.716) .0613474 (0.561) .4860761 (0.000)*** Yes Yes ACA Positive Only .1105369 (0.715) -.1404867 (0.364) -.0000426 (0.003)*** -.119416 (0.000)*** -.0969826 (0.157) 2.47e-09 (0.716) -.0973275 (0.314) .1762741 (0.061)* Yes Yes -.0280121 (0.273) 2084 294.20 (0.000)*** 4.13 .0125441 (0.839) 1082 163.62 (0.000)*** 4.88 -.0312377 (0.180) 964 159.32 (0.000)*** 3.85 Conclusion This study aims to shed lights on the impact of managers’ opportunism and the cost of debt financing. We use earnings management activities, as measured by income smoothing and abnormal accruals, to proxy for managers’ opportunism, while we use the corporate bonds costs and their ratings to proxy for debt financing costs. We hypothesize that firms where managers are more opportunist bear higher bond costs and have lower bond ratings. Our results support this conjecture since we find that firms that report small losses or profits have higher bond costs. Moreover, we find that abnormal accruals increase bonds costs and reduce their ratings. Finally, we split our sample into two subsamples; firms that overstate their reported income and those that understate it. This is to see whether bondholders and rating agencies have the same perception for both direction of manipulation (increasing or decreasing income). Our results show that bondholders and rating agencies are less tolerant with increasing income firms than with decreasing income firms. We interpret this result by the fact that increasing income activities through abnormal accruals is more likely to be considered by bondholders and rating agencies as an opportunistic behavior. References Aharony, J., Lin, C-J., Loeb, M.P., 1993, “Initial public offerings, accounting choices, and earnings management.” Contemporary Accounting Research, 10, pp. 61-81. Ahmad-Zaluki, Nurwati A. A., Campbell, K. and Goodacre, A., 2007, “Earnings Management and Post-IPO Performance: An Investigation of Economic Crisis and Other Determinants”. Available at SSRN: http://ssrn.com/abstract=963085. Anderson C. R., Mansi A. S. and Reeb, D. M., (2003), “Founding Family Ownership and the Agency Cost of Debt.” Journal of Financial Economics, 68, pp. 263-285. Berle, A., and Means, G., (1932), The Modern corporation and Private Property, New York: Macmillan. Bharath, S. T., Sunder, J., and Sunder S. V., 2004, “Do Sophisticated Investors Understand Accounting Quality? Evidence from Bank Loans.” Working Paper, University of Michigan. Burgstahler, D., I. Dichev, 1997, “Earnings management to avoid earnings decreases and losses.” Journal of Accounting and Economics, 24, pp. 99-129. Chin, C. L., Lin, T. T., Lee C. C., 2005, “Convertible Bonds Issuance Terms, Management Forecasts, and Earnings Management: Evidence from Taiwan Market.” Review of Pacific Basin Financial Markets and Policies, 8, 3, pp. 543-571. DeFond, M, Jiambalvo, J., 1994, “Debt covenant violation and manipulation of accruals.” Journal of Accounting Economics, 18, 2, pp. 145-176. Degeorge, F., J. Patel, R. Zeckhauser, 1999, “Earnings manipulation to exceed thresholds.” Journal of Business, 72, pp. 1-33. DuCharme L.L., Malatesta P.H., and Sefcik S.E., 2001, “Earnings management: IPO valuation and subsequent performance.” Journal of Accounting, Auditing and Finance, 16, pp. 369-396. DuCharme L. L., Malatesta P. H., and Sefcik S. E., 2004, “Earnings Management, Stock issues, and shareholder lawsuits.” Journal of Financial Economics, 71, pp. 27-49. Haw, I. M., Hu, B., Hwang, L., and Hw,. W., 2003, “Ultimate Ownership, Income Management, and Legal. and Extra-Legal Institutions.” Journal of Accounting Research, 42, pp. 423–62. 128 Healy, P. M., and Wahlen, J. M., 1999, “A review of the earnings management literature and its implications for standards setting.” Accounting Horizons, 13, pp. 365-383. Jensen, M. and W. Meckling, (1976), “Theory of the Firm: Managerial Behavior, Ownership Structure.” Journal of Financial Economics, 3, pp. 305-360. Kieschnik, R, Urcan, O., 2006, “Earnings Management and Convertible Debt Issuance.” Available at SSRN: http://ssrn.com/abstract=817164. Kimbro M.B., 2005, “Managing underpricing? The case of pre-IPO discretionary accruals in China.” Journal of International Financial Management and Accounting, 16, 231-262. Koch, B. S., 1981, “Income Smoothing: an Experiment.” The Accounting Review, 56, pp. 574-586. Kothari, S.P., Leone A.J et Wasley C.E., 2005, “Performance matched discretionary accrual measures.” Journal of Accounting and Economics, 39, pp.163-197. Leuz, C., Nanda, D. and Wysocki P. D., 2003, “Earnings management and investor protection: an international comparison.” Journal of Financial Economics, 69, 3, pp. 505527. Roosenboom P., Goot T.V.D., and Mertens G., 2003, “Earnings Management and Initial Public Offerings: Evidence from Netherlands.” The International Journal of Accounting, 38, pp.243-266. Schipper, K., 1989, “Commentary on earnings management.” Accounting Horizons, 3, pp. 91-102. Sercu P, Vander Bauwhede H, Willekens M, 2006, “Earnings management and debt.” FETEW Research Report AFI_0619, K.U.Leuven, 25 pp. Skinner, D. and Myers. L., 1999, “Earnings momentum and earnings management.” Unpublished working paper, University of Michigan and University of Illinois. Teoh, S. H., Welch, I. and Wong, T. J., 1998a, “Earnings management and the underperformance of seasoned equity offerings.” Journal of Financial Economics, 50, 1, pp. 63-99 Teoh, S. H., Welch, I. and Wong, T. J.,, 1998b, “Earnings management and the long-run market performance of initial public offerings.” Journal of Finance, 53, 6, pp. 19351974. Trueman, B., and Titman, S., 1988, “An Explanation of Accounting Income Smoothing.” Journal of Accounting research, 26, pp. 127-143. 129 ASAC 2008 Halifax, Nova Scotia Christian Cadiou Nathalie Cotillard Jérémy Morvan IAE de Bretagne Occidentale Université de Bretagne Occidentale SECURITIZATION BUY OUT : LE CAS FRANCAIS La securitization buy out combine reprise d’entreprise par endettement et titrisation d’actifs de la société acquise. L’objet du papier est de décrire l’architecture du montage et à en présenter l’intérêt dans le contexte français. Le choix d’une structure financière est une décision de répartition des risques entre apporteurs de ressources (actionnaires et créanciers). L’accès à de nouveaux financements vise à développer le portefeuille d’investissement et à créer de la valeur économique. La diversité des sources de financement, avec un spectre de plus en plus large, rend toutefois difficile de telles opérations dans la mesure où il semble impossible d’appréhender tous les risques inhérents aux montages financiers complexes. Sur le marché de la reprise d’entreprises, le LBO (Leverage Buy Out) est une architecture fondée sur l’effet de levier. Cette technique de financement, qui semblait avoir atteint sa maturité (Le Nadant, 2000), surprend par les innovations successives qui participent à la sophistication du montage (Cadiou et al., 2007). Sur ce segment très concurrentiel, où l’imagination est la clef de l’avantage compétitif, le risque affecte à différents niveaux des partenaires toujours plus nombreux associés à l’opération. De fait, le montage, qui est par définition tendu, serait de plus en plus attentif à la gestion des comportements de ces parties prenantes associées à un projet très risqué. Auparavant simple et centrée sur la société holding, la structuration plus complexe de la dette amène actuellement à reconsidérer le risque financier. En effet, à côté des risques propres au montage (risques de crédit, de faillite…), les opérations de refinancement des apporteurs de ressources ajoutent un nouveau risque lié à la liquidité sur les marchés. Par exemple, l’introduction de la titrisation sur le marché de la dette senior des montages LBO a favorisé l’augmentation et l’allongement de la maturité de la dette en permettant aux banques de se refinancer auprès d’investisseurs présents sur le marché et de transférer le risque de crédit. Ainsi, si les banques financent de la dette senior, elles externalisent dès que possible le risque sur le marché primaire lors de la syndication des opérations de LBO. Parmi les innovations récentes sur ce marché, les actifs de la cible économique peuvent ainsi être refinancés par une titrisation économique : c’est le SBO (Securitization Buy Out). En revanche, contrairement à la titrisation de la dette senior LBO qui s’est démocratisée en France depuis son introduction sur le marché européen, le SBO ne s’adresse pour l’instant qu’à des opérations d’envergure. Les questions sont donc : quelles sont les opérations réalisées sur le marché français ? Quelles réalités recouvrent ces opérations ? Quels en sont les enjeux pour l’avenir ? Après une présentation des montages financiers que sont la reprise par endettement, la titrisation et leur combinaison sous forme de SBO (I), nous analyserons les déterminants de ce montage (II), puis 4 cas français (III). Enfin, nous situerons l’avenir de ces montages complexes (IV). I. La reprise par SBO La reprise par effet de levier et la titrisation sont deux techniques financières d’inspiration anglo-saxonne. Elles font l’objet d’une application dans de nombreux pays, notamment en France qui 130 est le deuxième marché en Europe du Private Equity. Le LBO y atteint 80% des investissements en Private Equity soit 8 milliards d’euros en 2006, en hausse de 40% par an depuis 10 ans (Banque de France, 2007). Mais en se diffusant, le montage évolue. Le LBO associe ainsi la titrisation économique des actifs de la cible pour s’ouvrir à une nouvelle structure ingénierique. 1. Le financement de la reprise : le LBO Le LBO est une technique de financement de reprise d’entreprise reposant sur la création d’un holding fortement endetté destiné à acquérir le capital d’une société-cible. Le service de la dette est assuré par la remontée des flux de trésorerie dégagés par la cible vers le holding. Le LBO est conçu pour une durée déterminée qui correspond au délai nécessaire à l’apurement de la dette du holding. Le montage combine les leviers juridiques, financiers, sociaux et fiscaux. Le premier repose sur la création d’un holding de reprise destiné à abriter le montage financier et à optimiser le contrôle des repreneurs, notamment par l’émission de produits hybrides. Le deuxième levier permet de réduire l’apport de capitaux par les repreneurs en faisant appel à l’endettement. Il permet aussi d’améliorer la rentabilité financière par l’effet de levier de la dette. Le troisième levier mobilise les repreneurs par l’intéressement à la réussite du montage. Ceci peut prendre notamment la forme d’une quote-part sur le taux de rendement interne dégagé par le fonds d’investissement associé à la reprise. Enfin, le dernier levier utilise la fiscalité de groupe en jouant sur l’intégration fiscale qui réduit l’assiette d’imposition et le régime mère-fille qui exonère les revenus issus du versement d’un dividende. Un état des lieux des pratiques de financement au niveau du holding peut être avancé. On constate ainsi que les montages financiers tendent à être de plus en plus complexes, reposant sur l’empilement de plusieurs types de ressources financières (Banque de France, 2006). Projet de reprise Holding Immo. Structure financières financière Cash Montage financier Financements structurés Cash Fonds propres Partenaires financiers Titres Repreneurs Fonds d’investissement Dette PIK Notes Partenaires financiers Dette junior Fonds mezzanine Reprise Dette second lien Entreprise Hedge funds Cash Dette senior Pool bancaire Portefeuille Structure d’actifs financière Figure 1 : Les sources de financement du LBO Le schéma montre que la structure originelle du LBO que constitue le projet de reprise s’est enrichie d’une structure financière du holding plus complexe, avec un risque décroissant de haut en bas jusqu’à la dette senior elle-même en tranches de risque différent. Le montage est marqué par la pluralité des partenaires intéressés à un ou plusieurs niveaux dans le montage financier. La tension du montage repose sur le rapport entre dettes et capitaux propres. Toutefois, la tendance est à affiner l’endettement. En effet, l’ingénierie financière développe des financements de plus en plus fins, sur- 131 mesure adaptés aux besoins les plus précis. Ainsi, si la dette senior bénéficie des plus fortes garanties avec un remboursement prioritaire par rapport aux autres types de dettes, la structure financière est de plus en plus fréquemment décomposée en tranches classées par ordre croissant de risque. La tranche A est amortissable, c’est-à-dire remboursée par échéances successives. Les tranches B, C ou D lui sont subordonnées : non seulement leur remboursement intervient après celui de la tranche A, mais en plus elles sont remboursées in fine et les intérêts sont capitalisés. L’objectif de cette structuration est alors de limiter le service de la dette dans les premières années du montage. Cette dette in fine est principalement souscrite par un pool bancaire. Elle peut être titrisée (CDO) pour permettre aux apporteurs de ressources de se refinancer sur le marché. Par ailleurs, la dette junior est subordonnée aux différentes tranches de la dette senior. Cette « mezzanine » est constituée de titres financiers combinant titres de créances assortis de possibilités de conversion en titres de capital : ce sont des obligations convertibles (OC), remboursables en actions (ORA) ou assorties de bons de souscription d’actions (OBSA) principalement souscrites par des fonds spécialisés dits « mezzanines ». Le risque attaché à la dette junior est d’autant plus important que la rémunération est décalée dans le temps. La dette subordonnée peut également être constituée d’obligations high yield dont la notation est particulièrement faible (Baa pour Moody’s ou BBB pour Standard & Poors), rangeant cette classe de titres de créance dans la catégorie spéculative. Le risque, notamment à long terme, est particulièrement élevé. Enfin, entre dettes senior, dettes junior et fonds propres, s’intercalent des financements innovants. La dette « PIK notes » pour « payment in kind » (paiement en nature) est un financement qui implique un remboursement et un versement d’intérêts à échéance sous forme de prime. Cette dette assure la liaison entre fonds propres et dette junior. La dette « second lien » est un financement qui assure la liaison entre dette senior et dette junior. Elle est immédiatement subordonnée à la dette senior mais est prioritaire sur la dette subordonnée. Cette dette est notamment souscrite par les hedge funds. 2. La titrisation économique La titrisation est un montage ingénierique. Elle est « économique » car elle porte sur tout ou partie des actifs économiques de la cible dans un montage LBO. La titrisation consiste à organiser le transfert d’un portefeuille d’actifs à une structure ad hoc : un fonds commun de créances (FCC ou SPV pour Special Purpose Vehicle). Celui-ci finance alors l’acquisition en émettant des titres de créance négociables (TCN) souscrits par des investisseurs. Les fonds collectés sont reversés à l’entreprise cédante. Le remboursement et la rémunération des titres de créance sont assurés par les revenus générés par les actifs transférés qui peuvent être de toute nature (immobilisations corporelles ou financières, stocks, créances commerciales…). Ce mécanisme permet ainsi de transformer des actifs peu liquides en titres financiers librement négociables. Mais la titrisation constitue surtout une source de financement à moindre coût. En effet, ce financement n’est pas inscrit dans le passif de la société cédante, ce qui ne dégrade pas sa solidité financière. Enfin, la titrisation permet d’émettre des titres de créance dont la qualité est garantie par les actifs transférés : le taux d’intérêt servi est ainsi négocié avec une prime de risque réduite, l’objectif étant souvent d’atteindre la meilleure notation possible (AAA), au même niveau que la dette souveraine. 132 Arrangeur Emetteur de SPV Propriétaire des actifs Investisseurs – TCN AAA Entreprise SPV Garanties sur actifs Titres TCN AA TCN … Portefeuille Structure d’actifs financière Portefeuille Structure d’actifs financière Cash Risque Cash TCN BBB Equity + Transfert d’actifs Emission de titres financiers Cash Figure 2 : Mécanismes de la titrisation Comme le montre le schéma précédent, la titrisation est un outil supplémentaire de répartition des risques au sein du montage : les actifs constituent le sous-jacent d’une émission de titres dont le profil de risque est varié du moins risqué (TCN AAA) au plus risqué (equity). Le flux de trésorerie (cash) généré par l’opération s’apparente à un autofinancement. Son utilisation est libre par l’entreprise qui peut, sans contrainte, le consacrer à toute opération, de nature économique (investissement) ou de nature financière (remboursement anticipé de dettes, réduction de capital…). La titrisation génère ainsi une marge de manœuvre particulièrement utile dans le cas où le cédant ne peut accéder directement aux marchés financiers (qualité de signature dégradée), supporte une contrainte de financement (dépendance financière et rationnement des capitaux propres) ou cherche à élargir ses sources de refinancement en dettes financières (désintermédiation). 3. Le SBO Reprise par effet de levier et titrisation économique sont combinées pour former une opération financière complexe dite SBO. Initié dans les pays anglo-saxons, ce montage est également utilisé en France. De nouveaux acteurs intègrent le projet de reprise. La description synthétique de l’architecture SBO et l’identification des opérations mises en œuvre par l’opération permettent de situer les rôles des partenaires associés aux repreneurs dans le SBO (fonds d’investissement, apporteurs de dettes, SPV, investisseurs institutionnels…) et d’en analyser les effets sur les entités concernées. Projet de reprise Investisseurs Holding – Immo. Structure financières financière Cash Reprise Risque Entreprise cible SPV Portefeuille Structure d’actifs financière Portefeuille Structure d’actifs financière Transfert d’actifs Cash 133 TCN Cash + Figure 3 : Montage du SBO : LBO et titrisation des actifs de la cible Dans un SBO, la titrisation de reprise est économique : elle consiste en un transfert d’actifs de la cible (portefeuille de créances commerciales, stocks…) à un SPV qui finance l’acquisition grâce à une émission de TCN adossés aux actifs titrisés. La ressource financière dégagée est alors reversée à la cible. Dans ces processus de reprise où les montages sont particulièrement tendus, les partenaires financiers portent une attention aux actifs de la cible afin d’optimiser la remontée de trésorerie vers le holding. La titrisation se révèle particulièrement bien adaptée aux contraintes du LBO. En effet, au niveau du holding, l’endettement est important et les fonds propres sont limités pour assurer le contrôle du groupe aux repreneurs. Pour la société cible, les revenus sont largement mobilisés par le holding pour assurer la bonne fin de la reprise. La titrisation offre ainsi non seulement la possibilité d’accéder à une nouvelle source de financement mais aussi à un taux réduit : le remboursement et le versement d’intérêt des TCN émis par le FCC sont garantis par les actifs cédés. II. Les déterminants du SBO La recherche académique pose plusieurs jalons pour comprendre les mécanismes du LBO et de la titrisation. 1. Contexte théorique Au niveau du holding, le LBO pose la question du niveau d’endettement supportable. Sur ce point, différents apports s’opposent. Modigliani et Miller (1958) montrent que la valeur de la firme est indépendante de l’endettement. Cette neutralité implique qu’il n’existe pas de structure financière optimale : en théorie, il est possible d’endetter à l’infini une entreprise, sans impact sur sa valorisation. Toutefois, le niveau d’endettement est souvent une source d’inquiétude. Ainsi, Myers (1977) pose la théorie du compromis fondée sur un arbitrage entre actions et dettes. Myers et Majluf (1984) en déduisent une théorie du financement hiérarchique (Pecking Order) : plus coûteuses, les actions constituent la source de financement la moins utilisée. Ainsi s’explique la propension de l’ingénierie financière à développer des titres hybrides, entre capitaux propres (coût) et dettes (risque). De plus, Jensen et Meckling (1976), Kaplan (1989) et Miller (1990) notent que la création de valeur par LBO ne vient pas du montage mais de la tension qu’il exerce sur la cible en obligeant une optimisation du processus productif de la firme reprise. Jensen (1986) évoque ainsi l’effet disciplinaire de l’endettement sur le dirigeant de la firme d’autant plus incité à payer les créanciers qu’ils peuvent, en cas de défaut, prendre le contrôle de la cible et priver ainsi le dirigeant-repreneur de sa principale source de rémunération à échéance. Au niveau de la cible, la question qui se pose est alors celle de l’utilisation du flux financier généré par la titrisation. La titrisation est un mécanisme d’optimisation de la remontée de flux de trésorerie. Modigliani et Miller (1961) affirment la neutralité des dividendes sur la valeur de la firme. En effet, la distribution d’un dividende est une décision de répartition et non de création de valeur. Par contre, l’augmentation du dividende, si elle est le reflet d’une augmentation de la capacité bénéficiaire de l’entreprise, a une influence positive. On a ici une motivation de la titrisation qui permet de gonfler le dividende versé à la holding. Le SBO constitue ainsi un mécanisme d’alignement des intérêts des acteurs du holding, qu’ils soient actionnaires ou créanciers. En effet, le dividende permet de rembourser les créanciers plus rapidement, de sécuriser leurs prêts, et d’accélérer leur sortie au profit du repreneur. Il est également possible de considérer la titrisation comme un exemple de free cash 134 flow (Jensen, 1986) qui versé sous forme de dividende permet aux fonds d’investissement de sortir du capital de la holding. La titrisation d’actifs peut revêtir différents objectifs. Le premier est marqué par le souci de faciliter et accélérer la reprise tandis que le deuxième s’attache à assurer le développement économique de la cible : Dans le premier cas, le holding opte pour une cession des actifs de la cible destinée à accélérer la reprise en permettant la sortie rapide des capital-risqueurs. C’est une opération de recapitalisation qui vise à changer d’actionnaire de référence. La titrisation permet alors de disposer d’un flux de trésorerie mobilisable au profit des créanciers les plus pressés. Recapitalisation et restructuration entraînent une nouvelle répartition des risques entre les partenaires au sein du holding ; Dans un second cas, le holding peut utiliser le flux généré par la titrisation dans une logique de croissance interne en dédiant le produit de la titrisation au renforcement des capacités de développement économique de la cible. Le SBO est alors une opération de redéploiement économique dans l’objectif d’accélérer le développement de la cible afin de créer de la valeur économique au profit des partenaires financiers, dont les fonds d’investissement ; Dans un troisième cas, la titrisation s’enrichit d’un mécanisme de debt push down (descente de dette). La titrisation est alors au service de la restructuration en permettant le remboursement d’une partie des dettes du holding et en transférant cette nouvelle capacité d’endettement vers la cible. Il y a donc transfert vers la cible d’une partie du service de la dette contractée par le holding. Le SBO incite à poser un nouveau regard sur la politique de financement. La théorie s’est déplacée vers une compréhension des logiques de choix des moyens de financement prenant en considération les contextes et les relations entre agents associés au projet (Ross, 1977 ; Myers et Majluf, 1984). Y sont rattachées des perspectives organisationnelles et cognitives puisque le financement peut être abordé sous l’angle d’un arrangement financier où les questions de création et de répartition de valeur et celles de compromis dynamiques entre parties prenantes sont essentielles. L’ouverture du spectre des partenaires intéressés au partage de la valeur élargit la conception bilantielle. La structure de financement intègre d’autres parties prenantes au passif et introduit le concept de capital organisationnel à l’actif (Cornell et Shapiro, 1987). Le SBO est au cœur de cette problématique. Il se présente comme l’association d’une opération de titrisation à un montage LBO qui a pour objet de financer de manière optimale la reprise d’entreprise. Le noyau stratégique, association d’un fonds d’investissement et d’un dirigeant intéressé dans le cas du SBO, devient l’organe central de décision. Il a pour mission de rechercher la structure financière souhaitable et acceptable en adéquation avec la situation spécifique de l’organisation. L’opération s’appréhende comme un compromis dynamique à trouver associant un ensemble complexe de partenaires impliqués à différents niveaux et pour différentes durées dans le financement d’un projet organisant un rachat où chaque opération est appréhendée comme un cas d’espèce : « la meilleure politique de financement sera celle qui conduira les dirigeants à bien orienter la firme, à découvrir et entreprendre les opportunités créatrices de valeur, à les protéger, à bien coordonner les connaissances dans la production, à résoudre au mieux les conflits » (Charreaux, 2002). 2. Le contexte pratique L’analyse du fonctionnement financier du montage permet de mettre en évidence les missions du noyau stratégique, repreneur et fonds d’investissement, orientées sur la distribution des risques entre les apporteurs de ressources, la création de valeur permise par l’optimisation de l’activité opérationnelle de la cible et la répartition de la valeur créée entre les partenaires associés au montage. L’affectation des flux générés par la titrisation des actifs de la cible est au cœur de ces problématiques. La titrisation est-elle au service de la création de valeur économique de la cible ou est-elle un acte de répartition et de gestion des risques, au service d’un ou plusieurs acteurs du holding ? Plusieurs scénarios peuvent ainsi se dégager. 135 Figure 3 : Les facteurs explicatifs du SBO Holding de reprise (H) ACTIF IMMOBILISE Investis. financie r TITRES DE C Opération de contrôle financier Actions STRUCTURE FINANCIERE Dettes Capitaux propres Apports actionnaires Dettes financières Apports créanciers O é ti d Flux revenant au projet Cible (C) ACTIF ECONOMIQUE Projet opérationnel Actif immobilisé Opérations d'investisseme nt BFR d'exploitation Flux revenant aux actionnaires Flux revenant aux créanciers Répartition de valeur Investis. industrie l Accès aux ressources Flux de répartition PASSIF FINANCIER Noyau stratégique Arrangement contractuel RESULTAT Flux Exploit. ECONOMIQUE Opérations de répartition Marché Investisseurs Intermédiaire s financiers Actions Titres hybrides Créances financières Obligations convertibles Dette supersubordonée Distribution des risques Création de valeur Flux de cession FC C Flux de titrisation Parts Titrisation Cession Le SBO de redéploiement. L’optimisation de l’activité opérationnelle de la cible, qui développe des flux de trésorerie issus des cycles d’exploitation et d’investissement, est mise au service de la création de valeur économique au niveau de la cible. La titrisation constitue une option de croissance détenue par le noyau, dont la stratégie est la valorisation des actifs de la cible. Dans ce cas d’un SBO de redéploiement, le rachat par LBO et la titrisation sont des opérations conjointes où le coût de la titrisation doit être inférieur à la rentabilité économique des actifs pour qu’il y ait création de valeur économique. La titrisation permet ainsi de mettre en valeur des actifs inutilisés ou mal refinancés par les moyens de financement traditionnel. Elle assure la liquidité des actifs et offre des possibilités de financement d’investissements. Le flux de trésorerie, au service de la croissance économique, apporte une flexibilité de financement jusqu’à présent inexistante dans le cadre d’un LBO classique, où les flux dégagés par la cible sont affectés au service de la dette et n’ont pas vocation à être réinvestis dans la constitution de nouveaux actifs. Dans le LBO traditionnel, l’effet disciplinaire de la dette exercé sur les dirigeants incite à favoriser l’efficacité opérationnelle au détriment de l’investissement. La titrisation constitue ainsi une option de croissance. En revanche, bien que l’horizon du repreneur soit limité dans ces montages à effet de levier, est-il réellement incité à privilégier des investissements risqués en vue de maximiser la valeur de la société alors qu’il est associé au capital de la société holding ? Si oui, cette stratégie ne devrait pas rassurer les prêteurs dans la mesure où la titrisation des actifs de la cible réduit d’autant leur sécurité en cas de défaillance. Les conflits d’intérêt entre le noyau stratégique et les créanciers financiers sont d’autant plus renforcés que parfois la titrisation porte sur la totalité de l’actif de la cible (opération de type whole business). En effet, en cas d’échec du montage, le FCC est sécurisé grâce à l’instauration par la loi de sécurité financière d’un compte à affectation spéciale au profit du fonds, sur lequel le produit du recouvrement des actifs titrisés est enregistré et donc isolé. Les actifs titrisés ne peuvent être redistribuées au profit des prêteurs du montage : les programmes de titrisation se trouvent donc être seniors par rapport à la 136 dette LBO. Ce frein explique en partie la position des prêteurs seniors, qui préfèrent les titrisations partielles portant sur des montants raisonnables et de maturité ajustée à celle de la dette LBO. Le SBO de restructuration. La titrisation crée un flux unique mais nouveau de trésorerie qui alimente la remontée de dividendes vers le holding. Celui-ci peut les reverser aux créanciers, accélérant le remboursement de la dette ou peut offrir une possibilité de sortie pour les fonds d’investissement actionnaires. Ainsi, sous l’hypothèse d’une affectation du cash au bénéfice des partenaires de la société holding, il convient de distinguer les deux possibilités suivantes : soit celui-ci est mis au service de la dette existante, soit il est redistribué aux apporteurs de capital. Dans le cas d’un SBO de restructuration, la titrisation est mise au service de la structure financière de la société holding. L’effet disciplinaire de l’endettement contraint le noyau stratégique à affecter le cash au service de la dette de la holding. Avec la baisse des marges sur la dette constatée par les professionnels (Les Echos, 22 mai 2005), le refinancement de la dette senior du LBO par le biais de la titrisation, option détenue par les prêteurs seniors, implique un transfert de risque entre les partenaires du montage et une renégociation du financement LBO. Selon Wruck (1990), l’endettement précipite les faillites et donc les restructurations créatrices de valeur via la recherche d’une mutualisation des risques du montage (partage des risques avec le marché ou strip financing). La titrisation s’insère dans les techniques de gestion de bilan et offre des possibilités d’actions sur le couple rentabilité-risque. Pour le cédant, les actifs titrisables sont transformés en cash, ce qui engendre, au niveau du bilan, un allègement de l’actif et une amélioration de la liquidité. Au niveau de la structure financière, si le cash est mis au service de la dette, l’opération améliore fictivement l’image financière et réduit de fait le risque financier du montage, facilitant l’accès à d’autres financements. Au niveau du compte de résultat, le coût est inférieur au financement classique à moyen terme. Enfin, il y a aussi la possibilité de bénéficier d’exonérations fiscales. Les conséquences les plus immédiates sont donc l’amélioration des ratios (solvabilité, liquidité, indépendance financière), l’accroissement de la rentabilité économique et également financière, la diminution des risques supportés par les capitaux propres en cas de désendettement et l’augmentation des possibilités de refinancement. En revanche, si la titrisation, programme garanti sur les actifs de la cible et différé du rachat par LBO, constitue une opportunité pour restructurer le passif, alors la rémunération requise par les prêteurs devra refléter un niveau de risque élevé, avec l’ouverture de tranches plus risquées telles que les dettes mezzanines et l’introduction des nouveaux instruments de dette obligataire (PIK notes). Ce financement par émission de titres à risque vise donc à reconsidérer le risque des opérations de rachat d’entreprise par LBO, dont le score s’est très nettement dégradé depuis quelques années. Le nombre de LBO ayant obtenu une note d’investissement spéculatif B a sensiblement progressé entre 2003 et 2005 (de 16% à 75% respectivement). En définitive, l’opération implique un transfert de risque sur les nouveaux prêteurs ainsi que sur les apporteurs de capital que sont le repreneur et le fonds d’investissement. SBO de recapitalisation. La titrisation est mise au service du noyau. Elle constitue une option de retrait détenue par les actionnaires, dont l’exercice favorise un transfert de risque vers les nouveaux actionnaires et les prêteurs. Le programme de titrisation profite plus au fonds d’investissement, dont la part apportée lors du rachat par LBO serait en baisse depuis deux ans, et accélèrerait ainsi sa sortie du montage. La revente d’entreprise entre fonds d’investissement est ainsi devenue une pratique courante. Les rotations d’actifs montrent que les fonds d’investissement sortent facilement des montages dès lors que le TRI objectif est atteint. Ces LBO en série se négocient à des prix croissants entraînant des effets de levier plus conséquents. Ils s’accompagnent d’une réflexion stratégique visant à trouver de nouvelles sources de croissance et de rentabilité. Il semblerait en second lieu que l’intéressement des managers voire des salariés soit devenu l’élément clé des opérations de reprise. Le TRI revêt ainsi une importance pour les managers qui bénéficient de dispositifs de rétrocession, dont le principe repose sur la renonciation d’une partie de la plus-value des investisseurs à l’occasion de la sortie. Le succès des opérations de reprise réside en partie dans la performance et la motivation des repreneurs, si bien que le caractère essentiel du levier social s’accentue (Baffreau et Timsit, 2001). Les dirigeants bénéficient fréquemment d’une émission d’OC ou d’ORA dont la conversion ou le remboursement confère l’accès au capital du holding. En cas de 137 renonciation, le titre est assimilé à de l’endettement. Plus le TRI des investisseurs est élevé, plus le taux de conversion ou de remboursement est bas, et plus le prix de sortie par action dont bénéficient les managers est important. Ils bénéficient également de mécanismes d’intéressement qui doivent leur permettre de s’approprier une partie de la plus-value lors du débouclage de l’opération, puisque l’accroissement de la valeur est en partie de leur fait. L’accès au capital consenti par l’investisseur professionnel au repreneur prend alors la forme de promesse de cessions d’actions, d’attribution de BSA, d’actions de préférence, d’actions gratuites, d’options de souscription ou d’achat d’actions. De tels mécanismes instaurent un alignement des intérêts entre le groupe des actionnaires, impliquent fortement le management et focalisent les efforts sur la réussite du projet conformément aux objectifs des acteurs du holding. La titrisation en participant à l’accélération du débouclage d’une opération permettrait un déclenchement plus rapide de répartition de la valeur entre tous les actionnaires. Pour mieux distinguer les motivations du SBO, il est nécessaire de se pencher sur les pratiques des montages recensés en France. Ainsi, la question de l’utilisation des flux générés par la titrisation se pose. Le SBO, ayant un impact sur les risques supportés par les acteurs du montage, l’opération est-elle créatrice de valeur et si c’est le cas, qui en bénéficie ? Quelle est la nouvelle répartition des risques ? L’analyse de plusieurs cas en France permet d’avancer plusieurs éléments de réponse pour caractériser les SBO. Tableau 1 : Trois cas de SBO SBO de redéploiement SBO de restructuration SBO de recapitalisation Le cash sert à valoriser les actifs de la cible (croissance économique) Hypothèse : coût titrisation < rentabilité économique des actifs LBO et titrisations simultanées Le cash est au service de la dette LBO Hypothèse : titrisation (option détenue par les prêteurs) différée / LBO Objectifs : création de valeur économique Objectifs : renégociation du financement LBO, transfert de risque entre acteurs du LBO, nouvelle capacité d’endettement La titrisation constitue un outil de gestion des risques Le cash est au service des actionnaires Hypothèse : titrisation (option détenue par les actionnaires par les actionnaires historiques) différée / LBO Objectifs : sortie d’actionnaires, transferts de risques entre actionnaires, recomposition du capital La titrisation permet la gestion du pacte actionnarial La titrisation offre une flexibilité de financement de la croissance III. Etude des SBO français Si le LBO et la titrisation sont des opérations fréquemment mises en œuvre, les combinaisons sont plus rares. Nous constatons en France un balbutiement du SBO. De fait nous n’avons recensé que quatre opérations de SBO en France, même si plusieurs montages sont en cours d’analyse chez les arrangeurs. A partir de ces études de cas, nous essaierons, à travers une discussion, de situer l’intérêt de l’opération dans un contexte de dynamisation de la reprise d’entreprises. 1. Etude des cas Pour identifier les opérations de SBO, nous sommes partis de la revue hebdomadaire spécialisée Option Finance. Un dépouillement systématique a été réalisé sur une période allant du début 1999 au milieu de l’année 2006. L’analyse des numéros 529 (4 janvier 1999) à 895 (28 aout 2006) nous ont permis de détecter les 4 opérations françaises de SBO. Le SBO de BSN Glasspack (2000) est la première opération de ce type. Plus récemment, Cegelec, Fraikin et Franz Bonhomme 138 ont mis en place une telle procédure de refinancement de la dette LBO ce qui montre un regain d’intérêt de la technique. Entreprise concernée BSN Glasspack Cegelec Fraikin Franz Bonhomme Titre de l’article Opération suivi BSN Glasspack refinance sa dette SBO-LBU en titrisant ses créances d’une titrisation commerciales Cegelec conjugue titrisation et LBO suivi d’une titrisation dette senior Fraikin se finance avec une SBO Whole Business titrisation innovante dite whole business sur son parc et ses SBO simultané contrats de location Comment Franz Bonhomme a financé son 4ème LBO ? Tableau 1 : documentation de l’étude de cas Références Option Finance n° 622 , 18-12-2000, p.22-23 Option Finance n° 790 , 21-6-2004 Option Finance n° 817 , 17-1-2005 Option Finance n° 867 , 23-1-2006 Une collecte d’informations sur les opérations a été réalisée. Elle concerne les rapports d’activité des entreprises, les rapports d’agence de notation, des études de faisabilité et des analyses spécifiques. Les synthèses sont le résultat de la compilation de l’ensemble de ces informations. La démarche est donc exploratoire. Les opérations concernées sont étudiées selon un ordre chronologique et concernent respectivement BSN Glasspack, Cegelec, Fraikin et Franz Bonhomme. 2. Le SBO BSN Glasspack En 1999, BSN Glasspack (BSNG) est l’objet d’un rachat par LBO. L’activité emballage de Danone est ainsi acquise par le fonds d’investissement CVC Capital Partners. L’opération, d’un montant de 823M€, est financée pour 150M€ par un crédit d’exploitation au coût élevé (Euribor + 1,75%). La recherche d’un financement moins onéreux que ce crédit entraîne une opération de titrisation en 2000 ajustée à la durée du LBO. La titrisation repose sur la cession des créances commerciales de BSNG à un compartiment du FCC Securipack. Le FCC, mieux noté que la société cédante, a pu se refinancer à un taux plus intéressant sur le marché. Le montant et la nature des créances expliquent l’intérêt de l’opération car elles permettaient largement de couvrir les commissions importantes à acquitter pour ce type d’opération. En effet, l’activité de BSNG génère un encours important de créances, supérieur à 300M€ payées en moyenne à 90 jours. Le caractère récurrent de ces créances provenant de clients diversifiés de risque faible était susceptible d’intéresser les investisseurs. Le regroupement des créances clients pour toute l’Europe a amené à créer une centrale de facturation : BSNG Services. Sa fonction est d’acheter les emballages en verre des centres de production de BSNG et de les revendre aux clients. Les créances ainsi acquises sont cédées au FCC qui les acquiert par un règlement immédiat en liquidité. Pour couvrir le risque de défaut des clients, le portefeuille de créances est surdimensionné, ce qui constitue un coût pour la centrale de facturation. Le FCC, ne pouvant se refinancer directement sur le marché pour des raisons juridiques, émet des parts ordinaires notées AAA. Celles-ci sont souscrites par Phébus Finance qui émet alors des billets de trésorerie. La qualité de la notation permet à Phébus Finance de se refinancer à moindre coût (Euribor + 0,65%). Le risque d’illiquidité du marché des billets de trésorerie est garanti par une ligne de liquidités venant à échéance lors de la sortie prévue du LBO. Sur le risque de défaut des clients, le FCC n’a pas de recours sur le cash versé à BSNG Services. Le cash est mis au service du refinancement de la dette d’exploitation du LBO facturée Euribor + 1,75%. 139 L’intérêt du montage est triple. D’une part, les frais financiers sont réduits et la capacité de financement de BSNG se renforce. L’image financière de l’entreprise est ainsi améliorée au niveau des ratios financiers. C’est donc un SBO de restructuration. D’autre part, l’opération est directement créatrice de valeur. En effet, dans le montage du LBO, une clause de dégressivité précise qu’en cas d’amélioration des garanties de BSNG, il y a réduction des marges des apporteurs de ressources et donc des coûts pour l’entreprise sur les autres financements contractés lors du LBO. Enfin, au niveau organisationnel, la traçabilité des créances et la simplification de la gestion de la clientèle procurent un avantage, même si des coûts d’organisation spécifiques (création de BSNGS) ont été engagés. F ilia le s d e p r o d u c t i o n de BSN G C e s s io n d e m a r c h a n d is e s C lie n t s d e B S N G C ré a n c es V e n te s B S N G S e r v ic e s ( B S N G S ) P a ie m e n t à 9 0 jo u r s C e s s io n d e c r é a n c e s C a sh F C C S e c u r if a c t - C o m p a r t im e n t B S N G S G a r a n tie p r e m iè r e d em a n de P a r ts o r d in a ir e s N o te A A A L ig n e d e liq u id ité a d o s s é e a u x p a r ts o r d in a ir e s B S B G S P h e b u s F in a n c e A G F IA R T H SB C C C F E m is s io n d e b ille ts d e tr é s o r e r ie A - 1 / F 1 I n v e s tis s e u r s Figure 4 : Le SBO BSN Glasspack (montage simplifié) 3. Le SBO CEGELEC En 2001, Cegelec est racheté par LBO mené par un pool constitué de ses dirigeants associés à deux fonds d’investissement pour un total de 800 M€. Cette filiale d’Alsthom est spécialisée dans les services technologiques. Le financement du LBO s’est fait pour 550 M€ par dette bancaire et pour 250M€ de fonds propres et de quasi fonds propres sous forme d’obligations convertibles. La titrisation réalisée en 2003 est accompagnée d’un refinancement global du LBO. Elle a pour objet de réduire les frais financiers importants (45M€) alors que le résultat net atteignait à peine 1,8M€. Elle a nécessité le refinancement de la dette LBO du holding de reprise (dettes senior et mezzanine) et le remboursement des OC permettant la sortie des deux fonds d’investissement pour un total évalué à 668M€ (fin 2003). L’opération a également permis d’optimiser la création de valeur. Le coût du refinancement par titrisation se révèle moins onéreux que celui de la dette LBO auquel il s’est substitué. Le besoin de refinancement pour un total de 580M€ a été assuré pour 260 M€ par une titrisation de créances commerciales mise en place sous la forme d’une ligne de crédit renouvelable et remboursable in fine à 5 ans et pour 320M€ par dette senior remboursable sur 6 ans. La titrisation a concerné les créances clients ce qui a permis de déplacer une partie de l’endettement du holding vers la cible selon la technique de descente de dette par titrisation. La titrisation se met ainsi au service du remboursement de la dette de la tête de groupe. Le programme sur cinq ans a été ajusté à la durée du LBO. Sur le montage, Ester Finance, établissement de crédit, achète les créances des filiales de Cegelec grâce à un dépôt de Calyon. Cet établissement détient donc une créance sur Ester Finance. Il cède les créances notées AA au FCC qui émet des parts pour se financer. Les parts sont rachetées par LMA, une société commerciale, qui se finance à son tour sur le marché des billets de trésorerie. Pour faire remonter le 140 cash issu des filiales vers Cegelec SA, celle-ci a dû désinvestir les filiales étrangères qu’elle détenait en direct à ses filiales françaises et belges. Les plus-values constatées ont été compensées par des moins-values sur dépréciation de titres de participation. La cession interne s’est faite sans impôt. La remontée de cash de Cegelec SA vers Cegelec Holding a donné lieu à une distribution d’un dividende exceptionnel, toujours hors impôt dans le cadre du régime mère-fille. Ce SBO tient donc à la fois du SBO de restructuration et du SBO de recapitalisation (émission d’OC). Cegelec Holding Filiales Cegelec Paiement ( dépôt) Ester Finance (Ets de crédit) Créances achat Dépôt Vente Détention d’une Créance sur Ester Caylon (arrangeur) Cession de créances FCC Rehaussement De 25% Emission de parts LMA (Sté Ciale) Emission de billets de trésorerie Marché Figure 5 : Le SBO Cegelec L’intérêt immédiat du montage a été une réduction de la dette senior de 668 à 580M€ et une baisse des frais financiers de 45 à 20M€. Les nouvelles garanties sur la dette senior renégociée ont été assouplies. Par ailleurs, l’opération a permis d’accéder à des possibilités de croissance externe auparavant très difficilement réalisables. Le risque a ainsi évolué d’un risque de crédit dû à un montage financier particulièrement tendu vers un risque économique. La sortie des investisseurs a été anticipée afin d’éviter que la modification de l’actionnariat de référence ne s’accompagne d’un refinancement total de la nouvelle dette contractée. 4. Le SBO FRAIKIN En février 2003, grâce à un LBO, Eurazeo achète le groupe Fraikin à Fiat pour un montant de 764 M€. L’activité de location de véhicules étant consommatrice de capitaux, il a été envisagé de trouver un substitut au prêt-relais de 420M€ mis en place, un endettement amortissable n’étant pas envisageable. En effet, pour un résultat moyen de 250M€, la firme devait réinvestir de 80 à 100% dans de nouveaux véhicules. Le financement du renouvellement du parc de camions posait donc un problème dans le LBO. Ces dépenses d’investissement réduisaient d’autant le flux de trésorerie de la société opérationnelle, limitant la remontée de dividendes avec des effets induits sur la gestion de l’endettement du holding. La titrisation s’est effectuée en fin 2004, soit un peu plus d’une année après la reprise du groupe. Il s’agit de la mise en place de facilités de crédit adossées à une titrisation. Les actifs titrisés sont les loyers générés par les contrats de location et un parc de camions. Il s’agit d’une titrisation whole business réalisée sur 90% des actifs de Fraikin. Les revenus des contrats de location étant récurrents et la valeur du parc de camions étant prévisibles, ces deux classes d’actifs ont pu être concernées par la titrisation. La titrisation a permis de lever plus de financement que la dette classique et de s’adapter au plan d’affaires en prévoyant une croissance du chiffre d’affaires. En effet, le mécanisme prévoit que la trésorerie levée par Fraikin s’accroît avec l’augmentation de l’actif 141 économique. Le montage vise le refinancement immédiat du crédit relais de 420M€ par une titrisation. Une possibilité de rechargement est programmée pour un montant compris entre 350 et 600M€ pendant une période de 5 ans. L’achat de nouveaux camions fait l’objet d’une titrisation amenant à constater un rechargement au fur et à mesure des investissements. Fraikin continue à assurer la gestion des contrats de location et la maintenance du parc de camions comme souscontractant de Fraikin Assets. Le FCC refinance Fraikin avec une émission d’obligations AAA à 5 ans remboursables in fine et amortissables en 10 ans pour 315M€ et une émission d’obligations subordonnées BBB pour 35M€. S’y ajoute un programme de billets de trésorerie AAA de 70M€ au départ mais pouvant porter jusqu’à 250M€. Golden share Trust Fraikin SAS 99,99% 100% Management, administration Fraikin Services Fraikin Opcos Fraikin Assets Contribution de camions, de loyer, d’actifs Accord de sous contrat Loyers Garantie financière AAA Organisation substitutive capable de gérer les contrats KPMG corporate recovery Europe Assistance Facilité de crédit 600M€ MBIA Assignation de la facilité FCC EuroTruck lease Calyon - CIC Emission d’obligations et de billets de trésorerie Investisseurs Figure 6 : Le SBO Fraikin L’intérêt de l’opération est localisé dans l’obtention d’un prix de financement intéressant et de marges nettement inférieures au coût classique du LBO. C’est donc un SBO de restructuration. La titrisation prévoit un refinancement rechargeable sur la durée du LBO et adaptable avec la croissance économique du groupe. Il y a donc mise en place d’un financement pérenne de la flotte de transport au travers d’une titrisation globale d’entreprise par un mécanisme de descente de dette. C’est donc également un SBO de redéploiement. 5. Le SBO FRANS BONHOMME (FB) Frans Bonhomme a fait l’objet de LBO en série. Le premier a été initié par les fonds Partenaires et Apax. Un autre a été réalisé en 2000 par PAI, Cinven et Astorg pour un montant de 380 M€, un troisième en 2003 par Apax accompagné de plusieurs banques et fonds d’investissement pour 520 M€. La dernière opération s’est produite en 2005 sous l’égide de Cinven pour un prix de transaction de 900 M€. Celle-ci s’est complétée d’une titrisation. Nous sommes donc ici dans un contexte de serial LBO dont les montants n’ont cessé de croître. L’activité de FB, spécialisée dans la fabrication et la distribution à destination de professionnels de tuyaux plastiques, est très stable car dépendant à la fois du secteur de la construction et du BTP mais aussi de la rénovation. La transformation de l’EBIT en cash flow est importante, l’entreprise ayant un besoin en fonds de 142 roulement et des investissements modérés. La titrisation a porté sur des créances clients pour un montant de 70M€ sur une durée de 7 ans. La bonne qualité du portefeuille de clients a permis d’obtenir un financement aux conditions de financement d’une dette corporate, prix bien inférieur à celui d’une dette LBO. L’encours de la dette LBO en a été d’autant diminué. La titrisation a eu pour but de remplacer à court terme un crédit d’exploitation au niveau de la société opérationnelle. La facilité devant par ailleurs remplacer un crédit relais au niveau du holding d’acquisition. Le montage vise à refinancer le holding de reprise tout en assurant un financement du cycle d’exploitation au niveau de la société opérationnelle. Le refinancement du holding est réalisé par une dette senior à quatre tranches (respectivement de 160, 100, 100 et 50M€ pour un total de 410 M€) d’une durée de 7 à 10 ans. S’y ajoute une dette de second lien de 50M€ à 10 ans et demi et une mezzanine de 130M€ à 11 ans. La titrisation porte sur 70 M€ pour une durée identique à la tranche A de la dette senior. FB Holding Arrangeur Natexis Frans Bonhomme Cible Cession Dette senior A 160 B 100 C 100 D 50 Second lien 45 Mezzanine 130 70M€ FCC Emission billets de trésorerie AAA Investisseurs Figure 7 : Le SBO Frans Bonhomme L’intérêt de l’opération est localisé dans le coût de la titrisation bien plus intéressant que celui de la dette classique. La titrisation prévoit un refinancement sur la durée de la dette senior. La facilité obtenue est utilisée à la fois au niveau de la cible pour le financement de l’exploitation (debt push down) et au niveau du holding pour le refinancement de la dette senior. La titrisation est simultanée au montage LBO et s’intègre pleinement dans la structure financière du montage. La capacité d’endettement créée permet à la cible de se développer économiquement. L’opération est donc largement un SBO de restructuration. 6. Synthèse L’analyse de ces quatre cas permet de mieux apprécier la logique économique et financière de l’opération SBO en France. La synthèse suivante met l’accent sur les caractéristiques sur de chaque opération et permet des comparaisons. Elle donne lieu à une discussion. BSN Glasspack 1999 Cegelec 2004 Fraikin 2005 Opération SBO Montage Titrisation Parts Emission marché Décalé Durée du LBO Notation AAA Billets de Trésorerie SBO « Debt push down » SBO Whole business par titrisation « Debt push down » Décalé Décalé Durée 5 ans Durée 10 ans Notation AA Notation AAA Billets de trésorerie Billets de trésorerie et 143 Franz Bonhomme 2005 SBO « Debt push down » Simultané Durée 7 ans Notation AAA Billets de trésorerie obligations Actifs titrisés Activité Objet Intérêt Autre impact Caractéristiques financières Organisationnelle Nature SBO Créances commerciales Créances Commerciales Ensemble de l’activité de Créances commerciales contrats de location et parc de camions Emballages de verre Services technologiques Location de véhicules Tuyaux, canalisations Coût de la dette Refinancement du solde Refinancement de la dette Coût de la dette d’exploitation à réduire. de la dette LBO du existante et de la d’exploitation à réduire et Substitution et holding, renégociation de croissance par titrisation, de l’encours dette LBO, refinancement par la dette senior, obtention d’un crédit substitution et titrisation remboursement des OC relais refinancé par refinancement par des financiers. titrisation. titrisation Baisse des frais Baisse des frais Baisse du coût du Baisse du coût de financiers du financiers et de financement, marges financement du LBO, financement (réduction l’endettement du intéressantes, remplacement du crédit des coûts), amélioration holding, descente de la financement de la flotte d’exploitation sur cible, de la capacité dette vers la cible, relayé sur le holding d’endettement marges avantageuses Amélioration de l’image Renégociation des Refinancement Activité stable, et des ratios financiers covenants et amélioration rechargeable sur la durée sécurisation du (impact sur les marges de la marge de la dette du LBO, sécurisation du financement du dégressives des senior, distribution d’un financement de la développement. financements LBO), super dividende au croissance, limitation de effet rentabilité-risque holding. l’apport en fonds propres nul Créances diversifiées Refinancement global CF connus et récurrents CF importants, I et BFR importantes sur des Ouverture de la modérés, marges élevées, clients de qualité. possibilité de croissance fort potentiel de externe. développement Dette non exigible si entrée en Bourse Centralisation de la Création d’une structure facturation des clients de gestion substitutive Simplification de la capable de reprendre les gestion activités en cas de faillite Restructuration Restructuration et Restructuration et Restructuration recapitalisation redéploiement Tableau 2 : Synthèse sur la pratique des SBO Le marché du LBO a depuis l’origine vocation à s’adresser aux entreprises matures dégageant des résultats satisfaisants et relativement stables dans le temps pour limiter le risque des prêteurs associés au financement de la dette de reprise au niveau du holding. Plusieurs nouvelles tendances sont apparues dans le montage et mises en évidence dans les études de cas. Une première tendance montre un élargissement des modalités de financement qui se déplacent vers la société cible. Le debt push down, les techniques de financement des besoins en fonds de roulement (affacturage, reverse factoring…) et la titrisation mettent en avant l’intérêt à la fois de descendre la problématique du financement et d’optimiser la trésorerie à partir de la cible. Une seconde tendance va dans le sens d’une complexification de la structure financière du holding par empilement de ressources financières et ajout de dettes intermédiaires et par décomposition d’une ressource en tranches. A titre d’exemple, la dette senior de Franz bonhomme comprend quatre tranches, une amortissable et trois in fine. Une troisième tendance est apparue avec les valorisations qui s’accroissent et donc les montants de refinancement qui augmentent. Les dettes in fine font alors l’objet d’une titrisation financière par les banquiers du pool afin de se refinancer sur le marché financier via l’émission de CLO souscrits principalement par des hedge funds. En conséquence, les tensions qui s’exercent dans la mise en œuvre de telles structures augmentent sensiblement. De fait, les SBO ne s’adressent pour l’instant en France qu’à des opérations sur-mesure d’envergure qui s’imprègnent de cette complexité. 144 IV. Discussion L’analyse de quatre cas permet d’apprécier la logique économique et financière de l’opération SBO en France. La synthèse suivante met l’accent sur les caractéristiques sur de chaque opération et permet des comparaisons. Elle donne lieu à une discussion. Dans cette étude exploratoire, les cas représentent de manière quasi-exhaustive la réalité française en matière de SBO. Nous sommes en mesure de faire plusieurs constats et d’enrichir la réflexion sur les pratiques d’ingénierie financière. Dans la version économique du SBO, les actifs titrisés sont des actifs inscrits au bilan de la cible. La titrisation conjuguée au LBO est en général en concurrence avec plusieurs autres modes de refinancement. La titrisation est, dans trois des quatre cas répertoriés, décalée par rapport au LBO. Elle intervient en aval de l’opération, soit une année ou deux après la reprise. Elle est simultanée dans le dernier cas, dans le cadre d’un serial LBO ( 4ème LBO de Frans Bonhomme). Dans tous les cas, l’opération procure de la trésorerie au service du montage, provoque un subtil réaménagement des pouvoirs et des risques entre acteurs financiers et améliore fictivement l’image financière. Nous en déduisons que la titrisation économique est une opération financière qui offre de la flexibilité aux montages LBO classiques. Elle constitue ainsi une option de croissance potentielle. L’alternative d’une réallocation économique est parfois envisagée. Le montage organisé par Fraikin vise plus précisément le développement de l’activité de la cible. La titrisation du parc de camions et des contrats de location longue durée futurs renforce les cash-flows opérationnels, lesquels sont insuffisants pour financer les nouveaux camions. Ici la titrisation soutient la croissance de l’entreprise. Dans ce cas, l’externalisation des actifs de la cible permet donc de dégager des ressources au service de la croissance tout en favorisant également la remontée de dividendes vers les actionnaires de la société holding. Cependant, les motivations des repreneurs au sein du holding sont contrastées. La restructuration de la dette et la recapitalisation, rarement le redéploiement, sont inhérentes au montage. Par ailleurs, les mesures prudentielles visant à protéger les intérêts des prêteurs seniors, adoptées notamment par Cegelec, indiquent que la titrisation des actifs de la cible peut générer des conflits d’intérêt au sein de la société holding entre repreneurs-dirigeants et financeurs. Dans cet exemple, le programme de titrisation des créances de Cegelec, principal actif de la cible, a généré un problème de rang entre les prêteurs, lesquels se sont retrouvés dans une subordination structurelle, y compris les prêteurs seniors. Pour y pallier, les arrangeurs ajustent systématiquement la durée du programme de titrisation sur celle du LBO. De même, les opérations de titrisation étudiées sont généralement partielles, Fraikin étant le seul cas de titrisation complète (titrisation whole business) à dénombrer en France. De toute évidence, ces différentes mesures visent à rassurer les apporteurs de ressources du holding de reprise. A travers les exemples étudiés, il apparaît nettement que le programme de titrisation est un acte de répartition et de gestion des risques au service des partenaires de la société holding, prêteurs et actionnaires. L’objet de la titrisation vise généralement le remboursement d’une partie de la dette senior, le rachat d’OC, le réaménagement de la dette et le réendettement (restructuration). Plus précisément, notre étude nous conduit à affirmer que la titrisation est au service de la structure financière de la société holding. La perspective de titriser les actifs de la cible au service de la dette, option de refinancement détenue par les créanciers de premier rang, rassure le pool bancaire. Les principales motivations de la holding sont la réduction des frais financiers, qui est généralisée, et l’amélioration de la capacité de financement. Les actifs titrisés sont ciblés (stabilisés, récurrents…) et garantis (surdimensionnement) : ce sont les stocks, les créances commerciales, les parcs de camions ou d’automobiles et les contrats, afin de permettre au FCC d’obtenir une notation favorable. Le SBO vise donc à utiliser une source alternative de financement. Il dégage du cash et relève d’une diversification des ressources. En revanche, le coût intrinsèque du montage (long, coûteux, garanti, mobilisateur de compétences) peut largement compenser le gain. Le 145 SBO s’adresse à des entreprises qui ont des besoins de financement importants au moment du LBO ou en cours de LBO. Si les LBO successifs se développent ces dernières années en France, où l’attrait du marché non coté par les fonds d’investissement est une réalité, qu’en est-il de la possibilité d’utiliser le cash de la titrisation au service du noyau, repreneur et fonds d’investissement ? L’émergence des serial LBOs (Frans Bonhomme) a d’ailleurs renforcé le pouvoir de négociation des managers, lesquels sont de plus en plus associés au processus de sortie et au partage de la plus-value générée par le fonds, mécanismes prévus dans le pacte d’actionnaires. En contrepartie, le risque financier de ces opérations a tendance à augmenter avec la relance de l’endettement. La remontée des taux d’intérêt risque de changer la donne en fragilisant ces montages financiers « tendus ». Le programme de titrisation adopté par Cegelec s’inscrit dans cette logique, où les fonds provenant du véhicule de titrisation permettent un retrait prématuré des deux fonds d’investissement via le remboursement des OC. De même, après avoir envisagé une introduction en bourse début décembre 2006 pour Fraikin, Eurazéo a opté pour une cession en acceptant une offre ferme de CVC Capital Parners. Cette offre porte sur la totalité du capital de Fraikin pour une valeur de l’entreprise à 1350M€. Elle représente pour Eurazéo, actionnaire majoritaire, un produit de cession de 340M€ soit un multiple de 3,4 fois l’investissement initial et un TRI de plus de 35% sur la période courant de mars 2003 à février 2007. Le SBO économique est donc adapté a priori aux opérations où une réelle limite à l’endettement existe. La véritable création de valeur relève de l’utilisation finale du flux de trésorerie dégagé par la titrisation. Or il est principalement affecté à la restructuration et peu au redéploiement. S’il ne sert qu’à rembourser par anticipation des financements, c’est dans le coût du financement de substitution que la création de valeur financière peut être améliorée. Encore faut-il admettre que le SBO a un impact sur la réduction du risque pour les partenaires. S’il sert à ouvrir des possibilités de financements dédiés à des investissements nouveaux rendus impossibles par les tensions du montage LBO, c’est dans la rentabilité de ces investissements que se trouve la véritable création de valeur. Le SBO économique par la titrisation permet de mieux mettre en valeur les actifs économiques de la cible, d’ouvrir des solutions de financement du cycle d’exploitation, accélère l’extinction de la dette et participe à accélérer la sortie des fonds d’investissement. Les fonds d’investissement, à l’origine, avaient vocation à accompagner une évolution dans les opérations primaires en accélérant le développement et donc la valorisation. Les TRI réalisés montrent que les fonds spéculent en favorisant les opérations en série. Ils rachètent une cible, la revendent à un autre fond, la rachète parfois ensuite et ainsi de suite. A chaque nouveau montage, toujours plus tendu, c’est la réalisation du TRI par l’investisseur qui déclenche la cession. Le LBO est donc devenu un produit d’investissement. La titrisation économique participe au mécanisme de valorisation de la cible et à la sortie plus rapide des fonds. Nous sommes bien loin de l’idée de reprise d’un projet économique par un repreneur industriel. Mais, le refinancement de cette dette s’est également organisé. Ce dernier constat et l’ouverture de la titrisation à tout type de classe d’actifs élargissent le cadre d’analyse de la titrisation et pose la question de la marchéisation du LBO. Le SBO devient un produit financier à part entière où les investisseurs peuvent intervenir à différents niveaux et ont des motivations spécifiques parfois bien éloignées de la reprise d’entreprise. Pour sa partie financière, le SBO et donc la titrisation financière, s’intéresse plus particulièrement à la structure financière du holding et aux possibilités de refinancement de la dette sur le marché via le CLO. L’objet de notre étude n’était pas à l’origine d’aborder la titrisation de la dette in fine par les banquiers seniors inscrits dans le montage. Cependant de telles pratiques insistent sur le transfert du risque de crédit sur le marché et les investisseurs. La crise de la titrisation financière ne peut que se répercuter sur la titrisation économique. Elle engendre une crise de confiance vis-à-vis de tous les actifs titrisés et de l’ingénierie des montages. Elle développe une défiance vis-à-vis du secteur financier. Elle provoque une crise du refinancement et donc de liquidité qui devient rare et chère. Elle déclenche une crise de gouvernance dans les montages à durée déterminée, modifiant la structure du pouvoir entre parties prenantes impliquées, et infléchit les stratégies de ces agents intéressés à 146 différents niveaux du montage. Les SBO économiques sont donc rattrapés par la marchéisation des financements via les CDO. Ainsi, les LBO primaires d’envergure se raréfient. Bibliographie Baffreau M. et Timsit A. (2001), « Le management, un enjeu majeur pour la réussite d’un LBO », Option Finance, 659, 31-37 Banque de France (2006), « les mécanismes de transfert de risque sont-ils suffisamment robustes ? », Revue de la stabilité financière,9, 18-31 Banque de France (2007), « Private equity », Bulletin, 165, 115-117 Cadiou C., Cotillard N. et Morvan J. (2007), « Montages financiers complexes : le securitization buy out », Revue du Financier, 166, 4-15 Charreaux G. (2002), « Variation sur le thème : A la recherche de nouvelles fondations pour la finance et la gouvernance d’entreprise », Finance Contrôle Stratégie, 5(3), 5- 68 Cornell B. et Shapiro A. C. (1987), « Corporate Stakeholders and Corporate Finance », Financial Management, 5-14 Jensen M. C. et Meckling W. H. (1976), « Theory of the Firm : Managerial Behaviour Agency Costs and Capital Structure », Journal of Financial Economics, 3, 305-360. Jensen M. (1986), « The Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers », American Economic Review, 76, 323-329 Kaplan S. (1989), « The Effect of Management Buyouts on Operating Performance and Value », Journal of Financial Economics, 24(2), 217-254 Le Nadant A. L. (2000), La performance des LBO en France, doctorat nouveau régime, sld. C. Cadiou, Université de Rennes 1 Modigliani F. et Miller M. H. (1958), « The Cost of Capital, Corporation Finance and the Theory of Investment », American Economic Review, 48, 261-297 Miller M. H. et Modigliani F. (1961), « Dividend Policy, Growth and the Valuation of Shares », Journal of Business, 34, 411-433 Myers S. (1984), « The Capital Structure Puzzle », Journal of Finance, 39, 575-592 Myers S. et Majluf N. (1984), « Corporate Financing an Investment Decisions When Firms have Information that Investors do not have », Journal of Financial Economics, 13, 187-221 Option Finance (2000), « BSN Glasspack refinance sa dette en titrisant ses créances commerciales », 622 Option Finance (2004), « Cegelec conjugue titrisation et dette senior », n° 790 Option Finance (2005), « Fraikin se finance avec une titrisation innovante dite whole business sur son parc et ses contrats de location », 817 Option Finance (2006), « Aspects juridiques et financiers d’un LBO », hors série Option Finance (2006), « LBO : jusqu’où peut monter le marché ? », 908 Option Finance (2006), « Comment Franz Bonhomme a financé son 4ème LBO », 867 Ross S. (1977), « The Determination of Financial Structure: the Incentive-Signalling Approach, Bell Journal of Economics, 8, 23-40 Wruck K. (1990), « Financial Distress, Reorganization, and Organizational Efficiency », Journal of Financial Economics, 27, 419-444 147 ASAC 2008 Peter Jones Halifax, Nouvelle-Ecosse Mohammed CHARMOUH (Etudiant) Université de Lille 2 [email protected] CREATION DE VALEUR FINANCIERE ET STRATEGIQUE LORS DES FUSIONS ET ACQUISITIONS Les fusions et acquisitions sont un moyen nécessaire au développement externe des entreprises permettant ainsi la réalisation de différentes synergies. Ce travail permet de déceler la nature de la valeur créée lors des ces opérations. L’objectif, ici, est de mettre en avant la valeur stratégique et la valeur financière. Les fusions et acquisitions (F&A) sont un moyen nécessaire à la stratégie de croissance externe de l’entreprise, permettant ainsi de dégager directement ou indirectement des profits à court et à long terme. Ces fusions et acquisitions peuvent être considérées comme un mode de restructuration des groupes, dont les desseins ne sont pas toujours cohérents avec leur bonne marche, pour de redéployer tous les éléments matériels et immatériels, et distinguer le plus rentable du moins rentable. Dès lors, le développement stratégique externe impose une connaissance approfondie du passé pour expliquer le présent, ensuite, pour prévoir le futur. Généralement, l’objectif principal des fusions et acquisitions est la réalisation de synergies, qui dépendra souvent de la capacité de la nouvelle entité à déployer et gâter les économies potentielles ou inexploitées de chaque fonction. Dans une opération de fusion ou acquisition, on a au moins deux entreprises. Cette opération se déroule dans un marché occasionnel, aléatoire et imprévisible, appelé dans la littérature marché de contrôle d’entreprise. En conséquence, la réussite de l’opération ne dépend pas seulement de la volonté des entités concernées, mais aussi de la volonté de ce marché. Ensuite, vient l’étape cruciale d’intégration de l’entreprise acquise qui retiendra toute l’attention de l’équipe dirigeante et absorbera tous ses efforts. Ce type de croissance représente une importance capitale dans la stratégie de développement de l'entreprise. Il consiste en une acquisition totale ou partielle, par différents moyens, de titres d'une entreprise, en acquérant ainsi des actifs déjà combinés et organisés, prêts à fonctionner. Cependant, le rachat de tout ou partie des actions d'une autre entreprise est assujetti à des procédures très réglementées et contrôlées. Ainsi, ces opérations doivent être traitées comme une transaction dans un marché de contrôle d'entreprises (Steiner 1975). La théorie d'agence considère ce marché comme un mécanisme de contrôle externe permettant de réduire le comportement déviant des dirigeants et d'augmenter la valeur des entreprises. Ainsi, dans cette recherche, à travers une étude de cas, nous illustrons, tout d’abord, la stratégie adoptée pour réussir ce genre d’opérations. Ensuite, nous mettons la lumière, à l’aide de la méthode d’événement, sur la création de valeur financière. Enfin, à travers la méthode de comparaison, nous mesurons la création de la valeur stratégique. 148 1. Evolution du groupe Sanofi-Aventis 1.1. Le groupe Sanofi Le groupe Sanofi est le résultat de quelques 300 acquisitions dans le monde. Il a été crée suite à la prise de contrôle de la société pharmaceutique Labaz par le groupe Elf-Aquitaine en 1973. Elle devient ainsi la petite filiale santé de ce groupe. Dans les premières années qui ont suivi sa création, Sanofi a procédé aux acquisitions des sociétés Parcor, Clin-Midy et Choay. Ainsi, en 1978, la société lance Ticlid® son premier produit majeur sur le marché. En 1994, Sanofi fait une entrée significative sur le marché américain par l’acquisition de Sterling Winthrop, la branche pharmaceutique du groupe Eastman Kodak. Trois ans après, elle lance sur ce marché son premier produit Avapro® puis Plavix®. Ce dernier produit a toujours fait la fierté du groupe car il représentait le chiffre d’affaires le plus élevé de Sanofi. En 1999, Sanofi fusionne avec Synthélabo, la filiale de l’Oréal, propulsant le nouveau groupe à la 2ème place en France après Aventis, et à la 14ème place mondiale. Synthélabo a été créée en 1970 suite à la fusion des Laboratoires Dausse (fondés en 1834) et les Laboratoires Robert & Carrière (fondés en 1899). Elle lance en 1988 deux produits majeurs sur le marché français, Stilnox® et Xatral®. En 1993, elle lance Stilnox® aux Etats-Unis sous le nom commercial d’Ambien® qui devient, un an après, le premier médicament dans le traitement de l’insomnie. 1.2. Le groupe Aventis Il est issue du rapprochement, en 1999, entre l’allemand Hoechst et le français RhônePoulenc, lui-même fruit de la fusion en 1928 des sociétés chimiques des usines du Rhône et des établissements Poulenc Frères. Tout comme Sanofi, la société franco-allemande fait partie du CAC40. En 2003, elle était classée 1ère en France et 7ème mondiale dans le secteur pharmaceutique. Aventis est une société anonyme financée par action. Son capital social est détenu principalement par le public (80,2%), et par le Groupe Kuwait Petroleum (13,5%). Elle a une forte présence aux EtatsUnis où elle possède un important centre de recherche et plusieurs sites industriels. Malgré tous les efforts et manœuvres engagés par Aventis, il convient de constater que son capital reste toujours dispersé et exposé ainsi aux OPA, ce qui la rend très vulnérable et une proie potentielle pour les grands prédateurs. D’ailleurs, Boisson et al. (2001)1, en parlant de la fusion de Rhône-Poulenc et Hoechst en 1999 qui a donné naissance à Aventis, disaient qu’il est vraisemblable qu’une telle approche négociée soit le résultat d’un effort pour éviter de tomber sous la domination d’un tiers. Ils ajoutent que les deux groupes sont, en effet, confrontés à la concentration du secteur des sciences de la vie et sont, donc, potentiellement une cible pour un acquéreur hostile. Chose qui serait inévitable en 2004. 1.3. Le déroulement de l’OPA En 2003, Sanofi comptabilisait un résultat net de 2.079 millions d’euros contre 1.901 millions d’euros pour Aventis, avec seulement la moitié d’armes que cette dernière (capital, effectif et chiffre d’affaires). La trésorerie nette d’Aventis n’est que de 828 millions d’euros, alors que celle de Sanofi est trois fois supérieure (2.465 millions d’euros), pour un taux d’endettement total de 5,82% chez Sanofi contre 49,71% pour Aventis2. Pour couvrir ses besoins en liquidité, à cause de la baisse du CA, Aventis a doublé ses dettes à long terme en 2003, ce qui a influencé négativement son résultat final. 1 In Martinet A.C. et Thietart R.A., (2001), Stratégies, actualité et futurs de la recherche, Ed. Vuibert. L'endettement change la structure des pouvoirs au sein de la firme et influence fortement sa rentabilité. Zwiebel (1996) et Garvey et Hanka (2001) pensent que les managers utilisent l'endettement non pas en vue de poursuivre les objectifs des actionnaires mais bien pour se protéger des OPA hostiles. Ça peut être le cas d’Aventis ! 2 149 Ainsi, Aventis se trouvait dans une situation de vulnérabilité extrême (très endettée depuis 5 ans, dégradation de la situation générale et capital dispersé). En novembre 2003, Total et L’Oréal (principaux actionnaires de Sanofi) annoncent que leur alliance prendra fin en décembre 2004. Non protégée par son "pacte d’actionnaires" particulier, Sanofi-Synthélabo sera alors très exposée au risque d’OPA. Elle profite ainsi de la vulnérabilité d’Aventis et lance son OPA. Le 3 janvier 2004, Sanofi-Synthélabo transmet confidentiellement un projet à la Commission Européenne pour évaluer les éventuels problèmes concurrentiels d’une offre sur Aventis. Le 14 janvier, la rumeur d’une fusion s’étend. Le 26 janvier, Sanofi-Synthélabo annonce officiellement aux actionnaires d’Aventis une Offre Publique Mixte3. Pour financer cette OPA, Sanofi-Synthélabo a dû négocier un prêt de 12 milliards d’euros avec la BNP Paribas, le groupe Merrill Lynch et quelques sept autres établissements4. Pour contrer l’offre de Sanofi et gagner un peu de temps, en sachant qu’Aventis ne pouvait pas utiliser la stratégie du "pac-man" en lançant une OPA sur Sanofi-Synthélabo5, elle décide donc de déposer un recours en justice contre la recevabilité du 3 février et contre le visa du 13 février. Ceci lui a permis de gagner 3 mois et mieux organiser sa défense, en espérant influencer l’opinion de ses actionnaires et laisser fluctuer le cours de ses titres de façon à rendre l’offre peu attractive. Tout d’abord, elle a mis en place une nouvelle stratégie de communication6 pour renforcer la cohésion de ses actionnaires. Alors, le groupe présente ses bons résultats de 2003 (une hausse de son chiffre d’affaires de 8%, et du bénéfice net par action de 22%), et propose aux actionnaires en manque de liquidités le rachat intensifié de ses propres actions. Ensuite, avec le groupe américain Blackstone, Aventis a créé la société Aurora qui possède les produits "en queue de portefeuille", représentant un chiffre d’affaires de 1.500 millions d’euros en 2003. Enfin, elle utilise la stratégie du "spin-off" qui consiste à céder les activités non stratégiques pour améliorer sa rentabilité et augmenter son cours boursier. L’objectif principal, ici, est de démontrer que l’offre de Sanofi-Synthélabo est insuffisante. En parallèle, Aventis a essayé de trouver un chevalier blanc, comme Roche et Novartis. Le 15 mars, le groupe Suisse Novartis annonce pouvoir relever le prix de l’offre initiale. Le marché semble apprécier cette proposition puisque le cours d’Aventis a progressé de 4,8% le jour de son annonce. Au fait, sans leur intervention, Sanofi-Synthélabo aurait ratée cette OPA et payée cher son échec. En effet, grâce à l’entente amicale entre le PDG de Sanofi et le président français CHIRAC, le gouvernement a suivi de près le déroulement de cette opération et organisé plusieurs rencontres avec les responsables des deux groupes. Ainsi, afin d’éviter un conflit politique avec la France7, le suisse Novartis a retiré son offre. En plus, les deux principaux actionnaires de Sanofi-Synthélabo, Total et L'Oréal, ont apporté la caution financière, ce qui a joué auprès de ses partenaires financiers. Par conséquent, Sanofi a pu renégocier sa dette et l’augmenter de 3 milliards d’euros pour réévaluer son offre initiale sur Aventis. Finalement, la nouvelle offre devient amicale8 et l’absorption d’Aventis par Sanofi se finalise le 31 décembre 2004. 3 Elle combine OPE à 81% et OPA à 19%. L’offre porte sur 47,8 milliards d’euros. L’offre principale est mixte (5 actions Sanofi-Synthélabo et 69€ contre 6 actions Aventis = 60,43€ l’action Aventis). 4 Souvent, plusieurs banques financent en même temps une OPA afin de partager le risque d’échec. Par exemple Total a longtemps fermé sa porte à ceux qui ont soutenu Elf dans son offre hostile ratée sur TotalFina en 1999. 5 Le capital de Sanofi-Synthélabo était verrouillé par Total et L’Oréal jusqu’en décembre 2004. 6 Suite aux OPA, les dirigeants changent leur comportement vis-à-vis des actionnaires. Pour contrer l’OPA de B.S.N. sur Saint Gobain en 1969, la direction a mis en place un plan de communication avec les actionnaires pour leur faire prendre conscience de la valeur de l'entreprise et renforcer ainsi leur attachement à cette dernière (Hirigoyen, 1999). 7 Conformément à l’article 151-3 du Code Monétaire et Financier, le ministre de l’Economie peut bloquer un investissement ne venant pas de l’Union Européenne et concernant le secteur stratégique comme de la santé. 8 Avec une offre principale mixte (5 actions Sanofi et 120€ pour 6 actions Aventis = 68,93€ pour chaque action Aventis, au lieu de 60,43€ proposés auparavant). 150 2. Discussion théorique, méthodologie et données 2.1. Stratégie de F&A : réussites et échecs Généralement, "on peut distinguer les acquisitions qui s’inscrivent dans le cadre de l’analyse stratégique classique et visent à améliorer la position concurrentielle de l’entreprise en exploitant des synergies, de celles à caractère essentiellement opportuniste dont l’objectif et de réaliser des plusvalues financières à court terme, soit en redressant l’entreprise acquise, soit en exploitant une situation de sous-évaluation" (Strategor, 1997, p.186). Ajoutons à cela, le comportement égoïste des managers (Roll, 1986)9, et la recherche d’enracinement par les dirigeants (Shleifer et Vishny, 1988). Cependant, les bonnes attentions ne suffisent pas toujours pour réussir une OPA. Une opération de F&A est très risquée parce qu’elle se déroule dans un environnement très incertain avec une asymétrie d’information élevée. La présence, la curiosité et le poids de quelques acteurs, tels que le conseil d’administration, les financiers, les stratèges, les juristes, les actionnaires, et les partenaires financiers, influencent sur le succès de l’opération. A cela, il faut ajouter l'effet "Penrose"10 qui est lié aux coûts spécifiques de la croissance externe. Plus ces coûts sont élevés, plus l'initiative d'acquérir sera nulle. Dans le contexte français, Caby (1994) a réalisé une étude couvrant la période de 1970-1990 en utilisant des mesures comptables. Il conclut qu'il n'y a pas de création de richesse suite à une opération de F&A, ni en faveur de la cible ni en faveur de l'acquéreur. Souvent les F&A ne se traduisent que par une très faible création de valeur pour les actionnaires de l'entreprise acquéreuse, voire une destruction de valeur, quels que soient le secteur et l'époque (Broyles, 1977 ; Gregory, 1997). Ainsi, Ravenscraft et Scherer (1987) pensent que les désinvestissements, suite à un échec, sont de l'ordre de 20%. Chez Porter (1987) et Young (1981), ils sont de l'ordre de 50%. 2.1.1. Création de valeur financière. Fisher (1930) et Williams (1938) décrivent la valeur comme le revenu qu’un actif est susceptible de produire. La création de valeur est, donc, l’augmentation d’un revenu grâce à des évènements se déroulant entre deux périodes. Cette valeur met l’actionnaire spéculateur au sein de sa propre définition. Elle mesure ainsi le gain que procure une action sur une période courte. Son élément clé est l’investisseur qui achète des titres pour les revendre en misant sur l’augmentation de leur valeur dans le futur proche. Elle est constatée ainsi au sein des marchés financiers. Martin et McConnell (1991) ont mesuré la création de valeur actionnariale en se basant sur l'étude d'événement, ils déduisent que la rentabilité anormale est significativement différente de zéro et toujours positive autour de la date d'annonce (proposition 1a). En s'appuyant sur les travaux de Malatesta (1981) et de Bradley et al. (1988), Seth (1990) constate aussi que les opérations de croissance externe sont créatrices de richesse. Jensen et Ruback (1983) et Jensen (1993) concluent à un accroissement significatif de la valeur de l'acquéreur et de l'acquis. Citons aussi le travail de Barnes (1984), Cartwright et al. (1987), Huang et Walking (1987), Doukas et Travlos (1988) et Jennings et Mazzeo (1991) qui trouvent que les rendements anormaux autour de la date d'annonce sont positifs, pouvant aller jusqu'à +20% (Wansley et al., 1983). Si une entreprise annonce une OPA sur une autre, c’est parce que cette dernière est mal gérée et mal valorisée ainsi par le marché. Après l’acquisition, on suppose que la cible serait mieux gérée et mieux valorisée par le marché. En conséquence, lors d’une OPA, la valeur de l’action de la cible augmente et celle de l’acquéreuse baisse (Jensen et Ruback, 1983 ; Bradley et al., 1988). Cette baisse peut être expliquée par le fait que l’opération soit très risquée à cause des problèmes d’intégration et 9 L’orgueil des dirigeants (Hubris Hypothesis) les pousse à se lancer dans des politiques de croissance externe même si elles s’avèrent néfaste, désirant ainsi s’asseoir sur des trônes au sommet d’un empire. 10 Penrose (1959). 151 d’augmentation de l’endettement pour financer l’acquisition. En outre, Seth (1990), Martin et McConnell (1991), Healy et al. (1992), Jensen (1993), ont constaté que l'augmentation de la valeur profite plus aux actionnaires de la cible qu'à ceux de l'acquéreur (proposition 2b). 2.1.2. Création de valeur stratégique. Pour définir cette valeur stratégique, nous nous plaçons du côté de l’investisseur industriel qui a une optique de continuité de l’exploitation. Le droit aux dividendes et aux votes est, donc, le seul gain qu’il tire de cet investissement à long terme. On parle de synergies stratégiques lorsqu’elles sont générées au niveau de l’entreprise suite à des décisions opérationnelles repensant le redéploiement de ses moyens matériels et immatériels. Ainsi, la valeur stratégique est le résultat de l’efficience des centres opérationnels et des départements fonctionnels, et de l’optimisation du potentiel fiscal et financier de l’entreprise. Elle est, par conséquent, une mesure à long terme de la performance bilantielle et boursière de l’entreprise. Après la fusion, l’attitude de compétition se transforme en une exploitation commune des ressources et une collaboration étroite entre les deux entités. Les économies réalisées ainsi dans les acquisitions horizontales sont dues à la rationalisation des moyens matériels et immatériels, et la suppression des coûts liés aux comportements concurrentiels. Toutefois, ces synergies n’interviennent qu’après la mise en commun des fonctions du nouvel ensemble. Cependant, leur réalisation ne dépend pas seulement de cette mise en commun, mais aussi de la capacité de cette nouvelle entité à déployer et gâter les économies potentielles ou inexploitées de chaque fonction. Healy et al. (1992) confirment que les fusions conduisent à un meilleur taux d’utilisation des actifs. Quant à Kim et McConnell (1977), Asquith et Kim (1982) et Eger (1983), ils constatent que les regroupements d’entreprises permettent la réduction du risque d’illiquidité. Ainsi, les acquisitions liées créent de la valeur opérationnelle (Seth, 1990) grâce aux synergies industrielles (Chatterjee et Lubatkin, 1990), à la réduction de la pression concurrentielle qui est due à l’accroissement des barrières à l’entrée (Porter, 1983) et à la diminution du nombre d’acteurs dans le secteur (Mueller, 1992). Comme les fusions horizontales sont à l’origine des regroupements entre des entreprises concurrentes qui servent le même marché, leur réunion facilite le transfert de spécialité technique et managériale, les restructurations d'actifs (Rappaport, 1986) et permet une meilleure utilisation du management en place (Jensen et Ruback, 1983). Au final, les F&A peuvent êtres vues comme un mode de restructuration de la nouvelle entité, de rationalisation de ses services et personnes, et d’augmentation de son pouvoir vis-à-vis de ses partenaires. Tous ces réaménagements sont à l’origine de l’augmentation des recettes et de la diminution des dépenses, en générant de la valeur stratégique au niveau de l’entreprise (proposition 2). 2.2. Méthodologie et données 2.2.1. Justification de l’étude de cas. Une investigation qualitative et quantitative, puis descriptive et analyste avec des tests, était considérée comme la méthode la plus appropriée pour atteindre le but de cette recherche, celui de la valeur créée et de la stratégie adoptée pour réussir la fusion. Yin (1994) considère l’étude de cas comme une stratégie de recherche légitime, distincte de l’enquête, l’étude en laboratoire et l’étude historique. Donc, la méthode de l’étude de cas implique l’examen d’un phénomène contemporain au sein de son contexte réel, et peut être ainsi généralisable à des propositions théoriques. Cette généralisation est analytique et non statistique (Yin, 1994). Notre objectif est de déceler et découvrir la nature de la valeur créée lors d’une opération de fusion en étudiant une seule opération en profondeur, et mobilisant plusieurs données comptables et financières des entreprises concernées par l’opération. Notre ambition ne vise pas seulement à 152 clarifier et exposer la stratégie de développement externe adoptée par notre entreprise, mais aussi à décrire un phénomène rare qui consiste en une fusion-absorption d’une grande entreprise par une autre plus petite. Autrement-dit, l’absorption du groupe Aventis par la société Sanofi-Synthélabo en 2004 a marqué les esprits. Cette opération de fusion-absorption entamée par Sanofi et subie par Aventis, en sachant que cette dernière représente deux fois Sanofi, symbolise un phénomène rare, sinon unique en son genre. Au final, l’étude de cette opération de fusion qui s’est déroulée en 2004, se propose de nourrir et d’affiner les études récentes sur le déroulement de ce genre d’opérations d’un coté, et de l’autre coté d’expliquer la nature de la valeur créée lors d’une fusion. Dans ce qui suit, nous exposons les méthodologies de calculs et les variables servant à mesurer la création de valeur financière et stratégique. 2.2.2. Méthode d’événement. Le but de cette étude est de voir comment la publication d’un événement influence le cours boursier d’une société quelconque. Cette méthodologie s’inspire des travaux de Sharpe (1964) et Fama et al. (1969) qui s’appuient sur l’idée que le marché boursier réagit immédiatement à des annonces qui sont supposées affecter la performance future de l’entreprise. Ces auteurs ont établit une relation linéaire entre le rendement d’équilibre d’un titre sur une période et le rendement moyen du marché. Ainsi, l’élément principal de cette méthode est la définition avec précision de la date d’événement. La date d’événement Il est difficile de repérer avec certitude la date de la première rumeur que reçoit le marché boursier. Selon nos recherches, nous avons constaté qu’un rapport a été transmis le 3 janvier à la commission européenne, et qu’au 14 janvier une rumeur d’une éventuelle fusion entre les deux groupes touche le marché financier. Face à ces rumeurs et à la demande expresse de l’Autorité des Marchés Financiers, le 16 janvier Sanofi-Synthélabo précise qu’il n’y a aucune négociation en cours avec Aventis. Cependant, malgré ce message, les journaux persistent et parlent d’une probable fusion entre les deux sociétés. Enfin, le 26 janvier Sanofi-Synthélabo annonce son OPA hostile sur Aventis. Selon les réactions de marché, le début de la rumeur peut être le 14 ou le 16 janvier, voire le 22 janvier. Pour évaluer l’action Aventis et lui donner la valeur à laquelle elle est censée être vendue, en considérant que le 22 janvier est le premier jour de rumeur, Sanofi calcule la moyenne des cours entre le 1er et le 21 janvier en ajoutant une prime. La vraie réaction du marché nous convint d’admettre que le jour où la bourse a reçu la vraie rumeur était le 22 janvier, car si l’information se produit avant cette date, les rendements anormaux générés par cette acquisition doivent être observés avant la date d’événement. Pour envisager des analyses qui reflètent la réalité économique, nous supposons que le 22 janvier, considéré par Sanofi comme jour de rumeur et de réaction de marché, est le jour d’événement (date 0) qui sera pris en compte pour mesurer la création de valeur, et non pas le 26 janvier (la date d’annonce officielle). L’important est en effet de savoir quand les marchés sont susceptibles d’avoir pris connaissance de l’opération. Comme nous l’avons souligné auparavant, des fuites peuvent parfois se produire engendrant des mouvements précoces des marchés. Dans ce cas de figure, cette date ne peut être cernée avec fiabilité. Ainsi, nous avons choisi 10 jours avant la date d’événement et 10 jours après pour définir la fenêtre d’événement. Puis, les jours entre le 360ème jour avant la date 0 et le 60ème jour avant cette date, sont considérés comme la période d’estimation. Ces 300 jours nous permettent de calculer la rentabilité normale moyenne. Au total, nous allons exploiter 321 jours par entreprise. Les variables Nous avons choisi le CAC40 comme étant l’indice boursier qui sert à évaluer le rendement moyen du marché. Ce modèle de marché est exprimé sous la forme suivante : 153 Rit = ai + bi Rmt + ε it (1) Où : Rit et Rmt : le rendement du titre i et le rendement de marché sur la période t ; ai : mesure l’indépendance du titre i de la rentabilité du marché ; bi : mesure la dépendance du titre i de la rentabilité du marché ; εit : l’erreur résiduelle dont l’espérance est nulle. Si les deux coefficients ai et bi sont stables sur une période de temps dégagée de l’influence de l’événement, le modèle permet d’évaluer le rendement du titre i sur la période t de façon qu’il se serait établi si l’événement n’était pas produit. Puis, le test F de Ficher a été effectué pour juger la significativité statistique globale du modèle, et le test t de Student pour juger la significativité statistique des deux coefficients de la régression. Le rendement anormal est mesuré par l’écart entre la rentabilité observée (RO) de l’action et sa rentabilité normale estimée (R), qu’elle aurait dû afficher si aucun événement n’avait été survenu, pour chaque jour de la fenêtre d’événement. Alors, le calcul du rendement anormal de l’action i à la date t est le suivant : RAit = ROit − Rit (2) Après avoir constitués deux échantillons, le premier regroupant les données d’Aventis et le second regroupant les données de Sanofi-Synthélabo, et déterminés les rendements anormaux (RA) pour chaque titre, nous avons cumulé ces RA pour chaque titre sur toute la période d’événement (RAC). Ensuite, nous avons calculé la moyenne des rendements anormaux des deux actions (RAM) pour chaque jour de la fenêtre d’événement. Puis, nous avons cumulé ces rendements anormaux moyens (RAMC) sur toute la période de la fenêtre d’événement pour visualiser l’impact total de l’événement étudié. ⎛ 1 ⎞N RAM = ⎜ ⎟ ∑ RA it ⎝N⎠ i Et RAMCt = (3) +10 ∑ RAM t = −10 t (4) N est le nombre total de titres dont le rendement anormal a été estimé pour chaque jour de la période de l’événement de –10 à +10. Ici N est égale à 2. Dans cette étude la variable à expliquer est les rendements anormaux des entreprises Sanofi-Synthélabo et Aventis qui ont réalisé une fusion en 2004. Cette variable a pour objectif de montrer s’il y a une création ou une destruction de la valeur actionnariale lors de la fusion. Pour arriver à cette fin, seuls les rendements anormaux (RA) et les rendements anormaux cumulés (RAC) ont été mobilisés. Les RA ont pour but de visualiser l’impact de l’annonce sur un jour donné, que ce soit avant, pendant ou après l’événement. Par contre, les RAC du début jusqu’à la fin de la fenêtre d’événement, ont pour objectif d’expliquer l’impact global de l’événement sur toute la période considérée comme fenêtre d’événement. Dans une dernière étape, nous avons testé les résultats obtenus pour savoir s’ils sont statistiquement significatifs. Nous avons, donc, mobilisé pour cette fin le test t de Student afin de tester la significativité des RA, RAM, RAC et RAMC. Comme notre travail porte sur un seul échantillon, et conformément à Ruback (1982) et Bruner (1999), l’estimation de t est ajustée par l’autocovariance des gains. Tout d’abord, on a : t = RACt / Ecart typet Ensuite, on a ajusté : Ecart type = ((T *Var( RAt ) + 2(T − 1)Co var(RAt , RAt −1 ))1/ 2 (5) 154 (4) Où T est égale à la différence entre le premier jour d’accumulation et, le dernier jour plus 1. Et la variance et la covariance des deux entreprises ont été estimées sur une période de -360 à -60 jours avant la date d’évènement. 2.2.3. Méthode de comparaison. La technique la plus utilisée pour mesurer la performance à long terme est la méthode dite de "pairage" (Husson, 1988). Cette technique consiste à comparer des données comptables et financières des sociétés concernées avant l'opération avec celles des mêmes sociétés après l'opération. Le but ici est de voir s’il y a une influence de la fusion ou l’acquisition sur la rentabilité de l'entreprise, car le seul évènement qui s’est produit entre les deux périodes (avant et après) est la fusion. Ceci nous permettra de voir s’il y a un changement de la situation de l'entreprise grâce à cette stratégie de développement externe, et déduire la différence entre le fait que chaque entreprise travaille d’une façon indépendante de l’autre, et le fait que les deux entreprises travaillent ensemble avec un seul management et une seule stratégie. La méthode de calcul Comme les deux entreprises n’ont pas la même taille, il serait irrationnel de calculer simplement la moyenne de leurs deux ratios. Pour expliciter notre démarche, imaginons que le taux d’endettement de l’entreprise A est de 30% (27/90 avec 27 de dettes et 90 de fonds propres) et celui de l’entreprise B est de 40% (68/170). Après la fusion, la nouvelle entité devient AB et ce taux devient 37% (107/290). Pour comparer l’endettement des deux entreprises A et B avant la fusion avec celui de l’entité AB, il est préférable de calculer un nouveau taux d’endettement au lieu de prendre leur moyenne [(30%+40%)/2=35%], car ces deux entreprises n’ont pas la même taille (90 et 170). Il faut supposer qu’une entité est la filiale de l’autre et procéder à une intégration simple des comptes. C'est-à-dire qu’il faut additionner leurs dettes financières (27+68=95) et leurs fonds propres (90+170=260). Grâce à ce calcul, on va avoir un taux d’endettement (95/260=36,54%) différent de la moyenne (35%), et qui est plus proche de la réalité. Ainsi, comme les deux entreprises n’ont pas la même taille et ont procédé suite à leur fusion à une intégration globale de leurs comptes, dans notre démarche nous entamons nos calculs de cette façon, en supposant qu’avant la fusion, une entité est la filiale de l’autre, et procédons à une intégration globale de leurs comptes. Le problème de réévaluation du bilan Suite à l’OPA qui a été finalisée en août 2004, les deux groupes ont procédé à une fusion globale de leurs entités. Cette opération consiste en une fusion par absorption qui a conduit à la disparition totale du groupe Aventis. Ainsi, avant d’intégrer tous les comptes d’Aventis dans ceux de l’acheteur (qui est ici Sanofi), ce dernier a commencé par une réévaluation du bilan de la société acquise (qui est ici Aventis). Pour pouvoir comparer les données des deux entités avant et après la fusion, nous devons disposer des mêmes bases de comparaison. Cependant, le problème de la réévaluation des comptes de la société Aventis se pose à ce niveau là. A ce stade trois options se présentent. La première est la plus compliquée et la moins fiable car il faut disposer des données intimes de l’entreprise et suivre la même méthode de réévaluation qui a été utilisée pour la fusion. Elle consiste à réévaluer les comptes de la société Aventis avant la fusion. La deuxième est compliquée aussi mais plus fiable. Elle consiste à retraiter les comptes d’après la fusion de la société Aventis. Par contre si nous connaissons les comptes qui ont été réévalués, nous ne connaissons ni leur influence à long terme sur le résultat, ni leur méthode de réévaluation. Enfin la dernière solution est la moins compliquée et la plus simple des trois. Elle consiste à comparer les variations des performances des groupes Sanofi et Aventis avant la fusion avec celles du nouveau groupe Sanofi-Aventis après la fusion. Autrement-dit, nous prenons les variations de quelques ratios, que nous allons définir un peu plus loin, entre 2004 (l’année de la fusion) et 2006, pour les comparer avec les variations des mêmes ratios des deux groupes entre 2001 et 2003. 155 Nous avons choisi 2004 et 2006, car il faut au moins 2 ans après la fusion pour déduire ses effets (Seth (1990), Jensen (1992) et Singal (1993)). Comme la fusion est intervenue quatre mois avant la clôture de l’exercice (fin août 2004), nous supposons que le changement de management et la mise en commun des services n’auraient aucun impact sur la situation générale de l’entreprise au 31 décembre 2004 et ne seraient pas constatés au niveau du bilan. Donc, le bilan de 2004 est une image fidèle des états consolidés des deux groupes sans l’influence de leur nouvelle stratégie de regroupement. Certes, il y avait des réévaluations, mais nous allons les comparer avec les données de 2006, qui ont subi ces mêmes réévaluations, pour calculer la variation. Si on veut, on peut se contenter de comparer seulement ces deux années 2004 et 2006, mais nous avons préféré porter la comparaison sur plusieurs années pour élargir le champ de notre étude. De ce fait, nous allons traiter les variations entre 2004 et 2006 (après la fusion), et les variations entre 2001 et 2003 (avant la fusion). Les variables Vu la multitude des critères de performance, et le peu de travaux disponibles pour guider le chercheur dans la sélection d’une mesure appropriée de la performance (Keats, 1990), le choix des variables n’était pas un travail aisé. Toutes les méthodes de mesure et d’évaluation ont leurs avantages et leurs inconvénients. Elles ont été défendues par quelques auteurs et critiquées par d’autres. Sachant qu’il n’y a pas un consensus à ce niveau, nous avons choisi les variables comptables et boursières les plus explicatives et les plus cohérentes, à nos yeux, avec notre étude. Tableau 1 : Les critères de mesure de la performance Critères Formules Significations Indicateurs boursièro-comptables EVA (Economic Value Added) M/B (Market To Book) PER (Price to Earning Ratio) RO (1 − Tis ) − ( CMCP x CI ) valeur de marché capitaux propres Exprime la productivité du capital en mesurant la performance interne réalisée par l’équipe en place. Exprime la valorisation par le marché de la politique générale de l’entreprise. valeur boursière de l ' action bénéfice par action Montre la part de bénéfice que procure une action par rapport à sa valeur. résultat opérationn el actif total Exprime la capacité de la société à vendre un produit avec profit indépendamment de sa politique financière et fiscale. résultat net fonds propres Exprime la capacité de l’entreprise à rémunérer ses actionnaires. Indicateurs de rentabilité Rentabilité économique (ROCE) Rentabilité financière (ROE) Rentabilité des actifs (ROA) résultat actif Exprime la capacité de l’entreprise à créer de la richesse grâce à son efficacité opérationnelle. net total Indicateurs de situation Croissance du CA Endettement FCF (Free Cash Flow) CA − CA CA N − 1 N N −1 Dette à plus d ' un capitaux propres an résultat opérationnel + amortissements – IS corrigé – ΔBFR – Investissement net Mesure l’évolution de l’activité de l’entreprise en se concentrant seulement sur le cycle d’exploitation. Montre la politique de financement de la firme et sa capacité à contracter et rembourser d’autres dettes. Mesure le cash disponible dans l’entreprise permettant de rembourser des dettes et d’acquérir des entreprises. Concernant l’EVA, le RO représente le résultat opérationnel, Tis est le taux d’impôt sur les sociétés, CMCP correspond au coût moyen pondéré du capital, et CI représente les capitaux investis. 156 Pour calculer la valeur de marché, on multiplie le cours moyen annuel de l’action par le nombre d’actions au 31 décembre. En ce qui concerne les FCF, on prend le Résultat opérationnel avant frais financier et impôt, plus l’Amortissement net (dotations – reprises), moins l’Impôt sur le bénéfice (34,43%11 calculé sur la base du résultat opérationnel), moins la variation du BFR, moins l’Investissement net (acquisitions-cessions). 3. Résultats et analyse 3.2. La création de valeur financière 3.2.1. Les rendements anormaux. La première remarque que l’on peut déduire du graphique ci-dessous, est que les courbes de la rentabilité normale des titres Aventis et Sanofi, que nous avons estimée grâce à notre modèle de régression, suivent une distribution normale. Ces deux courbes de la rentabilité normale représentent la rentabilité des actions Sanofi-Synthélabo et Aventis que l’on aurait pu constater si aucun événement ne se s’était produit. Graphique 1 : L’évolution des rendements des actions Sanofi et Aventis 6% 4% 2% 10 9 8 7 6 5 4 3 2 1 0 -1 -2 -3 -4 -5 -6 -7 -8 -9 -10 0% -2% -4% -6% -8% Rté normale Aventis Rté normale Sanofi R A Sanofi RAM Av-Sa R A Aventis Ensuite, nous avons effectué des tests à la fois sur les rendements anormaux et sur les rendements anormaux moyens pour voir leur significativité aux seuils statistiques habituels (soit 1%, 5% et 10%). La figure ci-dessous récapitule tous les résultats que nous avons obtenus. La période d’analyse (21 jours) de l’impact de l’annonce, de l’OPA de Sanofi-Synthélabo sur Aventis, est de 10 jours précédents et 10 jours suivants la date de la rumeur. Cette période permet d’identifier des anticipations et des corrections éventuelles du marché boursier. Tableau 2 : Les rendements anormaux des actions Sanofi et Aventis DATE -10 -9 -8 -7 RA Aventis -1,15% *** 2,15% * -1,17% *** 0,97% RA Sanofi -3,02% *** -0,58% -0,41% -0,24% RAM Av-Sa -2,09% *** 0,78% -0,79% *** 0,36% 11 Le taux de l’IS est de 33,33% plus une contribution sociale de 3,3% qui est appliquée aux entreprises dont le montant de l’impôt dépasse 763.000€. 157 -6 2,47% ** -5 -0,37% ** -4 0,56% -3 1,60% -2 -1,43% *** -1 0,25% * 0 4,82% *** 1 1,83% ** 2 3,50% *** 3 -0,03% * 4 1,54% 5 1,88% ** 6 3,74% *** 7 -0,02% ** 8 -0,01% ** 9 0,16% ** 10 -0,29% *** *p<0,1 **p<0,05 ***p<0,01 0,30% -0,71% 2,52% *** -0,62% 0,40% 0,78% 2,94% *** -6,87% *** -5,19% *** 0,20% 0,95% * 0,28% 4,41% *** -2,41% *** -0,13% 1,66% *** 1,80% *** 1,39% -0,54% 1,54% 0,49% -0,52% 0,51% 3,88% -2,52% -0,85% 0,08% 1,25% 1,08% 4,07% -1,22% -0,07% 0,91% 0,75% ** ** *** ** *** *** *** ** * *** *** La rumeur de l’OPA de Sanofi-Synthélabo sur Aventis, a un impact significatif sur les rendements anormaux durant plusieurs jours. Tout d’abord chez la cible (Aventis), un impact global et très significatif pendant les 21 jours a été constaté, sauf pour les jours -7, -4, -3 et +4. Cependant, nous remarquons qu’il y a un impact significatif et négatif au dixième, huitième, cinquième et deuxième jour précédant l’annonce, et au troisième, septième, huitième et dixième jour après l’annonce. Soulignons aussi que le jour de la rumeur (le 22 janvier) a marqué l’augmentation la plus élevée (4,82% de rentabilité anormale), et qui est très significative au seuil de 1%. Nous constatons également un impact significatif et positif qui apparaît le jour avant l’annonce (0,25%), ainsi que les deux jours après l’annonce (1,83% et 3,50%). Ensuite chez l’initiateur, en l’occurrence Sanofi-Synthélabo, il est intéressant de remarquer qu’il y a un impact très significatif et négatif au premier et deuxième jour après le jour de la rumeur (respectivement -6,87% et -5,19%). Toutefois, le jour de la rumeur il réalise un rendement anormal significatif (2,94%), mais il reste moins important que celui de la cible. Comme chez Aventis, le dixième jour avant et le septième jour après la date 0, un impact très significatif et négatif (-3,02% et -2,41%) a été constaté. Cependant, les rendements anormaux montrent un impact important de l’événement pendant les jours 9, 10 et -4, mais le plus important et le plus élevé est constaté au sixième jour (4,41%). Concernant les rendements anormaux moyens, on remarque qu’il n’y a plus d’impact significatif le neuvième et le dixième jour après la date 0. Notons aussi que le rendement anormal moyen du sixième jour (4,07%) est le plus élevé de tous les rendements anormaux moyens. Soulignons aussi que les rendements anormaux de Sanofi influencent négativement la rentabilité totale. Mais globalement, il y a un impact positif et significatif autour de la date d’événement. Ceci correspond à notre proposition 1a, et est cohérent avec les résultats de Jensen et Ruback (1983), Barnes (1984), Cartwright et al. (1987), Huang et Walking (1987), Doukas et Travlos (1988) et Jennings et Mazzeo (1991). Nous voudrions apporter une explication à deux variations significatives et exceptionnelles, à savoir les jours -6 (RA=2,47%) et 6 (RA=3,74%) chez Aventis, et -4 (RA=2,52%) et 6 (RA=4,41%) chez Sanofi-Synthélabo. Le -6 correspond au mercredi 14 janvier, la date -4 correspond au vendredi 16 janvier et le jour 6 correspond au vendredi 30 janvier 2004. Soulignons que le 14 janvier est la date de la première rumeur d’une éventuelle OPA. Ainsi, ce jour là, les gens achetaient les actions de la cible (RA=2,47%), et attendaient la fin de la semaine, qui est ici le vendredi 16 158 janvier, pour acheter celles de l’initiateur (RA=3,52%). Notons que ces deux RA sont les plus élevés avant la date 0. Ensuite, ce phénomène s’est répété une deuxième fois. Le 22 janvier est le jour de la seconde et vraie rumeur, Aventis réalise ainsi un RA de 4,82%. Puis, il a fallu attendre la fin de la semaine (vendredi 30 janvier) pour que Sanofi réalise le RA (4,41%) le plus élevé après l’événement12. Il est frappant de conclure que la cible profite le plus de cette augmentation. Le jour de la rumeur (22 janvier), Aventis réalise un rendement anormal de 4,82%, alors que Sanofi-Synthélabo ne réalise que 2,92%, voire même une rentabilité anormale négative au deuxième et troisième jour (-6,87 et -5,19%). Cela rejoint notre proposition 1b et est conforme aux conclusions de Jensen et Ruback (1983), Bradley et al. (1988), Seth (1990), Martin et McConnell (1991), Healy et al. (1992), et Jensen (1993). 3.2.2. Les rendements anormaux cumulés. Rappelons que les rendements anormaux cumulés expriment le gain global qu’un investisseur peut tirer durant toute la période d’évènement. Le graphique ci-dessous montre les rendements anormaux moyens, les rendements anormaux moyens cumulés et les rendements anormaux cumulés de chaque titre. Graphique 2 : L’évolution des rendements anormaux 25,00% 20,00% 15,00% 10,00% 5,00% 10 8 6 4 2 0 -2 -4 -6 -5,00% -8 -1 0 0,00% -10,00% -15,00% RAC AVENTIS RAC SANOFI RAM Av-Sa RAMC Av-Sa Pour l’annonce de l’OPA de Sanofi-Synthélabo sur Aventis, les RAC d’Aventis n’ont pas cessé de grimper durant toute la fenêtre d’évènement et surtout après l’annonce. Par contre, les RAC de Sanofi ont été négatifs sur toute la période d’évènement sauf le premier jour suivant le jour de la rumeur. Toutefois, les rendements anormaux moyens cumulés ne s’éloignent pas trop des rendements anormaux moyens, car ils ont été influencés par la négativité des rendements anormaux cumulés de la société Sanofi. Le tableau ci-dessous affiche plus clairement ces résultats. 12 L’étrange postulat que les humains préfèrent le loisir au travail, explique ainsi ce phénomène. Vu la proximité du week-end, les investisseurs ont le cœur léger et sont plutôt acheteurs le vendredi. A l’inverse le lundi, à la perspective d’avoir quatre jours de labeur avant le prochain week-end, les investisseurs sont peu enjoués et donc plutôt vendeurs. Chen et Singal (2003) ont une autre explication qui reste dans le domaine de la rationalité : les vendeurs à découvert ne voulant pas garder une position ouverte le week-end la solde en rachetant leur position le vendredi et la réouvrent le lundi en vendant de nouveau à découvert. Ce phénomène a été étudié par Harris (1986), Smirloks et Starks (1986), Hamon et Jacquillat (1994). 159 Tableau 3 : Les rendements anormaux cumulés DATE RAC AVENTIS -10 -1,15% *** -9 0,99% -8 -0,18% ** -7 0,79% -6 3,26% *** -5 2,89% * -4 3,45% *** -3 5,05% *** -2 3,62% * -1 3,86% *** 0 8,68% *** 1 10,52% *** 2 14,02% *** 3 13,99% *** 4 15,53% *** 5 17,41% *** 6 21,16% *** 7 21,13% *** 8 21,12% *** 9 21,29% *** 10 20,99% *** *p<0,1 **p<0,05 ***p<0,01 RAC SANOFI -3,02% *** -3,60% *** -4,01% *** -4,25% *** -3,95% *** -4,66% *** -2,14% -2,75% ** -2,35% ** -1,58% * 1,36% -5,51% *** -10,70% *** -10,51% *** -9,56% ** -9,27% *** -4,87% ** -7,28% *** -7,41% *** -5,75% * -3,95% * RAMC Av-Sa -2,09% *** -1,31% -2,10% -1,73% * -0,34% -0,88% 0,66% 1,15% ** 0,63% 1,14% ** 5,02% ** 2,50% ** 1,66% ** 1,74% * 2,99% *** 4,07% *** 8,14% *** 6,93% * 6,85% *** 7,77% *** 8,52% * D’après ce tableau, nous constatons un impact significatif et positif de l’évènement sur les rendements anormaux cumulés d’Aventis, qui apparaît pendant tous les jours de la fenêtre sauf au 7ème et 9ème jour avant la rumeur. L’évolution de ces rendements anormaux cumulés montre un impact important et très significatif de l’OPA. Ces rendements n’ont pas cessé d’augmenter durant toute la période après l’évènement. Par contre, chez Sanofi, cet impact est négatif mais il n’est pas significatif le jour de la rumeur et 4 jours avant. Globalement, il reste négatif et très significatif. A part le jour de la rumeur (jour 0), Sanofi n’a réalisé que des rendements anormaux cumulés négatifs durant les 20 jours de la fenêtre d’évènement. Pour l’OPA de Sanofi sur Aventis, un impact significatif sur les rendements anormaux moyens cumulés a été mis en évidence à partir du jour précédant la date de la rumeur (-1) jusqu’au 10ème jour suivant cette date. La significativité positive de ces rendements a été fortement influencée par la forte évolution des rendements anormaux de la société Aventis. Notons aussi que le rendement anormal moyen cumulé le plus significatif et le plus élevé a été réalisé le dernier jour de bourse qui correspond au vendredi 30 janvier (ici le 6ème jour après la rumeur = 8,14%). Selon ces résultats, qui sont généralement significatifs durant les dix jours avant l’évènement, nous pouvons conclure qu’il y avait des fuites de l’information concernant l’OPA. Rappelons que Sanofi a envoyé le 3 janvier un dossier détaillant cette OPA, et le 14 janvier la rumeur d’une éventuelle fusion circulait dans le marché boursier. Ainsi, quelques investisseurs vendaient les actions Sanofi dans l’optique d’acheter les Actions Aventis. Ils savaient pertinemment que ce n’était pas une fusion organisée mais plutôt une OPA de Sanofi sur Aventis. Ils savaient aussi que les actions de cette dernière allaient être vendues plus chères. La forte réaction boursière n’a été constatée qu’après la vraie rumeur du 22 janvier. Ceci nous incite à penser que l’information relative à cette OPA est parvenue progressivement aux acteurs boursiers. Voire même avant sa première annonce publique (le 26 janvier). La prise en compte par les acteurs de ce caractère progressif de l’information apparaît clairement, tout d’abord dans le fait 160 qu’aucun autre évènement n’est présent à cette période, ensuite dans le fait que les rendements anormaux cumulés sont significatifs durant plusieurs jours avant la date 0. Pour conclure, nous confirmons qu’un investisseur détenant des actions de la cible d’une OPA (ici Aventis) tire un rendement cumulé plus élevé (ici 20,99%) qu’un investisseur détenant des actions de l’initiateur de la même OPA (ici Sanofi-Synthélabo). Toutefois, dans notre cas étudié, l’actionnaire de l’initiateur ne gagne pas seulement moins que l’actionnaire de la cible, mais il perd carrément de la valeur cumulée (ici -3,95). 3.3. La création de valeur stratégique Le tableau ci-dessous montre les résultats de notre étude en mettant en avant la variation, de nos critères de comparaison, précédent et suivant la fusion. Tableau 4 : Les variations des performances de Sanofi et Aventis Critères Boursièro-comptables EVA (Economic Value Added) M/B (Market To Book) PER (Price to Earning Ratio) De rentabilité Rentabilité économique (ROCE) Rentabilité financière (ROE) Rentabilité des actifs (ROA) De situation Chiffre d’affaires en M d’€ Endettement FCF (Free Cash Flow) en M d’€ Sanofi plus Aventis 2001 Sanofi plus Aventis 2003 Variation avant fusion Variation après fusion Groupe Sanofi Aventis 2006 Groupe Sanofi Aventis 2004 772,01 6,56 37,50 1464,82 4,46 18,36 90,00% -32,00% -51,03% 145,00% 71,84% -4,89% 525,19 2,15 24,36 -1160,40 1,25 25,61 0,12 0,18 0,07 0,18 0,24 0,11 51,91% 32,97% 62,07% 118,96% 76,81% 115,97% 0,06 0,10 0,06 0,03 0,05 0,03 29 429 0,27 1385,06 25 863 0,19 3083,55 -12,12% -28,55% 122,63% 90,79% -53,17% 199,44% 28 373 0,10 7527,12 14 871 0,21 -7569,42 Avant de commencer nos commentaires, nous rappelons qu’il nous était impossible de comparer les résultats, de chaque ratio des différentes années, entre eux, parce que les comptes ont subi, après la fusion, une réévaluation à leur juste valeur. Dans une telle perspective, nous avons pensé qu’il serait pertinent de comparer leur variation et non leur valeur. 3.3.1. Critères boursièro-comptables. A partir de ces résultats, nous constatons que la variation de l’EVA est très forte entre 2004 et 2006. Elle a augmenté de 145%. Par contre, entre 2001 et 2003, elle n’a augmenté que de 90%. Sachant que l’EVA est une mesure de la qualité de l’équipe en place et de la performance interne de l’entreprise, son augmentation peut être expliquée par le fait que travailler ensemble en mettant en commun plusieurs services crée plus de valeur stratégique que de travailler séparément. L’évolution du ratio Market To Book est plus forte après la fusion. Sa croissance est de 71,84% contre une régression de 32% avant la fusion. Notons que les faibles valeurs de ce ratio en 2004 et en 2006 (respectivement 1,25 et 2,15 contre 6,56 pour 2001 et 4,46 pour 2003) sont dues à la réévaluation des capitaux propres de la société Aventis. Nous pouvons conclure, alors, que les investisseurs continuent, suite à la fusion, à faire confiance dans la stratégie générale de l’entreprise, car ce ratio exprime le rapport entre sa stratégie future et sa stratégie passée. En ce qui concerne le Price to Earning Ratio, nous remarquons une forte diminution entre 2001 et 2003 de 51%, alors qu’entre 2004 et 2006, cette diminution n’était que de 4,89%. Nous 161 déduisons grâce à ce ratio que dans les deux situations, avant ou après la fusion, il y avait une baisse de la création de valeur, mais cette baisse était plus maîtrisée après la fusion. Ces ratios correspondent à un mélange entre la traduction bilantielle d’une stratégie industrielle de développement facilement maîtrisable, et la traduction boursière d’une stratégie dont les revenus ne sont pas contrôlables directement par le management. Ainsi, nos résultats sont en conformité avec ceux obtenus par Healy, Palepu et Ruback (1992), et Pilloff et Stephen (1996) utilisant des données boursièro-comptables pour mesurer la richesse créée, et déduisant un accroissement significatif de la performance suite à une fusion ou acquisition. 3.3.2. Critères de rentabilité. Pour la rentabilité économique, une augmentation de 119% a été constatée entre 2004 et 2006, contre 52% entre 2001 et 2003. Une des explications de l’envolée de ce ratio après la fusion, est l’augmentation du chiffre d’affaires entre 2004 et 2006, mais aussi la maîtrise des coûts du nouveau groupe après la fusion. Ceci s’est traduit par un accroissement rapide du résultat opérationnel. Comme ce ratio exprime la capacité de l’entreprise à vendre ses produits avec profit indépendamment de sa politique de financement et de fiscalisation, nous remarquons que cette fusion a permis une réduction des coûts d’exploitation, peut être grâce à la suppression des doublons en machines, locaux et personnes. En commentant la rentabilité financière, nous constatons que sa variation après la fusion (77%) est plus élevée que celle avant la fusion (33%). Cette rentabilité des fonds propres est le résultat net qui reste aux actionnaires après la rémunération de tous les autres partenaires. Logiquement une augmentation de cette rentabilité est un bon signe de la capacité de l’entreprise à rémunérer ses actionnaires. Soulignons aussi qu’en 2003 on enregistre un résultat courant avant impôt de 5.961 millions d’euros et de 4.748 millions d’euros en 2006, pour arriver, à la fin, au même résultat net (à peu près 4.000 millions d’euros en 2003 et 2006). Comme le résultat est égal à la différence entre le résultat courant et l’impôt sur les sociétés, force est de constater que le nouveau groupe a réalisé une économie d’impôt de plus d’un milliard d’euros suite à la fusion13. Concernant la rentabilité des actifs, la différence entre les deux variations est très claire. Une augmentation de 116% pour la période après la fusion contre 62% pour la période précédente la fusion. Cette augmentation est due à l’accroissement du résultat net et à la diminution de l’actif total en 2006. Donc le nouveau groupe crée, suite à la fusion, plus de richesse grâce à la restructuration et l’opérationnalisation de ses actifs. Nos résultats correspondent aux conclusions, tout d’abord, de Rappaport (1983) et Healy et al. (1992) confirmant que les fusions conduisent à un meilleur taux d’utilisation des actifs, ensuite, de Seth (1990) et Gaughan (1991) déduisant que les fusions horizontales permettent de bénéficier des synergies opérationnelles et des gains d’efficience. 3.3.3. Critères de situation. Une augmentation du chiffre d’affaires, dans le cas où l’entreprise adopte les mêmes méthodes comptables, ne peut être expliquée que par deux hypothèses. La première est celle où l’entreprise augmente ses prix, et la seconde est celle où l’entreprise augmente ses quantités vendues. La première représente l’augmentation du pouvoir de négociation vis-à-vis des clients et la seconde représente l’augmentation du pouvoir de marché. Donc, après une baisse de 12% avant la fusion, le nouveau groupe a réalisé une augmentation de son chiffre d’affaires de 91% après la fusion. Ceci peut être expliqué par cette nouvelle taille critique et ce nouveau pouvoir acquis grâce à la fusion. 13 Cela rejoint les résultats de Jensen et Ruback (1983) et Rappaport (1986) qui stipulent que les fusions permettent une meilleure utilisation du potentiel fiscal des deux entités. 162 Ensuite, la baisse du taux d’endettement se traduit par la politique suivie par le groupe Sanofi. Il a toujours essayé de s’endetter le moins possible contrairement à la politique financière d’Aventis. Suite à la fusion, Sanofi qui s’est fortement endetté en 2004 pour financer l’OPA, a réussi à réduire son taux d’endettement de 21% à 10%. Nous constatons que cette baisse est plus importante après la fusion qu’avant la fusion. Elle est due à la nouvelle stratégie et au nouveau style de management des nouveaux dirigeants. Enfin, la forte augmentation du Free Cash Flow entre 2004 et 2006 est beaucoup plus importante que celle entre 2001 et 2003. Sachant que le FCF est la différence entre, d’une part la somme du résultat opérationnel et des amortissements nets, et d’autre part la somme des investissements nets, la variation du BFR et l’impôt corrigé sur les sociétés, logiquement l’absence en 2006 d’un investissement équivalent à celui entrepris en 2004 pour acquérir Aventis influence fortement la variation de ce résultat. En plus, l’amortissement pratiqué en 2006, qui est de 5,2 milliards (contre seulement 1,9 milliard d’euros en 2004), vient gonfler ce FCF. Ces flux de liquidité disponibles expliquent le remboursement massif des dettes du nouveau groupe et le financement de sa croissance. Finalement, ces conclusions ressemblent aux résultats trouvés, tout d’abord, par Stillman (1983), Porter (1987) et Mueller (1992) qui pensent que la création de valeur est due à l’augmentation du pouvoir de marché. Ensuite, par Jensen et Ruback (1983) et Perdreau (1998) qui conçoivent aussi que la création de valeur dans les fusions et acquisitions résulte de la réduction des coûts de faillite et l’accroissement de la capacité d’endettement. Enfin, par Kim et McConnell (1977), Asquith et Kim (1982) et Eger (1983), qui constatent que le regroupement d’entreprises permet de réduire le risque d’illiquidité. Ainsi, tous ces résultats ont contribué à conforter, notre proposition 2. Après la fusion des deux entités et la mise en commun de leurs moyens humains et matériels, le nouveau groupe a réalisé des économies sur ses dépenses en se restructurant, a conquis de nouveaux marchés en augmentant ses ventes, et a soigné sa stratégie générale en se dotant d’un nouveau mode de gestion. Conclusion L’objectif de ce travail consistait à illustrer, grâce à une étude de cas, le rôle du marché de contrôle des entreprises et la nature de la valeur créée suite à une fusion ou acquisition. Pour arriver à cette fin, une approche quantitative permettant de mesurer la création de valeur, et une approche qualitative permettant de comprendre les implications des opérations de fusions et acquisitions, ont été mises en avant. Ainsi, notre étude permet, tout d’abord, de cerner le rôle jouer par le protagoniste et illustrer le soutien infaillible de ses partenaires. Ensuite, elle montre une nette création de valeur financière suite à l’annonce de l’OPA, et stratégique suite à la mise en commun des activités des deux entreprises fusionnées. Toutefois, il convient de signaler la portée limitée de nos résultats dans la mesure où un cas ne permet pas la généralisation statistique. Néanmoins, à notre avis, les études de cas contribuent à la compréhension en profondeur d’un phénomène contemporain. Cependant, une question, à notre connaissance, reste toujours sans réponse, en l’occurrence, les sources et l’origine de la valeur créée. L’augmentation de la valeur ne peut être due qu’à l’accroissement des produits, ou la diminution des charges, ou bien les deux en même temps. Une étude approfondie de ces trois hypothèses de sources de création de valeur serait un apport intéressant dans ce domaine. 163 Annexes Données financières et comptables des sociétés Sanofi Aventis Eléments (M=millions d’euros) Sanofi 2001 Chiffre d’affaires M Amortissement net M Résultat opérationnel M Résultat net M Capitaux propres M Dettes à long terme M Dettes finan. totales M Capitaux investis M BFR M BFR N-1 M Variation BFR M Investissements nets M Total actif M Effectif total au 31/12 Total actions au 31/12 Bénéfice par action Dividende par action Cours boursier au 01/01 Cours boursier au 31/12 Cours annuel moyen Bêta Taux d'imposition moyen Taux d'intérêt moyen Coût réel de la dette Rendement du marché Coût des fonds propres CMCP 6 488 312 2 106 1 586 5 768 119 404 9 172 4 827 3 622 1 205 113 9 967 30 514 732005084 2,17 0,66 69,2 83,8 69,33 0,5927 0,35 0,0650 0,0423 0,0530 0,0491 0,0587 Aventis 2001 Sa+Av 2001 22 941 29 429 -724 -412 3 639 5 745 1 633 3 219 12 021 17 789 4 652 4 771 10 710 11 114 32 731 28 903 -306 4 521 -318 3 304 12 1 217 720 833 39 234 49 201 91 729 122 243 787553585 1519558669 1,91 0,58 93,5 79,75 83,66 0,7814 0,319 0,1326 0,0903 0,0504 0,0489 0,0775 Sanofi 2003 8 048 314 3 075 2 079 6 323 53 368 15 691 4 599 4 211 388 350 9 749 33 086 732848072 2,95 1,02 58,25 59,7 52,28 1,0106 0,34 0,0650 0,0429 0,0630 0,0632 0,0668 Méthode de calcul de l’EVA (Pour 2006) : Capitaux propres = 45.820 millions d’euros Dettes financières = 6.944 millions d’euros Charges financières = 455 millions d’euros Rendement des emprunts d’Etat = 4,35% Rendement moyen du marché = 6,77% Bêta de la société = 0,8447 ( Rit = ai + β i Rmt + ε it ) Taux d’IS moyen = 21% Taux d’intérêt = 455/6944 = 6,55% Coût réel de la dette : 6,55% (1 – 0,21) = 5,17% Coût de fonds propres : 4,35 + 0,8447(6,77 – 4,35) = 6,39% Coût moyen des capitaux pondéré : (45820 x 6,39% + 6944 x 5,17%) / (45820 + 6944) = 6,23% EVA = [4828 (1 – 0,21)] – [(45820 + 6944) x 6,23%] = 525,19 164 Aventis 2003 Sa+Av 2003 17 815 25 863 -859 -545 3 670 6 745 1 953 4 032 10 434 16 757 3 158 3 211 5 085 5 453 25 519 22 210 -556 4 043 -192 4 019 -364 24 284 634 28 277 38 026 75 567 108 653 785905944 1518754016 2,42 0,82 51,8 52,4 46,32 0,8798 0,385 0,0832 0,0512 0,0612 0,0591 0,0695 Sanofi Sanofi Aventis Aventis 2004 2006 14 871 28 373 1 918 5 217 2 426 4 828 2 241 4 399 41 272 45 820 8 654 4 499 16 042 6 944 57 314 52 764 2 333 3 174 5 005 2 460 -2 672 714 14 173 790 85 557 77 763 96 439 100 289 923286539 1359434683 2,18 2,97 1,2 1,75 60,4 75,3 58,8 69,95 55,83 72,34 0,8396 0,8447 0,17 0,21 0,0476 0,0655 0,0395 0,0518 0,0650 0,0677 0,0615 0,0639 0,0554 0,0623 Bibliographie Andrade G., Mitchell M., Stafford E. (2001), "New Evidence and Perspectives on Mergers", Journal of Economic Perspectives, vol. 15, n° 2, p. 103-120. Ansoff H.I. (1984), "Stratégie du développement de l’entreprise", Hommes et Techniques : Traduction française de Ansoff H.I. (1965), "Corporate Strategy", MacGraw Hill. Barnes P. (1984), "The Effect of a Merger on the Share Price of the Attacker", Accounting and Business Research, vol. 15, n° 57, p. 45-50. Bradeley M., Desai A., Kim E.H. (1983), "The rational behind interfirm tender offers : information or synergy?", Journal of Financial Economics, vol. 11, p. 183-206. Caby J. (1994), "Motivations et efficacité des offres publiques d'achat et d'échange en France de 1970 à 1990", Thèse pour le doctorat ès Sciences de Gestion, IAE de Nancy. Cartwright S., Cooper C.L. (1992), "Mergers and Acquisitions : The Human Factor", Ed. Butterworth-Heinemann, Oxford. Chatterjee S. (1986), "Types of synergy and economic value : the impact of acquisitions on merging and rival firms", Strategic Management Journal, vol. 7, n° 2, p. 119-140. DeLong, Gayle L. (2001), "Gains from Focusing versus Diversifying Bank Mergers", Journal of Financial Economics, vol. 59, p. 221–252. Dierickx I., Cool K. (1989), "Asset stock accumulation and sustainability of competitive advantage", Management Science, vol. 35, p. 1504–1511. Doukas J., Holme´ n M., Travlos N. (2002), " Diversification, ownership and control of Swedish corporations", European Financial Management, Vol. 8, p. 281–314. Eger C.E. (1983), "An Empirical Test of the Redistribution Effect in Pure Exchange Mergers", Journal of Financial and Quantitative Analysis, vol. 18, p. 547-572. Fama E.F., Eugene F., Fisher L., Jensen M.C., Roll R. (1969), "The Adjustment of Stock Prices to New Information", International Economic Review, Février, vol. 10, n°l, p. 1-21. Fredrickson J.W. (1985), " Effects of decision motive and organizational performance level on strategic decision processes", Academy of Management Journal, vol. 28, n°4, p. 821-843. Gaughan P.A. (1991), "Mergers and Acquisitions", Ed. Harper-Collins Publishers, New York. Gregory A. (1997), "An Examination of the Long Run Performance of UK Acquiring Firms", Journal of Business Finance & Accounting, vol. 24, n° 7, p. 971-3. Hamon J., Jacquilla B. (1992), "Le marché français des actions : études empiriques 1977-1999", Collection Finance, PUF. Harris L. (1986), "A Transaction Data Study of Weekly and Intradaily Patterns in Stock Returns", Journal of Financial Economics, vol. 69, p. 27-59. Haspeslagh P.C., Jemison D.B. (1987), "Acquisitions : myths and reality", Sloan Management Review, Winter, p. 53-58. Healy P.M., Palepu KG. et Ruback RS. (1992), "Does Corporate Performance Improve after Mergers ?", Journal of Financial Economies, vol. 31, p. 135-175. Huang Y.S. et Walkling RA. (1987), "Target Abnormal Returns Associated with Acquisition Announcements", Journal of Financial Economies, vol. 19, p. 329-349. Hunt J. W. (1990), "Changing pattern of acquisition behavior in takovers and the consequences for acquisition processes", Strategic Management Journal, vol. 11, p. 69-77. Husson B. (1990), "La Prise de Contrôle d'Entreprises", Ed. P.U.F., 2ème Edition, Paris. Jaeger A.M. (1987), "La culture organisationnelle : un élément à ne pas oublier dans les acquisitions et fusions", Revue Internationale de Gestion, septembre, vol. 12, n° 3. Jennings R.H., Mazzeo M.A. (1991), "Stock Price Movements around Acquisition Announcements", The Journal of Business, vol. 64, n° 2, p. 139-164. Jensen C., Ruback RS. (1983), "The market for Corporate Control - The Scientific Evidence", Journal of Financial Economies, vol. 11, p. 5-50. Jensen M.C. (1993), "The Modern industrial Revolution, Exit, and the Failure of internaI Control Systems", Journal of Finance, n° 48, p. 831-880. Kazanjian R. (1988), "Relations in Dominant Problems to Stage of Growth in Technology-Based New Venture", Academy of Management Journal, vol. 31, p. 257-279. 165 Malatesta P.H. (1983), "The wealth effect of merger activity and the objective functions of merging firms", Journal of Financial Economics, vol. 11, p. 155-181. Martin K.J., McConnell J.J. (1991), "Corporate Performance, Corporate Takeovers, and Management Turnover", The Journal of Finance, vol. XIVI, n°2, june, p. 671-687. Martinet A.C., Thietart R.A. (2001), "Stratégies, actualité et futurs de la recherche", Ed. Vuibert. Mathur I., De S. (1989), "A Review of the Theories of and Evidence on Returns Related to Mergers and Takeovers", Managerial Finance, vol. 15, n° 4, p. 1-12. Mueller D.C. (1992), "Mergers", in The New Palgrave Dictionary of Money and Finance, vol. 1113, p. 700-707. Napier N.K. (1989), "Mergers and acquisitions, human resource issues and outcomes: a review and suggested typology", Journal of Management Studies, vol. 26, n° 3, p. 271-89. Penrose E. (1959), "The Theory of the Growth of the Firm", Wiley, New York. Perdreau F. (1998), "Désengagements et recentrage en France: 1986-1992 ", Finance, Contrôle et Stratégie, vol. 1, n°2, p. 137-165. Pettway R.H., Trifts J.W. (1985), "Do Banks Overbid when Acquiring Failed Banks ?", Financial Management, vol. 14, n° 2, p. 5-16. Pilloff A., Steven J. (1996), "Performance changes and shareholder wealth creation associated with mergers of publicly traded banking institutions", Journal of Money, Credit and Banking, vol. 28, n°3, p. 294-310. Rappaport A. (1986), "Creating shareholder value : the new standard for business performance", The Free Press, Mac Millan, New York. Rau P.R, Vermaelen T. (1987), "Glamour, Value and the Post-Acquisition Performance of Acquiring Firms", Journal of Financial Economies, vol. 49, p. 223-253. Ravenscraft D.J., Scherer F.M. (1987), "Life after Takeover", Journal of Indus trial Economies, vol. 36, n°2, p. 147-156. Roll R. (1986), "The Hubris Hypothesis of Corporate Takeovers", Journal of Business, vol. 59, p. 197-216. Romelaer P. (1997), "Changement d'organisation et ressources humaines", in SIMON, Y., JOFFRE, P., "Encyclopédie de gestion", Economica, 2ème édition, p. 306-324. Schwert, G. William (2000), "Hostility in takeovers: In the eyes of the beholder ?", Journal of Finance, vol. 55, p. 2540–2599. Seth A. (1990), "Sources of value creation in acquisitions : an empirical investigation", Strategic Management Journal , vol. 11, p. 431-446. Seth A. (1990), "Value creation in acquisition : a rexamination of performance issues", Strategic Management Journal, vol. 11, n° 2, p. 99-115. Sharpe W.F. (1964), "Capital Asset Prices : A Theory of Market Equilibrium under Conditions of Risk", Journal of Finance, vol. 40, n° 3, p. 425-442. Shleifer A., Vishny R.W. (1988), "Value Maximization and the Acquisition Process", Journal of Economie Perspectives, Winter, Vol. 2, n°1, p. 7-20. Simon H.A. (1945), "Administrative Behavior", Macmillan, New York. Simon, H. A. (1969), "The Sciences of the Artificial", MIT Press, Cambridge, MA. Singh A.H., Montgomery C.A. (1987), "Corporate Acquisition Strategies and Economic Performance", Strategic Management Journal, p. 377-386. Smirlock M., Starks L. (1986), "Day-of-the-week and Intraday Effects in Stock Returns", Journal of Financial Economics, vol. 17, p. 197-210. Stillman R. (1983), "Examining Antitrust Policy towards horizontal Mergers", Journal of Financial Economics, vol. 11, p. 225-240. Thiétart R.A. (1990), "Positions compétitives", Encyclopédie du Management, Tome 1 Vuibert, p. 251-257. Travlos N.G. (1987), "Corporate Takevoer Bids, Methods of Payment, and Bidding Firm's Stock Returns", The Journal of Finance, sept, vol. XIII, n°4, p. 943-963. Wansley J.W., Lane W.R., Yang H.C. (1983), "Abnormal Returns to Acquired Firms by Type of Acquisition and Method of Payment", Financial Management, vol. 12, n° 3, p. 16-23. Wansley J.W., Roenfeldt R.L., Cooley P.L. (1983), "Abnormal Returns from Merger Profiles", Journal of Financial and Quantitative Analysis, vol. 18, n° 2, p. 149-163. 166 Wilkins A.L., Ouchi W.G. (1983), "Efficient culture: Exploring the relationship between culture and organizational performance", Administrative Science Quartely, vol.28, p.468-481. Wilkinson B. (1994), "Labour and Industry in the Asia-Pacifc : Lessons from Newly Industrialized Countries", Berlin : de Gruyter, 1994. Williams O.E. (1979), "Transaction cost Econoinies : the Governance of contractual Relations", Journal of Law and Economics, n°22, p. 233-261. Williamson O.E. (1975), "Markets and Hierarchies", The Free Press, New-York. Williamson O.E. (1981), "The Modern Corporation, Origins, Evolution, Attributes", Journal of Economic Litterature, December, vol. XIX, p. 1537-1568. Yin R. (1994), "Case Study Research, Design and Methods", 2nd Ed., Sage Publications. Young D. (1997), "Economic Value Added : A Primer for European Managers", European Management Journal, vol. 15, n°4, p. 335-343. 167 ASAC 2008 Halifax, Nova Scotia Iraj Fooladi Dalhousie University Real and Nominal Duration: A Multi-Dimension Hedging Strategy1 Abstract In this study we consider two different durations: (1) the real duration and (2) the nominal duration. The real duration is a measure of the sensitivity of asset or liability value to changes in the real rate of interest. The nominal duration is a measure of the sensitivity of an asset or liability value to changes in the expected rate of general price inflation. These two durations arise because the nominal interest rate is divisible into a real rate and the expected inflation rate. Thus, when inflation is present, a duration measure depends on the source of the change in the interest rate. The real and nominal durations can be utilized, in principle, to hedge simultaneously the impact of changes in the real interest rate and changes in the expected inflation rate. I. Introduction: This paper mainly concerns with hedging the equity portion of balance sheets. Usually, the main concern in hedging balance sheets is the derivation of real duration for financial and physical assets. Real durations are measures of the sensitivity of the corresponding value of assets to changes in real rates of interest. Using these durations, Financial (and perhaps non-financial) institutions may be able to devise hedges that prevent the institution’s net worth from fluctuating with real interest rates. In this study, we also develop nominal durations. These durations are measures of the sensitivity of the corresponding value of assets to changes in nominal interest rates induced by changes in expected inflation rates. Using these durations, the institution may be able to devise hedges that prevent the firm's net worth from fluctuating with changes in expected inflation rates. The distinction between real and nominal durations is first mentioned in a comprehensive study conducted for a government’s treasury. A summary of this study is presented in Bierwag, Folladi, and Roberts (2000). Here we demonstrate that it is possible in principle to establish joint hedges of net worth against the risk of fluctuations in real interest rates as well as expected inflation rates. The hedges considered here may be constructed by choosing the cash flow and maturity structure of the assets and liabilities so as to meet variously specified duration conditions. Offbalance sheet activities that produce similar hedges are also possible. Conventional duration and its uses do not distinguish between inflation induced interest rate changes and real interest rate changes. The basic assumption in traditional duration analyses is either that the rate of inflation is zero or that the promised cash flows are independent of inflation rates. Our analysis distinguishes between the sensitivities of an asset's value to changes in real and nominal interest rates. By real duration we mean the sensitivity of an asset's price with respect to shifts in the real interest rate. Real duration is calculated using real cash flows and a real discount rate. The nominal duration measures the asset's price sensitivity to changes in the expected rate of inflation, holding the real rate of interest fixed. In effect, the change in the expected inflation rate changes the nominal interest rates and possibly the cash flows. In our analysis it turns out that the nominal and the real durations are equal to each other when the promised cash flows are insensitive to changes in expected inflation rates. The development of these two duration Measures (real and nominal duration) is essential because they can be utilized, in principle, to hedge simultaneously the impact of changes in the real interest rate and changes in the expected inflation rate. 1 The idea for this paper was developed based on a consulting work done with my colleagues Late Gerry Bierwag and Dr. Gordon Roberts. 168 Inflation may affect the valuation and duration of assets and liabilities in two different ways. First, as stated earlier, nominal interest rates theoretically includes an adjustment for expected inflation to account for the decline in expected purchasing power. Second, for many assets and liabilities, the future cash flows (or other benefits or costs) may be a function of inflation. This effect typically arises because future revenues and costs include a price or wage component which rises with the level of inflation. When these two effects of inflation are present at the same time, they may or may not completely offset one another. It is possible that the offsetting effects will balance perfectly with the result that the value of the asset or liability will be unaffected by changes in inflationary expectations. Inflation-adjusted bonds are theoretically supposed to behave that way. This case is represented formally in Proposition 1 in the next section. As we point out there, the proposition is derived using inflationary expectations and does not require that expected and actual inflation be equal. Also in the next section, we present a series of other propositions to develop a formula for numerical duration. We then suggest a multipurpose hedging strategy in other sections. II. Duration and Inflation As mentioned in the last section, inflation may affect the valuation and duration of assets and liabilities in two different ways (the effect on cash flows and on the discount rate), and they have a tendency to (partially or totally) offset one another. This tendency for the cash flow adjustment and the discount rate adjustment to cancel one another can be illustrated. Consider an enterprise in which the projected future net cash flows can be represented as Ct(1+h)t where Ct is the net cash flow at time t in the absence of inflation and where h is the annual expected rate of inflation. Here, Ct can be regarded as the real cash flow at future date t and it is measured in current dollars reflecting today's purchasing power. Multiplying Ct by (1+h)t gives the cash flows in terms of future dollars so that Ct(1+h)t has the same purchasing power at time t as Ct has today. Here, the future cash flows are changing in order to reflect the changing purchasing power of the dollar. The present value of the future cash flow can be written as Ct(1+h)t/(1+i)t = Ct/(1+r)t (1) where i is the current nominal interest rate and r is the current real interest rate. By the Fisher (1966) effect, we have (1+i)t = (1+r)t(1+h)t so that a cancellation occurs in equation (1) and the current value of Ct(1+h)t is also equivalent to the real value, Ct/(1+r)t. For an asset or liability with many such flows one can write its value as v = Σtct(1+h)t/(1+i)t = ΣtCt/(1+r)t, (2) and the value V is unaffected by the expected future rate of inflation, h. This development can be summarized in proposition 1. Proposition 1: The present value of a stream of future cash flows is independent of expected future inflation rates if the future cash flows and the nominal discount rate are fully adjusted to the expected rate of inflation. It is not necessary in this proposition to assume that the expected rate of inflation equals the actual rate of inflation. It is only necessary that expected inflationary adjustments to future cash flows cancel with the expected inflationary adjustments to the discount rate. Examples to which this proposition would apply are physical assets having cash flows that grow or contract with inflation rates and fully inflation indexed bonds or mortgages. 169 If we follow mechanically the definition of Macaulay (1938) or Fisher and Weil (1971) duration, we can write the duration of the asset or liability in equation (2) as D = ΣttCt(1+h)t/(1+i)tV = ΣtCtt/(1+r)tV (3) and so conclude that the duration (represented by the term on the left) is equivalent to the real duration (represented by the term on the right) but such is not the case. Duration is a measure of the sensitivity of the value of the asset or liability to changes in interest rates. Looking at equation (2), we see that if the nominal rate increases simply because the expected rate of inflation changes, the value is unaffected. We are compelled to argue that the nominal duration (the expression on the left) is zero if the source of its change is the expected inflation rate. On the other hand, if the real rate of interest changes, cet. par., the valuation of the appropriate duration with respect to r is thus given in equation (3). It is clear that the asset or liability in Proposition 1 has two different measures of duration. Thus, we have the following proposition. Proposition 2: If the present value of a stream of future cash flows is independent of expected future inflation rates, the nominal duration is zero if nominal interest rates change because of changes in expected inflation rates and the real duration is given by equation (3) and is greater than zero. This result in Proposition 2 is not a new one, but it has not been expressed quite in this way before. A major feature of the valuation equation in (2) is that the cash flows change as the nominal interest rate changes. Typically, this is associated with adjustable rate mortgages or notes that have cash flows indexed to some nominal interest rate. In proposition 2, we simply identify the source of change in the nominal interest rate as the expected rate of inflation. It should be pointed out that changes in the real rate of interest, cet. par., also cause changes in the nominal discount rate. The nominal duration defined in Proposition 2 is with respect to changes in nominal interest rates induced by changes in the expected rate of inflation. In utilizing duration measures, one must be careful to distinguish the source of interest rate changes. Expressing all assets or liabilities in terms of some base year price level does not alter the duration computations. This is a simple computational result that we might keep in mind. Suppose we wish to express the value of all assets and liabilities in terms of the base year prices of k years earlier. If h has been the average rate of inflation since date k, and if V is the current value of assets or liabilities, then V-k = V/(1+h)k (4) is the value of the current asset or liability in terms of the prices of k years ago. It similarly follows that the durations do not change because of the valuations in a base year. We have, for example, D-k = ΣtCtt(1+r)-t/V = ΣtCtt(1+h)-k(1+r)-t/V-k = D. (5) Summarizing this result, Proposition 3: The duration of an asset or liability is unaffected by expressing the asset or liability values in terms of the prices of any base year. Sometimes in financial analysis, real values are defined as the expression of values in terms of constant purchasing power or in the prices of a base year. It seems appropriate here to point out that a real duration corresponds to the sensitivity of an asset with respect to the real rate of interest and does not simply mean that the values in the duration formulas are expressed in real terms. 170 However, by Proposition 3, it is clear that we can use current or base year prices. Durations are unaffected by such measurement decisions. Of course, not all securities have the adjustable characteristics as noted in Proposition 1. It is worthwhile to consider some major exceptions and to inquire as to a distinction between a corresponding real and nominal duration for them. Consider a security issued with a fixed coupon rate. Most bonds are of this type and have cash flows that are fixed and do not change in response to changes in expected inflation rates. The value of these securities can be written as V = ΣtCt/(1+i)t = ΣtCt/(1+h)t(1+r)t, (6) where Ct are the fixed dollar cash flows. Here, we immediately observe that V changes by the same amount whether there is a unit change in h, cet. par., or a unit change in r, cet. par. We cannot distinguish between the two effects. In this case the bond has a nominal duration exactly equivalent to its real duration. For some assets or liabilities, the inflationary adjustment to the discount rate is a reflection of an entire market for financial and other assets and liabilities. The inflation adjustments to the cash flows for a particular asset may only reflect the expectations of traders or investors in a smaller market. The inflationary adjustment to the cash flows in the smaller market may not be equivalent to the inflationary adjustments to the discount rate in the broader financial markets. This means that the value of the asset or liability may exceed or fall short of the value that would be generated with full equivalence. One might argue that this phenomenon reflects only temporary disequilibrium, because higher values than justified by the market as a whole will attract investors and lower values than justified may discourage investors. Nonetheless, the phenomenon may occur and the length of time that it will last can be anyone's guess. Here, the nominal duration for such a security is greater than zero if any change in expected inflation rates do not equally apply to the cash flows and the discount rate. However, as long as there is some change in cash flows induced by changed inflationary expectations, the nominal duration will be less than the real duration. The expected rate of inflation used in the valuation formulas is an average rate of price change. The price changes for some goods and services may be larger or smaller than for others. Price changes are rarely uniform across goods and services, but reflect lagged inflationary adjustments as well as continually shifting demands and supplies. It would not be unusual for the cash flows for many firms to increase over time at a rate that is smaller or larger than the average rate for the economy as a whole. This feature can alter the nominal durations and may make them even negative. A general formula for the nominal duration can be derived. For this purpose, let us write the value of an asset or liability as V = ΣtCt(1+h′)t/(1+h)t(1+r)t (7) where h′ is the expected inflation rate on the cash flows for a particular firm and h is the expected inflation rate for the economy as a whole. Differentiating V with respect to h we have dV/dh = ΣtCtt(1+h′)t-1(∂h′/∂h)/(1+h)t(1+r)t - ΣtCt(1+h′)tt/(1+h)t+1(1+r)t or (1/V)dV/dh = (∂h′/∂h)(1+h′)-1ΣttCt(1+h′)t/(1+h)t(1+r)tV - (1+h)-1ΣtCtt(1+h′)t/(1+h)t(1+r)tV. Let Dr be the real duration, we can then re-write the above expression as -(1/V)(1+h)dV/dh = Dr{1 - (∂h′/∂h)(1+h)/(1+h′)} = Dn 171 (8) (9) (10) which is the usual way to define a duration with respect to a variable like h. Several observations of this relationship can be noted. If ∂h′/∂h = 0, then Dr = Dn. This requires that changes in the inflation expectations have no effect on the cash flows. If ∂h′/∂h = 1 and h′ = h, then Dn = 0. This is the case where we have a full inflationary adjustment to changes in expected inflation rates. Unusual (and perhaps pathological cases) are possible. If ∂h′/∂h = 1, but if h′<h, then Dn<0. This situation arises when the cash flows initially increase at a lower rate than generally expected inflation rates but where both rates increase on a one-to-one basis after that. This can occur, for example, when h′ = .02, h = .03, but when h increases to .04, h′ increases to .03. A negative duration arises occasionally in the duration literature. It means that dV/dh > 0 or that increased expected inflation rates result in the asset or liability increasing in value. The result in equation (10) can be stated as Proposition 4: The nominal duration of an asset or liability, Dn, can be expressed in terms of the real duration, Dr, as Dn = {1 - (∂h′/∂h)(1+h)/(1+h′)}Dr, where h′ is the expected inflation rate on the cash flows and h is the expected inflation rate for the economy as a whole. The key to the distinction between Dn and Dr is clearly the ratio ( ∂h′/∂h)(1+ h) . (1+ h′) The nominal and real durations will differ from each other as long as this ratio differs from zero. This ratio encompasses the levels of h and h′ as well as their relative changes. Even if the levels take on positive values, Dn and Dr can be equal as long as ∂h′/∂h = 0. For Dr and Dn to differ, changes in the expected rate of inflation for the economy must induce changes in the expected cash flows for this particular asset or liability. To illustrate the distinction between real and nominal durations, we develop a numerical example for an asset which has cash flows over a ten year period. The asset has a real cash flow of $100 in period zero growing at a real rate (g) of 2.5 percent per year to $102.50 (in real dollars) at the end of year 1 and so on. The real discount rate (r) is 4.5 percent giving this asset a real duration of 5.34 years as shown at the bottom of the column labelled "D (real)" in Table 1. For some assets and liabilities, the cash flows may not change by the same amount as the change in the expected inflation rate. In such cases, the nominal duration of the asset is no longer equal to zero because a change in the expected inflation rate does not apply equally to the cash flows and the discount rate. Equation (10) provides a formula for computing the nominal durations when the cash flow adjustments differ from the discount rate adjustments because of changes in expected inflation rates. A practical example, borrowed from Bierwag, Fooladi, Roberts (2000) and discussed in Section V, is electric company’s assets which may have cash flows which inflate at a rate different from CPI. For example, long-term contracts for electricity may cause cash flows to lag inflation. On the other hand, electricity prices may increase faster than the overall inflation rate in some periods. Table 1 shows how nominal durations are computed for such assets. In the middle of the table we construct a case in which an asset's cash flows grow at a real rate of 2.5 percent and inflate at a different rate, h' = 4 percent, for a total growth rate of 6.60 percent. Since the overall inflation rate (h) is only 2 percent as in the first example, this asset is inflating faster than the general economy. The nominal discount rate is 6.59 percent as in the first example. As a result, the price is higher than the base case and nominal duration D(NOM1) is no longer zero. It is .10. 172 A third case occurs when an asset's cash flows inflate at a rate (h') below the overall inflation rate (h)---in this case at 1 percent when the overall inflation rate is 2 percent. As seen in the bottom panel of Table 1, the present value of the nominal cash flows is lower and nominal duration is negative because the cash flows initially increase at a lower rate than inflation (h'<h) but both rates increase on a one-to-one basis. III. Hedging with Real and Nominal Durations It is well known in the duration literature that some effective hedges can be achieved by managing duration gap (matching the durations of assets and liabilities, taking into consideration an adjustment factor for capital structure). Matching nominal durations insulates net worth from shifts in expected inflation rates. Matching real durations insulates net worth from unanticipated shifts in real interest rates. Given any information on how changes in expected inflation may affect cash flows, Equation (10) provides the formula for determining nominal durations. Hedging implementation of our formulas for real and nominal durations involves potential complications because, in practice, real interest rates and inflationary expectations may shift. The discussion of real and nominal durations suggests that, in this case of joint shifts, the sensitivity of an asset's price will depend on both its real and nominal durations. As shown in the appendix the combined sensitivity will also depend on the covariance between shifts in inflationary expectations and shifts in real rates. The simplest case is a zero correlation between inflation and real rates. We believe that this correlation is low, but testing is not possible without reliable data on ex ante real rates and inflationary expectations. When the promised cash flows on a security are adjusted by a formula in response to changes in nominal interest rates, an adjustment to the duration formula is possible. This adjustment has been developed for adjustable rate mortgages (ARMs) typically issued in the U.S. and Canada. In this study, the nominal durations allow for adjustments to the promised cash flows for changes in expected inflation rates. In effect, an additional term is subtracted from the traditional duration to obtain the ARM duration just as an additional term is subtracted from the real duration to obtain the nominal durations. (See Proposition 4). Although the nominal durations are similar to the ARM durations, there are major differences because of mortgage amortization requirements. Nonetheless, ARM durations are a precedent for nominal durations. As long as the precise form of the cash flow adjustment can be specified, the traditional duration can be adjusted to accommodate this special feature. If inflationary conditions are controlled in the long-run so that inflationary expectations are zero, then nominal durations can be discarded. In other words, if inflation is a short-run phenomenon, the measure of nominal duration is also a short run phenomenon. Meanwhile in an inflationary world, an accurate estimate of the nominal duration depends on an accurate estimate of the inflationary impact on cash flows. However, one can derive estimates for real durations of the FI's equity as well as for the nominal durations for its assets. This will result in suggesting a corresponding liability structure that minimizes the risk of changes in the net worth of the FI due to changes in inflationary expectations. In the next section, we formally develop a proposition that suggests a strategy for hedging against changes in both real interest rate and the expected inflation. IV. Duration-Based Hedging --- Real or Nominal? 173 The maturity and cash flow structure of the assets and liabilities on the balance sheet of a firm can in principle be constructed so that the market value of net worth will be invariant to changes in expected inflation rates or real interest rates. If the net worth invariance is achieved by controlling the real and nominal durations of assets and liabilities, a duration-based hedge is effectively in place. The net worth of a firm, the difference between the market value of the assets and liabilities, A-L=N, my be differentiated with respect to the expected inflation rate, to obtain dN/dh = dA/dh - dL/dh = (DLnL - DAnA)(1+h)-1, (11) where DAn and DLn are the nominal durations of the assets and liabilities respectively. If dN/ dh = 0, then DAn = (L/A)DLn, (12) so that the net worth is unaffected by a change in the expected inflation rate, if the assets and liabilities have a maturity and cash flow structure for which DAn = (L/A)DLn. In terms of first order changes, equation (11) is approximated as ΔN = -A[DAn - (L/A)DLn]Δh = -ADGAPNnΔh, (13) where DGAPNn is a coefficient of sensitivity. (For greater accuracy one may wish also to devise a convexity measure). A similar development, of course, can be undertaken with respect to the real rate, r, so that one obtains dN/dr = [DLrL - DArA](1+r)-1, (14) and ΔN = A.DGAPNrΔr. (15) Thus, in terms of hedge targets, we have the two coefficients DGAPNn and DGAPNr that constitute targets, and the controlling entity may wish to set each of these DGAPs equal to zero in order to achieve a net worth hedge with respect to both the expected inflation rate and the real rate of interest. [If instead the firm wanted to keep the ratio, (N/A), invariant with respect to h and r, the relevant DGAPs become DLn - DAn and DLr - DAr respectively.] We can pose this result as: Proposition 5: If a firm wishes to immunize its net worth against changes in expected inflation rates and changes in the real rate of interest, then it should set DGAPNn and DGAPNr equal to zero. Although Proposition 5 is stated with respect to an entire balance sheet, one may decompose the balance sheet into segments and attempt DGAP hedging with respect to these segments. For example, one might pick a single asset and estimate the durations for this enterprise as DAn and DAr. Next, the debt structure used to finance this enterprise would have the liability durations DLn and DLr. It is conceivable that for this particular asset cash flows might well increase with expected inflation rates so that DAn may be close to zero. In that case, one may wish to finance the assets with debt for which DLn is close to zero. A case for financing the enterprise with fully inflation-indexed debt is easily conceptualized in this context. There remains the question of immunization with respect to the real interest rate. There are likely to be sufficient degrees of freedom here, to arrange the maturity structure of the debt so as to retain a zero nominal duration but where DLr can be set so as to keep it in line with DAr and a zero DGAPNr. This example illustrates three main ideas: (1) One can decompose the balance sheet in manner so as to hedge it in segments, (2) An appropriate hedge, 174 when inflation is a significant economic problem, may involve both nominal and real durations and the corresponding DGAPs, and (3) Setting a pair of DGAPs equal to zero involves the simultaneous determination of the debt cash-flow structure and its maturity in order to set two DGAPs simultaneously equal to zero. One may view nominal and real hedging as attempting to satisfy two target objectives simultaneously --- DGAPNn and DGAPNr. As in many policy contexts, the achievement of multiple targets may involve the use of multiple tools. In the example, these multiple tools consist of the many ways in which one can attempt to arrange the cash-flow and maturity structure of outstanding debt. In partitioning the assets of the balance sheet so as to immunize net worth, there are a variety of possibilities that may be kept in mind. For example, we may have Enterprise I having assets AI and liabilities LI and Enterprise II having assets AII and liabilities LII. Immunization would require that the duration of the assets taken together be equal to (L/A) times the duration of the liabilities taken together. Thus, although we may not immunize NI = AI - LI or NII = AII - LII separately, we may immunize N = NI + NII in the aggregate. V. A practical Example: As mentioned earlier, the concept of real versus nominal duration is also useful for hedging balance sheets of non-financial companies. For example Electric company’s assets and equity provide a good opportunity to discuss the distinction between real and nominal durations in a practical hedging context. We begin with an assumption that it is known with certainty that the prices and cash flows of these assets grow with the overall inflation rate. In this case, nominal duration with respect to a shift in inflationary expectations is zero as the present value of the stream of services is independent of inflation. In this case, electricity assets and equity could be financed with index bonds for a perfect hedge. So, if we compare an historical measure of electricity prices (PPI) with overall inflation rate measured by the Consumer Price Index (CPI), and find that they are very closely tied, financing the electric company’s assets with index bonds with real duration gap of zero would have produced a close hedge measured ex post. On the other hand, if electricity price changes and CPI diverge, a different hedge is needed. For example, suppose the product Price Index PPI is 7.36 percent and CPI is only 4.3 percent, reflecting a ratio of PPI/CPI of 1.7. Projected into the future, this suggests that if inflation measured by CPI is 2 percent in the near future as many expect in Canada, then PPI will inflate by 1.7 x 2 percent = 3.4 percent. Because the inflationary portion of the growth rate of electric company’s cash flows exceeds the general inflation rate, the nominal duration with respect to a shift in inflation rates is no longer zero in this case. To show how nominal duration differs from real duration we use the same assumptions made by Bierwag et. Al (2000), where the growth rate ranges from 1.5 percent to 3 percent and the discount rate ranges from 4 to 7 percent. We repeat their calculation of real duration and add our calculation of Nominal Duration (Columns 5 and 6). Nominal duration now adjusts to the difference between inflationary expectations in capital markets (assumed to be 2 percent) and inflationary expectations for electric company cash flows (assumed to be 3.4 percent). The numbers for nominal duration for this case are shown in Columns 5 and 6 of Table 2. Each pair of nominal duration corresponds to one real duration in Column 4 and calculated based on the assumption presented in the same line Table 2. 175 Assuming ∂h'/∂h = 1, the nominal durations are small and range from .16 to .20 (refer to Table 2) with a mean of .18 and standard deviation of .01. When ∂h'/∂h = 0, the nominal and real durations are equal, as expected. Similar logic can also transform the real equity durations into the nominal equity durations displayed in Table 3. Here again, we used the same assumptions undertaken by Bierwag et. al, to calculate real duration and added our calculation of nominal durations (columns 4 and 5). The real equity durations in Table 3 have a mean of 16.66 and a standard deviation of 2.44. For the nominal equity durations, the mean is .23 and the standard deviation is .03. As for assets, the nominal duration equals the real duration when ∂h'/∂h = 0. This example illustrates the distinction between real and nominal durations for the electric company assets, and shows that when cash flows inflate faster than the general inflation rate, nominal durations for electricity assets and equity are not zero. The analysis demonstrates how duration measures, real and nominal, can be calculated in parallel for an important class of assets. An organization wishing to hedge against both inflation and real interest rate shocks needs to pay attention to both duration measures. VI. Summary and Conclusion: In this study we consider two different durations: (1) the real duration and (2) the nominal duration. The real duration is a measure of the sensitivity of asset or liability value to changes in the real rate of interest. The nominal duration is a measure of the sensitivity of an asset or liability value to changes in the expected rate of general price inflation. These two durations arise because the nominal interest rate is divisible into a real rate and the expected inflation rate. Thus, when inflation is present, a duration measure depends on the source of the change in the interest rate. We argue that both the real and nominal durations can be utilized, in principle, to hedge simultaneously the impact of changes in the real interest rate and changes in the expected inflation rate. To implement this in practice, one might pick a single enterprise and estimate the nominal and real durations for the assets of this enterprise as DAn and DAr. Next, the debt structure used to finance this enterprise would have the nominal and real liability durations DLn and DLr. For some enterprises, it is conceivable that public charges might well increase with expected inflation rates so that DAn may be close to zero. In that case, one may wish to finance the assets with debt for which DLn is close to zero. A case for financing the enterprise with fully inflation-indexed debt is easily conceptualized in this context. There remains the question of immunization with respect to the real interest rate. There are likely to be sufficient degrees of freedom here, to arrange the maturity structure of the debt so as to retain a zero nominal duration but where DLr can be set so as to keep it in line with DAr and a zero real duration gap, DGAPNr. This example illustrates two main ideas: (1) An appropriate hedge, when inflation is a significant economic problem, may involve both nominal and real durations and the corresponding DGAPs; and (2) Hedging both nominal and real DGAPs involves the simultaneous determination of the debt cash flow structure and its maturity in order to set two DGAPs simultaneously equal to zero. One may view nominal and real hedging as attempting to satisfy two target objectives simultaneously --- DGAPNn and DGAPNr. As in many policy contexts, the achievement of multiple targets may involve the use of multiple tools. These multiple tools consist of the many ways in which one can attempt to arrange the cash flow and maturity structure of outstanding debt. Our discussion of DGAPs examines hedging of the balance sheet as a whole. However, one may decompose the balance sheet into segments and attempt hedging with respect to these segments. In the case of multi facet companies, hedging by segments may be preferable as it allows fuller use of available information. 176 The concept can also be applied to hedging cash flows that are sensitive to interest rate changes. The hedges can consist of futures positions or interest rate put options. If consistent with other objectives, interest rate swaps that transform these flows into fixed cash flows are also possible. Some of the cash outflows may be sensitive to changes in inflation rates. It may be possible to match changes in these flows with inflation sensitive cash inflows on the asset side. Interest rate swaps in which the interest rate index may be closely correlated with nominal interest rates are another possibility. References: Bierwag, Gerald O., I. Fooladi, and G. Roberts,2000. Risk Management with Duration: Potential and Limitations, Canadian Journal of Administrative Sciences.17,2,126-142. Fisher, Lawrence, 1966. An Algorithm for Finding Exact Rates of Return; I. Introduction. The Journal of Business, Chicago. 39,1, 111-118. Fisher, Lawrence, and Roman L. Weil, 1971. Coping With the Risk of Market-Rate Fluctuations: Returns to Bondholders from Naive and Optimal Strategies. Journal of Business.44, 408431. Macaulay, Frederick, R., 1938. Some Theoretical Problems Suggested by the Movement of Interest Rates, Bonds, Yields, and Stock Prices in the United States since 1856. National Bureau of Economic Research, New York, Columbia University Press. 44-53. 177 Table 1: Examples of Real and Nominal Duration t 1 2 3 4 5 6 7 8 9 10 CF0 Price g h h' r Ct(real) 102.50 105.06 107.69 110.38 113.14 115.97 118.87 121.84 124.89 128.01 100 0.025 0.02 0.02 0.045 Ct(NOM) 104.55 109.31 114.28 119.48 124.92 130.60 136.54 142.76 149.25 156.04 t 1 2 3 4 5 6 7 8 9 10 CF0 Price g h h' r Ct(real) 102.50 105.06 107.69 110.38 113.14 115.97 118.87 121.84 124.89 128.01 100 0.025 0.02 0.04 0.045 Ct(NOM) 106.60 113.64 121.14 129.13 137.65 146.74 156.42 166.75 177.75 189.48 t 1 2 3 4 5 6 7 8 9 10 CF0 Price g h h' r Ct(real) 102.50 105.06 107.69 110.38 113.14 115.97 118.87 121.84 124.89 128.01 100 0.025 0.02 0.01 0.045 Ct(NOM) 103.53 107.17 110.95 114.86 118.91 123.10 127.44 131.94 136.59 141.40 PV(real) 98.09 96.21 94.37 92.56 90.79 89.05 87.35 85.68 84.04 82.43 REAL 900.56 dh'/dh D(real) 5.34 PV(NOM) 98.09 96.21 94.37 92.56 90.79 89.05 87.35 85.68 84.04 82.43 NOMINAL 900.56 1.00 D(NOM1) 0.00 PV(real) 98.09 96.21 94.37 92.56 90.79 89.05 87.35 85.68 84.04 82.43 REAL 900.56 dh'/dh D(real) 5.34 PV(NOM) 100.01 100.02 100.03 100.04 100.05 100.06 100.07 100.08 100.08 100.09 NOMINAL 1000.52 1.00 D(NOM1) 0.10 Wt.t( r ) 0.11 0.21 0.31 0.41 0.50 0.59 0.68 0.76 0.84 0.92 0.00 D(NOM2) 5.34 Wt.t( r ) 0.11 0.21 0.31 0.41 0.50 0.59 0.68 0.76 0.84 0.92 0.00 D(NOM2) 5.34 PV(real) PV(NOM) Wt.t( r ) 98.09 97.12 0.11 96.21 94.33 0.21 94.37 91.62 0.31 92.56 88.98 0.41 90.79 86.43 0.50 89.05 83.94 0.59 87.35 81.53 0.68 85.68 79.18 0.76 84.04 76.91 0.84 82.43 74.69 0.92 REAL NOMINAL 900.56 854.74 dh'/dh 1.00 0.00 D(real) D(NOM1) D(NOM2) 5.34 -0.05 5.34 178 Table 2: Real and Nominal Duration of Electric Company’s Assets g 0.015 0.015 0.015 0.015 0.015 0.015 0.015 0.018 0.018 0.018 0.018 0.018 0.018 0.018 0.020 0.020 0.020 0.020 0.020 0.020 0.020 0.028 0.028 0.028 0.028 0.028 0.028 0.028 0.030 0.030 0.030 0.030 0.030 0.030 0.030 r 0.040 0.045 0.050 0.055 0.060 0.065 0.070 0.040 0.045 0.050 0.055 0.060 0.065 0.070 0.040 0.045 0.050 0.055 0.060 0.065 0.070 0.040 0.045 0.050 0.055 0.060 0.065 0.070 0.040 0.045 0.050 0.055 0.060 0.065 0.070 N 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 Real DUR 13.69 13.35 13.00 12.67 12.34 12.02 11.70 13.87 13.52 12.18 12.84 12.51 12.18 11.87 14.05 13.70 13.36 13.01 12.68 12.35 12.03 14.63 14.28 13.92 13.58 13.24 12.90 12.57 14.78 14.42 14.07 13.72 13.38 13.04 12.71 NOMINAL D(NOM1) D(NOM2) 0.19 13.69 0.18 13.35 0.18 13.00 0.17 12.67 0.17 12.34 0.16 12.02 0.16 11.70 0.19 13.87 0.18 13.52 0.18 12.18 0.17 12.84 0.17 12.51 0.16 12.18 0.16 11.87 0.19 14.05 0.19 13.70 0.18 13.36 0.18 13.01 0.17 12.68 0.17 12.35 0.16 12.03 0.2 14.63 0.19 14.28 0.19 13.92 0.18 13.58 0.18 13.24 0.17 12.90 0.17 12.57 0.2 14.78 0.2 14.42 0.19 14.07 0.19 13.72 0.18 13.38 0.18 13.04 0.17 12.71 179 Table 3: Real and Nominal Duration of Electric Company’s Assets g 0.0175 0.0175 0.0175 0.0175 0.0175 0.0175 0.0175 0.0178 0.0178 0.0178 0.0178 0.0178 0.0178 0.0178 0.0200 0.0200 0.0200 0.0200 0.0200 0.0200 0.0200 r 0.070 0.075 0.080 0.085 0.090 0.095 0.100 0.070 0.075 0.080 0.085 0.090 0.095 0.100 0.070 0.075 0.080 0.085 0.090 0.095 0.100 Real DUR 20.38 18.70 17.28 16.07 15.03 14.13 13.33 20.50 18.79 17.36 16.15 15.10 14.18 13.38 21.40 19.55 18.00 16.69 15.57 14.60 13.75 NOMINAL D(NOM1) D(NOM2) 0.28 20.38 0.25 18.70 0.23 17.28 0.22 16.07 0.20 15.03 0.19 14.13 0.18 13.33 0.28 20.50 0.25 18.79 0.24 17.36 0.22 16.15 0.20 15.10 0.19 14.18 0.18 13.38 0.29 21.40 0.26 19.55 0.24 18.00 0.23 16.69 0.21 15.57 0.20 14.60 0.19 13.75 180 Appendix Minimum Variance Techniques Using the linear duration approximations, one may write the percentage price change for the ith asset as ΔPi = - PiDiΔr, i = 1,2,3,...,N, (1) where Pi is the price, Di is the real duration of asset i, and Δr is a real interest rate change. If Xi is the number of units of the ith asset held, then the change in the value of the holdings of the ith asset is (2) ΔPiXi = - PiXiDiΔr. Let V = ΣPiXi. We can write the proportional change in the value of the assets as ΔV(r)/V = ΣΔPiXi/V = -ΣαiDirΔr = - DrΔr, (3) where αi is the proportion PiXi/V and Dr is the real duration of the assets taken as a whole. The variance of the proportional change in value is then Var[ΔV(r)/VF] = D2rσ2r (4) where σ2r is the variance of the change in the real rate. In a similar development, we have Var[ΔV(h)/V] = D2nσ2h (5) where ΔV(h) is the change in the value of the asset induced by a change in the inflation rate, Dn is the nominal duration, and σ2h is the variance of the change in the inflation rate. Similarly the co-variance of the change in values with respect to h and r is COV[ΔV(h)/V,ΔV(r)/V] = DrDnσhr (6) where σhr is the covariance between the real rate and the inflation rate. When both h and r can change simultaneously, it follows that the total proportional change in asset value is [ΔV(h) + ΔV(r)]/V, and the variance of the total proportional change is Var [(ΔV(h) + ΔV(r))/V] = D2rσ2r + 2DrDnσhr + D2nσ2n. (7) One may then establish an optimization procedure in which this variance is minimized with respect to X1,X2,..., and XN under the assumption that ΣXiPi = V. Such a minimization can be undertaken subject to constraints on minimal values that some of the assets can take on. One may also add liabilities to the problem so that we are minimizing the variance of net worth. Moreover, one can add other random variables to the problem so as to take into account factors that can affect value. Such variables might include exchange rates were applicable. The analysis can be deepened if there are many different real rates that are perhaps maturity specific, as well as different inflation rates that can affect the growth rate of specific cash flows. 181 ASAC 2008 Halifax, Nova Scotia Haibo Jiang (student) John Molson School of Business Concordia University AN EMPIRICAL COMPARISON: TWO SPECIAL CASES OF CEV OPTION PRICING MODEL AND BLACK-SCHOLES MODEL ON S&P CANADA 60 INDEX CALL OPTIONS1 This paper uses real market data of S&P Canada 60 Index call options to compare with the Black-Scholes model, the Absolute CEV model (α=0) and the Square Root CEV model (α=1). Two CEV models outperform the Black-Scholes for in-the-money options based on all four measures. It seems that CEV models could perform better than the Black Scholes model. Introduction Given that the options market is now very large and significant part of the trade of financial instruments, the evaluation of pricing of these derivatives becomes very important for regulators as well as market participants. The value of an option can be estimated by using a variety of quantitative techniques based on the concept of risk neutral pricing. In the famous Black-Scholes pricing formula it is assumed that the underlying stock price returns follow a lognormal distribution. However, empirical studies have shown that this assumption does not perfectly hold. Rubinstein (1994) examines the S&P 500 index option market and finds that Black-Scholes implied volatilities have a “smile” pattern prior to October 1987 market crash and a “sneer” after the crash. Consequently, several kinds of modification of the variances have been tried. The constant elasticity of variance (CEV) diffusion process (Cox, 1975; Cox and Ross, 1976), for example, is an alternative because it allows for the volatility of the underlying asset to be linked to its price level. Beckers (1980) compares with the Black-Scholes, Square Root and Absolute CEV models, and finds that the two special cases of CEV models yield higher option prices than the Black-Scholes model for in-the-money and at-the-money call options. Thus, it is interesting to further test these three models based on actual market data, especially on Canadian options market, which is less studied compared to its counterpart in US. 1 I would like to acknowledge the guidance and feedback of Dr. Stylianos Perrakis, and thank him for assisting in the completion of this paper. I give special thanks to Lora Dimitrova for her participation in the first stage of developing this paper. I also thank the Institute de Finance Mathématique de Montréal (IFM2) for their funding and support. 182 This paper carries out an empirical study of the three different models based on real market data on European style call options on S&P Canada 60 index (SXO), which is one of the most important options products trading on the Montreal Exchange. Similar to the S&P 500 index in the U.S., the S&P/TSX 60 index is an important benchmark index in Canada. As empirical research proposes that there is an inverse relationship between the level of the stock price and its volatility (Beckers, 1980), in fact, a loosely reverse relationship exists between S&P/TSX 60 index level and implied volatilities from SXO options, as illustrated in Figure 1. In addition, a volatility “smile” pattern is also exhibited, as shown in Figure 2. These two facts indicate that the underlying assumption of the Black-Scholes formula, a constant volatility, doesn’t hold. Consequently, the constant elasticity of variance (CEV) model is expected to value options better, thus give option prices that are closer to the actual market prices, than the Black-Scholes model. 183 Based on real market data on European style call options on S&P Canada 60 Index – SXO traded on the Montreal Exchange for the period of January 2006 until December 2007, results show that on average, the Black-Scholes model outperforms the CEV models. However, for the in-the-money options, the two CEV models outperform the Black-Scholes model based on all four measures. The article is organized as follows. The next section provides a brief review of the three models. The following sections provide a description of data, methodology, and report results and conclusions. Constant Elasticity of Variance (CEV) Models Black and Scholes (1973) derive their famous option pricing formula based on an assumption that underlying stock prices returns follow a lognormal diffusion process (1) where the variance is not a function of the stock price. It is well known that the assumption of constant variance is unrealistic, given that various volatility “smile” patterns have been observed. Cox (1975) and Cox and Ross (1976) derive option pricing formula based on the constant elasticity of variance (CEV) diffusion process. In CEV models, the stock price follows the diffusion process (2) where α (0 ≤ α <2) is a constant, known as the elasticity factor. The volatility of CEV models is (3) where σ0 is positive constant. Based on the CEV model, stock volatility tends to move in an opposite direction as stock prices move up and down. The Black-Scholes model is a special case for which α = 2, and therefore the variance is not a function of the stock price. In the case of the Absolute model (α=0) and Square Root model (α=1) a closed-form formula for pricing of European style call options exists. Cox (1975) derives an approximation formula for the Square Root model, while Cox and Ross (1976) introduce closed form solution for the Absolute model. The following are the formulas that will be later applied in the evaluation process of call options. Absolute Model (α=0) C ( S ,τ ) = ( Se − qτ − Ke − rτ ) N ( y1 ) + ( Se − qτ + Ke − rτ ) N ( y 2 ) + v (n( y1 ) − n( y 2 )) v =σ( e −2 qτ − e −2 rτ 1/ 2 ) 2(r − q ) (4) Se −qτ − Ke −rτ v − qτ − Se − Ke −rτ y2 = v y1 = 184 where, N(.) = cumulative unit normal distribution function n(.) = unit normal density function q = the dividend yield Square Root Model (α=1) C ( S ,τ ) = Se − qτ N (q(4)) − Ke − rτ N (q(0)) (5) 4( r − q ) S σ (1 − e −( r −q )τ ) 4( r − q ) K z = 2 ( r − q )τ − 1) σ (e 2 h( w) = 1 − ( w + y )( w + 3 y )( w + 2 y ) − 2 3 w + 2y z (w + 2 y)2 − − − − 1 + h(h − 1)( h h h h ) ( 1 )( 2 )( 1 3 )( )−( )h 2 4 + w y ( ) (w + y) 2( w + y ) q ( w) = 1/ 2 ⎡ 2 w + 2y w + 2y ⎤ )(1 − (1 − h)(1 − 3h)( )) ⎢2h ( 2 2 ⎥ + + w y w y ( ) ( ) ⎣ ⎦ y= 2 where, N(.) = cumulative unit normal distribution function n(.) = unit normal density function q = the dividend yield w = a parameter, which takes on the values of 0 or 4. The above formula for the Square Root CEV model is an approximate formula, which is very convenient to implement. As pointed out by Beckers (1980), however, the approximate formula gives very close prices to those estimated by the exact formula for the Square Root model. Data The data of closing bid and ask prices for S&P Canada 60 Index call options (SXO) are downloaded from the website of the Montreal Exchange.2 The SXO options are European style options, and mature in the third Friday of the contract month. Contract months include the nearest three months plus the next two months in the quarterly cycle, March, June, September and December. The sample period is two years, from January 2006 to December 2007. The sample data contains not only options contract specific information, such as strike prices and expiration dates, but also trading information, such as trade volume and open interest. It is observed that not all options have transactions on each trading day. This phenomenon may result from thin trading of the SXO products in the Montreal Exchange. Consequently, the middle of bid-ask option prices is used to estimate implied volatilities. The total number of call options in the sample is 29,653. 2 http://www.m-x.ca/nego_fin_jour_en.php 185 The daily S&P/TSX 60 index closing prices are derived from S&P/TSX 60 index futures (SXF) prices, which are also downloaded from the website of Montreal Exchange. Observed closing prices of the index from Toronto Stock Exchange is not used because the closing time of call options in Montreal Exchange is different from the S&P/TSX 60 index in Toronto Stock Exchange, which is 4:00 PM, but the same as the S&P/TSX 60 index Futures, which is 4:15 PM. In order to derive the future-based index, corresponding dividend yields and risk free interest rates are collected. Dividend yields for the S&P/TSX 60 index are collected from Bloomberg. Although index net daily dividends are provided, annual dividend yields are used to estimate implied volatilities. The annual dividend yields are updated on daily basis to reflect the net daily dividend. Unlike Dumas et al. (1998), who use the index level net of present value of expected dividends paid over the option’s life, this paper uses the annual dividend yield because it would be convenient while predicting options prices based on implied volatilities afterwards. Canadian Treasury Bills rates are used as risk free interest rates and obtained from the Bank of Canada’s website3. There are four series of daily T-Bills rates: 1-month, 3-month, 6-month, and 1-year. Proper risk free interest rates for different maturities are calculated by interpolating from the rates of the two nearest T-bills. For example, the risk free rate for a 2-month option is calculated by interpolating from the 1-month and 3-month rates. Finally, six inputs (the derived index S, strike price X, risk free rate r, time to maturity T, dividend yield q, and the market middle bid-ask call price c) are ready to calculate implied volatilities for call options. Methodology Implied volatilities are calculated once each week for the sample period. Similarly to Dumas et al. (1998), Wednesdays are selected for these calculations due to the fact that fewer holidays fall on a Wednesday than any other trading day. If a particular Wednesday is a holiday, the first available preceding trading day is used. In addition, options whose moneyness (defined as X/S) is less than 0.90 or great than 1.10 are eliminated since extremely deep in- and out-of-the-money options have relatively small time values. Furthermore, option contracts with fewer than 14 days to expiration are also eliminated because option prices are estimated in two days, one week, and two weeks later, respectively. Like Dumas et al. (1998), this paper eliminates options whose time-to-maturities are longer than 90 days. Finally, 1,981 Wednesday observations are qualified and included in the sample. Two programs are coded in MATLAB to implement the Absolute CEV model and the Square Root CEV model, and two additional programs to estimate implied volatilities based on the two CEV models. For the Black-Scholes model, standard procedures provided by MATLAB Finance Toolbox are utilized to calculate option prices and implied volatilities. Algorithm of finding implied volatility is based on try-and-error with a range of volatility of [0, 10]. The search process will stop when the absolute difference of actual option price and calculated option price is less than . For the same inputs (S, X, r, T, c, q), three implied volatilities are calculated based on the BlackScholes, the Absolute CEV and the Square Root CEV model, respectively. Then, it is assumed that the 3 http://www.bankofcanada.ca/en/rates/tbill-look.html 186 implied instantaneous volatilities of the S&P/TSX 60 index for each option contract do not change over the following 7 days. On each of next Wednesdays, those three implied volatilities, along with the same strike price and newly derived index level, time-to-maturity, risk-free rate, and dividend yield, are used to calculate three theoretical option prices based on these three models, respectively. Finally, these three predicted option prices are compared with actual market option prices. Four measures are designed to compare with the three models. Based on MacBeth and Merville (1980), the first measure is defined as the percentage difference of market option prices and predicted prices by the models. The lower (in absolute term) the average of percentage differences for a specific model, the better the model predicts. One disadvantage of this measure is that some positive overestimations and negative underestimations could cancel each other out. Thus, a second measure is introduced: the absolute percentage difference, which treats both over- and under-pricing equally as estimation errors. The second measure consequently overcomes the above disadvantage. The third measure is the dollar difference between predicted prices and actual market option prices, and the fourth measure is the absolute dollar difference. These last two measures tell us the economic magnitude of estimation errors. However, the last two dollar measures are not as good as the percentage differences. For example, the estimation error of $9 between actual price $20 and predicted price of $29 is smaller than the estimation error of $9 between actual price of $5 and predicted price of $14. Based on the absolute percentage difference and absolute dollar difference, pair-wise t-tests among the three models are conducted in order to investigate whether differences of their predicted prices are significant. In addition, the three models are compared with for three subsamples of out-of-the-money, at-themoney and in-the money options. This classification follows Rubinstein (1985). Based on the ratio of the strike price to the current index price, three subgroups are formed: 1. 2. 3. Out-of-the-money (0.90 to 0.98) At-the-money (0.98 to 1.02) In-the-money (1.02 to 1.10) Further, the three models are also compared with by dividing options into subsamples of different time to maturities. Based on time to maturity, another three subgroups are formed: 1. 2. 3. Very near maturity (2 to 30 days) Near maturity (31 to 60 days) Middle maturity (61 to 90 days) Finally, in order to investigate how the prediction error of the different models changes as the estimation time span changes, additional two time frames are considered: forecast of two days ahead and forecast of two weeks forward. Results This section mainly focuses on analyzing the comparison results of the three different models based on one-week forward predictions. Results for the two-day and two-week forecasts are reported as well. 187 Starting with Table 1, in terms of the percentage differences of actual market option prices and predicted prices by the three models, all three models give very close option prices, ranging on average from −1.90% to −2.26%. In terms of the absolute percentage differences, all three models also give close options prices, varying from 7.90% to 8.28%. Recall that the absolute percentage difference is the strictest measure. In terms of dollar differences, all three predicted prices are on average 13 to 14 cents lower than the actual market option prices, and the absolute dollar differences range from $0.88 to $0.89 per option. It is clear that all three models, on average, underestimate option prices relative to market prices. Based on the pair-wise comparisons between the models, all three models are actually significantly different from each other at least at 10% level. Overall, the Black-Scholes model predicts the price better than the Absolute CEV and the Square Root CEV models. In addition, the Square Root CEV model gives closer prices than the Absolute CEV model. However, the magnitude of differences between models is not big, as Beckers (1980) finds that all three models give very close options prices given the same inputs of stock price, strike price, time to maturity, risk free rate and instantaneous volatility. Table 1: Comparison of one-week forward estimation The second column refers to the percentage difference of market option prices and the predicted prices by three models. The third column is the measure of absolute percentage difference. The last two columns provide the dollar difference and absolute dollar difference between market option prices and the predicted prices by three models. p-values are reported in brackets. The symbols *, **, *** denote statistical significance at the 0.10, 0.05, and 0.01 levels, respectively. % Difference |% Difference| $ Difference |$ difference| Each Model vs. Market Prices Black‐Scholes ‐1.8984% (0.0002)*** 7.8960% (0.0000)*** ‐0.139171 (0.0000)*** 0.884013 (0.0000)*** Absolute CEV ‐2.2630% (0.0001)*** 8.2828% (0.0000)*** ‐0.132668 (0.0000)*** 0.890630 (0.0000)*** Square Root CEV ‐2.0730% (0.0001)*** 8.0751% (0.0000)*** ‐0.136038 (0.0000)*** 0.886396 (0.0000)*** Comparison Between Models BS vs. Absolute ‐0.3867% ‐0.006617 BS vs. SquareRoot (0.0000)*** ‐0.1790% (0.0173)** ‐0.002383 SquareRoot vs. Absolute (0.0000)*** ‐0.2077% (0.0905)* ‐0.004234 (0.0000)*** (0.0024)*** 188 Table 2 presents the results of the three subsamples of out-of-the-money, at-the-money and inthe-money options. For out-of-the-money options, in terms of the absolute percentage difference, prediction errors of all three models are relative big, ranging from 20.40% to 21.92%, which may be caused by many almost zero options price for out-of-the-money options. On the other hand, the absolute dollar differences are not that big, varying from $0.72 to $0.74. Interestingly, all three models overprice out-of-the-money options. Except for the measure of dollar difference, the Black-Scholes gives better prediction than the two CEV models. For the at-the-money options, two CEV models outperform the Black-Scholes model in terms of percentage difference and dollar difference but underperform in two absolute measures. In terms of the absolute percentage difference, prediction errors of all three models decrease dramatically to about 6.80%. The difference between the Black-Scholes model and the Square Root CEV model diminishes based on the pair-wise test of the absolute dollar difference. For in-themoney options, the Absolute CEV and the Square Root CEV models predict closer option prices to the market prices than the Black-Scholes model in terms of all four measures. Particularly, the Absolute CEV model performs the best, followed by the Square Root CEV model. Nevertheless, differences between prices predicted by the three models are insignificant. For the in-the-money options, all three models perform best given that the absolute percentage differences are lowest, around 2.15%. This is consistent with the findings of Beckers (1980) that two CEV models outperform the Black-Scholes model for atand in-the-money call options. The magnitude of the percentage difference for in-the-money options (−0.62% ~ −0.68%) is the smallest compared to out-of- and at-the-money options (−5.22% ~ −6.94% and −1.15% ~ −1.17%, respectively). As options move from out-of-the-money to in-the-money, both the percentage difference and absolute percentage difference of the three models decrease dramatically. Table 2: Comparison of Models Based on Moneyness The second column refers to the percentage difference of market option prices and the predicted prices by three models. The third column is the measure of absolute percentage difference. The last two columns provide the dollar difference and absolute dollar difference between market option prices and the predicted prices by three models. p-values are reported in brackets. The symbols *, **, *** denote statistical significance at the 0.10, 0.05, and 0.01 levels, respectively. Panel A: Out of the Money Options (X/S > 1.02) % Difference Each Model vs. Market Prices Black‐Scholes ‐5.2245% (0.0124)** Absolute CEV ‐6.9365% (0.0033)*** Square Root CEV ‐6.0491% (0.0064)*** Comparison Between Models BS vs. Absolute BS vs. SquareRoot SquareRoot vs. Absolute |% Difference| $ Difference |$ difference| 20.4016% (0.0000)*** 21.9251% (0.0000)*** 21.1160% (0.0000)*** 0.069441 (0.1647) 0.037577 (0.4724) 0.053336 (0.2962) 0.719070 (0.0000)*** 0.736668 (0.0000)*** 0.726476 (0.0000)*** ‐1.5235% (0.0000)*** ‐0.7144% (0.0000)*** ‐0.8092% (0.0000)*** 189 ‐0.017598 (0.0036)*** ‐0.007407 (0.0152)** ‐0.010191 (0.0008)*** Panel B: At the Money Options (0.98 < X/S < 1.02) % Difference |% Difference| $ Difference |$ difference| Each Model vs. Market Prices Black‐Scholes ‐1.1726% (0.0079)*** 6.7947% (0.0000)*** ‐0.121261 (0.0235)** 0.985078 (0.0000)*** Absolute CEV ‐1.1456% (0.0126)** 6.8985% (0.0000)*** ‐0.114202 (0.0383)** 0.995221 (0.0000)*** Square Root CEV ‐1.1579% (0.0101)** 6.8377% ‐0.117858 0.988817 (0.0000)*** (0.0298)** (0.0000)*** ‐0.1038% (0.0081)*** ‐0.010144 (0.0698)* ‐0.0430% (0.0289)** ‐0.003739 (0.1879) Comparison Between Models BS vs. Absolute BS vs. SquareRoot SquareRoot vs. Absolute ‐0.0608% ‐0.006405 (0.0021)*** (0.0225)** Panel C: In the Money Options (X/S < 0.98) % Difference |% Difference| $ Difference |$ difference| Each Model vs. Market Prices Black‐Scholes ‐0.6824% (0.0000)*** 2.1510% (0.0000)*** ‐0.261160 (0.0000)*** 0.897410 (0.0000)*** Absolute CEV ‐0.6248% (0.0000)*** 2.1472% (0.0000)*** ‐0.234997 (0.0000)** 0.895743 (0.0000)*** Square Root CEV ‐0.6536% (0.0000)*** 2.1481% (0.0000)*** ‐0.248162 (0.0000)*** 0.896188 (0.0000)*** 0.0038% (0.6778) 0.001667 (0.6392) 0.0029% (0.5296) 0.001222 (0.4792) 0.0009% (0.8458) 0.000445 (0.8022) Comparison Between Models BS vs. Absolute BS vs. SquareRoot SquareRoot vs. Absolute 190 The findings of the different time-to-maturity options, presented in Table 3, suggest that the Black-Sholes model outperforms the two CEV models based on most measures. As time to maturity increases, the absolute percentage prediction errors of all three models gradually decrease, from 12.62% to 5.50%. This improvement of prediction power is due to the fact that options with longer time to maturity have relatively large time value component in option prices than options with shorter time to maturity. Table 3: Comparison of Models Based on Maturity The second column refers to the percentage difference of market option prices and the predicted prices by three models. The third column is the measure of absolute percentage difference. The last two columns provide the dollar difference and absolute dollar difference between market option prices and the predicted prices by three models. p-values are reported in brackets. The symbols *, **, *** denote statistical significance at the 0.10, 0.05, and 0.01 levels, respectively. % Difference |% Difference| $ Difference Panel A: Time to Maturity < 30 days Each Model vs. Market Prices -5.1188% 11.9571% -0.1793 Black-Scholes (0.0010)*** (0.0000)*** (0.0000)** -5.6322% 12.6167% -0.1691 Absolute CEV (0.0012)*** (0.0000)*** (0.0001)** -5.3643% 12.2740% -0.1743 Square Root CEV (0.0000)*** (0.0011)*** (0.0000)** |$ difference| 0.7266 (0.0000)*** 0.7301 (0.0000)*** 0.7282 (0.0000)*** Panel B: 30 days < Time to Maturity < 60 days Each Model vs. Market Prices -0.5099% Black-Scholes (0.3099) -0.8374% Absolute CEV (0.1534) -0.6659% Square Root CEV (0.2185) 6.7558% (0.0000)*** 7.0250% (0.0000)*** 6.8754% (0.0000)*** -0.1104 (0.0064)*** -0.1040 (0.0121)** -0.1074 (0.0087)*** 0.8948 (0.0000)*** 0.8940 (0.0000)*** 0.8934 (0.0000)*** Panel C: 60 days< Time to Maturity < 90 days Each Model vs. Market Prices -0.8526% Black-Scholes (0.0421)** -1.1212% Absolute CEV (0.0164)** -0.9831% Square Root CEV (0.0261)** 5.4954% (0.0000)*** 5.7933% (0.0000)*** 5.6296% (0.0000)*** 191 -0.1464 (0.0178)** -0.1437 (0.0250)** -0.1452 (0.0209)** 1.0364 (0.0000)*** 1.0593 (0.0000)*** 1.0461 (0.0000)*** For the two days forward estimations and two weeks forward estimations, similar results to the one week forward estimations are found. In Table 4, as estimation period moves from 2 days to 1 week and then to 2 weeks, the absolute percentage difference of the three models increases gradually, from small percentage (7.76% ~ 8.28%) to a little big percentage (8.49% ~ 8.97%). In other words, the estimation error increases slowly as the estimation period increases, and, most importantly, relative performances among the three models remain the same. Table 4: Comparison of Models for Different Prediction Periods The second column refers to the percentage difference of market option prices and the predicted prices by three models. The third column is the measure of absolute percentage difference. The last two columns provide the dollar difference and absolute dollar difference between market option prices and the predicted prices by three models. p-values are reported in brackets. The symbols *, **, *** denote statistical significance at the 0.10, 0.05, and 0.01 levels, respectively. % Difference |% Difference| $ Difference |$ difference| ‐0.015215 (0.6126) ‐0.006571 (0.8334) ‐0.010969 (0.7203) 0.848809 (0.0000)*** 0.871113 (0.0000)*** 0.859421 (0.0000)*** ‐0.139171 (0.0000)*** ‐0.132668 (0.0000)*** ‐0.136038 (0.0000)*** 0.884013 (0.0000)*** 0.890630 (0.0000)*** 0.886396 (0.0000)*** ‐0.140967 (0.0000)*** ‐0.120979 (0.0001)*** ‐0.131194 (0.0000)*** 0.949186 (0.0000)*** 0.957782 (0.0000)*** 0.952284 (0.0000)*** Panel A. Two Days Later Each Model vs. Market Prices Black‐Scholes Absolute CEV Square Root CEV ‐1.3369% (0.2321) ‐1.6562% (0.1994) ‐1.4897% (0.2147) 7.7610% (0.0000)*** 8.2772% (0.0000)*** 8.0079% (0.0000)*** Panel B. One Week Later Each Model vs. Market Prices Black‐Scholes Absolute CEV Square Root CEV ‐1.8984% (0.0002)*** ‐2.2630% (0.0001)*** ‐2.0730% (0.0001)*** 7.8960% (0.0000)*** 8.2828% (0.0000)*** 8.0751% (0.0000)*** Panel C. Two Weeks Later Each Model vs. Market Prices Black‐Scholes Absolute CEV Square Root CEV ‐1.4936% (0.0036)*** ‐1.9771% (0.0012)*** ‐1.7184% (0.0020)*** 8.4870% (0.0000)*** 8.9666% (0.0000)*** 8.6989% (0.0000)*** 192 Conclusions From all the above, this paper has carried out empirical studies in pricing options with three models for the S&P/TSX 60 index call options with the most recent 2-year data. For the implied volatilities, this paper identifies a loosely inverse relationship between implied volatilities and the S&P/TSX 60 index, and presents a volatility “smile” pattern in the Canadian option data. These two facts indicate that the CEV models could outperform the Black-Sholes model because the CEV model assumes that stock volatility moves in an opposite direction as stock prices move up and down. Based on findings of the comparison analyses, the Black-Scholes model works better as a whole, especially for out-of-the-money and at-the-money options and across different time to maturities options. However, for the in-the-money options, the two CEV models outperform the Black-Scholes model in all four measures. This is consistent with the findings of Beckers (1980). In addition, the three models give more accurate estimations for longer time to maturity options than for short time to maturity. Further, for all of the three examined models, the estimation error slowly increases as the estimation period increases. In short, the two special cases of CEV option pricing model are promising, particularly for in-themoney options, in terms of both theoretical explanations and empirical studies. Further research could be to apply a general CEV model with more comprehensive data. 193 References Beckers, S. (1980). The Constant Elasticity of Variance Model and Its Implications For Options Pricing. The Journal of Finance, 65(3), 661-673. Black, F., & Scholes M. (1973). The Pricing of Options and Corporate Liabilities. The Journal of Political Economy, 81(3), 637-654. Boyle, P.P., & Tian, Y. (1999). Pricing Lookback and Barrier Options under the CEV Process. The Journal of Financial and Quantitative Analysis, 34(2), 241-264. Cox, J. (1975). Notes on Option Pricing I: Constant Elasticity of Variance Diffusions. Working Paper Stanford University. Cox, J.C., & Ross, S.A. (1976). The valuation of options for alternative stochastic processes. Journal of Financial Economics 3, 145-166. Dumas, B., Fleming, J., & Whaley, R.E. (1998). Implied Volatility Functions: Empirical Tests. The Journal of Finance, 53(6), 2059-2106. Macbeth, J.D., & Merville, L.J. (1980). Tests of the Black-Scholes and Cox Call Option Valuation Models. The Journal of Finance, 35(2), 285-301. Rubinstein, M. (1985). Nonparametric Tests of Alternative Option Pricing Models Using All Reported Trades and Quotes on the 30 Most Active CBOE Option Classes from August 23, 1976 Through August 31, 1978. The Journal of Finance, 40(2), 455-480. Rubinstein, M. (1994). Implied Binomial Trees. The Journal of Finance, 49(3), 771-818. Schroder, M. (1989). Computing the Constant Elasticity of Variance Option Pricing Formula. The Journal of Finance, 44(1), 211-219. 194 ASAC 2008 Halifax, Nova Scotia Paul Kalyta Telfer School of Management University of Ottawa MAKING IT PERSONAL: THE IMPACT OF CEO RETIREMENT PLAN ON FIRM RISK The study explores the argument that the structure of CEO retirement plan affects CEO risk-taking behavior and therefore firm risk. Analytical and empirical results provide strong support of the argument, thereby making an important contribution to the literature on the impact of executive compensation on firm risk. Introduction In most developed countries, payouts from regular defined benefit and defined contribution retirement plans are limited by income tax regulations. Supplemental Executive Retirement Plans (hereafter, SERPs) are non-contributory pension plans offered by companies to executives as a top-up to regular pension benefits. In essence, SERP benefits are similar to post-retirement salaries. SERP benefits represent a widespread and economically significant form of executive compensation, with most U.S. and Canadian firms offering SERPs to their CEOs and lower-level executives (Sundaram and Yermack, 2007; Kalyta and Magnan, 2008). The non-contributory nature of SERPs means that SERP benefits are entirely sponsored by the employer. In other words, an executive does not have to save for his or her SERP benefits: shareholders will bear all the costs. For this reason, SERPs are conceptually different from regular defined benefit and defined contribution retirement plans, in which both the employer and the employee contribute to the retirement fund. There is, however, a downside effect for executives. Since SERP benefits are entirely sponsored by the firm, in most cases accumulated SERP benefits may not be transferred from one employer to another. Consequently, in case of employment termination prior to the normal retirement age, an executive loses up to 100% (depending on terms of the SERP) of accumulated SERP benefits. Sundaram and Yermack (2007) suggest, inter alia, that CEO SERP arrangements should impact CEO’s risk-taking behavior. The rationale is as follows: if CEO has not reached the age when he or she is entitled to receiving SERP benefits, a positive association between the value of accumulated CEO SERP benefits and CEO’s risk aversion is expected. A CEO will avoid undertaking risky decisions to decrease the probability of job termination (in case of unfavorable outcome of the decision) and preserve accumulated SERP benefits. However, Sundaram and Yermack (2007) fail to provide empirical support of the argument. The study finds no significant impact of the value of accumulated CEO SERP on firm capital expenditures and frequency of debt rating upgrades – two proxies for firm riskiness. The results in Sundaram and Yermack (2007) are inconclusive due to several major conceptual and methodological limitations. First, the authors implicitly assume that SERPs are homogeneous, whereas in reality SERPs vary significantly from one executive to another. One important source of variation which is expected to have a key impact on CEO’s risk tolerance is the contingency of SERP benefits on firm performance. Second, Sundaram and Yermack (2007) do not account for the probability of early employment termination. For instance, the probability of early termination is likely to be low when the CEO approaches the age of normal retirement, i.e., the age at which the CEO is entitled to receiving 195 SERP benefits – as the payouts from golden parachutes are likely to be significantly more costly for the employer than keeping the CEO in the office until he or she retires. I develop an explicit analytical model according to which the relationship between CEO’s risktolerance and CEO’s SERP benefits varies according to the performance-contingency of SERP benefits and the probability of early employment termination. To verify the predicted relationships, several hypotheses are tested. In general, the results provide empirical support of the analytical predictions. Specifically, CEOs whose SERP benefits are contingent on performance appear to be more risk-tolerant than CEOs whose SERP benefits are not contingent on performance. Also, CEO’s risk tolerance is negatively associated with the size of accumulated SERP benefits when SERP benefits are not contingent on performance. The results are substantially different from those reported in Sundaram and Yermack (2007) who suggest that a relationship between CEO SERP benefits and risk tolerance is possible but fail to provide empirical support for such relationship. The results provide an important contribution to the literature on the impact of executive compensation on executive risk tolerance and, ultimately, firm decisions. The rest of the paper is organized as follows. Section 2 provides an overview of SERPs. Section 3 develops analytical predictions. Section 4 describes research methodology. Results are reported in Section 5. Section 6 concludes. SERP Supplemental retirement arrangements exist due to governmental regulations that limit the retirement income under regular pension schemes. In Canada, the Income Tax Act sets the limit on the annual income from Registered Pension Plans (RPP) at $2,333 per year of pension plan membership, which leaves generously-paid executives with but a modest fraction of their pre-retirement income. Kalyta and Magnan (2008) investigate a sample of CEOs of S&P/TSX60 firms and find that RPP benefits do not, on average, exceed five percent of CEO’s pre-retirement cash compensation – due to the Income Tax Act limitations. Similar limitations are in place in the U.S. According to the Internal Revenue Code, the limit on pensionable earnings under a qualified plan is set at $225,000. Consider a CEO who retires in 2007 with a $1,000,000 pre-retirement base salary, 35 years of pensionable service and the pension plan that calls for 2% of the last base salary multiplied by the number of years of service to be paid to CEO annually upon retirement. Under a regular pension scheme, CEO’s retirement benefits are limited to $157,500 per annum (35 * 2% * $225,000) because of the cap. A SERP is a non-contributory pension plan (i.e., it is completely funded by the employer) that permits to increase executive’s post-retirement income beyond the regular pension limit. In essence, SERP benefits are similar to a post-retirement salary. Under a SERP, a firm enters into a long-term contractual obligation with an executive to make ongoing retirement payments in excess of the regular pension cap until the death of the executive, or sometimes until the death of the surviving spouse. If there is a reasonable expectation that a firm may not honor the contract – e.g., in case of a hostile takeover or bankruptcy – it can be guaranteed by a letter of credit. Most often, the design of a SERP reflects the design of a regular pension plan: a certain percentage (multiplier) of pensionable earnings for each year of pensionable service to be paid annually to a retiree upon reaching the retirement age. In some cases, retirement benefits under SERPs are not associated with the number of years of pensionable service and are determined by pensionable earnings and multiplier only. However, while the general design of a SERP formula is straightforward, the way its components are determined and valued differ significantly from one executive to another. For instance, in some cases, pensionable earnings are limited to the base salary. In other cases, performance bonuses are also taken 196 into account. The design of other SERP components is also subject to alternatives: whether to provide retirees with survivor benefits, whether to limit the size of SERP benefits, whether to adjust SERP benefits to inflation, whether to allow early retirement, whether to impose tenure requirements, whether to fund retirement benefits, etc. Development of Research Hypotheses SERPs came to the attention of academic researchers in the last several years due to persistent anecdotes on excessive SERP benefits in individual cases and due to extremely limited disclosure of SERPs in publicly available corporate statements (prior to 2007, the companies were not required to disclose the value of actual or expected SERP benefits of top executives). Due to the lack of SERP disclosure, Murphy (1999) referred to SERP benefits as “stealth”, or hidden, compensation. Bebchuk and Fried (2004) suggested that SERPs were used to increase executive compensation off the radar screen of shareholders. The “stealth” nature of SERP benefits made them an attractive choice for managers with power to extract rents (i.e., receive compensation above the level that would have been received under optimal contracting), as the opposition from shareholders was likely to be minimal. Kalyta and Magnan (2008) investigated a sample of CEOs of S&P/TSX60 firms and provided empirical support of the latter arguments. The study found strong positive associations between the incidence and magnitude of CEO SERP benefits on one side and CEO power over the board on the other, consistent with the rent extraction hypothesis. Sundaram and Yermack (2007) provide a comprehensive but largely descriptive study on SERPs, their determinants, consequences, and relationship with other forms of executive compensation such as stock option grants. The authors suggest, inter alia, that CEO SERP arrangements impact CEO’s risktaking behavior. In case of employment termination prior to the normal retirement age, a CEO loses a certain percentage (up to 100%, depending on terms of the SERP) of accumulated SERP benefits. Consequently, the more significant is the value of accumulated CEO SERP benefits, the more risk-averse the CEO should be – if the CEO has not reached the age in which he or she is entitled to receiving SERP benefits yet. Sundaram and Yermack (2007) provide some graphical and statistical analysis on the association of the value of CEO’s SERP with firm’s risk. One graph shows that capital investments appear to decline as the value of CEO’s SERP increases. Another graph shows that increased SERP benefits also lead to a greater frequency of debt rating upgrades. However, using multivariate regressions, Sundaram and Yermack (2007) fail to find significant support of these relationships. The analysis in Sundaram and Yermack (2007) is limited for several reasons. First, it implicitly assumes that SERPs are homogeneous. In reality, SERPs significantly differ from one CEO to another. A critical characteristic is the performance-contingency of SERP benefits. While some SERP benefits are contingent on firm performance, others are not. Second, the study ignores that the probability of early employment termination and, consequently, a loss of accumulated SERP benefits due to the termination is different at different points of CEO career. For instance, the probability of early termination is likely to be low when the CEO approaches the age of normal retirement, i.e., the age at which he or she is entitled to receiving SERP benefits – as the payouts from golden parachutes are likely to me significantly more costly for the employer than having the CEO in the office for another year or so. The models below demonstrate that the performance-contingency of SERP benefits and the probability of losing accumulated SERP benefits in case of early employment termination should have a key impact on CEO’s risk-taking behavior. According to the nature of pension arrangements in place in a given year, CEOs can be classified into three groups: (1) NOSERP, which includes CEOs without SERP arrangements; (2) SERPSAL, which includes CEOs with SERP arrangements in which SERP benefits are determined based on salary only, and are therefore not contingent on CEO performance; and (3) 197 SERPBON, which includes CEOs with SERP arrangements in which SERP benefits are determined based on salary and bonus, and are therefore contingent on CEO performance. Suppose that the present value of expected CEO’s SERP benefits is v at the beginning of a given year. Hence, v = 0 for NOSERP, and v 0 for SERPSAL and SERPBON.1 Also, suppose that CEO performance in this year can be either good (denoted by superscript +) or poor (denoted by superscript –). Then, expected values of CEO SERP benefits at the end of the year are: E(v–)NOSERP = 0(1 – Ȇ) + 0Ȇ = 0 E(v+)NOSERP = 0 E(v–)SERPSAL = v(1 – Ȇ) + pvȆ = v – vȆ(1 – p) E(v+)SERPSAL = v E(v–)SERPBON = v(1 – Ȇ) + pvȆ = v – vȆ(1 – p) E(v+)SERPBON = v + bv where: Ȇ P B = = = (1a) (1b) (1c) (1d) (1e) (1f) probability of being fired or forced to retire due to poor performance percentage of SERP benefits received due to early retirement; 0 p < 1 percentage increase in the present value of SERP benefits due to good performance and increase in the bonus compensation As such, in a given year, accumulated SERP benefits of SERPSAL increase by 0 in case of good performance, and decrease by vȆ(1 – p) when the performance is poor:2 E(v)SERPSAL – vSERPSAL = [- vȆ(1 – p); 0] (2) Similarly, in a given year, accumulated SERP benefits of SERPBON increase by bv in case of good performance, and decrease by vȆ(1 – p) in case of poor performance: E(v)SERPBON – vSERPBON = [- vȆ(1 – p); bv] (3) Finally, regardless of the performance, the change in the value of accumulated SERP benefits of NOSERP is zero: E(v)NOSERP – vNOSERP = [0; 0] (4) According to Equations (2) and (3), when the performance is poor, the expected loss of the CEO whose SERP benefits are performance-contingent equals the expected loss of the CEO whose SERP benefits are based on the base salary only. However, the expected gain of the former CEO is higher (by the amount bv), when the performance is good. Ceteris paribus, higher reward incites risk-taking. Therefore, CEOs with performance-contingent SERPs are expected to be more risk-tolerant than CEOs whose SERPs are not contingent on performance.3 The following prediction is tested empirically: 1 v = 0 for SERPSAL and SERPBON at the beginning of Year 1 of executive’s employment. 2 For brevity, I ignore normal increases in the present value of SERP benefits - i.e. technical increases due to the accumulation of the additional year of credited service. The value of these technical increases is marginal; besides, their values are equal for SERPSAL and SERPBON (a certain proportion of v) and therefore do not affect most model predictions. For similar reasons, changes in base salary due to good (poor) performance are ignored. 3 The relationship between CEO’s risk-tolerance and the performance-contingency of CEO’s SERP is not expected to be endogenous. The reason is grounded in the design of existing SERP arrangements: in all cases, performancecontingent SERP benefits are calculated based on bonuses in addition to base salaries, not on bonuses instead of 198 H1: CEOs whose SERP benefits are contingent on firm performance are more risk-tolerant than CEOs whose SERP benefits are not contingent on firm performance. The comparison of CEOs whose SERP benefits are performance-contingent with CEOs who have no SERP arrangements is not that straightforward. According to Equation (4), the expected payoff (loss) is zero for NOSERP regardless of performance. According to Equation (3), the expected payoff of SERPBON is bv when the performance is good, while the expected loss is vȆ(1 – p) when the performance is poor. This means that no general prediction with respect to the relative risk-aversion of the two groups of CEOs can be made. On the one hand, contrary to CEOs with no SERP arrangements, CEOs with performance-contingent SERPs have something [vȆ(1 – p)] to lose in case of poor performance, and thus should be relatively more risk averse. On the other hand, contrary to CEOs with no SERP arrangements, CEOs with performance-contingent SERPs have something to gain (bv) in case of good performance, and thus are incited to take additional risk. Clearly, the risk-tolerance of CEOs with performance-contingent SERPs would differ according to values attached to bv and vȆ(1 – p). However, while it appears unfeasible to construct a justifiable general hypothesis that would compare CEOs with no SERP benefits and CEOs whose SERP benefits are contingent on performance, it is possible to make relative risk-aversion comparisons in a specific setting: during the final year prior to the expected CEO retirement, when the probability Ȇ of early termination approaches zero. The expected loss of all three groups of CEOs is equal (and equals zero) if the performance is poor in the final year as the CEO will retire at the end of that year anyway.4 At the same time, if the performance is good, CEOs with performance-contingent SERPs face higher expected payoffs (bv) than CEOs with no retirement arrangements and CEOs whose SERP benefits are determined based on the base salary only (zero). Since, ceteris paribus, higher reward incites risk-taking, the following hypothesis is examined: H2: In the final year prior to expected retirement, CEOs whose SERP benefits are contingent on firm performance are more risk-tolerant than CEOs with no SERP benefits. One would also expect CEO risk-tolerance to vary according to the size of already accumulated SERP benefits. However, the relationship is not that straightforward as in Sundaram and Yermack (2007), who hypothesize that a direct positive association between the two exists. Specifically, the association between risk preferences and the value of SERP should vary according to the performance-contingency of SERP benefits and the probability of early retirement, as showed by the examples that follow. Suppose, that the present value of SERP benefits of two CEOs at the beginning of a year is v1 and v2 respectively, such that v1 > v2. Also, suppose that SERP benefits of the two CEOs are determined based on salary only, and both CEOs face same Ȇ and p. According to Equation (2), expected annual changes in accumulated SERP benefits at the end of the year for the two CEOs are, respectively: E(v1)SERPSAL – v1SERPSAL = [- v1Ȇ(1 – p); 0] (5a) base salaries. Consequently, regardless of his risk preferences, a CEO would prefer a performance-contingent SERP in which pensionable earnings are based on salary and bonus to a SERP in which pensionable earnings are based on the salary only. SERP benefits are at least as high in the first case (if the performance is poor and the bonus is zero throughout the determination period) and likely to be higher (if the bonus is higher than zero in at least one determination year), but can never be lower. In other words, there is no downside effect or risk involved. As such, CEO’s initial risk preferences are not expected to impact the performance-contingency of SERP benefits. 4 The key point is that the expected loss of all CEOs is equal, and not that it equals zero. Certainly, the true expected loss due to the poor performance is higher than zero for all CEOs. For example, a CEO faces costs related to the loss of reputation. However, such losses are assumed to be homogeneous across the three groups of CEOs. 199 E(v2)SERPSAL – v2SERPSAL = [- v2Ȇ(1 – p); 0] (5b) The gain of the two CEOs when the performance is good is the same. However, since v1 > v2, the loss of the CEO with more accumulated SERP benefits is higher in case of poor performance than the loss of the CEO with less accumulated SERP benefits. Risk-aversion is therefore expected to be positively associated with the size of accumulated SERP benefits when SERP benefits are not contingent on performance. The only exception is the last year prior to CEO expected retirement when – as discussed above – the expected loss of both CEOs in case of poor performance is the same since Ȇ approaches zero. The following hypothesis is tested empirically: H3: Except in the final year prior to expected retirement, the risk-tolerance of CEOs whose SERP benefits are not contingent on firm performance is negatively associated with the size of accumulated SERP benefits. If SERP benefits are contingent on performance, the risk behavior of two CEOs with different values of accumulated SERP benefits is not that straightforward. Consider the abovementioned example for the two CEOs whose SERP benefits are performance-contingent. According to Equation (3), expected annual changes in accumulated SERP benefits at the end of the year for these two CEOs are, respectively: E(v1)SERPBON – v1SERPBON = [- v1Ȇ(1 – p); bv1] E(v2)SERPBON – v2SERPBON = [- v2Ȇ(1 – p); bv2] (6a) (6b) On one hand, the CEO with more accumulated SERP benefits loses more in case of poor performance, and thus is expected to be more risk-averse than the CEO with less accumulated SERP benefits. On the other hand, when the outcome is positive, the CEO with more accumulated SERP benefits gains more (bv1) than the CEO with less accumulated benefits (bv2), and thus is more incited to take risks. The risk-tolerance of the two CEOs would differ according to specific values attached to b, Ȇ and p. Consequently, justifying a general prediction about the impact of accumulated SERP benefits on CEO risk-tolerance when SERP benefits are performance-contingent appears problematic. However, during the final year prior to expected retirement, a CEO with accumulated SERP benefits v1 is expected to be more risk-tolerant than the CEO with v2. If the performance is poor in that year, both CEOs face the same consequences, due to Ȇ approaching zero. If the performance is good, the CEO with higher accumulated SERP benefits enjoys higher returns, and thus is incited to take additional risks. Therefore, the following hypothesis is examined: H4: In the final year prior to expected retirement, the risk-tolerance of CEOs whose SERP benefits are contingent on firm performance is positively associated with the size of accumulated SERP benefits. Table 1 summarizes research hypotheses on the relative association between CEO risk preferences and SERP benefits (i.e., Hypotheses 1 to 4). To the best of my knowledge, none of the hypotheses has been examined in the literature. Investigating the links between CEO SERP benefits on one side and CEO risk preferences on the other makes important analytical and empirical contributions to the stream of research on the impact of executive compensation on business practices. The only existing study on the impact of CEO SERP benefits on CEO risk preferences, Sundaram and Yermack (2007), finds no significant association between the two concepts. I extend the research by Sundaram and Yermack (2007) by developing an explicit analytical model to study the relationship between SERP benefits and risk preferences and relaxing some critical assumptions of SERP homogeneity. 200 Table 1 Summary of research hypotheses Comparison SERPBON SERPBON SERPSALHIGH SERPBONHIGH vs. vs. vs. vs. SERPSAL NOSERP SERPSALLOW SERPBONLOW Expected risk-tolerance relationship in a given year Years prior to Final year final year > > ? > < = ? > Hypothesis H1 H2 H3 H4 The table above summarizes research hypotheses on the association between CEO risk-tolerance and SERP benefits (Hypotheses 1 to 4). SERPBON are observations when CEO SERP benefits are contingent on firm performance. SERPSAL are observations when CEO SERP benefits are not contingent on performance. NOSERP are observations when CEO’s have no SERP arrangements. Subscripts HIGH and LOW refer to the size of accumulated SERP benefits. Methodology Econometric Models Four separate multivariate models that control for fixed firm- and CEO-level effects are estimated to investigate Hypotheses 1 to 4. The main limitation of the models is common to most empirical studies on risk preferences: it is difficult to estimate a reliable measure of risk tolerance. The study alleviates the problem by employing two alternative publicly-available proxies for CEO risk tolerance: firm’s distanceto-default, and firm’s capital and R&D expenditures. The distance-to-default is defined as the number of standard deviations of decline in a firm’s asset value that would push it into default.5 Capital and R&D expenditures are deflated by the beginning of period total assets to control for firm size.6 Separate multivariate regressions are run for each proxy. Since no measure represents a perfect estimation of CEO risk-tolerance, using several proxies that have been justified and employed in prior studies (e.g., Hamada, 1972; Bowman, 1979; Ross, 2004; Sundaram and Yermack, 2007) is beneficial for the reliability of results. Specifically, to test Hypothesis 1, the following model is estimated (firm and period subscripts are omitted for brevity): RISKTOL = Ȗ0 + į1BONSAL + Ȗ1SIZE + Ȗ2LEV + Ȗ3PERF + Ȗ4AGE + İ (7) where SIZE, LEV, and PERF are as identified previously and: = proxy for CEO’s risk-tolerance, either of the following variables: RISKTOL DTD: firm’s distance-to-default CAPEXP: firm’s capital and R&D expenditures deflated by lagged total assets = dummy variable equal to one if CEO’s SERP benefits are performance-contingent, BONSAL and zero if CEO’s SERP benefits are not contingent on performance = natural log of CEO’s age AGE 5 The default point and the distance-to-default are estimated using standard procedures found in the literature (among others, Crouhy et al., 2001; Sundaram and Yermack, 2007). 6 The results are not affected if log capital and R&D expenditures is used as a risk-tolerance proxy, instead of capital and R&D expenditures deflated by total assets. 201 Hypothesis 2 is tested using the following model: RISKTOL = Ȗ0 + į2BONNO + į3BONNO*LAST + Ȗ1SIZE + Ȗ2LEV + Ȗ3PERF + Ȗ4AGE + İ where: BONNO LAST (8) = dummy variable equal to one when CEO’s SERP benefits are performancecontingent, and zero when CEO has no SERP benefits = dummy variable equal to one if the year is the last year prior to retirement, zero otherwise Hypothesis 3 is examined using the following model: RISKTOL = Ȗ0 + į4PVSAL + į5PVSAL*NOTLAST + Ȗ1SIZE + Ȗ2LEV + Ȗ3PERF + Ȗ4AGE + İ where: PVSAL NOTLAST (9) = present value of accumulated SERP benefits of a CEO whose SERP is not contingent on performance = dummy variable equal to one if the year is not the last year prior to retirement, zero otherwise Finally, Hypothesis 4 is tested by running the following regression: RISKTOL = Ȗ0 + į6PVBON + į7PVBON*LAST + Ȗ1SIZE + Ȗ2LEV + Ȗ3PERF + Ȗ4AGE + İ where: PVBON = (10) present value of accumulated SERP benefits of a CEO whose SERP is performancecontingent If predicted associations between SERP benefits and CEO risk preferences are supported, coefficients on į1 (Hypothesis 1), į3 (Hypothesis 2), and į7 (Hypothesis 4) are expected to be positive and significant, while the coefficient on į5 (Hypothesis 3) is expected to be negative and significant. Regressions are controlled for common firm-level factors that potentially impact managerial propensity to undertake risky projects: size, leverage, and past performance (proxied, respectively, by log total assets, debt-assets ratio, and ROA).7 Specifically, the influence of firm size on capital and R&D spending is well-documented in the literature. Larger firms are expected to have greater resources to exploit innovations and develop sustained R&D programs (Schumpeter, 1942). A number of empirical studies support the prediction by confirming a positive association between capital and R&D expenditures with firm size (e.g., Baysinger and Hoskisson, 1989; Baysinger et al., 1991). In contrast, the relationship between capital and R&D expenditures and firm leverage is expected to be negative. High leverage prioritizes current cash-flows for debt service, thereby discouraging managers from investments into riskier long-term projects. In general, empirical studies support the negative association between firm’s leverage and capital and R&D expenditures (e.g., Long and Ravenscraft, 1993; Barker and Mueller, 2002), however this relationship is not always statistically or economically significant (Hitt et al., 1991). Finally, the evidence on the association between past performance and capital and R&D expenditures is somewhat mixed. Cyert and March (1963) suggest that poor past performance incites experimenting with innovative activities. Hitt et al. (1991) support this perspective empirically by confirming a negative association between past financial performance and R&D spending. The majority of later studies, however, find a positive relationship between the two variables (e.g., Hundley et al., 1996; Barker and Mueller, 2002). A possible 7 Using alternative proxies for firm size (log assets, revenue, log revenue), leverage (debt-equity ratio), and accounting performance (net income, ROE) does not qualitatively affect the results. 202 explanation is that past profitability justifies managerial actions, gives managers confidence and encourages undertaking even riskier long-term projects. The age of the CEO is included in regressions as another control variable. However, the evidence on the link between CEO’s age and risk-tolerance proxies is mixed. Some prior research suggests that younger CEOs are concerned about being disciplined by the managerial labor market in case of poor results, whereas for CEOs approaching retirement such career concerns are less relevant (e.g., Fama, 1980; Gibbons and Murphy, 1992). In addition, older CEOs are likely to possess more personal wealth that younger CEOs and may therefore be less risk-averse (Lewellen et al., 1987). However, CEOs approaching retirement may prioritize short-term performance and reduce capital and R&D expenditures (Dechow and Sloan, 1991). According to Lundstrum (2002) and Barker and Mueller (2002), CEO age has a negative association with R&D spending. Data The sample of 60 firms that comprised the S&P/TSX60 index in 1997 is used for the analysis.8 The S&P/TSX60 Index is comprised of Canada's largest publicly traded firms listed on the Toronto Stock Exchange and more than 60% of which are cross-listed in the USA. The proportion of CEO SERP benefits to cash compensation in larger firms is expected to be more pronounced due to a greater disproportion between pensionable earnings and post-retirement income from the RPP. Since any form of executive compensation is positively associated with firm size (among others, Lambert et al., 1991; Core et al., 1999; Craighead et al., 2004; Kalyta and Magnan, 2008), CEOs in larger firms are better remunerated. Higher pre-retirement cash compensation translates into greater disproportion between pensionable earnings and RPP benefits and – therefore – a greater role of a supplemental pension plan in preserving pre-retirement cash-inflows for a CEO.9 Since supplemental retirement plans are most important for better-compensated CEOs, associations of SERP benefits with CEO risk preferences are expected to be especially pronounced in larger firms. The sample encompasses the seven-year period between 1997 and 2003 and includes 60 firms, 116 CEOs, and 395 observations. Descriptions of SERPs are retrieved directly from annual proxy statements. When CEO’s age is not disclosed in the proxy statement, it is retrieved via Lexis-Nexis databases and Internet search engines. Death probability tables are retrieved from the Statistics Canada publications. Financial data is collected from Compustat. Missing observations are retrieved from Report on Business Top 1000 publications and corporate financial statements. To account for inflation, all monetary values are converted into 2003 dollars using historic CPIs. SERP Estimation Following prior studies (Masson, 1971; Sundaram and Yermack, 2007; Kalyta and Magnan, 2008), I use a two-step procedure to estimate the present value of accumulated retirement benefits. In the first step, the value of the annual pension already accumulated by the CEO is calculated. SERP formula, multiplier and years of pensionable service accumulated are disclosed in the proxy statement and do not have to be estimated. Pensionable earnings are typically not disclosed. Instead, a firm discloses information on 8 See Kalyta and Magnan (2008) for the list of firms. 9 Consider two retiring CEOs with SERPs, with 35 years of credited service, whose annual pension is to be determined by the product of pre-retirement cash compensation, years of credited service and a multiplier of 2%. Pre-retirement cash compensation of CEO A is $200,000. Pre-retirement cash compensation of CEO B is $1,000,000. In that case, CEO A is entitled to annual pension of $140,000, of which 52.8% ($73,885) is from the RPP, and 47.2% ($140,000 - $73,885) is from the SERP. CEO B is entitled to annual pension of $700,000, of which only 10.6% ($73,885) is from the RPP, and 89.4% is from the SERP. 203 components of pensionable earnings (i.e., base salary or some combination of base salary and other incentives) and the period over which pensionable earnings should be estimated (i.e., annual pensionable earnings immediately prior to retirement or the average of highest annual pensionable earnings over a longer period). Thus, in most cases, to estimate pensionable earnings in year t, the information on CEO’s compensation in prior years must be collected from earlier proxy statements. For any given CEO at any given year t, compensation data for at least three prior years, t–1 to t–3, is available.10 For any prior year in which the information on CEO compensation is unavailable, I assume that the salary is equal to the salary in the earliest year for which the compensation information is available, while the bonus is equal to the average bonus in years for which the compensation information is available.11 In the second step, the annual pension already accumulated by the CEO and death probability tables published by Statistics Canada are used to estimate the present value of the SERP. For simplicity, it is assumed that pension benefits are paid annually at the end of a year. When the age of the CEO is not disclosed in the proxy statement, it is retrieved via the Blue Book of Canadian Business, Who’s Who in Canadian Business, LexisNexis or Google. Any survivorship benefits are conservatively ignored as I am unable to gather information on the marital status of CEOs and the age of their spouses. I assume that current CEOs with SERPs will retire upon reaching the age at which they qualify for unreduced SERP benefits, which in most cases is the age of normal retirement specified in proxy statements. In one case, in which the CEO has already reached the age of normal retirement but continues serving in his position in 2003 (the last year in the sample), a retirement in 2004 is assumed. Results Univariate Analysis The univariate analysis indicates that three of the four hypotheses on the association between CEO’s risk preferences and SERP benefits are supported.12 Table 2 compares mean distances-to-default and capital and R&D expenditures across observations of interest. The only hypothesized relationship rejected by univariate analysis is the predicted link between the value of performance-contingent SERP benefits and risk preferences in the last year prior to CEO’s retirement (Hypothesis 4). The comparison between SERPBON observations with the highest value of accumulated SERP benefits (top 50%) and SERPBON observations with the lowest value of accumulated SERP benefits (bottom 50%) displays no significant difference in either risk-tolerance proxy. All other results provide evidence in favor of predicted links. On average, CEOs whose SERP benefits are performance-contingent display higher risk tolerance than CEOs whose SERP benefits are not contingent on performance (Hypothesis 1). Mean DTD and CAPEXP are significantly higher in the former subsample. Similarly, CEOs with performance contingent SERPs appear to be more risk-tolerant 10 The first year of observations is 1997 but I also have access the 1996 proxy statements. Since a proxy statement contains information on executive’s compensation during last three years, for any 1997 observation there are three years of prior remuneration data. 11 In two cases in the sample, firms reported the projected value of CEO’s annual pension. However, as actuarial and other assumptions were never made transparent, I ignored that information and estimated the value of the annual pension already accumulated by the CEO in the manner described in the text. This ensures uniformity of assumptions across all observations. In addition, a robustness test shows that using the reported values of projected CEO’s annual pensions in these two cases does not affect any results qualitatively. 12 Descriptive statistics and correlation coefficients are not reported due to brevity and value-added considerations. The reader is referred to Kalyta and Magnan (2008) for detailed information. 204 than CEOs with no supplemental retirement arrangements in the last year prior to retirement. Both proxies for risk-tolerance display the expected relationship. The difference in capital and R&D expenditures is especially pronounced: $841.9 million vs. $444.7 million. Finally, Hypothesis 3 is also supported by univariate results. Risk-tolerance appears to be associated with the size of SERP benefits that are not contingent on performance in the years preceding the last year prior to CEO’s retirement. The comparison between SERPSAL observations with the highest value of accumulated SERP benefits (top 50%) and SERPSAL observations with the lowest value of accumulated SERP benefits (bottom 50%) reveals that the distance-to-default in the later group is on average higher. Capital expenditures are marginally higher in the bottom 50% observations as well. Table 2 Risk tolerance by CEO’s SERP structure Hypothesis Period Observations H1 NOTLAST and LAST H2 LAST H3 NOTLAST H4 LAST SERPBON SERPSAL SERPBON NOSERP SERPSALUPPER SERPSALLOWER SERPBONUPPER SERPBONLOWER Predicted Sign > > < > DTD 2.94 2.35 ** 2.80 2.12 ** 2.02 2.76 *** 2.97 2.68 CAPEXP 939.4 745.4 ** 841.9 444.7 *** 705.8 855.0 * 865.1 814.9 *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level The table reports means for observations grouped according to CEO SERP structure in the sample of TSX/S&P60 firms for the 1999-2003 period. Significance levels are for one-tailed t-tests for differences in means. Monetary values are in 2003 dollars. DTD is firm’s distance-todefault. CAPEXP is firm’s capital and R&D expenditures. NOTLAST are observations that correspond to years that precede the final year prior to CEO’s termination. LAST are observations that correspond to the final year prior to CEO’s termination. SERPBON are observations that correspond to CEO’s with performance-contingent SERPs. SERPSAL are observations that correspond to CEO’s with SERPs that are not contingent on performance. NOSERP are observations that correspond to CEO’s with no SERP. Subscripts UPPER and LOWER correspond to top (bottom) 50% of observations divided according to the value of accumulated SERP benefits. Multivariate Analysis Four separate models (Equations 7 to 10) are run to test the hypotheses concerning associations between risk-tolerance proxies and CEO’s SERP benefits. The results of the regressions are presented in Table 3. The adjusted R-squared values range from 0.198 to 0.454 for models in which the independent variable is firm’s capital expenditures, and from 0.360 to 0.545 for models in which the independent variable is firm’s distance to default. Most control variables display expected relationships with dependent variables. Risk-tolerance proxies are negatively associated with firm’s leverage and CEO’s age, and positively – with size. The association with past performance is positive as well - a result in line with the proposition that profitability justifies past actions, gives managers confidence and encourages undertaking riskier projects. In general, the relationships hold across the models. The test of Hypothesis 1 reveals that the explanatory variable BONSAL which partitions the sample of CEOs with SERP according to performance-contingency of their SERP benefits has positive and significant associations with distance-to-default (0.240; p<0.05) and capital and R&D expenditures (0.020; p<0.05). As such, CEOs whose SERP benefits are contingent on performance appear to be more risk-tolerant than CEOs whose SERP benefits are not contingent on performance. A broader implication 205 Table 3 Determinants of CEO risk tolerance Variable SIZE LEV PERF AGE BONSAL BONNO BONNO*LAST PVSAL PVSAL*NOTLAST PVBON PVBON*LAST Adjusted R2 H1 DTD 0.249 -0.498 0.233 -0.785 0.240 ** *** * * ** CAPEXP 0.021 *** -0.012 0.073 *** -0.113 *** 0.020 ** H2 DTD CAPEXP 0.447 * 0.009 *** -0.581 *** -0.007 0.407 * 0.041 ** -1.003 -0.026 0.026 0.212 ** H3 DTD 0.163 -0.694 0.590 -0.816 * *** * * 0.371 0.360 DTD 0.308 ** -0.470 ** 0.501 * -0.763 CAPEXP 0.015 *** 0.013 0.026 * -0.091 ** -0.012 0.008 ** 0.097 -0.328 *** 0.463 H4 CAPEXP 0.025 *** -0.016 0.463 *** -0.123 ** 0.198 0.545 -0.002 -0.011 * 0.454 0.085 0.126 * 0.529 0.001 -0.001 0.396 *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level The table displays parameter estimates of the determinants of risk tolerance proxies for the sample of TSX/S&P60 firms for the 1999-2003 period. Models are estimated using OLS with robust HuberWhite standard errors. Dependent variables are firm’s distance-to-default (DTD) and capital and R&D expenditures deflated by lagged total assets (CAPEXP). SIZE is the natural log of total assets, LEV is the ratio of debt to total assets, PERF is the return on assets, AGE is the natural log of CEO’s age. BONSAL is a dummy variable equal to one for observations when CEO’s SERP is performance-contingent, and zero for observations when CEO’s SERP is not contingent on performance. BONNO is a dummy variable equal to one for observations when CEO’s SERP is performance-contingent, and zero for observations when CEO’s have no SERP. PVSAL is the value of accumulated SERP benefits of the CEO whose SERP benefits are not contingent on performance. PVBON is the value of accumulated SERP benefits of the CEO whose SERP benefits are performance-contingent. LAST are observations that correspond to the final year prior to CEO’s termination. NOTLAST are observations that correspond to years that precede the final year prior to CEO’s termination. To mitigate any influence from outliers all variables are winsorized at the 1% level. 206 is that SERPs are indeed not homogenous and their structure significantly impacts risk preferences The results also provide empirical support for Hypothesis 2. Coefficient on the BONNO*LAST interaction is positive and significant when either risk-tolerance proxy is used. In isolation, the explanatory variable BONNO displays no significant association with either distance-to-default or capital and R&D expenditures. In other words, CEOs whose SERP benefits are contingent on performance appear to be more risk-tolerant than CEOs with no SERP benefits, but only in the final year prior to expected retirement. Similarly, the analysis provides with evidence in favor of Hypothesis 3. As predicted, the coefficient on the PVSAL*NOTLAST interaction is negative. The association is weakly significant (-0.011; p<0.10) when capital and R&D expenditures are used as the dependent variable, but strongly significant (-0.328; p<0.01) when distance-to-default is used as a proxy for risk preferences. When the time frame is ignored, the association between accumulated SERP benefits and risk-tolerance proxies is marginally significant in one model and not significant in the other. As such, the risk-tolerance of CEOs whose SERP benefits are not contingent on performance appears to be negatively associated with the size of accumulated SERP benefits specifically in the years preceding the final year prior to CEO’s retirement, but not in the final year. In other words, on average, a CEO with higher accumulated SERP benefits is more risk-averse than a CEO with lower accumulated SERP benefits, when SERPs are not contingent on performance and when CEOs are not about to retire. Finally, the support of Hypothesis 4 is limited. The association between capital and R&D expenditures on one side and the interaction of interest (PVBON*LAST) on the other is statistically insignificant. When distance-to-default is used as an alternative risk-tolerance proxy, the association is only weakly significant, albeit positive, as predicted (0.126; p<0.10). As such, the results fail to support the proposition that the risk-tolerance of CEOs whose SERP benefits are contingent on performance is positively associated with the size of accumulated SERP benefits in the last year prior to CEO’s retirement. In years other than the last year, the size of accumulated performance-contingent SERP benefits does not appear to be linked to risk-tolerance proxies as well: coefficients on PVBON are statistically insignificant. To summarize, the results support three of the four hypotheses on links between SERP benefits and risk preferences. First, CEOs whose SERP benefits are contingent on performance appear to be more risktolerant than CEOs whose SERP benefits are not contingent on performance. Second, CEOs whose SERP benefits are contingent on performance are more risk-tolerant than CEOs with no SERP benefits but only in the last year prior to retirement. Otherwise, no significant difference in risk preferences of the two groups exists. Third, risk-tolerance is negatively associated with the size of accumulated SERP benefits when SERP benefits are not contingent on performance. The only exception is the last year prior to CEO’s retirement when the size of SERP benefits does not affect risk preferences. In general, the results provide a strong support to the assertion that the relationship between CEO’s SERP and risk preferences exists, it is not homogenous and varies according to the performance-contingency of SERP benefits. Research Limitations The study is subject to several limitations that are not expected to have substantial impact on the results. First, SERP estimation methodology does not account for survivor benefits, which may or may not be significant depending on CEO’s marital status and the age of the spouse. Clearly, ceteris paribus, the expected value of a SERP of a single CEO would be substantially lower than the expected value of a SERP of a married CEO whose spouse is 30 years younger. However, being unable to find out the marital status of CEOs using publicly available sources, I assume – conservatively – no survivor benefits. Another possibility is to assume that all CEOs are married to persons of opposite sex and that spouses are 207 of the same age as CEOs. As expected, redoing the analysis using this assumption yields very similar – univariate and multivariate – results. According to most SERPs, survivor benefits are limited to about 60% of executive’s pension and the life expectancy of females at age 60 is longer than that of males by about four years only. As a result, the actuarial value of the incremental pension benefit due to survivor benefits is insignificant when the abovementioned assumption is used. In addition, survivor benefits would be partially cancelled out if the probability of divorce was also accounted for. Including the dummy variable for financial institutions in all models does not qualitatively affect any results. However, the lack of control for industry effects is a potential limitation of the study and an interesting opportunity for future research. For example, it is possible that the impact of SERP benefits on CEO risk tolerance differs across sectors (e.g., due to heterogeneous demand for and supply of qualified CEOs and hence different probability of early employment termination). In addition, the results of this study should not be generalized to the population of public firms, as the analysis concentrates on larger firms. SERP benefits are expected to be more prevalent and sizable in larger firms due to a greater disproportion between CEO’s pre-retirement cash pay and regular pension. Kalyta and Magnan (2008) confirm that the presence and magnitude of SERP benefits is related – among other factors – to firm size. Also, caution should be exercised when extending results to firm’s officers other than the CEO. The impact of SERP benefits of lower level officers on firm’s riskiness may not be as strong as the impact of CEO SERP benefits. A study on SERPs of lower-level officers would constitute an interesting research extension. Conclusions The study investigates the impact of CEO SERP benefits on CEO’s risk preferences. Existing evidence in this area is limited. Prior literature investigates the impact of SERP’s value on risk preferences but suffers from important conceptual and methodological limitations and fails to capture statistically significant associations (Sundaram and Yermack, 2007). I develop an explicit analytical model according to which the relationship between CEO’s risk-tolerance and CEO’s SERP benefits varies according to the performance-contingency of SERP benefits and the probability of early employment termination. To verify the predicted relationships, four hypotheses are tested empirically. Controlling for commonly-used explanatory factors, proxies for risk-tolerance are regressed on variables that partition the sample according to the nature of CEO’s retirement arrangements and the period with respect to CEO’s retirement date. Since it is difficult to estimate the reliable measure of CEO’s risk tolerance (the main limitation of most empirical studies on risk preferences), two alternative and conceptually different proxies are used to alleviate the problem: firm’s distance-to-default and capital and R&D expenditures deflated by total assets. Higher values in the dependent variables correspond to more risk tolerance and lower values correspond to more risk aversion. All empirical models are run separately for each proxy. In general, the results confirm analytical predictions. Specifically, CEOs whose SERP benefits are contingent on performance appear to be more risk-tolerant than CEOs whose SERP benefits are not contingent on performance. Also, CEO’s risk tolerance is negatively associated with the size of accumulated SERP benefits when SERP benefits are not contingent on performance. To summarize, empirical findings suggest that the relationship between CEO SERP benefits and risk preferences does exist and varies according to the performance-contingency of SERP benefits. The result is important in light of the study by Sundaram and Yermack (2007) that does not take into account the heterogeneity of SERPs with respect to their performance-contingency and finds no association of CEO SERP benefits and risk preferences. More generally, the result provides important analytical and empirical contribution to the academic literature on associations between executive compensation arrangements and firm’s decisions. 208 References Barker, V., & Mueller, G. (2002). CEO characteristics and firm R&D spending. Management Science, 48(6), 782-801. Baysinger, B., & Hoskisson, R. (1989). Diversification strategy and R&D intensity in multiproduct firms. Academy of Management Journal, 32(2), 310-332. Baysinger, B., Kosnik, R., & Turk, T. (1991). Effects of board and ownership structure on corporate R&D strategy. Academy of Management Journal, 34(1), 205-214. Bebchuk, L., & Fried, J. (2004). Stealth compensation via retirement benefits. Berkeley Business Law Journal, 1(2), 294-326. Bowman, R. (1979). The theoretical relationship between systematic risk and financial (accounting) variables. The Journal of Finance, 34(3), 617-638. Brick, I., Palmon, O., & Wald, J. (2006). CEO compensation, director compensation and firm performance: Evidence of cronyism? Journal of Corporate Finance, 12(3), 403–423. Core, J., Holthausen, R., & Larcker, D. (1999). Corporate governance, Chief Executive Officer compensation, and firm performance. Journal of Financial Economics, 51(3), 371-406. Craighead, J., Magnan, M., & Thorne, L. (2004). The impact of mandated disclosure on performancebased CEO compensation. Contemporary Accounting Research, 21(2), 369-398. Crouhy, M., Galai, D., & Mark, R. (2001). Risk management. New York, NY: McGraw-Hill. Cyert, R., Kang, S., & Kumar, P. (2002). Corporate governance, takeovers, and top-management compensation theory and evidence. Management Science, 48(4), 453-469. Dechow, P., & Sloan, R. (1991). Executive incentives and the horizon problem: An empirical investigation. Journal of Accounting and Economics, 14(1), 51-89. Fama, E. (1980). Agency problems and the theory of the firm. Journal of Political Economy, 88(2), 288307. Gibbons, R., & Murphy, K. (1992). Does executive compensation affect investment? Journal of Applied Corporate Finance, 5(2), 99-109. Hamada, R. (1972). The effect of the firm’s capital structure on the systematic risk of common stocks. The Journal of Finance, 27(2), 435-452. Hitt, M., Hoskisson, R., Ireland, R., & Harrison, J. (1991). Effects of acquisitions on R&D inputs and outputs. Academy of Management Journal, 34(3), 693-706. Hundley, G., Jacobson, C., & Park, S. (1996). Effects of probability and liquidity on R&D intensity: Japanese and U.S. companies compared. Academy of Management Journal, 39(6), 1659-1674. Kalyta, P., & Magnan, M. (2008). Executive pensions, disclosure quality, and rent extraction. Journal of Accounting and Public Policy, 27(2), 133-166. 209 Lambert, R., Larcker, D., & Weigelt, K. (1991). How sensitive is executive compensation to organizational size? Strategic Management Journal, 12(5), 395-402. Lewellen, W., Loderer, C., & Martin, K. (1987). Executive compensation and executive incentive problems: An empirical analysis. Journal of Accounting and Economics, 9 (3), 287-310. Long, W., & Ravenscraft, D. (1993). LBOs, debt and R&D intensity. Strategic Management Journal, 14(Special Issue), 119-135. Lundstrum, L. (2002). Corporate investment myopia: A horserace of the theories. Journal of Corporate Finance, 8(4), 353-371. Masson, R. (1971). Executive motivations, earnings and consequent equity performance. Journal of Political Economy, 79(6), 1278-1292. Murphy, K. (1999). Executive compensation. In O. Ashenfelter & D. Card (Eds.), Handbook of Labor Economics (pp. 2485-2563). New York, NY: North Holland. Ross, S. (2004). Compensation, incentives, and the duality of risk aversion and riskiness. The Journal of Finance, 59(1), 207-225. Schumpeter, J. (1942). Capitalism, socialism and democracy. New York, NY: Harper and Row. Sundaram, R., & Yermack, D. (2007). Pay me later: Inside debt and its role in managerial compensation. The Journal of Finance, 62(4), 1551-1588. 210 ASAC 2008-01-31 Halifax, Nova Scotia Lawrence Kryzanowski Sana Mohsni (student) Department of Finance, John Molson School of Business, Concordia University CAPITAL COSTS FOR DOMESTIC AND CROSS-LISTED CANADIAN FIRMS 1 The internal rates of return are estimated for domestic and cross-listed Canadian firms and nine GICS sectors. Results show that, similar to the U.S., the Canadian corporate sector creates value. As expected, crosslisted firms enjoy, on average, lower costs of capital and equity than domestic firms. Although the cost of capital varies considerably across GICS sectors, value is created almost equally by all GICS sectors. Introduction The cost of capital and its components (such as equity) are among the most important numbers in financial economics. An accurate measure of capital costs is necessary for financial managers and investors to make optimal asset allocation (investment) decisions. A vast literature in finance deals with the estimation of the cost of capital/equity and the equity risk premium. While traditional studies may use asset-pricing models such as the CAPM and APT to provide such estimates, most of these asset-pricing approaches provide insufficient guidance on how to estimate the cost of equity or the market risk premium. Also, empirical investigations find that the equity risk premium on the US stock market over the last century has been considerably higher than predicted by standard equilibrium models (see Kocherlakota (1996), Cochrane (1997) and Siegel and Thaler (1997) for summaries). Due to the somewhat disappointing results suggested by equilibrium asset pricing models, a recent thrust in the literature focuses on ex-ante and/or fundamental information such as earnings or dividends or analysts’ earnings forecasts and variations of the Discounted Cash Flows (DCF) to estimate the cost of capital and/or the cost of equity.2 These models usually rely on fewer assumptions than the standard equilibrium models and provide estimates which may be more consistent with firm and market fundamentals. The cost of capital is estimated as the internal rate of return (IRR) that equates current stock prices to future cash-flows. Although analysts’ forecasts of earnings are usually used to estimate the exante cost of capital (Gebhardt, Lee and Swaminathan, 2001), this method requires analysts’ forecasts for all firms if the aggregate cost of capital is to be estimated. It also requires the estimation of terminal values and long-term (LT) earnings growth rates. Fama and French (1999) use an ex post approach which 1 Financial support from the Concordia University Research Chair in Finance, SSHRC and SSQRC-CIRPÉE are gratefully acknowledged. Please do not quote without the authors’ permission. 2 Examples include Blanchard (1993), Claus and Thomas (2001), Gebhardt, Lee and Swaminathan (2001), Fama and French (1999, 2002), Easton and Monahan (2003), Gode and Monahan (2003) and Botosan and Plumlee (2005). 211 is less fraught with measurement errors, does not rely on any estimated variables, and uses readily available financial statement and market data to infer the cost of capital for the nonfinancial corporate sector. In this paper, we use the IRR model of Fama and French (1999) to estimate the costs of capital and equity for domestic and cross-listed Canadian firms as well as 9 GICS sectors from 1960 to 2003. An advantage of the IRR methodology is that it provides an independent estimate of the aggregate cost of capital/equity that can be compared to historical or realized values, or those emitted by the various asset pricing models such as the CAPM. At a minimum, this helps in understanding the likely ranges for historical risk premia and how to relate that range to the expected risk premia. It also provides an assessment of investors’ returns from investing in all corporate securities. When applied to individual firms or industries, the estimated IRRs allow investors to better match their return requirements with the past profitability of these individual firms or industries. An interesting application of the IRR approach is that it allows for the computation of two types of IRRs. These are the IRR on value which is the return required by investors, and the IRR on cost which is the return delivered by the corporate sector. A comparison of the two values provides evidence on whether corporate investment adds value on average. This paper makes a number of contributions to the literature. First, similar to Fama and French (1999), we provide accurate estimates of the historical IRRs, specifically IRR on value for the Canadian nonfinancial corporate sector, and produce interesting insights regarding initial pricing of firms, postentry investments, earnings on investment and terminal values at the aggregate level. Our results suggest that corporate investments produced an average return that exceeded the cost of capital. Over the 19602003 sample period, the IRR on value or the real return on value for the nonfinancial corporate sector is 5.90 percent which is lower than the IRR on cost or the real return on cost which is 6.58 percent. The IRR on value [cost] for equity over the period 1960-2003 is 6.58 [7.78] percent which suggests that equitybased investments add higher value to the corporate sector. This is due only in part to the use of the historical book-values of assets when computing the IRR on cost. To mitigate this effect, we adjust bookvalues for replacement costs which reduce the IRR on cost for equity to 7.51 percent. The IRR on value for equity can also be interpreted as a measure of the equity risk premium in the non-financial Canadian corporate sector. Second, our findings shed light on the difference between rates of return and cost of capital estimates using market models with those based on fundamental values. We find that sectors such as IT suffer from a huge discrepancy when fundamental instead of market values are used to estimate their profitabilities and costs of capital. Third, we show how the costs of capital as estimated using the IRR on value and IRR on cost differ between domestic and cross-listed firms. We also explicitly examine the impact of cross-listing on the cost of equity for Canadian firms without assuming any market model.3 Our results confirm that cross-listing reduces both the cost of equity and the cost of capital for the cross-listed firms. Fourth, by applying the IRR procedure to 9 GICS sectors, we assess the historical performance and profitability across Canadian industrial sectors. By examining how the IRR on value and on cost differ by sectors, we are able to determine which sectors provide the highest [lowest] value added.4 Fifth, by correcting the IRR on cost estimates for the replacement costs of assets, we confirm that investments in the corporate non-financial sector still produce, on average, a return which exceeds their cost of capital. 3 Previous work such as Jorion and Schwartz (1986), Alexander et al. (1987, 1988), Foerster and Karolyi (1993) and Errunza and Miller (2000) study the impact of cross-listing on the cost of equity but assume a market model to infer the cost of capital. Recently, Hail and Leuz (2005) use forward looking residual income models to examine the costs of equity for cross-listed firms but do not include Canadian companies. 4 The GICS (Global Industry Classification Standard) classification was jointly developed by Morgan Stanley Capital International (MSCI) and Standard and Poor’s (S&P). Since its inception in 2000, the GICS is widely used in sector-based investing in global financial markets. 212 We infer that replacement cost adjustments, which reduce the IRR on cost, do not alter the positive value creation conclusion for the aggregate market and industries. Since some GICS sectors may experience large and persistent extraordinary items, we test the sensitivity of our IRR estimates to the inclusion of such items. This leads to somewhat lower IRR values, which indicates that the extraordinary items are, on average, negative across the studied sample period. The remainder of the paper is organized as follows. The methodology is presented in the next section. The sample and data collection are discussed in the third section. Section 4 presents the empirical results when using the IRR approach to measure the cost of capital. Section 5 presents the empirical results when replacement cost adjustments and extraordinary items are considered. Section 6 concludes the paper. Methodology The internal rate of return is defined as the discount rate that equates the aggregate market value (or book value) to the future cash flows generated by the non-financial Canadian corporate sector for the years 1960-2003. The cost of capital as computed herein describes the actual cost that the nonfinancial corporate sector bears to make use of investors’ funds. Initially, we make the assumption that the market value of equity is a good proxy of the intrinsic value of the firm for shareholders. Since the estimations use realized and not forecasted values, this approach rests on the validity of assuming that realized and expected returns are equal, on average, over longer periods of time. The specific model used to compute the cost of capital (or equity) is as follows: T ∑ IV0 = t =1 T ∑ ICo = t =1 T IVeq ,o = ∑ T t =1 IC0 ICeq,0 Xt Xeq,t It Ieq,t FSt X t − It (1 + r ) t + c , eq (1 + r ) (1 + r ) t =1 +∑ t + c (1 + rv,eq ) T +∑ t =1 TVT (1 + r ) (1) T TVT (1 + r ) (2) T c FSeq ,t − FBVeq ,t t =1 t + v FS t − FBCt ∑ T t v T + X eq ,t − I eq ,t (1 + r ) ∑ t =1 c t FS t − FBVt T v (1 + rv,eq ) ICeq ,o = ∑ IVeq,0 (1 + r ) t X eq ,t − I eq ,t t =1 where, IV0 X t − It t + FSeq ,t − FBCeq ,t (1 + r ) c ,eq t TVeq ,T (3) (1 + rv,eq ) T + TVeq ,T (1 + r ) T (4) c ,eq is the aggregate initial market value of firms that enter the sample at the beginning of the IRR estimation period; is the aggregate initial market value of equity of firms that enter the sample at the beginning of the IRR estimation period; is the aggregate initial total book value of the firms that enter the sample at the beginning of the IRR estimation period; is the aggregate initial total book value of equity of firms that enter the sample at the beginning of the IRR estimation period; is aggregate cash earnings for the firms in year t; is aggregate cash earnings for shareholders in year t; is the aggregate gross investment by the firms in year t; is the aggregate gross investment by shareholders in year t; is the terminal market value of firms that leave the sample in year t; 213 FSeq,t FBVt FBVeq,t FBCt FBCeq,t TVT TVeq,T is the terminal market value of equity of firms that leave the sample in year t; is the initial market value of firms that enter the sample in year t; is the initial market value of equity of firms that enter the sample in year t; is the book value of firms bought at cost during year t; is the book value of equity of firms bought at cost during year t; is the terminal market total (equity) value of firms that exist at the end of the sample period (2003 herein); and is the terminal market value of equity of firms that exist at the end of the sample period (2003 herein). Equation (1) computes the IRR on value, rv , which represents an estimate of the overall cost of capital on the whole nonfinancial corporate sector, and is computed using the market values of the firms in the sample (initial, entering and leaving) and the value of any other inflows or outflows of cash. The IRR on value is an indicator of the profitability of the public corporate sector and to a lesser extent the profitability of the economy as a whole. Equation (2) computes the IRR on cost, rc , which represents the actual IRR generated by the investment in the corporate sector and is computed using the book value of the sample at the initial period and upon subsequent entrance in the sample, as well as the market value of leaving firms and the value aggregate cash-earnings net of investments costs. Equation (3) computes what we call the IRR on equity, rv,eq,, which represents the cost of equity invested in the whole nonfinancial corporate sector or the return realized on equity, or the market equity risk premium. This IRR is computed using the market value of the sample equity at the initial period, the market value of equity of firms entering or exiting the market over the period, and any other inflows and outflows of cash. Equation (4) computes what we call the IRR on the cost of equity, rc,eq, which represents the actual return on equity that the nonfinancial corporate sector realizes if equity is assessed at their book values. Equations (1), (2), (3) and (4) are similar to the standard IRR expression for an investment project, although we are using slightly different notations. X t − I t is the aggregate annual cash-flow net of the aggregate annual gross investment, FSt − FBVt is the cash inflow from firms sold or being liquidated net of cash outflows for firms bought or which have just entered the market. FSt − FBCt is the cash inflow from firms sold or being liquidated net of the book value of firms bought or which have just entered the market. TVT is the terminal market value of firms at the end of the sample period and firms which leave the market at the end of 2003. When computing the IRR on equity, we restrict the measures to book and market value of equity. Therefore, X eq ,t − I eq ,t is the aggregate annual shareholders’ cashflow net of aggregate annual shareholders’ gross investment.5 FSeq ,t − FBVeq ,t is the market value of equity of firms sold or being liquidated net of the market value of equity of firms bought or which have just entered the market. FSeq ,t − FBCeq ,t is the market value of equity of firms sold or being liquidated net of the book value of equity of firms bought or which have just entered the market. TVT ,eq is the terminal market value of equity for firms that exist at the end of the sample period (2003 herein). Sample, Data Collection and Evolution in Number of Firms The sample consists of all publicly traded nonfinancial firms listed on the Toronto Stock Exchange (TSX) during the period of 1960 to 2003 and for which data are available on the Financial Post 5 We use net changes in book value of equity plus depreciation to estimate the value of annual gross investment of shareholders, It,, and earnings after taxes plus depreciation to estimate the shareholders cash-flow, Xt. 214 (FP) database. The IRRs are calculated using fundamental FP data for two periods; from 1960 to 2003 and from 1980 to 2003. The first period encompasses a longer estimation period which should theoretically provide more reliable estimates since we are using realized returns to estimate expected returns. The second period is chosen to reduce the impact of survivor bias since FP coverage prior to 1979 is not complete and is tilted towards large successful firms. Incidentally, empirical findings also show that realized returns are closer to expected values in recent years in the US market. Therefore, we believe that the shorter and more recent sample still represents a reasonable framework for our study. The evolution in the number of nonfinancial Canadian firms through time is summarized in Table 1. The number of firms grows from an average of 104 per year in 1960-1964 to 1112 in 2000-2003. The combined book capital of the firms in the sample also grows steadily and expands from 51.13 billion dollars in 1960-1964 to 1799.64 billion dollars in 2000-2003 in constant 1992 dollars. The number of firms entering and exiting the sample on an annual basis is also provided in Table 1. The number of firms grows by an average of more than 7 percent per year, and is highest during the 1985-1989 and 1990-1994 time periods, where the percentage of entering firms is around 19 percent and 12 percent, respectively. This increase in the number of firms corresponds to the beginning of the market boom following the 1982 recession. However, this does not coincide with the period of highest entering firms’ book-value which is the 1980-1984 period, and which represents an increase of 9.84 percent in firms’ book-values. The average annual increase in book capital during the overall period is around 3 percent, which indicates that firms which enter the market are usually small in size. [Please insert table 1 about here.] No firms exit the sample during the first 15 years, which is evidence of data backfilling. The rate of departures increases steadily to reach a peak of 7.64 percent of the firms and 12.97 percent of capital book value during the 1995-99 period. These are followed by the 2000-2003 period with 5.94 percent of firms and 7.74 percent of the capital book value of firms leaving the market. This is not surprising since it corresponds to the burst of the IT bubble. The average annual departure rate over the studied period is 2.05 percent of the firms in number, which is slightly lower than the corresponding departure rate of book capital of 2.83 percent. A firm is deemed to have departed if it has no data available on the FP database for three successive years. Similar to Fama and French, firms for which we have incomplete data (specifically, firms for which we do not have data on Income before Extraordinary Items and change in Assets), are considered to be missing since we are not able to compute their annual net cash-flows. A firm also is considered as being missing if no data are available to determine its market value for that year; that is, if the end-of-year closing price or the number of outstanding shares is missing. Missing firms are not considered in the computation of cash-flows for the years during which their data are missing. Table 1 shows that over the sample period only a small fraction of the firms is missing the critical data needed. The average during the 1960-2003 period is 0.33 percent of the firms in number and 0.29 percent of the firms in terms of book capital. Table 2 reports the GICS composition of the firms as well as the GICS composition of the entering and leaving firms. Over the sample period, Energy, Materials and Consumer Discretionary represent the sectors with the highest rates of firm entry. The rates are mostly stable or increasing between the successive five-year sub-periods for Energy, bell shaped for Materials, and decreasing for Consumer Discretionary. Energy, Materials and Consumer Discretionary also represent the sectors with the highest rate of departing firms throughout the overall time period. The sectors with the highest firm stability when it comes to entering and exiting the market are Utilities and Telecom, which both have rates lower than 2 [3] percent of firms in number entering [leaving] the sample. Book-value results are similar, with Energy, 215 Materials and Consumer Discretionary as the sectors with the highest book values for entering firms, and Industrials, Telecom and Utilities as the sectors with the highest book values of departing firms. The average percentage number of firms per GICS sector shows an increasing trend or a stable value for Energy, Materials, Industrials, Health Care and IT versus a declining trend in Consumer Discretionary, Consumer Staples, Telecom and Utility. Overall, the GICS sectors with the highest presence in the aggregate nonfinancial market during the 1960-2003 sample period are unsurprisingly Materials, Energy and Consumer Discretionary. [Please insert table 2 about here.] Empirical Estimates of Costs of Capital based on the FF methodology IRR on Value and on Cost, and Cost of Equity for Domestic and Cross-listed Nonfinancial Firms For comparison purposes, we begin by providing IRR estimates based on the same assumptions made by Fama and French, while extending their work by providing estimates for the cost of equity and the costs of capital for 9 GICS sectors. One important assumption is that market values correctly reflect the fundamentals. The nominal and real rates of return on the market values of nonfinancial firms and on the costs of their investments are reported in table 3.6 The nominal IRR on value [cost] of total capital for nonfinancial Canadian firms for the 19602003 period is 10.44 [11.22] percent. Thus, as expected, Corporate Canada added value over the studied period. Similarly, the nominal IRR on value [cost] of equity is 10.99 [12.19] percent. As expected, IRR on equity is higher than the overall IRR on capital for both value and cost, which is indicative of the higher risk assumed by equity holders. These estimates are comparable to cost of capital and cost of equity estimates documented in the literature for various time periods (e.g., Kryzanowski and He, 2007; Athanassakos, 1997), and are indicative of a profitable corporate sector, on average. The corresponding real values are 5.90 [6.64] for IRR on capital at value [cost] and 6.58 [7.78] for IRR on equity at value [cost]. For the 1980-2003 time period, nominal and real rates of return are lower for all costs but they are still indicative of profitability in the nonfinancial corporate sector since IRR on cost is consistently higher than the IRR on value for both total capital and equity costs. The decrease in the cost of capital seems to be indicative of a slight decrease of overall market risk or expected improvements in overall economic outcomes. Since we use the Fama and French methodology, our work bears the same limits as theirs. Furthermore, our estimates of the cost of equity are understated since we use the total amount of depreciation when computing the annual equity investments. This leads to an overstatement of gross investments in each year and thus an underestimate of the cost of equity. Compared to the U.S. results for the sample period 1950-1996, the IRR at value is comparable between the two markets whereas the IRR at cost is lower for Canadian firms. Although the sample period that we use is different from the Fama and French sample period, our results show at a minimum 6 Despite the presence of negative cash-flows in the IRR equations during the estimation periods 1960-2003, there is only one positive IRR for each set of cash-flows (except for the IT sector). Thus, the multiple IRR problem is not an issue here. 216 that the U.S. firms create, on average, more added value than their Canadian counterparts.7 Contrary to the US market where the costs of capital seem to increase by all measures when the more recent 19731996 period is considered, the Canadian market shows lower capital costs when we focus on the more recent 1980-2003 period. This trend discrepancy between the two markets could be explained by our inclusion of the IT bubble period which is not considered in the Fama and French sample that stops at 1996, and which exhibited large losses. [Please insert table 3 about here.] Table 4 reports rates of return on total capital and on equity at value and at cost for cross-listed non-financial firms.8 The number of nonfinancial Canadian firms listed in U.S. venues at any time between 1960 and 2003 is 314, and ranges from 13 during the 1960-1964 subperiod to 192 during the 2000-2003 subperiod. For each year, we consider the sample of cross-listed firms as our basis to compute the different IRR values. The nominal IRR on capital at value [cost] for nonfinancial cross-listed Canadian firms for the 1960-2003 period is 10.07 [10.52] percent. Similarly, the nominal IRR on equity at value [cost] is 10.75 [11.36] percent. As expected and consistent with previous findings by Jorion and Schwartz (1986) and Forester and Karolyi (1993), these values are lower than the costs of capital for domestic nonfinancial firms for the same time period. Our findings are also consistent with Hail and Leuz (2005) who document a decrease in the cost of equity following cross-listing. These findings may be due to a lower risk for cross-listed firms or better cash-flow performance related to cross-listing. The difference in IRR values is usually less than 1 percent and indicates that firms which are cross-listed enjoy a lower cost of capital and equity, albeit at a lower added value as measured by the difference between the cost and value measures of IRR. This may be due to the more competitive nature of markets for firms trading in more integrated markets. [Please insert table 4 about here.] IRRs on Value and Cost, and Equity Costs for GICS Sectors We now measure the costs of total capital and equity for Canadian nonfinancial GICS sectors. Table 4 reports the rates of return on capital and equity at value and at cost for 9 GICS sectors.9 The nominal IRR on capital at value for the 1960-2003 sample period varies between -2.04 percent and 12.55 percent with IT [Consumer Discretionary] having the lowest [highest] realized cost of capital. The dramatically lower IRR for IT can be partially explained by huge investments by Nortel Networks in the late 1970s and the beginning of the 1980s coupled with a series of low or negative earnings.10,11 If we 7 This could be linked to the productivity literature which finds that productivity growth in Canada has been lower than its U.S. counterpart. Although recent research shows that Canadian productivity has been showing higher growth rates in the more recent years. 8 Despite the presence of negative cash-flows in the IRR equations during the estimation period 1960-2003, there is only one positive nominal IRR for each set of cash-flows. The only exception is for the IT sector which exhibits negative IRRs. 9 To be consistent with our methodology, we exclude the Financials sector. 10 The net value of PPE almost quadrupled between 1975 and 1979. Income before extraordinary items did not show the same performance and was consistently lower than the change in PPE value. 11 Interestingly, starting from 1992 and as a clear indication of the stock market boom, the combined market value of IT firms entering the market was at least twice their book value. At the end of 1997, the market value of entering IT firms represented around 8 times their 217 discard the last years of the IT bubble by removing data from 1999 and 2000, the sector shows a nominal IRR on capital at value of 14.48 percent over the 1960-2003 sample period. This performance remains however very sensitive to the terminal value which indicates that most of the required return is linked to future expected growth rates and not to the current or past fundamental performance of the sector. When the IRR of capital at cost is considered, the highest cost is assumed by Consumer Staples followed very closely by Consumer Discretionary. The IRR on cost is systematically higher than the IRR on value for all GICS sectors, with the exception of Industrials which has a lower IRR on cost than on value during the 1960-2003 period. This may be partially explained by the discrepancy between the market and book values of assets in place, and may indicate that the market value is lower than the book value in certain cases. Therefore, all sectors except Industrials create positive value over the 1960-2003 sample period.12 Sectors with high barriers to entry such as Energy and Utilities show the highest added values. When the shorter and more recent sample subperiod of 1980-2003 is considered, IRRs at cost and at value are lower for 5 out of 9 sectors. For example, Energy, Materials and Industrials exhibit decreasing trends in their costs of capital, whereas sectors such Consumer Discretionary, Telecom and Utilities exhibit increases in their costs of capital. Value added increases in Energy, Consumer Discretionary and Health Care, and decreases in Materials, Consumer Staples, Telecom and Utilities. Consistent with the cost of capital theory, the cost of equity is higher than the overall cost of capital for all but Industries and IT. The negative cost of equity, which is lower than the overall cost of capital for IT, is due to value-destroying investments in the period up to and including the IT bubble. The sector with the highest [lowest] cost of equity during the 1960-2003 period is Consumer Staples [IT]. When comparing the IRR on value with that on cost for both overall capital and equity, we note that sectors create, on average, slightly higher added values from their equity financings. We now compare our cost of equity estimates to those of He and Kryzanowski (2007) who use a CAPM approach to generate cost of equity estimates for 10 GICS sectors in Canada and the U.S. over the period 1988-2005. Not surprisingly, we find some differences between our estimates and their estimates for some GICS sectors. For instance, our results show slightly higher cost of equity estimates for Consumer Staples, Consumer Discretionary, Telecom and Utilities, and somewhat lower estimates for Industrials and Health Care over the sample period 1980-2003. The most interesting difference is for the IT sector, which shows the highest cost of equity when estimates are based on the CAPM as opposed to the lowest and conspicuously negative cost of equity when estimates are based on the IRR approach. Such discrepancies show once more that risk premium/cost of capital estimates inferred from approaches using fundamental variables may differ from those using traditional asset pricing models. The IRR approach avoids the reliance on the use of industry betas and market risk premiums, both of which are subject to large estimation errors (Fama and French, 1997; He and Kryzanowski, 2007).13 As noted by Fama and French (1997): “[E]stimates are distressingly imprecise. Standard errors of more than 3% per year are typical when we use the CAPM or the three-factor model to estimate industry costs of equity. These large standard book value. However, this was not enough to reverse the overall sector performance due to the huge losses registered following the market crash at the beginning of 2000. 12 The IT sector does not create positive value per se but the value destruction is lower using the IRR at cost. 13 Also, we would expect our estimates to differ from theirs since we do not assume that Canadian/U.S. markets are integrated as in He and Kryzanowski (2007). 218 errors are driven primarily by uncertainty about true factor risk premiums, with some help from imprecise estimates of period-by-period risk loadings” (p. 178). Overall, our findings suggest that estimates of the cost of capital and equity for industries are still unresolved. Thus, practitioners are well advised to compare estimates based on market models to those based on fundamental variables before making major investment decisions. [Please insert table 5 about here.] IRRs on Value and Cost Per Year To mitigate the discrepancy between the use of a non-stochastic discount rate and the true, underlying rates of return (and risk premiums) expected by investors, we apply the IRR approach to different estimation horizons. This reduces the discrepancies between internal and expected rates of return, and provides a description on how the cost of capital evolves over time.14 Using a rolling window approach which calculates the IRR initially based on data from 1960 to 1965, and then adds data as the estimation period is extended from 1960 to 2003, we examine how the various IRR values and the costs of equity vary over the sample period. Figure 1 depicts the IRR at value and at cost in real terms for the costs of capital and equity for the nonfinancial corporate sector. Returns decrease from 1965 to 1977, with a slight spike in 1973 following the oil crisis, then increase and become more or less stable during the 1980-98 period, rise dramatically with the peak of the IT boom in 1999, and then decrease sharply thereafter due to the burst of the bubble and the beginning of the economic slowdown in the early 2000’s. These multiple IRR results also show the sensitivity of the overall sample IRRs to the terminal values and to economic news. [Please insert figure 1 about here.] EMPIRICAL ESTIMATES BASED ON THE ADJUSTED FF METHODOLOGY In this section, we provide empirical adjustments to better estimate the cost of equity. Although the IRRs on value have few problems since they are based on accurate market values, the IRRs on cost which are based on book-values could be improved if better measures of asset values are introduced.15 As a correction, we use replacement cost of reported assets in lieu of historical book values to calculate more accurate estimates of the IRR on costs. As a second investigation, we examine the sensitivity of our IRR estimates to the inclusion of Extraordinary Items as part of aggregate cash-flows. This assessment not only provides better estimates of actual realized returns but also provides information on the impact of Extraordinary Items on the returns of the Canadian Corporate sector. It also provides information on the 14 Strictly speaking, the nonstochastic discount or internal rate of return used in the literature differs from the true underlying and stochastic expected rate of return (Samuelson, 1965), and such differences increase with increases in the variance of the expected rates of return. 15 We do not correct for pre-entry investments in intangible assets (which are expensed but do not appear in the balance sheet) since we do not have the appropriate data. This would maintain the upward bias in the IRR on cost but an important part of this bias should disappear once we correct for replacement cost. Other problems are not corrected for are the impact of mergers which understates the IRR on cost and the impact of bankruptcy which usually leads to the overestimation of the terminal value of the firm leaving the sample. But since both phenomena are small in size, they should not have a large impact on the IRRs. 219 actual added value achieved in sectors such as Energy and Materials, which are known for the reoccurrence of extraordinary items. IRR on Cost using Replacement Costs The use of reported book values to measure the values of the assets that firms hold when entering or leaving the market understates the value of the assets of existing firms and consequently overstates the value of the IRR on cost. In the following, we use replacement costs to compute the IRR on cost. This allows us to examine whether corporate investments still generate a positive value on average once we correct for the replacement cost of the assets. To correct for replacement costs, we use the Ritter and Warr (2002) estimates of annual investments in assets needed to replace depleted assets. This estimate is computed as follows: X= Book ⎡ (i / n) ⎤ ⎢ ∑ n −i ⎥ i =1 ⎢ (1 + G ) ⎥⎦ ⎣ , (6) n 1/ n ⎛ GDPt ⎞ (7) π =⎜ ⎟ −1 , ⎝ GDPt − n ⎠ (8) G = ROE × (1 − div ) , where Book is the book value of equity; div is the dividend payout ratio; X is an estimate of the annual investments in assets to replace depleted assets that grows them at the nominal rate G; ROE is return on equity; π is the average rate of inflation over the estimated life of the assets; GDPt is the level of the GDP deflator at year t; and n / 2 is accumulated depreciation over depreciation costs which represents an estimate of half the depreciable life of the assets. Table 6 reports the IRR on cost estimates for domestic and cross-listed firms as well as the IRR on cost for 9 GICS sectors, once we correct for replacement cost. In each case, the IRR on cost equation [(2) and (4)] is solved using the replacement cost of assets, which is equal to the book value of assets plus, X , the annual investments in assets needed to replace the depleted assets. As expected the replacement cost-adjusted IRR at cost for both capital and equity is slightly lower than the book-value IRR at cost for all firms and for all 9 GICS sectors. However, all IRR at cost estimates remain higher than the corresponding IRRs at value, which confirms that the Canadian corporate sector creates value even if replacement costs are considered in computing the rates of return. [Please insert table 6 about here.] IRR Sensitivity to Extraordinary Items Table 7 reports estimation results for the IRRs on capital and equity at value and at cost when replacement costs and Extraordinary Items are accounted for. Interestingly, the IRR on capital and equity at value and at cost for both domestic and cross-listed firms are reduced when Extraordinary Items are considered, which indicates that the aggregate value of extraordinary items is negative on average. This result occurs for most industries with the exception of Consumer Discretionary, IT and Consumer Staples. The first two sectors show higher IRR values when Extraordinary Items are included in the aggregate cash-flows for both sub-periods 1960-2003 and 1980-2003, and Consumer Staples shows such a tendency mainly for the more recent sub-period. The percentage changes in the real IRR on capital at value [cost] 220 for the 9 GICS sectors vary between -37 [-29] percent and 4 [8] percent over the 1980-2003 time period, with Health Care and Consumer Discretionary having the most extreme changes in their cost of capital values. The percentage changes for the real IRR on equity at value [cost] for the 9 GICS sectors vary between -7 [-11] percent and 11 [3] percent, with Materials, Industrials and Consumer Discretionary having the highest changes in their cost of equity values over the time period 1960-2003. The overall results show that, although some industries have their costs of capital increase when Extraordinary Items are considered, Extraordinary Items are negative on average over the 1960-2003 time period, which leads to a decrease in the realized returns when these items are considered. [Please insert table 7 about here.] Conclusion The IRR approach of Fama and French (1999) was used to estimate the costs of total capital and of equity for domestic and cross-listed Canadian firms and nine GICS sectors. The IRR approach enabled us to distinguish between the IRR at value or the cost of capital as expected by stakeholders and the IRR at cost which represents the return delivered by the corporate sector. Since the IRR on cost is persistently higher than the IRR on value, we infer that the Canadian nonfinancial corporate sector creates value. As expected, cross-listed firms have lower costs of total capital and of equity, which confirms similar previous findings. Rates of return based on fundamentals are slightly different form the expected returns based on the market when individual GICS sectors are examined, especially for the IT sector. This suggests that certain bubbles (such as the IT bubble) could have been avoided if fundamental values were considered. Both IRR at value and at cost decrease on average but all our qualitative conclusions are maintained after applying adjustments for replacements costs and considering extraordinary items in the computations. This indicates that the corporate sector creates added value even if replacement costs are used instead of historical book values. Not only is there a high persistence of extraordinary items in certain Canadian sectors but their aggregate impact is negative over the 1960-2003 time period. 221 REFERENCES Alexander, G., C. Eun, and S. Janakiramanan, 1987. Asset pricing and dual listing on foreign capital markets: A note, The Journal of Finance 42: , 151-158. Alexander, G., C. Eun, and S. Janakiramanan, 1988. International listings and stock returns: Some empirical evidence, Journal of Financial and Quantitative Analysis 23: , 35-151. Athannassakos, George, 1997. Estimating the cost of equity and equity risk-premia of Canadian firms, Multinational Finance Journal 1:3, 229-254. Blanchard, Olivier J., 1993. Movements in the equity premium, Brookings Papers 24:2, 75-118. Botosan, C. A. and M. A. Plumlee, 2005. Assessing alternative proxies for the expected risk premium. The Accounting Review 80:1, 21-54. Campbell, John Y. and Tuomo Vuolteenaho, 2004. Inflation illusion and stock prices, The American Economic Review 94:2, 1923. Claus, James and Jacob Thomas, 2001. Equity premia as low as three percent? Evidence from analysts' earnings forecasts for domestic and international stocks, Journal of Finance 56:5, 1629-1666. Cochrane, John H., 1997. Where is the market going? Uncertain facts and novel theories, Economic Perspectives 21, 3–37. Easton, P. D. and P. Monahan, 2003. The empirical relation between realized return and accounting based measures of expected returns. University of Chicago Working Paper, Chicago, IL. Errunza, V. R. and D. P. Miller, 2000. Market segmentation and the cost of equity capital in international equity markets, Journal of Financial and Quantitative Analysis 35, 577–600. Fama, E. and K. French, 1997. Industry costs of equity, Journal of Financial Economics 43: , 153-194. Fama, E.F. and H. Babiak, 1968. Dividend policy: An empirical analysis, Journal of the American Statistical Association 63: , 1132-61. Fama, Eugene F. and Kenneth R. French, 1999. The corporate cost of capital and the return on corporate investment, Journal of Finance 54:6, 1939-1967. Fama, Eugene F. and Kenneth R. French, 2002. The equity premium, Journal of Finance 57:2, 637-659. Foerster, S. R. and G. A. Karolyi, 1998. The effects of market segmentation and investor recognition on asset prices: Evidence from foreign stock listings in the U.S., Working Paper, Ohio State University. Gebhardt, William R., Charles M. C. Lee and Bhaskaran Swaminathan, 2001. Toward an implied cost of capital, Journal of Accounting Research 39:1, 134-176. Gode, D. and P. Monahan, 2003. What affects the implied cost of equity capital?, Review of Accounting Studies 8, 399-431. Hail, Luzi and Christian Leuz, 2005. Cost of capital and cash-flow effects of U.S. cross-listings, Working Paper, He, Zhongzhi and Lawrence Kryzanowski, 2007. Cost of equity for Canadian and U.S. sectors, North American Journal of Economics and Finance 18:2, 215-229. Jorion, P. and E. Schwartz, 1986. Integration vs segmentation in the Canadian stock market, Journal of Finance 41: , 603-616. Kocherlakota, Naryana, 1996. The equity premium: It’s still a puzzle, Journal of Economic Literature 34, 42–71. Lee, C.F., C. Wu, and M. Djarraya, 1987. A further empirical investigation of the dividend adjustment process, Journal of Econometrics 58: , 267-85. Lintner, J., 1956. Distribution of incomes of corporations among dividends, retained earnings and taxes, American Economic Review 61: , 97-113. Modigliani, Franco and Richard A. Cohn, 1979. Inflation and the stock market, Financial Analyst Journal 35:2, 24-44. Ritter, Jay R. and Richard S. Warr, 2002. The decline in Inflation and the bull market of 1982 to 1999, Journal of Financial and Quantitative Analysis 37:1, 29-61. Sharpe, Steven A., 2002. Reexamining stock valuation and inflation: The implications of analysts’ earnings forecasts, The Review of Economics and Statistics 84:4, 632-648. Siegel, Jeremy J. and Richard H. Thaler, 1997. The equity risk premium puzzle, Journal of Economic Perspectives 11, 191–200. Taggart, Robert A., Jr., 1985. Effects of regulation on utility financing: Theory and evidence, Journal of Industrial Economics 33:3, 257-276. 222 Table 1. Number of Firms and Total Assets in the Sample of Nonfinancial Firms, 1960-2003 This table reports averages of the percent of firms and book capital entering, departing and missing for various five-year periods beginning with 1960-64 and finishing with 2000-03. The sample includes all publicly traded nonfinancial firms in the Financial Post database that are incorporated in Canada and for which data on market and book values of capital for any two years between 1960 and 2003 are available. A firm enters the sample at the end of the first fiscal year for which market and book value data are available, and leaves the sample at the end of the last fiscal year for which such data are available. “Firms” is the number of firms at the beginning of each calendar year. A firm is allocated to calendar year t if its fiscal year-end is between July 1 of year t-1 and June 30 of year t+1. Book capital is the total end-of-year book value of long-term debt, short-term debt and equity (in billions of 1992 dollars) for firms in the sample. Net long-term debt is used to measure long-term debt and current long-term debt, and when not available current liabilities are used to measure short-term debt. Book equity is total assets minus total liabilities. If total liabilities are not available, book equity is computed as common share equity plus preferred share equity plus retained earnings. “Percent of Firms” refers to the number of firms: (i) entering the sample, (ii) leaving the sample, or (iii) missing either income before extraordinary items or the change in assets, each divided by the number of firms in the sample at the beginning of that year. “Percent of Book Capital” refers to the comparable ratio of the total book capital of the firms in one of the three categories divided by the total book capital of all firms in the sample. Period 1960-1964 1965-1969 1970-1974 1975-1979 1980-1984 1985-1989 1990-1994 1995-1999 2000-2003 1960-2003 Firms 104 131 172 198 272 587 1130 1310 1112 532 Percent of Firms Entering Departing 3.76 0.00 3.02 0.00 6.03 0.00 2.93 0.21 9.73 0.06 19.73 0.47 12.09 5.28 6.79 7.64 3.01 5.94 7.75 2.05 Missing 0.00 0.00 0.00 0.00 0.34 1.05 0.90 0.32 0.22 0.33 Book Capital 51.13 71.02 95.00 115.52 279.23 731.87 1712.89 1607.78 1799.64 667.93 Percent of Book Capital Entering Departing 0.51 0.00 0.10 0.00 0.86 0.00 1.02 0.13 9.84 0.00 5.01 0.12 1.67 6.51 2.76 12.97 7.20 7.74 3.02 2.83 Missing 0.00 0.00 0.00 0.00 0.08 0.05 0.01 2.35 0.01 0.29 Table 2. Number of Firms, Firms Entering and Firms Departing from the Sample of Nonfinancial Firms by GICS Sector, 1960-2003 This table reports the average percentage compositions in 9 GICS sectors and book capitals entering and departing for various five-year periods beginning with 1960-64 and finishing with 2000-03. A firm enters the sample at the end of the first fiscal year for which market and book value data are available, and leaves the sample at the end of the last fiscal year for which such data are available. “Firms” is the number of firms at the beginning of each calendar year. A firm is allocated to calendar year t if its fiscal year-end is between July 1 of year t-1 and June 30 of year t+1. Book capital is the total end-of-year book value of long-term debt, short-term debt and equity (in billions of 1992 dollars) for firms in the sample. 223 Table 2 Cont’d Energy Materials Industrials Cons. Disc. Cons. Staples Health Care Period Firms Ent Dep Ent Dep Ent Dep Ent Dep Ent Dep Ent Dep Panel A: Percentage distribution of number of firms entering (Ent) and departing (Dep) in 9 GICS sectors 1960-1964 104 13.33 0.00 13.33 0.0 13.33 0.00 20.00 0.00 20.00 0.00 0.00 0.00 1965-1969 131 5.26 0.00 31.58 0.00 21.05 0.00 42.10 0.00 0.00 0.00 0.00 0.00 1970-1974 172 20.93 0.00 23.26 0.00 20.93 0.00 20.93 0.00 4.65 0.00 4.65 0.00 1975-1979 198 40.74 50.00 18.52 0.00 14.81 50.00 11.11 0.00 0.00 0.00 0.00 0.00 1980-1984 272 24.14 0.00 26.72 0.00 14.66 0.00 14.66 0.00 7.76 0.00 0.86 0.00 1985-1989 587 27.72 23.08 37.78 30.77 9.86 7.69 11.09 15.38 3.70 7.69 2.05 0.00 1990-1994 1130 22.41 34.57 27.84 31.97 11.37 9.67 12.19 14.13 3.62 3.35 5.60 0.00 1995-1999 1310 30.48 33.40 19.40 24.95 8.55 9.94 9.70 12.47 2.31 3.81 7.62 1.90 2000-2003 1112 23.44 35.03 16.41 15.57 9.38 11.08 7.81 12.57 1.56 2.99 10.16 3.89 1960-2003 532 26.08 33.68 27.59 24.79 10.88 9.61 11.85 13.53 3.56 3.62 5.01 1.96 Panel B: Percentage distribution of book capital of entering (Ent) and departing (Dep) firms in 9 GICS sectors 1960-1964 51.13 2.26 0.00 16.33 0.00 0.00 0.00 0.55 0.00 5.95 0.00 0.00 0.00 1965-1969 71.02 5.49 0.00 54.51 0.00 5.23 0.00 20.08 0.00 14.69 0.00 0.00 0.00 1970-1974 95.00 13.17 0.00 52.01 0.00 17.62 0.00 6.98 0.00 1.18 0.00 1.35 0.00 1975-1979 115.52 38.01 10.65 2.01 0.00 6.25 89.35 10.08 0.00 0.00 0.00 0.00 0.00 1980-1984 279.23 7.96 0.00 5.57 0.00 1.88 0.00 74.52 0.00 1.51 0.00 0.74 0.00 1985-1989 731.87 52.08 34.71 15.90 52.84 13.07 1.78 11.37 9.96 1.39 0.54 0.09 0.00 1990-1994 1712.89 16.27 7.60 14.28 8.29 7.76 60.86 44.36 0.32 1.97 14.06 0.15 0.00 1995-1999 1607.78 65.68 0.38 17.78 0.84 4.41 0.47 3.53 1.65 1.15 0.57 0.66 0.46 2000-2003 1799.64 4.94 0.51 12.45 18.41 0.68 2.54 56.43 3.54 0.29 1.65 0.72 12.19 1960-2003 667.93 26.72 2.79 13.23 6.36 4.13 17.62 34.00 1.22 1.95 4.38 0.51 2.25 Panel C: Percentage of number of firms in each of the 9 GICS sectors 1960-1964 104 11.15 26.35 11.15 19.62 13.08 0.00 1965-1969 131 10.38 26.26 12.06 21.68 12.21 0.00 1970-1974 172 12.09 25.81 13.72 22.21 10.47 0.93 1975-1979 198 15.35 24.14 14.14 21.11 8.89 1.01 1980-1984 272 18.01 25.37 13.24 20.07 7.35 1.10 1985-1989 587 21.29 27.67 11.82 16.87 5.93 1.50 1990-1994 1130 22.09 30.32 11.13 13.40 4.51 2.80 1995-1999 1310 20.96 27.91 11.18 12.17 4.32 4.98 2000-2003 1112 16.49 25.29 10.61 11.01 4.03 6.26 1960-2003 532 19.53 28.07 11.64 14.68 5.60 3.47 224 Ent IT Dep Telecom Ent Dep Utilities Ent Dep 6.67 0.00 2.33 11.11 4.31 4.93 13.34 19.40 12.50 11.58 0.00 0.00 0.00 0.00 0.00 15.38 3.72 8.67 13.17 8.47 6.67 0.00 0.00 3.70 0.86 1.23 2.80 1.39 0.78 1.78 0.00 0.00 0.00 0.00 0.00 0.00 0.74 3.38 2.99 2.38 6.67 0.00 2.33 0.00 6.03 1.64 0.82 1.15 3.13 1.67 0.00 0.00 0.00 0.00 0.00 0.00 1.86 1.48 2.69 1.96 53.78 0.00 0.61 1.54 0.18 2.10 1.38 4.87 20.15 8.98 0.00 0.00 0.00 0.00 0.00 0.16 0.18 0.41 1.41 0.36 11.12 0.00 0.00 42.12 0.00 2.30 3.85 1.56 0.04 1.23 0.00 0.00 0.00 0.00 0.00 0.00 0.30 38.63 31.44 26.65 10.02 0.00 7.08 0.00 7.63 1.70 9.97 0.36 18.92 9.23 0.00 0.00 0.00 0.00 0.00 0.00 8.39 56.60 28.31 38.38 2.31 2.29 1.86 3.33 3.68 4.46 7.01 11.27 12.73 8.02 6.92 5.95 4.65 4.14 3.09 2.08 2.18 2.24 1.51 2.54 8.65 7.48 6.40 5.56 5.00 3.75 2.44 2.02 2.01 3.16 Table 3. Rates of Return on Capital and on Equity at Value and at Cost for Canadian Nonfinancial Firms, 1960-2003 This table reports the rates of return on total capital and on equity capital at value and at cost using the estimation approach of Fama and French. The IRR on value [cost] estimates of the cost of capital estimate the return on corporate investments under the assumption that firms are acquired at their market [cost] values when they enter the sample, and are sold at their market value either when they leave the sample or when the sample is liquidated in 2003. The IRR on value estimates the return on equity under the assumption that the firms pay interest and principal at market [book] values for all firms in the sample, and firms are acquired and sold at their market [book] values. Time Period 1960-2003 1980-2003 Cost of Capital (%) IRR on Value IRR on Cost Nominal Real Nominal Real 10.44 5.90 11.22 6.64 8.98 4.95 9.92 5.85 Cost of Equity (%) IRR on Value IRR on Cost Nominal Real Nominal Real 10.99 6.58 12.19 7.78 9.20 5.23 10.30 6.30 Table 4. Rates of Return on Capital and Equity at Value and at Cost for Cross-listed Canadian Nonfinancial Firms, 1960-2003 This table reports the rates of return on capital and on equity capital at value and at cost using the estimation approach of Fama and French. The IRR on value [cost] estimates of the cost of capital estimate the return on corporate investments under the assumption that firms are acquired at their market [cost] values when they enter the sample, and are sold at their market values either when they leave the sample or when the sample is liquidated in 2003. The IRR on value estimates the return on equity under the assumption that firms pay interest and principal at market [book] values for all firms in the sample, and firms are acquired and sold at their market [book] values. Time Period 1960-2003 1980-2003 Cost of Capital (%) IRR on Value IRR on Cost Nominal Real Nominal Real 10.07 5.65 10.52 6.07 8.62 4.71 9.44 5.51 Cost of Equity (%) IRR on Value IRR on Cost Nominal Real Nominal Real 10.75 6.41 11.36 7.06 8.97 5.10 9.83 5.94 Table 5. Rates of Return on Capital and Equity at Value and at Cost for Canadian GICS Sectors, 1960-2003 This table reports the rates of return on capital and on equity capital at value and at cost using the unmodified estimation approach of Fama and French for the 9 nonfinancial Canadian GICS Sectors. The IRR on value [cost] estimates of the cost of capital estimate the return on corporate investments under the assumption that firms are acquired at their market [cost] values when they enter the sample, and are sold at their market values either when they leave the sample or when the sample is liquidated in 2003. The IRR on value estimates the return on equity under the assumption that firms pay interest and principal at their market [book] values for all firms in the sample, and firms are acquired and sold at their market [book] values. Sector Energy Materials Industrials Cons. Disc. Cons. Staples Health Care IT Telecom Utilities Time Period 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 Cost of Capital (%) IRR on Value IRR on Cost Nominal Real Nominal Real 10.65 6.13 12.29 7.78 8.93 5.15 11.31 7.50 8.77 4.87 9.90 5.57 6.65 3.02 7.52 3.85 11.08 7.43 10.85 7.14 10.03 6.76 10.32 7.08 12.55 7.83 13.27 8.24 13.15 9.04 14.47 10.20 12.36 7.53 13.40 8.55 14.64 10.43 15.52 11.34 9.40 6.40 9.75 6.57 8.61 5.76 9.13 6.29 -2.04 -4.37 -1.57 -3.92 -2.34 -4.75 -1.73 -4.12 10.33 5.49 11.00 6.12 12.06 7.98 12.16 8.02 10.14 5.35 11.74 6.87 11.06 6.86 12.01 7.75 225 Cost of Equity (%) IRR on Value IRR on Cost Nominal Real Nominal Real 11.71 7.16 14.64 10.21 9.05 5.24 11.94 8.12 9.20 5.03 11.23 7.13 6.83 3.25 7.77 3.99 12.95 8.69 12.82 8.56 12.58 9.02 13.30 9.77 12.78 8.13 13.68 9.04 13.26 9.19 14.58 10.45 13.53 8.62 14.77 9.86 16.46 11.97 17.32 12.93 11.72 9.05 12.33 9.62 10.25 7.25 11.10 8.11 -5.96 -8.17 -5.38 -7.61 -6.09 -8.34 -5.42 -7.67 11.46 6.63 12.88 8.04 13.73 9.39 14.03 9.71 10.65 6.03 11.93 6.98 11.29 6.99 12.27 7.91 Table 6. Rates of Return on Capital and Equity at Cost for Domestic Canadian Nonfinancial Firms, Cross-listed Canadian Nonfinancial Firms and 9 GICS Sectors using Replacement Costs, 1960-2003 This table reports the rates of return on capital and on equity capital at cost using the estimation approach of Fama and French adjusted for replacement costs. The IRR on cost estimates of the cost of capital estimate the return on corporate investments under the assumption that firms are acquired at their cost values when they enter the sample, and are sold at their market values either when they leave the sample or when the sample is liquidated in 2003. The cost value of entering firms is corrected for replacement costs using the Ritter and Warr (2002) procedure. Firm/Sector Time Period Domestic Firms Cross-listed Firms Energy Materials Industrials Cons. Disc. Cons. Staples Health Care IT Telecom Utilities 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 Cost of Capital (%) IRR on Cost Nominal Real 11.02 6.48 9.71 5.68 10.42 5.99 9.34 5.43 12.00 7.55 11.00 7.25 9.68 5.38 7.36 3.71 10.61 6.94 10.12 6.90 13.13 8.25 14.06 9.87 13.27 8.45 15.25 11.11 9.52 6.53 8.92 6.10 -1.75 -4.09 -1.90 -4.28 10.88 6.10 12.12 7.72 11.42 6.63 11.55 7.38 226 Cost of Equity (%) IRR on Cost Nominal Real 11.86 7.51 9.97 6.04 11.19 6.87 9.67 5.81 14.24 9.88 11.55 7.79 10.85 6.79 6.94 3.40 13.47 9.20 12.93 9.46 13.35 8.79 14.01 9.98 14.60 9.72 16.87 12.56 11.99 9.30 10.82 7.85 -5.55 -7.78 -5.59 -7.83 12.58 7.85 13.82 9.50 10.82 6.33 11.47 7.55 Table 7. Rates of Return on Capital and Equity at Value and at Cost for Domestic Canadian Nonfinancial Firms, Cross-listed Canadian Nonfinancial Firms and 9 GICS Sectors using Replacement Costs and Extraordinary Items, 1960-2003 This table reports the rates of return on capital and on equity capital at cost using the estimation approach of Fama and French adjusted for replacement costs. The IRR on cost estimates of the cost of capital estimate the return on corporate investments under the assumption that firms are acquired at their cost values when they enter the sample, and are sold at their market values either when they leave the sample or when the sample is liquidated in 2003. The cost value of entering firms is corrected for replacement cost using the Ritter and Warr (2002) procedure. Extraordinary Items are included when computing aggregate cash earnings at the end of each year. Firm/Sector Time Period Domestic Firms Cross-listed Firms Energy Materials Industrials Cons. Disc. Cons. Staples Health Care IT Telecom Utilities Figure 1. 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 1960-2003 1980-2003 Cost of Capital (%) IRR on Value IRR on Cost Nominal Real Nominal Real 10.36 5.84 10.93 6.42 8.84 4.84 9.55 5.55 9.97 5.58 10.31 5.91 8.46 4.58 9.17 5.29 10.59 11.92 7.49 6.09 5.08 7.15 8.85 10.89 8.59 9.47 5.22 4.31 2.87 3.54 6.48 7.16 10.81 10.35 6.78 7.27 6.47 6.60 9.70 9.77 12.88 13.72 8.87 8.10 9.50 10.43 13.66 14.69 12.18 13.07 8.32 7.41 10.93 10.62 15.06 14.67 8.63 8.75 4.66 4.02 5.17 5.47 7.98 8.25 -2.07 -1.78 -4.12 -4.40 -4.78 -4.32 -2.37 -1.94 10.30 10.96 6.08 5.46 7.87 7.92 12.01 12.06 6.49 9.99 5.24 11.23 6.64 7.14 10.79 11.26 Cost of Equity (%) IRR on Value IRR on Cost Nominal Real Nominal Real 10.81 6.45 11.67 7.38 8.91 4.99 9.66 5.77 10.53 6.26 10.96 6.71 8.37 4.55 9.32 5.51 11.65 7.11 14.17 9.82 8.96 5.15 11.43 7.68 4.68 10.26 6.33 8.75 2.57 7.27 3.69 6.64 13.59 9.69 12.60 8.65 11.35 7.95 11.61 8.29 12.93 8.30 13.91 9.25 13.94 9.90 14.98 10.86 13.26 8.44 14.28 9.52 18.08 13.80 16.70 12.56 11.68 9.01 11.94 9.25 10.24 7.25 10.80 7.85 -6.00 -8.21 -5.60 -7.82 -6.13 -8.38 -5.64 -7.87 11.43 6.60 12.85 8.01 14.62 10.26 13.89 9.56 10.40 5.83 11.57 6.96 10.82 6.24 11.87 7.50 Rates of Return on Capital and Equity at Value and at Cost for the Sample Period Beginning in 1960 and Ending in 1965 to 2003 .35 .30 .25 .20 .15 .10 .05 65 70 75 80 85 90 95 Real IRR on Equity at cost Real IRR on Equity at Value Nominal IRR on Capital at Cost Nominal IRR on Capital at Value 227 00 ASAC 2008, 24-27 May 2008 Halifax, Nouvelle-Écosse Imen LATROUS Assistant Professor Université du Québec à Chicoutimi Samir TRABELSI Assistant Professor Brock University DO FAMILY FIRMS USE MORE OR LESS DEBT? This paper investigates whether the identity of controlling shareholders influences the financing decision of the firm. In particular, we explore the impact of family control on firm debt levels. We also study the effect of family involvement in management on firm leverage. Using a sample of firms listed on the French stock market, our results show that family firms use less debt than non-family firms. Our findings are consistent with the hypothesis that family-controlled-shareholders prefer less debt as a mean to reduce firm risk. Furthermore, our results show that family firms that have a family member as CEO use more debt than family firms with outside CEOs. 1. Introduction Recent empirical studies in corporate governance suggest that the most important conflict of interest, in several countries around the world, is between controlling shareholders and minority shareholders rather than between managers and shareholders (La Porta et al. 1999; Claessens et al. 2000; Faccio et al. 2002). Furthermore, these studies show that the dominant form of controlling ownership in the world is by families (La Porta et al. 1999; Claessens et al. 2000; Faccio and Lang 2002; Andersson and Reeb, 2003a). Indeed, La Porta et al. (1999) find that 35 percent of large publicly traded firms around the world are family-controlled. Faccio and Lang (2002) analyze the ultimate controlling owners of Western European corporations at the 20% cut-off and find that 44.29% are family controlled. For American firms, Anderson and Reeb (2003a) show that almost one third of S&P 500 firms are family-owned. Family-controlling shareholders represent an important and distinctive class of large shareholders. They are more risk averse than non-family shareholders due to their less diversified human and financial capital. They generally invest a large part of their personal wealth into the capital of the family-controlled shareholders. They are also concerned with the firm’s long term survival and reputation. Familycontrolling shareholders are usually involved in the operational management of the firm as a CEO/director, which gives them a large discretion over firm’s decisions. Frequently, family controlling shareholders own more voting rights than cash flow rights, principally by using dual-class shares, pyramids and cross-holdings (La Porta et al. 1999; Claessens et al. 2000; Faccio and Lang, 2002). The 228 significant presence of family firms in several countries around the world and their particular characteristics make them an interesting type of blockholder. Several recent studies have looked at the impact of family control on firm performance (Anderson and Reeb, 2003b; Claessens et al., 2000; Cronqvist and Nilsson, 2003). Little is known about the financing decisions of family-controlled firms. Whether leverage of family firms differs from that of non family firms remains an open question. For US firms, Mishra and McConaughy (1999) show that family firms use less debt than non family ones while Anderson and Reeb (2003b) show that family firms use similar levels of debt as non family firms. In other countries, Harijono et al. (2004), Wiwattanakantang (1999) and Gallo and Vilaseca (1996) find contrasting results. Our study attempts to add to this literature by exploring the impact of family control and management on firm leverage. First, do family-controlled firms use high or low levels of leverage? Family controlling shareholders own poorly diversified portfolios and have incentives to reduce the risk to the firm. Debt is used as a means of reducing firm risk because less debt decreases the probability of financial distress (Friend and Lang, 1998). Nevertheless, to avoid the dilution effect of equity financing, family controlling shareholders should use more debt to inflate their power and to become more entrenched (Stulz, 1988). Second, are debt levels related to the presence of a family member as CEO? Family management can reduce agency problem (Berle and Means, 1932; Jensen and Meckling, 1976). Agency theory would predict that debt is not used as a disciplinary device limiting managerial opportunism (Jensen, 1986). On the other hand, family members having CEO positions may influence firm policies to meet family owninterests (Villalonga and Amit, 2006). In such corporations, a family member serving as CEO may use his/her controlling position to extract private benefits at the expense of the outside shareholders. Using data on all Fortune-500 firms during 1994–2000, Villalonga and Amit (2006) find that when descendants act as CEOs firm value is destroyed, in particular, for second-generation heirs. Since debt permits the family to dominate more resources without diluting control stake, family members acting as CEOs have incentive to use debt to enhance the control power of family shareholders and to further expropriate small shareholders (Faccio et al. (2001a)). Consistent with the opportunistic behaviour of family management, Anderson and Reeb (2003a) find that when family members hold the CEO position, the cost of debt financing is higher relative to family firms with outside CEOs. We use a sample of firms listed in the French stock market over the period 1998-2002 to explore these questions. Several studies have focused on these questions for US family firms (Anderson and Reeb, 2003b; Mishra and McConaughy, 1999). We examine French data, since the French institutional environment and ownership structure differs from those in the U.S (La Porta et al., 1999). The French context provides an especially good platform to study whether family firms employ more or less debt than non family ones as family firms are prevalent in France. Indeed, Faccio and Lang (2002); La Porta et al. (1999) show the importance of family firms in France. Faccio and Lang (2002) find that 70.92% of non financial French firms in their sample are dominated by families. They also show that 13.16% of French family firms are controlled via pyramids. This device permits families to separate ownership from control and allow controlling shareholders to dominate a firm through a cascade of firms owning only a small fraction of its capital. Consequently, important private benefits of control are took out at the expense of minority shareholders of the firm at the bottom of the pyramid. Furthermore, the management team of family controlled firms is often controlled by family members. Faccio and Lang (2002) show that as many as 62.2% of French firms have a family member in management as CEO, chairman, vice chairman or honorary chairman. La Porta et al. (1999) find that 75% of French family controlling shareholders belong to firm management. 229 Our results show that the use of debt by family firms is significantly lower than that of non family firms. Family firms prefer to use financing forms with low probabilities of default, suggesting a lower reliance on debt financing. On the other hand, our findings indicate that family firms having family members as CEOs use more debt than family firms with outside CEOs. Family CEOs prefer to employ more debt to limit the dilution of their power and to extract valuable private benefits of control at the expense of outside shareholders. Yet, firms placing family members as CEOs employ less debt when an outside blockhlder is present. Then, outside blockholdings appear to have an impact on the opportunistic behavior of family CEO. Our research contributes to the extant literature in several ways. First, our results show that distinguishing between family control and management is crucial to understand the financing decisions of family firms. Secondly, the relationship between family control and debt usage show that family controlling shareholders, as large, undiversified blockholders, seek to reduce firm risk by influencing capital structure decisions. Thirdly, in line with recent research supporting that family blockhodlers can further influence agency conflicts by placing one of their members in the CEO position; we find evidence that family management could be detrimental to minority shareholders (Anderson and Reeb, 2003a). Our findings show that family firms having family member as CEO employ more debt than family firms with an outside CEO. Nevertheless, minority shareholders who fear expropriation by family controlling shareholders may prefer a low level of leverage. The remainder of this paper is organized as follows. Section 2 reviews the related literature and presents our hypothesis. Section 3 describes research design, sample and data. Section 4 presents the empirical results on French firm data and Section 5 concludes the paper. 2. Family firms and leverage 2.1. Family control and debt financing Family controlling shareholders represent a special class of large shareholders that tend to have a very long horizon on their investments and have a unique power and incentive structure in the firm (Andersson et al., 2003)). Family shareholders hold generally undiversified portfolios relative to atomistic shareholders because they invest a large part of their wealth in the firm. They also face a situation where their reputation is strictly related to that of the firm. Moreover, family controlling shareholders want to pass the firm to subsequent generations. Thus, they are very interested in the firm’s long term survival. Casson (1999) and Chami (1999) suggest that founding families view their firm as an asset to pass to family members or their descendants rather than wealth to consume during their lifetimes. Therefore, they have strong incentives to minimize firm risk by reducing debt usage. This suggests that family firms tend to use less debt than non family ones. Indeed, family controlling shareholders prefer a low level of debt to reduce the risk of their undiversified financial and human capital investments (Friend and Lang, 1988). Furthermore, family controlling shareholders have incentives to protect their private benefit of control. So, they use lower debt to reduce the probability of bankruptcy which leads to the loss of their private benefits of control. Family controlling shareholders prefer also employ less debt in order to limit monitoring by creditors (Harris and Raviv, 1988). For US firms, Mishra and McConaughy (1999) show that family firms use less debt than non family ones. Based upon this, we test the following hypothesis: 230 Hypothesis 1: If family controlling shareholders want to reduce firm risk, then we expect that family firms will use less debt than non-family firms In the other hand, through their power position, family controlling shareholders exert dominant influence on the firm’s decisions. They are able to extract valuable private benefits of control at the expense of outside shareholders. Johnson et al. (2000) referred to this case as tunneling which is defined as the transfer of assets and profits out of firms for the benefit of their controlling shareholders. Cronqvist and Nilsson (2003) show that family controlling minority shareholders are associated with the largest discount on firm value. They document that private benefits of control are relatively important for family firms. For firms dominated by controlling shareholders, debt is used to enhance the voting power of controlling shareholders and to further expropriate outside shareholders. Within a corporate pyramid, increased debt level by an affiliate can be recycled into external loans guaranteed by other affiliates (Faccio and Lang, 2001a). Debt permits family controlling shareholders to limit the dilution of their voting power. Consequently, they prefer to use more debt to protect themselves from unfriendly takeovers (Harris and Raviv, 1988; Stulz, 1988). Furthermore, family controlling shareholders have incentives to employ more debt in order to enhance their power and dominate more resources without diluting their control stake (Faccio and Lang, 2001a). Indeed, Holmen et al. (2002) show that the Swedish firms controlled by families use more debt than non family firms. In addition, using a sample of Thai firms, Wiwattanakantang(1999) explores the determinants of firm capital structure and finds a high debt level of family firms relative to non family ones. This finding supports the argument which states family controlling shareholders use debt to inflate their voting power and reduce the discipline of the market for corporate control (Stulz, 1988; Harris and Raviv, 1988). Recent studies suggest that high debt level of family firms is related to a lower cost of debt financing. They argue that the divergence of interests between shareholders and bondholders is potentially less severe in family firms than in non family firms (Anderson et al., 2003a). Using a sample of 252 US industrial firms, Anderson et al. (2003a) find that family ownership is associated with a lower agency cost of debt. We test the following prediction: Hypothesis 2: If family controlling shareholders seek to enhance their power and to further entrench themselves, then we expect that family firms will have higher levels of debt than their counterparts. In sum, we can distinguish between two competing hypothesis regarding the relationship between family control and debt usage. Therefore, empirical testing appears to be necessary to establish which of the two is empirically valid. 2.2. Family management and leverage A common characteristic of family firms is that family members often have a managerial role (La Porta et al., 1999; Faccio et al., 2002; Claessens et al., 2000). Indeed, they usually serve as the firm’s CEO or fill other top management positions. This can have two implications. First, by holding the role of CEO, families can more easily align the firm’s interest with those of the family, thus they can reduce the traditional owner-manager conflict. However, choosing the CEO among family members can harm firm performance by excluding more qualified and talented outside professional managers. Furthermore, 231 outside managers may have skills and experiences that family members may not have. Family members serving as CEOs may influence firm policies to meet their interests. Burkart et al. (2002) argue that family management, especially by descendants is associated with poor decision-making. Thus, family controlled firms are more exposed to managerial entrenchment. Gomez et al. (2001) agree with this argument and suggest that family CEO are likely less accountable to shareholders and directors than hired managers. The evidence on this topic is mixed. On the one hand, Palia and Ravid (2002) and Adams et al. (2004) show that firms with founder-CEO trade at premium, indicating that this type of CEO reduces agency conflicts inside the firm. On the other hand, Smith (1999) and Perez-Gonzalez (2002) find a negative stock market reaction for firms that have founding members as CEO. Futhremore, Morck et al. (1988) show a low performance for firms with founder CEO. Villalonga and Amit (2006) find a non monotonic effect of generation on firm value. They show that family ownership only creates value when the founder is still active in the firm either as CEO or as chairman with a hired CEO. Nevertheless, descendant-CEOs destroy value, especially second generation family firms. The effect of the third generation descendant-CEOs on firm value is non significant. But, the fourth generation CEOs affect positively the firm value. Andersson et al. (2003a) investigate the impact of founding family ownership structure on the agency cost of debt. They find that having a family member in management lead to more severe debt agency costs. Thus, bondholders view placing family members as CEO as detrimental to their wealth and, consequently they require a higher cost of debt. The presence of family CEO can then exacerbate agency conflicts with outside shareholders and creditors. Families can exert the greatest influence on the firm by placing one of their members in the position of CEO. Indeed, family members acting as CEO possess sufficient power to meet family interests, however, these interests need not necessarily be in the interests of the firm or outside shareholders. Family CEOs may engage in non-profit maximizing objectives, self dealing transactions, excessive compensation or special dividends. Thus, they are able to derive important private benefits of control to the detriment of outside shareholders and thereby result in minority shareholder wealth expropriation. Thus, we posit that family CEOs prefer use more debt due to its non-dilutive effect. Debt permits family CEOs to inflate their power and further entrench themselves with more resources under their control (Faccio and Lang , 2001a). Hypothesis 3: We expect that family firms use more debt than non family firms, in particular when family member is serving as CEO. 3. Research design 3.1. Sample Our data consist of listed firms on the French stock exchange for the period 1998-2002. Financial companies are excluded from the sample because such companies have to comply with very stringent legal requirements. Those firms that were subject to mergers or acquisitions, or those that were not listed on the stock exchange for a given year, were also eliminated. We also removed firms with negative book equity values (Kremp et al., 1999). Our sample is trimmed by applying a methodology similar to that of Kremp et al. (1999). This yields 553 firm-year observations on 118 firms for the period 1998 through 232 2002. Firm-level accounting data and market equity value data are extracted from Thomson Financial data base. We collected ownership structure and voting rights data from financial reports. Different measures of leverage were considered. 3.2. Leverage measures We consider three measures of leverage according to whether book value or market value is used. First, we look at total debt divided by total assets (L1). The second measure is the total debt scaled by the book value of total invested capital of the firm (book value of total debt plus book value of equity) (L2) (Frank and Goyal, 2003). Finally, we use the ratio of total debt to market value of total capital of the firm (book value of total debt plus market value of equity) (L3). Data limitations confine us to measure debt only on book value. 3.3. Family firm definitions One of our primary concerns is the identification of family firms. Various studies use the term ‘family firm’ differently. Andersson and Reeb (2003b) use the fractional equity ownership of the founding family and (or) the presence of family members on the board of directors to identify family firms. Villalonga and Amit (2006) restrict the term ‘family’ to either the founder’s family or to an individual or family that becomes the largest non institutional shareholder in the firm through the acquisition of a block of shares. In this study, the typology of control as proposed by Le Maux (2003) is used. Therefore, we split up our sample according to whether firms are dominated by controlling minority shareholders, controlling majority shareholders or are widely held. Controlling majority shareholder(s) own alone or with other shareholders (family members or other shareholders involved in shareholder agreements) 40%1 and more of cash flows or voting rights. Controlling minority shareholders hold less than 40% of cash flow rights or voting rights but they dominate the board with members who are affiliated with them. Board members affiliated with a firm’s controlling shareholders are managers, family members, banks, insurances, employees, government, other shareholders involved with the controlling shareholders in a shareholder agreement. Thus, controlling shareholders may form with several allies a controlling coalition. According to Le Maux (2003), shareholder agreements are very common in France. In fact, in 45% of firms belonging to the CAC 40 and in 29% of firms belonging to the SBF 120, there are one or more agreements between shareholders. A firm is regarded as being widely held when it is not dominated either by controlling majority shareholders or by controlling minority shareholders. According to this typology of control, we consider two definitions of a family firm. First, we define family firm when the largest shareholder is an individual or members of the same family2 and holding 40% and more of cash flows or voting rights. In this case, family is controlling majority shareholders. On the other hand, a firm is classified as family controlled when there is an individual or family members who hold less than 40% of cash flow rights or voting rights but they dominate the board with members who are affiliated with them. This second definition suggests that some families are able to 1 2 French regulations require 40% of voting rights as cut-off level for presumed control. Family members having the same last name. 233 exert control with minimal fractional ownership. In these firms, the family is controlling minority shareholder. Therefore, we use a dummy variable (FamFirm) to denote family controlled firms. That is FamFirm= 1, if the firm is family controlled, = 0, otherwise. Furthermore, families are usually involved in the operational management of the firm. Thus, we use a dummy variable FamCEO to measure the family involvement in management. That is, FamCEO=1, if family member serves as CEO, = 0, otherwise. 3.4. Control variables We also refer to control variables that are usually considered in the literature as influencing the firm’s capital structure, such as growth opportunities, firm size, the nature of assets, profitability, the operational risk, non debt tax shields and industry classification. In line with Rajan and Zingales (1995), we use the market to book ratio to proxy for growth opportunities. Rajan and Zingales (1995) find a negative relation between debt and market to book ratio (MTB) for a sample of large American, German, French, British and Canadian firms. Titman and Wessels (1988) found a negative relation between leverage and other proxies of growth opportunities. Myers’ (1977) underinvestment problem suggests a negative relationship between growth and debt. Indeed, Myers (1977) shows that firms with growth opportunities may invest sub-optimally, and thus creditors will be more unwilling to lend for long term. We measure firm size by the logarithm of total assets (T) (Faccio et al., 2001a). Rajan and Zingales (1995) propose that the firm size may proxy for the probability of bankruptcy, which is high for small firms. Large firms are more transparent, suffer less from informational asymmetry and have easier access to financial markets. So, large firms should use more debt financing. Several empirical studies find ambiguous results on the relationship between leverage and firm size ( Kim and Sorensen, 1986; Rajan and Zingales, 1995). We use the ratio of tangible assets to total assets for the tangibility attribute (S) (Kremp et al., 1999). Rajan and Zingales (1995) assert that tangible assets can be pledged as collateral for loans, and therefore reduce debt agency costs. Myers (1977) suggests that underinvestment problem due to debt financing is weaker for the firms with more tangible assets. We then expect a positive relation between leverage and tangible assets. Firm profitability is measured by the ratio of earnings before interest, taxes and depreciation to total assets (RO) (Fontaine et al., 1996). Myers and Majluf (1984) suggest that more profitable firms use less debt because they have sufficient internal funds. Firms will turn to debt only after they exhaust internal funds. Several empirical studies find negative relationship between profitability and leverage (Friend and Lang, 1988; Jensen et al., 1992). 234 Another possible explanatory variable is net income volatility. This measure was used in a number of empirical papers (Titman and Wessels, 1988; Friend and Lang, 1988). An increase in income volatility is considered as a serious threat for the creditors. Therefore, higher income variability may lead to a lower credit supply (Bradley et al., 1984). Then, income volatility should be negatively related to leverage. We calculate income volatility by the standard deviation of firm’s accounting profitability. We use the previous ten years when estimating standard deviation. The NDTS variable is used to capture the non debt tax shields argument put forward by De Angelo and Masulis (1980). They state that firms with a high level of non debt tax shields are expected to receive lower tax benefits associated with leverage and hence will use less debt financing. According to De Angelo and Masulis (1980), non debt tax shields are negatively associated with leverage. We calculate NDTS variable as the ratio of annual depreciation scaled by total assets. The industry feature is also seen as important in explaining corporate leverage. Firms belonging to the same industry face similar market conditions and have the same risk characteristics. Titman and Wessels (1988) suggest that industrial companies use less debt because they are exposed to high liquidation costs. They use a zero-one dummy variable for industry classification. Based on the FTSE classification, we created dummy variables to control for the effect of industrial classification on the level of debt ratios. We use four dummy variables to control whether the company belongs to industry, consumer goods, services and new technologies sectors. 3.5. Testing methodology First, we test whether leverage of family controlled firms differs from that of non family firms. We use the following specification to test the relation between family control and corporate debt levels. L(k)it = β0 + β1 FamFirm + β2 Tit + β3Sit + β4 ROit + β5 Rit + β6 Qit + β7 NDTSit +β8 ( industry dummy variables it)+ εit (1) Where i denotes the cross-sections and t denotes time-period with i= 1…118 and t= 1...5. We have yearly observations from 1998 to 2002). L (k) represent different leverage measures (L1, L2 and L3) with k= 1, 2 and 3. Finally, εit is the ‘normal’ error term. We use panel data regression analysis. The fixed effect regression cannot be used in our first specification Eq.(1)3. Indeed, the fixed effect estimator cannot estimate the effect of any time invariant variable like FamFirm variable (Baltagi, 1995). Thus, the pooled regression estimation technique is employed. To investigate the impact of placing family members as CEO, we repeat the testing in Eq.(1) by adding FamCEO dummy variable. 3 And for all our specifications. 235 L (k)it = β0+β1 FamFirmit +β2 FamPDGit++β3 Tit+β4Sit+ β5 ROit+β6 Rit+ β7 Qit +β8 NDTSit +β9 ( variables dummy secteurit)+ εit (2) 4. Empirical results 4.1. Descriptive and univariate statistics Table I (Panel A) shows the percentage of firms with controlling shareholders and reports (Panel B) the identity of controlling shareholders. Interestingly, we find that 64.7% of firms are dominated by controlling majority shareholders and 19.9% of firms are dominated by controlling minority shareholders, at the 40% cut-off level. Only 15.4% of firms are widely held. We note that firms dominated by controlling majority shareholders are strongly present in our sample. Panel B of table II shows that the family control represents 57.1% of the firms. Another frequent controlling shareholders category is ownership by corporations, which is corresponds to 35.9% of the total number of firms. The other ownership categories are state (2.8%) and financial institutions (4.3%) ownerships. Therefore, Panel B shows the predominance of family control in the French context. This corroborates the findings of Faccio and Lang (2002) and La Porta et al. (1999). For instance, Faccio and Lang (2002) find that 70.9% of French firms of their sample are controlled by families. La Porta et al. (1999) show that 50% of French medium-sized firms are family-controlled. In panel C, we can observe that in 92% of family firms the CEO is a member of the family shareholders. In sum, Table I shows that, for French listed firms, (i) controlling majority shareholders are predominant, (ii) family control of firms appears to be common, (iii) families are strongly involved in management, Table I. Sample description N (firm-year Proportion observations) (%) Panel A: Type of control Widely held Controlling minority shareholders Controlling majority shareholders Total Panel B: type of controlling 85 110 15.4 19.9 358 64.7 553 100 236 shareholders Financial institutions State Corporations Family Total Panel C: Family involvement in management FamCEO OutsideCO Total 20 4.3 13 168 267 468 2.8 35.9 57.1 100 246 21 267 92 8 100 The panel A of Table II shows that family controlling shareholders hold on average 55.5% of cash flow rights. This result highlights the prevalence of ownership concentration for French family firms. Furthermore, panels B and C show that the firms dominated by families have, on average, a lower leverage, than non-family firms (20.7% versus 25.2% for L1, 36.6% versus 41.8% for L2, 24.1% versus 30.6% for L3). This finding suggests that family controlling shareholders use less debt in order to limit the risk of their undiversified capital and human investments. Table III. Descriptive statistics N (firm-year observations) Minimum Maximum Panel A : family firms L1, book leverage L2, book leverage L3, market leverage ParAc,controllingshare holders ownership NDTS, non debt tax shield RO, profitability R, Risk T, size S, collateral Q, Growth opportunities Panel B: Non family firms L1, book leverage L2, book leverage L3, market leverage 260 Mean Standard deviation 0 0.583 0.207 0.130 260 255 0 0 0.813 0.87 0.366 0.241 0.209 0.192 258 0.128 0 .898 0.555 0.166 265 0 0.361 0.05 0.03 265 266 265 265 -0.259 0 0.75 0 0.484 0.216 4.448 0.534 0.087 0.032 2.649 0.175 0.075 0.029 0.635 0.124 264 0.085 7.94 2.177 1.408 196 196 191 0.001 0.002 0 0.584 0.946 0.868 0.252 0.418 0.306 0.139 0.204 0.207 237 ParAc,controllingshare holders ownership NDTS, non debt tax shield RO, profitability R, Risk T, size S, collateral Q, Growth opportunities 185 0.163 0.988 0.536 0.208 201 0 0.661 0.057 0.051 201 201 201 201 -0.638 0 0.654 0 0.417 0.349 4.243 0.751 0.063 0.0368 2.966 0.261 0.083 0.037 0.624 0.197 201 0.179 7.935 1.848 1.392 Table III provides a univariate analysis between family controlled firms and non family firms Panel A of Table IV shows that difference in indebtedness (L1, L2 and L3) between the two controlling owner categories is significant. Moreover, the difference in indebtedness is significant between firms placing family member as CEO and firms having outside managers. Table III. Tests of differences in median debt ratios (Z-statistics) Variables Type of owners Median P (z) difference (zstatistics) L1 Non family firms 3.47 0.0005*** Family firms L2 Non family firms 2.64 0.0083*** Family firms L3 Non family firms 3.365 0.0008*** Family firms L1 FamCEO 0.83 0.208 Outside CEO L2 FamCEO 0.026** 2.21 Outside CEO L3 FamCEO 0.017** 2.38 Outside CEO At last, a correlation analysis of the independent variables was performed. Cross-correlations are generally low, except for growth opportunities and size. To check whether these two variables are collinear, we perform a VIF test. Our VIF test results are considerably lower than 3. Thus, multicollinearity among the independent variables should not constitute a problem. 238 4.2. Multivariate regression results The primary variable of interest is FamFirm and its coefficient, which indicates the difference in debt usage between family and non family firms. Table IV gives the regression results of the effect of family control on debt levels (M1, M2, M3). Each regression corresponds respectively to the three measures of leverage L1, L2 and L3. For all models (M1, M2, M3), we find that the coefficient on FamFirm is negative and significant at 1%, 5% and 10% level for M1, M2 and M3 models, respectively. This result suggests that family firms have lower leverage than non family firms. This negative relation between family control and leverage indicates that family firms use less debt than non family firms. This finding rejects the argument which states that family controlling shareholders employ more debt for entrenchment purposes (Stulz, 1988). Generally, family controlling shareholders invest a large fraction of their personal wealth in the firm. Thus, they seek to reduce risk through their financing choices, especially by using less debt in their capital structure. Moreover, families are strongly concerned with firm’s long term survival and want to pass their firm to subsequent generations. Therefore, they have incentives to employ financing forms with low probabilities of default, suggesting a lower reliance on debt financing. Our results are consistent with the findings of Allouche and Amann (1995), Mishra and McConaughy (1999), Gallo and Vilaseca (1996). Nevertheless, our results are contrary to Harijono et al. (2004) and Wiwattanakantang (1999) findings. Allouche and Amann (1995) show that French family firms of their sample are reluctant to use debt financing. Gallo and Vilaseca (1996) show that Spanish family firms have lower leverage. On the other hand, Harijono et al. (2004) and Wiwattanakantang (1999) find a positive relation between family control and leverage for a sample of Australian firms and a sample of Thai firms, respectively. This supports the hypothesis that family firms employ more debt to enhance their power, and is consistent with Bebchuk’s (1999) argument. Indeed, Bebchuk (1999) argues that private benefits of control are important in family controlled firms. Thus, family controlling shareholders have incentives to conserve control and therefore use more debt due to its non dilutive effect. However, Anderson and Reeb (2003b) show that family firms use no less and no more debt than non family firms. This finding suggests that families may maintain holdings in low-risk businesses, thus they have less need to engage in risk reducing activities. Turning to the control variables, we find that most of them have statistically significant explanatory power and their signs are consistent with predictions4. Indeed, the estimated coefficient of firm size is positive and statistically significant. This is consistent with the argument which states that firm size serves as an inverse proxy for the probability of bankruptcy, which implies that larger firms should be more highly leveraged. The positive impact of size on leverage is consistent with the results of many empirical studies (Rajan and Zingales, 1995; Booth et al., 2001; Frank and Goyal, 2002). The coefficient of the firm risk variable is negative and significant at 1% level. The result supports the view that firms with higher earning volatility use less debt due to higher bankruptcy risks. Higher profitability is associated with a lower leverage level. This finding is consistent with the pecking order hypothesis of Myers and Majluf (1984) and the empirical results of Titman and Wessels (1988), and Friend and Lang (1988). Lastly, the coefficient on Q variable, which proxies for growth opportunities, is negative and significant at 1% level. Therefore, firms with growth opportunities use less debt financing (Bradley et al., 1984; Titman and Wessel, 1988). 4 Except for variable measuring the tangibility of assets (S) and non debt tax shields (NDTS). 239 In sum, our results confirm our first hypothesis which states that if family controlling shareholders desire to limit the risk of their undiversified human and financial capital investments, then they will use less debt than non-family firms. Several studies show that family firms perform better than non family firms (Andersson and Reeb, 2003c; Barontini and Caprio, 2005). Thus, family controlling shareholders are more likely concerned with value maximizing objectives rather than extracting private benefits of control. They have little incentives to employ more debt in order to expropriate outside shareholders. Moreover, due to the undiversified nature of their holdings and their desire for firm survival and reputation, family controlling shareholders have strong incentives to reduce firm risk by limiting debt usage. Table IV. Regression results on the relationship between leverage and family control Variables L1 FamFirm NDTS RO R T S Q R² N Dependant variables L2 L3 M1 -0.04 (0.003)*** 0.18 (0.416) -0.18 (0.033)** -0.71 (0.000)*** 0.02 (0.012)** 0.03 (0.493) -0.019 (0.000)*** 0.18 456 M2 -0.04 (0.02)** -0.07 (0.785) -0.466 (0.001)*** -0.21 (0.000)*** 0.08 (0.000)*** -0.08 (0.269) -0.023 (0.001)*** 0.24 456 M3 -0.03 (0.05)* 0.089 (0.746) -0.27 (0.023)** -0.76 (0.004)*** 0.09 (0.000)*** 0.01 (0.783) -0.06 (0.000)*** 0.47 446 We next investigate the influence that family members serving as CEOs have on firm debt levels. We include a dummy variable for a family member as CEO into our regressions. We report the regression results in Table V. Interestingly, we find that the coefficient on FamCEO is positive and significant for all regressions (M4, M6, M8) using either book value or market value of debt. This suggests that having a family member as CEO is associated with a significantly higher debt levels. This finding confirms our univariate analysis (Table IV). Family firms use more debt when a family member takes the position of CEO. Family CEOs prefer to contract more debt to limit dilution of their power. Indeed, debt permits to family CEOs to protect and enhance their control for future extraction of private benefits. Our findings are consistent with the results of Morck et al. (1988); Sraer and Thesmar (2004), Villalonga and Amit (2006) and Yeh (2005) which show that family member CEOs lead to poor performance, relative to outside CEOs. Furthermore, our results are consistent with Anderson and Reeb (2003a) who find that having family member in management leads to more severe debt agency costs. Bondholders view placing one or more family members in the position of CEO as detrimental to their wealth and thus require higher yields. 240 However, our results indicate that even after controlling for family involvement in management, family firms still have lower debt levels than non family firms. Lastly, we study the effect of the presence of an outside blockholder for firms with family CEOs. We use a dummy variable SdBlock to indicate outside blockholder, defined as entities holding 10% and more of the firm’s shares, and having no relationship to family controlling shareholders. Other large shareholders may have strong incentives to monitor and discipline firm managers. We repeat the testing in Eq.(2) by adding the product of FamCEO and SdBlock (M5, M7, M9). From Table VI, we notice that the coefficient on FamFirm is negative and significant for models M7 and M9. The coefficient on FamCEO is positive and significant for models M7 and M9. The coefficient on the product of FamCEO*SdBlock is negative and significant for models M7 and M9. To sum up, models M5, M7 and M9 show that family firms have lower leverage than non family firms. Furthermore, firms with family member CEOs use more debt than family firms with outside CEOs. Nevertheless, firms placing a family member as CEO employ less debt when an outside blockholder is present. Family member CEOs tends to reduce debt usage for entrenchment objectives after controlling for an outside blockholder. This is consistent with the argument which states that outside blockholder plays a monitoring role over family CEOs (Faccio et al., 2001b). Table V. Regression results on the relationship between leverage and family control Variables L1 FamFirm NDTS RO R T S Q R² N Dependant variables L2 L3 M1 -0.04 (0.003)*** 0.18 (0.416) -0.18 (0.033)** -0.71 (0.000)*** 0.02 (0.012)** 0.03 (0.493) -0.019 (0.000)*** 0.18 456 M2 -0.04 (0.02)** -0.07 (0.785) -0.466 (0.001)*** -0.21 (0.000)*** 0.08 (0.000)*** -0.08 (0.269) -0.023 (0.001)*** 0.24 456 241 M3 -0.03 (0.05)* 0.089 (0.746) -0.27 (0.023)** -0.76 (0.004)** * 0.09 (0.000)** * 0.01 (0.783) -0.06 (0.000)** * 0.47 446 5. Conclusion This paper explores whether leverage of family controlled firms differs from that of non family firms. There are two competing hypotheses about the relationship between family control and debt levels. If family controlling shareholders desire to limit the risk of their poorly diversified human and financial investments, we expect consequently that family firms will use less debt than non-family firms. In contrast, if family controlling shareholders need to conserve control and to entrench themselves further, we expect that family firms will have higher levels of debt than their counterparts. Moreover, we study the effect of the family involvement in management on firm leverage. Using a sample of 118 firms listed on the French stock market over the period 1998-2002, our results show that the use of debt by family firms is significantly lower than that of non family firms. This significant negative association between debt and family control rejects the hypothesis which states that family controlling shareholders have incentives to expropriate minority shareholders wealth, and should use more debt to further entrench themselves. However, our findings are consistent with the hypothesis which indicates that family controlling shareholders issue less debt to reduce firm risk. Furthermore, our results show that family firms having a family member as CEO use more debt than family firms with outside CEOs. When family member serves as CEO, the conflict of interest between shareholders and managers became less relevant. On the other hand, by dominating the management, family shareholders are able to take decisions according to their objectives which may harm firm performance. They have incentives to satisfy family interests which are always different to those of the external shareholders. Family member CEOs prefer to use more debt in order to increase their control and extract more private benefits of control. Nevertheless, firms placing a family member as CEO employ less debt when an outside blockholder is present. References 1. Adams, R.B., Almeida H., and Ferreira, D. (2005) ‘Understanding the relationship between founderCEO and firm performance’, Working Paper, New York University. 2. Allouche, J., and Amman, B. (2000) ‘L’entreprise familiale : Un état de l’art’, Finance, Contrôle, stratégie, Mars 2000. 3. Anderson, R.C., S.A. Mansi, and Reeb, D.M. (2003a) ‘Founding family ownership and the agency cost of debt’, Journal of Financial Economics, 68, pp. 263-285. 4. Anderson, R.C. and Reeb, D.M. (2003b) ‘Founding family ownership, corporate diversification and firm leverage’, Journal of Law and Economics , October, pp. 653-684 5. Baltagi, B.H., 1995, Econometric analysis of panel data, Jhon Wiley & Sons Ltd. 6. Barontini, R., and Caprio L. (2005) ‘The effect of ownership structure and family control on firm value and performance: Evidence from Continental Europe’, ECGI Finance Working Paper n° 88. 7. Bebchuk, L, (1999) ‘A rent-protection theory of corporate ownership and control’, Working Paper, 7203, National Bureau of Economic Research, Cambridge, MA. 242 8. Bebchuk, L., R. Kraakman and Triantis, G. (1999) ‘ Stock pyramids, cross-ownership, and the dual class equity: the creation and agency costs of separating control from cash flow rights’, National Bureau of Economic Research, NBER Working paper No. 6951. 9. Bennedsen M., and Wolfenzon D. (2000) ‘The balance of power in closely held corporations’, Journal of Financial Economics, 58, pp.113-139. 10. Berle, A., and Means, G. (1932) The Modern Corporation and Private Property. (New York, MacMillan) 11. Booth, L., Aivazian,V., Demirguc-Kunt, A. and Maksimovic,V. (2001) ‘Capital structure in developing countries’ , Journal of Finance, 56, pp.87-130. 12. Burkart, M., Panunzi F., and Shleifer A. (2002) ‘Family firms’, CEPR No3234 13. Bradley, M., Jarrell G.A., and Kim, E.H.(1984) ‘On the existence of an optimal capital structure: theory and finance’, Journal of Finance 39, pp. 857 879. 14. Casson, M. (1999) ‘The economics of the family firm’, Scandinavian Economic History Review, 47, pp. 10-23. 15. Chami, R. (1999) ‘What’s different about family business?’, Working paper, University Notre Dame et IMF. 16. Claessens, S., Djankov, S. and Lang, L.H.P. (2000) ‘The separation of ownership and control in East Asian corporation’, Journal of Financial Economics, 58, pp.81-112. 17. Cronqvist, H. and Nilson M. (2001) ‘Agency costs of controlling minority shareholders’ Journal of Financial and Quantitative Analysis, 38, pp.659-719. 18. De Angelo, H. and Masulis R. (1980) ‘Optimal capital structure under corporate and personal taxation’, Journal of Financial Economics, 8, pp. 3-29. 19. Faccio, M., Lang L.H.P and Young L. (2001a) ‘Debt and corporate governance’, Meetings of Association of Financial Economics in New Orleans. 20. Faccio, Mara, Larry H.P. Lang, and Young, L. (2001b) ‘Dividends and Expropriation’, American Economic Review, 91, 54-87. 21. Faccio, Marra. and Lang. Larry H.P (2002) ‘The ultimate ownership of Western European Corporations’, Journal of Financial Economics, 65, 365-395. 22. Fontaine, P., and Nijokou, C. (1996) ‘Les déterminants de la structure financière : Une comparaison internationale’, Banque et Marchés, n°24, Septembre –Octobre. 23. Frank, M., and Goyal V. (2003) ‘Testing the pecking order theory of capital structure’, Journal of Financial Economics, 67, pp. 217-248. 24. Friend, I. and Lang L. (1988) ‘An empirical test of the impact of managerial self interest on corporate capital structure, Journal of Finance, 43(2), pp.271-281. 25. Gallo, M. A. and Vilaseca, A. (1996), ‘Finance in family business’, Family Business Review, 9(4), pp. 387-401. 26. Gomes A., and Novaes W. (2001) ‘Sharing of control as a corporate governance mechanism’, PIER Working paper 01-029, University of Pennsylvania Law School. 27. Harijono, Ariff M., and Tanewski G.A, (2004) ‘The impact of family control of firms on leverage: Austarlian evidence’, Working paper, Faculty of Business and Economics, Monash University. 28. Harris, M. and Raviv, A. (1988), ‘Corporate Control Contests and Capital Structure’, Journal of Financial Economics, 20, pp. 55-86. 29. Hausman, J. (1978) ‘Specifications tests in econometrics’, Econometrica, 46, pp.297-355. 30. Hsiao, C. (1986) Analysis of panel Data ( England: Cambridge University Press, Cambridge) 31. Jensen, M., and Meckling, W. (1976) ‘Theory of the firm: Managerial behavior, agency costs, and ownership structure’, Journal of Financial Economics, 3, pp. 305-360. 32. Jensen, G., Solberg, D. and Zorn T. (1992) ‘Simultaneous determination of insider ownership, debt and dividend policies’, Journal of Financial and Quantitative Analysis, 27(2), pp.247-263. 33. Johnson, S., La Porta R., Lopez-de-Silanes, F. and Shleifer A. (2000) ‘Tunneling’ American Economic Review Papers and Proceedings, 90, 22-27. 243 34. Kim,W and Sorensen, E. (1986) ‘Evidence on the impact of agency costs of debt on corporate debt policy’, Journal of Financial and Quantitative Analysis, 21(2), pp.131-144. 35. Kremp, E., Stöss E., and Gerdesmeier D. (1999) ‘Estimation d’une fonction d’endettement. Résultats à partir d’un panel d’entreprises françaises et allemandes’, Projet de recherche commun de la Deutshe Bundesbank et de la Banque de France, sous la direction de Sauvé, A., et Scheuer M. 36. La Porta, R., Lopez-de-Silanes, F. and Shleifer, A. (1999) ‘Corporate ownership around the world’, Journal of Finance, 54(2), pp. 471-518. 37. Lehman, E., and Weigand J. (2000) ‘Does the Governed Corporation Perform Better? Governance Structures and Corporate Performance in Germany’, European finance Review, 4, 157-195. 38. Le Maux, J. (2003) ‘Les bénéfices privés: une rupture de l'égalité entre actionnaires’, Revue Finance, Contrôle, Stratégie, 6, pp.63-92. 39. Mishra, C.S. and McConaughy, D.L. (1999) ‘Founding family control and capital structure: The risk of loss of control and the aversion to debt’, Entrepreneurship Theory and Practice, summer, pp.53-64. 40. Morck, R., Shleifer, A. and Vishny, R. (1988) ‘Management ownership and market valuation: an empirical analysis’, Journal of Financial Economics, 20, pp. 293–315. 41. Myers, S.C. and Majluf, N.S. (1984) ‘Corporate financing and investment decisions when firms have information that investors do not have’, Journal of Financial Economics, 13(2), pp. 187-221. 42. Sraer, D., and Thesmar D. (2004) ‘Performance and behaviour of family firms: Evidence from the French stock market’, CEPR Discussion Paper No4520. 43. Stulz, R. (1988) ‘Managerial control of voting rights: Financing policies and the market for corporate control’, Journal of Financial Economics, 20, pp. 25-54. 44. Rajan, R. and Zingales, L. (1995) ‘What do we know about capital structure? Some evidence from international data’, Journal of Finance, 50, pp.1421-1460. 45. Titman, S. and Wessels, R. (1988) ‘The determinants of capital structure choice’ Journal of Finance, 43, pp.1-19. 46. Yeh Y. (2005) ‘Do Controlling Shareholders Enhance Corporate Value’ Corporate governance, 13, pp.313-325. 47. Villalonga B., and Amit R. (2006) ‘How do family ownership, management, and control affect firm value’, Journal of Financial Economics, 80, pp. 385–417. 244 ASAC 2008 Halifax, Nova Scotia David Rankin (student) Faculty of Law University of Toronto Jun Yang School of Business Administration Acadia University Eric Wang School of Business Athabasca University RECONSTRUCTING THE HISTORICAL PERFORMANCE OF MERGED ECOMMERCE MUTUAL FUNDS 1 In the current Canadian regulatory environment, a mutual fund company is able to delete the history of an underperforming fund from public record by merging it with another fund. This paper addresses the effects of this rule on the super-concentrated group of Canadian mutual fund mergers that involved eCommerce portfolios between 2001 and 2004. Introduction In the late 1990s and the first part of the year 2000, eCommerce stocks were hot, and, much like their American compatriots, Canadian mutual fund companies wasted no time creating portfolios to take advantage of the heat. Between November 1998 and December 2000, a total of 17 ‘specialty’ Internet sector mutual funds appeared on the Canadian market. Through these funds, Canadians investors funnelled billions of dollars into the Internet sector before the bubble burst in late 2000. Before the lion’s share of these funds were dissolved, terminated, merged, or repositioned, hundreds of millions of dollars of investors’ capital simply disappeared into the swill of market forces.2 While the industry may not want to talk about this unfortunate time in investment history, several lessons can be learned about the functioning of the Canadian legal system as it pertains to mutual funds. When things start to go wrong, mutual fund companies in Ontario – where most of the major players are located – are given a considerable amount of flexibility with regard to how they deal with underperforming funds. Subject to unit-holder approval, National Instrument 81-102 (the main set of 1 We are grateful for comments from participants of Atlantic School of Business 2006 conference in Sackville, NB. Jun Yang (the corresponding author) can be reached at (902) 585-1791 or [email protected]. 2 Consistent with previous studies that address deceased mutual funds, the abovementioned failed funds are referred to as ‘non-survivors’ and the funds into which they are merged are identified as ‘survivors’. 245 regulations surrounding mutual funds as adopted by the Ontario Securities Commission) permits mutual funds to be terminated, renamed, or merged into other funds at the discretion of management. While these provisions are necessary given that some mutual funds eventually out-live their intended life-spans and have to be written off, several fine points in the sections relating to fund mergers allow companies to make any embarrassing performance that may be on their record books all but disappear. Entirely in accordance with NI 81-102, after two funds merge, the past performance of the underperformer is virtually eliminated from the public record. Not only do mergers increase the asset base of the surviving fund and prevent current unit-holders from liquidating, but they also allow the company to hide its funds’ poor historical performance. Post-merger prospectuses and sales communications are not required to include non-survivor performance, and, 12 months after the date of the merger, the survivor is permitted to report performance as if it were an entirely new fund. This does, of course, create a discrepancy. In the prospectus of the surviving fund, the listed date of inception and the first indication of performance may be several years off. NI 81-102 essentially allows companies to ‘start fresh’ and disregard the cumulative performance their fund managers – while retaining all of their original investors’ capital. To illustrate the effect on this practice on investor perceptions of a fund, take the Royal eCommerce Fund (REF) and its partner fund, the Royal Global Technology Sector Fund (RGTSF) – which together managed $30.30 million at the time of their merger in July 2002. At the end of 2000, the REF and the RGTSF posted cumulative losses of -53.60% and -41.8% respectively (Black, 2002). In 2002 the RGTSF was merged into REF and the REF was renamed the Royal Global Technology Sector Fund. Four years of extremely mediocre performance later, the merged fund claimed to have a positive compound return since inception of -0.84% on July 12th, 2005 – which is mathematically impossible given the fund’s complete history (GlobeFund, 2005). While this performance communication is entirely legal given the current wording of NI 81-102 §15.9, it effectively allows companies to ‘sweep underperformers under the rug’ when they are performing below company standards. Given the potential for mutual fund companies to mislead investors by using the provisions of NI 81-102 to cover up past poor performance, the purpose of this paper is to propose a technique to reporting merged fund performance that does not allow companies to cover up poor performance. The eCommerce funds mentioned above that were merged out of existence between 2001 and 2004 in Canada are used as a sample in an empirical analysis using the proposed technique to determine whether or not this has a material impact on the communication of performance. Literature Review Despite the wide body of knowledge regarding the underlying structural causes and macroeconomic effects of the high-tech boom and bust, very little research has been done to address its impact on the investment industry – with the exception of a high profile study of hedge funds by Brunnermeier and Nagle (2004) that concludes that hedge funds were “riding the technology bubble”. While this work is relevant to a study of the investment industry, however, mutual funds and hedge funds 246 are too structurally dissimilar to draw any viable conclusions. The eCommerce funds addressed in this paper have not been the focus of very much academic research. Many aspects of the Canadian regulatory environment have been studied in great detail with regard to how they affect the mutual fund industry. Most importantly, Erlichman (2000) conducted a study of mutual fund governance and concluded that an amendment to the National Instrument was required to ensure that mutual fund shareholders had a right to oversee management via a board of governors. Erlichman’s proposal formed the basis of an entirely new National Instrument (NI 81-107) currently being recommended to the provincial SECs by the Canadian Securities Administrators (CSA). Erlichman’s work is relevant to this paper – as it addresses the misaligned objectives of management and unit-holders. As indicated above, research seems to suggest that mutual fund managers change an underperforming fund’s name or merge it with a better performer for their own ends rather than for the benefit of shareholders. As noted by Cooper et al. (2005), a mutual fund can gain a positive cumulative abnormal flow of investors’ capital simply by changing its name to something more ‘fashionable’. When this is done in conjunction with a merger, it benefits the company (by generating more investment flows and making the sponsor’s portfolio of funds look more attractive) while it hurts unit-holders (in that it restricts the information available to new investors). While neither Erlichman (2000) nor the proposed NI 81-107 address this issue specifically, the boards of governors that they recommend may (given more research on this issue) curb this behaviour by holding management to account. With regard to the importance of favourable raw returns for mutual fund companies, a number of interesting studies have emerged to indicate that issuers often engage in questionable activities to push up quarterly and annual compound rates of return. According to a study by Carhart et al. (2002), there is solid evidence that a number of major mutual funds have been known to “paint the tape” (i.e. aggressively trade stocks that the fund already owns just before the end of each fiscal quarter to push up prices) to increase their quarterly raw performance by as little as 0.5%. While this does not benefit investors, it does illustrate just how critical favourable raw performance can be to mutual fund companies.3 The techniques employed in the empirical sections of this paper are derived from the study of the well documented phenomenon of survivorship bias (e.g., Grinblatt & Titman, 1989; Brown et al., 1992; Malkiel, 1995; Brown & Goetzmann, 1995; Elton et al., 1996; Carpenter & Lynch, 1999; Carhart et al. 2002). Survivorship bias – a concept originating in the late 1980s – refers to the problem of sampling bias in past studies of mutual fund performance such as the classic work of Jensen (1968). In these studies, only funds that were in existence at the end of the investigation period were included in the sample. Funds that were terminated or merged because of poor performance were not included – resulting in aggregate α (excess performance) values that were biased by the non-inclusion of non-survivors. While estimates for the level of this bias vary greatly, arguably the most sophisticated study of survivorship bias was completed by Elton et al. (1996) – where survivorship bias between 1976 and 1993 was estimated to be 77.16 basis points.4 The key to this study was the use of a “follow the money” technique for tracking the capital invested in non-survivors before and after the merger. By taking to account the terms of mergers, Elton at al. (1996) were able to account for risk differentials between non-survivors and the funds into which they merged. More importantly, however, they made a case for the reinvestment assumption 3 It should be noted that this practice is technically illegal. This being said, however, according to Carhart et al. (2002), it is done surprisingly often in the American jurisdiction. There is no estimate of the prevalence of this practice in Canada. 4 This is done by using a three-index model and a re-investment assumption. An estimate of 90.69 basis points was determined using a three-index model and no re-investment assumption. 247 critical to the study of non-survivors. Under this framework, Elton et al. (1996) were able to calculate α values for non-survivor funds from the date of their inceptions until the end of the sample period. This was done by making the reasonable assumption that all the investors in the non-surviving fund accepted units of the merged fund and did not opt to take their portion of net asset value (NAV) in the form of cash, which is something that fund sponsors go to great lengths to ensure. As this paper focuses on nonsurvivors, the reinvestment assumption of Elton et al. (1996) forms a crux of the empirical analysis. Methodology for Empirical Analysis With regard to pre-merger reconstruction, the logical technique would be to calculate the monthly percentage returns of a combination portfolio that consists of all of the funds that were eventually merged into a survivor. To maintain the per unit price of the continuing fund, units of the non-survivor are converted at an appropriate ratio. As it is rarely ever the case that all of a product fund’s components had equal asset bases prior to a merger, a weighted average must be taken to adjust for relative fund size. Because of the nature of mutual fund unit prices, component funds should be weighted by the relative number of their outstanding units. This weighted average NAV (WANAV) can be used to determine the returns of a combination fund. Appendix A shows the details of the proposed methodology on reconstructing the historical performance of merged funds. Once the pre-merger performance records of the sample funds are all reconstructed, this paper will use the data to identify any discrepancies that may exist between how the sample mutual funds currently convey their past performance and how they would report their historical performance if the weighting technique outlined above was employed. Because mutual fund companies are only legally required to report their raw returns (usually in terms of annual compound rates), this paper will put a heavy influence on both yield since inception (YSI) and compound annual rate since inception. As noted by Carhart, et al. (2002), mutual fund companies will often go to extreme lengths to upwardly bias the returns reported in their prospectuses (and communicated through their sales representatives). Prospectus statistics are therefore critical to the analysis. By merging during a recovery in securities prices, it is hypothesised that a fund can capture the most favourable YSI and compound annual rates. It furthermore allows it to conceptually truncate below industry average compound rates. Both of these behaviours are tested for in the empirical analysis of this paper. In addition, because mutual funds are often rated by institutions (such as Morningstar and GlobeInvestor) on measures other than raw returns, this paper will also describe the pre-merger, postmerger, and full-history fund in terms of a number of performance and risk measures to determine whether there is a notable difference in the way post-merger mutual fund information is communicated in today’s regulatory environment versus how it would be reported if pre-merger performance was taken into account. Fund specific raw-return data used in this study was acquired from different sources depending on availability. Monthly percentage returns of non-survivor funds were calculated from the net asset value per unit (hereafter referred to as NAV) in archived issues of the Globe and Mail between the fund’s inception and the date of its merger. Monthly returns for surviving funds were taken from the Gold Investor archives of GlobeFund.com. See Appendix B for the merger families. 248 Empirical Analysis Pre- and Post- Merger Performance Before addressing the effects of weighted averaging on pre-merger performance, it is interesting to note the differences between pre- and post- merger performance and risk. As shown in Table 1, after a merger mutual funds report higher performance and lower risk than their original component funds. Without exception, reported post-merger raw performance for each product fund is higher than the premerger performance of any of its components. The differences can be staggering, particularly with regard to yield since inception (YSI). Even more importantly, there was a major discrepancy between the preand post- merger average annual compound rate of return, which must be reported in the prospectuses of mutual funds. Due to the highly competitive nature of mutual fund market fund management companies go to great lengths to report favourable returns, and it appears as if merging mutual funds may be one technique to omit unfavourable annual compound returns from future publications. The average for every other performance metric, furthermore, is higher for the post-merger funds when their histories are truncated than for any of their components. In addition, the average risk related metrics are lower for each post-merger fund than their pre-merger components. Table 1: Pre- and Post-Merger Metrics Pre-Merger Post-Merger -66.99% 11.98% YSI(raw) -34.78% 6.09% i(inception) -0.9762 -0.8193 α(J) -0.4339 0.0790 Sh -1.3030 0.4298 Tr β(J) 2.0462 1.4649 11.61% 6.36% σ Where YSI(raw)=yield sine inception, i(inception)=annual compound return since inception, α(J)=excess return based on Jensen’s model, Sh=Sharpe Ratio, Tr=Treynor’s Measure, σ=standard deviation (variability), and β(J)=beta (volatility) based on the Jensen regression. TSX Composite Index is used as the market index. Reconstructed Histories After averaging the full and truncated metrics of the eCommerce merger families (using the reconstruction technique outlined in the methodology section), it was found that, in general, performance per unit of variability, performance per unit of volatility, yield-since-inception, and compound annual return since inception appear higher for the post-merger funds (as shown in Table 2). In addition, volatility and variability are both reported at reduced levels when pre-merger performance was abridged. 249 Table 2: Full and Truncated History Metrics Full-History Post-Merger -58.16% 11.98% YSI(raw) -16.90% 6.09% i(inception) -1.2406 -0.8193 α(J) -0.0918 0.0790 Sh -0.5405 0.4298 Tr β(J) 1.9909 1.4649 11.81% 6.36% σ Mutual fund companies always try to make the historical performance of their portfolios seem favourable (ostensibly to make ill-informed investors think that their funds are ‘hot’). It is paramount that they are able to report performance that is at least comparable to group averages. When the pre-merger performance of the sample funds was reconstructed, it was found that in 2006 the 5-year annual compound return on Royal Global Technology Sector Fund (RGSTF), Mackenzie Universal Internet Technologies Fund (MUETCC), and Investors Group Global Science and Technology Fund (IGGSTC) were all considerably below the group average (as determined by GlobeInvestor). This represents a discernable blemish for these funds when their pre-merger histories are reconstructed. Table 3 compares the funds’ period compound returns with group averages. Altamira eBusiness fund, the only eCommerce fund that did not undergo a merger, is included to show the importance of levelling the playing field. 1 –year 3-year 5-year since inception Table 3: Period Annual Compound Rates MUETC IGGST Alt. RGSTF C C SSFTF eBusi 4.64% -2.91% -7.50% * -7.73% 3.30% 6.44% 2.01% * -3.30% -19.53% -18.53% -16.55% * -14.58% -24.50% -23.46% -2.49% -17.15% -6.55% Group Avg. -2.15% 5.87% -8.46% n/a † RGTSF - Royal Global Technology Sector Fund; MUITF - Mackenzie Universal Internet Technologies Fund; IGGSTF - Investors Group Global Science and Technology Fund; SSFTF - Sentry Select Focused Technologies Fund. * Not calculated as the 1-year, 3-year, and 5-year annual compound rates refer to different time periods as the other funds and group averages (as the product fund was terminated in May, 2004). In addition to having notably poor performance past the three-year mark, each of the above funds reports a negative annual compound return since inception. On average, the eCommerce funds in the sample returned -16.9% per year since they were formed, compounded annually. When the pre-merger performance of these funds was truncated, however, they report an average annual return of 6.09%, compounded annually. It is apparent that the fund mergers in the sample took place around the time of a rebound in securities prices. 250 All of this being said, however, it is possible to argue that the post-merger funds are more than simply combinations of their components, and consequently it does not make sense to pool pre-merger data with post-merger data. To address this concern, a standard Chow test was performed for each merger family. For three of the merger families encompassing 12 of the 14 funds addressed in this paper, there was no evidence of a parametric shift - meaning that the pre- and post- merger data sets were substantially similar. For the RBC merger, however, there seems to be a notable difference in how the fund performed after the merger date, as determined by the residual sum of squares (RSS) from the pooled data versus the sum of the RSS of the pre- and post- merger periods taken separately. Conclusion In the May 2002 merger documents of the continuing fund MUETCC, the preparing lawyer explicitly stated that “in accordance with regulatory requirements, no performance information is shown for the Fund, as any past performance information prior to the date of the acquisition of these assets would be misleading” (Behrman, 2001). While the lawyer was correct that including the past performance of just the continuing fund (the survivor) would be confusing, the omission of any sort of indication of pre-merger performance is equally detrimental to investors. As shown in this paper, monthly (or daily or even yearly) returns of a combination of component funds before a merger can be calculated using a weighted average based on the number of units outstanding. Using this technique, the eCommerce mutual funds in the sample appear considerably different when their pre-merger histories are taken into account, both in terms of prospectus reporting and in the financial metrics compiled by investment institutions. Even if mutual fund companies are not engaging in merger activities for the purpose of covering up unfavourable statistics, it is possible to use a merger as a vehicle to achieve more favourable prospectuses. Without exception, the eCommerce mergers sampled in this paper reported higher compound returns after their histories were truncated. It is asserted here that mutual fund companies should be legally required to maintain and report their merged funds’ pre-merger histories using the reconstruction technique outlined in this paper, or, at the very least, separately for each component (if there is a structural shift in return metrics as determined by a Chow Test). The marginal cost of such a provision would be relatively small, and it would unquestionably add informational value to investors in the form of fuller disclosure. Appendix A: Methodology Details With regard to pre-merger reconstruction, the logical technique would be to calculate the monthly percentage returns of a combination portfolio that consists of all of the funds that were eventually merged into a survivor. As it is rarely ever the case that all of a product fund’s components had equal asset bases prior to a merger, a weighted average must be taken to adjust for relative fund size. Because of the nature of mutual fund unit prices, component funds should be weighted by the relative number of their outstanding units. Weighting the NAVs collected from archived periodicals by units outstanding is mathematically equivalent to summing the asset bases of both funds and dividing by the sum of the 251 outstanding units.5 This weighted average NAV (WANAV), while not useful in any absolute sense, can be used to determine the monthly returns of a combination fund. It can be easily shown that the percentage change in WANAV is equal to the percentage change in the value of the holdings of the funds that it combines. Making the assumptions delineated in the body of the paper and defining γ as the asset base of a fund and u as the number of units outstanding, one can show that for two funds (A and B) %ΔγAB|u’=u* = %ΔWANAV: Therefore %ΔWANAV = %ΔγAB|u’=u. *u’ = u because we are only interested in the percentage change resultant from changes in the net holdings of the mutual funds, not changes in the asset base resulting from investors either buying or liquidating units. **Because of the simplifying assumption that the ratio between units outstanding between the funds does not change. Note that WANAV on any particular date is not equivalent to what NAV would be if the two funds merged. This is because of the nature of a merger. To maintain the per unit price of the continuing fund, units of the non-survivor are not exchanged with units of the survivor on a 1:1 basis. They are rather converted at an appropriate ratio as to maintain the per-unit price of the continuing fund. WANAV The proof for why this works is extremely straightforward as it involves nothing more complicated than simple fractions. Remembering that the net asset base of a fund = number of units outstanding * NAV, the total net asset value per unit of a combination of two funds would be: 5 where u is the number of units outstanding. 252 is only calculated to capture the aggregate monthly percentage change in the value of a number of mutual funds’ holdings. By weighting the monthly per-unit prices of each component fund as described above, the return that would have been had on a merged fund can be estimated prior to the merger date. This is, of course, assuming that the sample funds do not substantially alter their investment structures after a merger and that the ratio of the units that each fund has outstanding relative to each other remains somewhat constant. It is not necessary that every fund has the same number of units outstanding throughout the entire period, but it is necessary that increases and decreases happen proportionally. In addition to this, the technique outlined above does not account for transaction costs. Appendix B: Merger Families Merger Family # 1 2 3 4 Non-Survivor(s) Survivor Royal e-Commerce Fund Keystone Altamira e-Business Capital Class Keystone Altamira Science and Technology Capital Class Keystone Altamira RSP e-Business Fund Keystone Altamira RSP Science and Technology Mackenzie Universal RSP Emerging Technologies Fund Mackenzie Universal Internet Technologies Fund Investors Global e.Commerce Class Sentry Select E-Commerce and Internet Technology Fund 1999 Sentry Select Internet Technology Fund Royal Global Technology Sector Fund Mackenzie Universal Emerging Technologies Capital Class Investors Global Science & Technology Class Sentry Select Wireless Communications Fund References Behrman, B.D. (2001). Application to the securities regulatory authorities of each province and territory of Canada (the “authorities”) under mutual reliance review system for exemptive relief (“MRRS”) for securities legislation relief and exemption application under national instrument 81-102 (“NI 81-102”). [November 25, 2005 from SEDAR] 253 Black, J. (2002). Application under section 5.5(1)(b) of National Instrument 81-102 (the “Instrument”) for approval of the merger of each of the Terminating Funds into each of the applicable Continuing Funds under the Mutual Reliance Review System. [November 23, 2005 from SEDAR] Brown, S.J., & Goetzmann, W.N. (1995). Performance persistence. Journal of Finance, 50(2), 679-698. Brown, S.J., Goetzmann, W.N., Ibbotson, R.G., & Ross, S.A. (1992). Survivorship bias in performance studies. Review of Financial Studies, 5(4), 553-580. Brunnermeier, M.K., & Nagel, S. (2004). Hedge funds and the technology bubble. The Journal of Finance, 59(5), 2013-2040. Carhart, M.M., Kaniel, R., Musto, D.K., & Reed, A.V. (2002). Leaning for the tape: Evidence of gaming behavior in equity mutual funds. Journal of Finance, 57(2), 661-693. Carhart, M.M., Carpenter, J.N., Lynch, A.W., & Musto, D.K. (2002). Mutual fund survivorship. Review of Financial Studies, 15(5), 1439-1463 Carpenter, J.N., & Lynch, A.W. (1999). Survivorship bias and attrition effects in measures of performance persistence. Journal of Financial Economics, 54, 337-374. Cooper, M.J., Gulen, H., & Rau, P.R. (2005). Changing names with style: Mutual fund name changes and their effects on fund flows. Journal of Finance, 60(6), 2825-2858. Elton, E.J., Gruber, M.J., & Blake, C.R. (1996). Survivorship bias and mutual fund performance. The Review of Financial Studies, 9(4), 1097-1120. Erlichman, S.I. (2000). Making it mutual: Aligning the interests of investors and managers. Toronto: Canadian Securities Administrators. Fama, E.F., & French, K.R. (1992). The cross-section of expected stock returns. Journal of Finance, 47(2), 427-465. GlobeFund. (2005, November 30). Fund Profile: RBC Global Technology Sector. http://globefunddb.theglobeandmail.com/gishome/plsql/gis.fund_pro?fundname=RBC+Global+Technolo gy+Sector&pi_universe=PUBLIC_FUND&product_id= [December 7, 2005] Grinblatt, M., & Titman, S. (1989). Mutual fund performance: An analysis of quarterly portfolio holdings. Journal of Business, 62(3), 393-416. Jensen, M.C. (1968). The performance of mutual funds in the period 1945-1964. The Journal of Finance, 23(2), 389-416. Malkiel, B.G. (1995). Returns from investing in equity mutual funds. Journal of Finance, 50(2), 549-572. Mutual Funds, National Instrument 81-102 (2000). Sharpe, W.F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk. Journal of Finance, 19, 425-442. 254 Jacques Saint-Pierre (Professeur Titulaire) Chawki Mouelhi (Étudiant au Ph.D.) Faculté des sciences de l’administration Université Laval ASAC 2008 Halifax, Nouvelle-Écosse Sélection de portefeuilles et prédictibilité des rendements via la durée de l’avantage concurrentiel1 En tenant compte de la dynamique concurrentielle et de la circulation bidirectionnelle de l’information, cette recherche examine le contenu informationnel et le pouvoir prédictif de la durée de l’avantage concurrentiel sur les rendements boursiers. INTRODUCTION La finance et la stratégie ont été, pendant longtemps, développées en toute indépendance, de telle sorte qu’on parle d’une dichotomie entre les deux disciplines. L’analyse stratégique a essayé de fournir des réponses aux questions suivantes : Quels sont les avantages concurrentiels possédés par l’entreprise ? Combien du temps vont-ils durer ? Comment peut-on, dans un univers compétitif, les protéger, les conserver ou les transformer ? Toutefois, elle n’a pas essayé de les quantifier. L’importance accrue de l’approche de la création de valeur a constitué le point de départ du processus de convergence entre l’analyse financière et l’analyse stratégique, ou de manière plus large, entre finance et stratégie Rappaport et Mauboussin (2001) ont constaté que l’analyse de la stratégie concurrentielle est la clé d’une analyse consistante de la valeur actionnariale. D’après ces auteurs, les investisseurs ont besoin d’anticiper les variations dans la dynamique de création de valeur des firmes. Ces variations permettent de réviser les différents déterminants de la valeur, dont la durée de l’avantage concurrentiel de la firme (CAP : «Competitive Advantage Period»). Il convient de signaler que dans un marché financier efficient, le concept de la durée de l’avantage concurrentiel est étroitement lié à la valeur intrinsèque. En effet, dans un tel marché, tous les actifs sont évalués par l’utilisation de trois déterminants : Le rendement, le risque et la période durant laquelle la firme aura un rendement sur son capital supérieur à son coût du capital, i.e. la durée de l’avantage concurrentiel. Alors que les deux premiers déterminants ont été explicitement utilisés dans la construction de portefeuilles d’actions à travers le cadre habituel rendement-risque, la durée de l’avantage concurrentiel de la firme a été complètement ignorée. En pratique, la plupart des firmes utilisent pour leurs plans stratégiques, une période de prévision très courte. Celle-ci est habituellement différente de la durée estimée de leur avantage concurrentiel. Une telle pratique augmente considérablement le pourcentage de la valeur terminale dans la valeur intrinsèque de la firme, de telle sorte que le calcul de cette valeur devient hautement sensible aux hypothèses de croissance implicites au-delà de la période de prévision qui sont incorporées dans la valeur terminale. Plusieurs études ont montré théoriquement que le concept de la durée de l’avantage concurrentiel se manifeste clairement dans le mécanisme de création de valeur. Ces études reposent essentiellement sur la 1 Nous remercions M. Richard Grizzetti de Stern Stewart & Co., New York et M. Carl Simard de StockPointer, Montréal pour nous avoir fourni des données, les Professeurs Gilles Bernier et Stéphane Chrétien de l’Université Laval pour leurs commentaires et le LABVAL pour nous avoir fourni du financement. 255 pertinence de la CAP comme mesure approchée de l’avantage concurrentiel durable de la firme et sur l’hypothèse de la circulation bidirectionnelle de l’information (Mauboussin et Johnson (1997), Booth (1998), Mills (1998), Brealy et Myers (2000), Rappaport et Mauboussin (2001), Mills et Dahlhoff (2003)). Selon ces auteurs, une mesure issue du marché boursier comme la MICAP (Market Implied Competitive Advantage Period), pourrait être un bon indicateur de la CAP. Toutefois, ces études ne sont pas sorties de leur cadre théorique, et à notre connaissance, il n’y a aucune étude qui a abordé empiriquement la pertinence de la CAP quant à sa contribution dans la sélection de portefeuilles. Dès lors, l’objectif de la présente recherche consiste à mesurer et à analyser le contenu informationnel et la capacité prédictive de la variation du MICAP quant à la création de valeur boursière. Ce travail sera subdivisé en quatre sections : Dans une première section, nous donnons les diverses définitions de la CAP. Dans la deuxième, nous présentons une revue de la littérature concernant les principales études effectuées sur la CAP. La méthode du calcul du MICAP, l’échantillon, la période d’étude, la méthodologie et les hypothèses de cette recherche sont présentées dans la troisième section. Les résultats et leur analyse sont rapportés dans la quatrième section, suivie d’une conclusion. 1. Qu’est ce que la durée de l’avantage concurrentiel ? La durée de l’avantage concurrentiel est considérée par plusieurs auteurs comme étant le meilleur indicateur de la performance future de la firme. En effet, un avantage concurrentiel durable crée des barrières puissantes autour de l’activité de la firme, ce qui lui permet d’écarter les concurrents et de réaliser des performances économiques supérieures sur une longue période. Ainsi, la clé du succès à long terme d’une firme est largement déterminée par sa capacité de saisir un avantage concurrentiel, mais aussi de le conserver. La notion de durée de l’avantage concurrentiel a été introduite depuis plus de quarante ans par Modigliani et Miller (1961). Il s’agit de l’intervalle de temps durant lequel la firme présente un taux de rendement sur ses investissements additionnels qui excède son coût marginal du capital. Les forces concurrentielles conduiront les rendements vers le coût de capital à travers le temps, i.e. vers un équilibre concurrentiel tel que décrit dans les manuels d’économie industrielle. 2. La durée de l’avantage concurrentiel : une revue de la littérature Il convient de signaler que les études en finance sur la durée de l’avantage concurrentiel ne sont pas très nombreuses. C’est grâce à celles de Mauboussin et Johnson (1997) et Rappaport et Mauboussin (2001) que cette notion a été mise en lumière. Mauboussin et Johnson (1997) ont constaté qu’une grande partie de la performance des actions des firmes américaines est due aux anticipations du marché sur la CAP de ces firmes. D’après les deux auteurs la CAP représente un indicateur de performance courant et futur de la firme. Autrement dit, la longueur et le changement relatif de la CAP peuvent avoir un impact substantiel sur la valeur de l’action et sur la création de valeur pour les actionnaires. Une étude de Boston Consulting Group (2000) montre que l’écart, entre la valeur marchande et la valeur fondamentale des firmes les plus performantes au monde, a considérablement augmenté. BCG a appelé cet écart « la prime d’anticipation ». En effet, cet écart est perçu comme étant les anticipations du marché concernant les opportunités de croissance future de la firme. Autrement dit, on prévoit que les firmes, avec des « primes d’anticipations » élevées auront une bonne performance dans le futur proche et éloigné. En effet, les investisseurs anticipent que de telles firmes seront capables de générer des flux monétaires libérés qui ne sont pas encore justifiés par leurs actifs en place. Pour Rappaport et Mauboussin (2001) la valeur marchande de l’action incorpore l’anticipation du marché quant à la vraie durée de l’avantage concurrentiel de la firme. Les deux auteurs ont constaté que les « primes d’anticipations » 256 élevées sont dues essentiellement à la vision très longue du marché et que la plupart des firmes ont besoin d’au moins dix ans de création de valeur pour justifier leur prix courant. Mills et Dahlhoff (2003) ont constaté que la différence de la « prime d’anticipations » à travers les firmes s’explique par le concept de la CAP. En effet, ils ont considéré que l’approche par l’avantage concurrentiel est parfaitement appropriée pour déterminer la vraie valeur des firmes. Selon la théorie financière, seule la firme qui dispose d’un avantage concurrentiel par rapport à ses concurrentes est capable de générer des rendements supérieurs. Les deux auteurs ont constaté que l’étude de Mauboussin et Rappaport (2001) a enrichi considérablement le débat sur la pertinence du CAP. D’après Mills et Dahlhoff (2003), si les hypothèses de Mauboussin et Rappaport (2001) sont vraies, alors la CAP peut être utile pour la prise des décisions d’investissement. En plus, elle peut être un indicateur pertinent pour les dirigeants en les aidants à mieux comprendre comment les marchés financiers apprécient l’avantage concurrentiel de leurs firmes. Toutefois, les deux auteurs ont mentionné que la faiblesse majeure de ce nouveau concept réside dans le fait qu’il n’existe aucune évidence empirique qui prouve la relation entre la CAP et la valeur de la firme. Dans son étude annuelle sur les firmes créatrices de la valeur, BCG (2003) a analysé les performances financières et boursières de plus de 5000 firmes, les plus performantes au monde, afin d’identifier les déterminants de leur succès. Les résultats de cette étude ont montré que dans les industries en récession, seules les firmes possédant un avantage concurrentiel durable (donc une CAP longue) sont encore capables de créer de la valeur pour les actionnaires. Sur le plan pratique, le concept de la CAP gagne de plus en plus d’intérêt chez les analystes institutionnels et les gestionnaires de fonds de placement. À titre d’exemples, aux États-Unis, Morningstar Inc., propriétaire d’Ibbotson Associates Inc., une des plus importantes sociétés d’informations financières au monde, utilise maintenant, explicitement, le concept de CAP, sous le nom de « moat », dans sa méthodologie de recherche pour évaluer les entreprises. Il en est de même de Rochdale Securities LLC de New York qui utilise ce concept pour ses clients institutionnels. Au Canada, Stockpointer, qui fournit de l’information financière à plus de 300 analystes financiers, utilise explicitement ce concept. 3. Méthodologie et hypothèses de recherche 3.1. Méthode de calcul du MICAP Pour calculer la MICAP d’une firme, on se base sur la formule de Modigliani et Miller (1961). Les équations 1 et 2, ci-dessous, représentent le point de départ de notre méthode. Dans l’équation 1, on considère l’IICAP (Intrinsic Implied Competitive Advantage Period) comme étant la durée de l’avantage concurrentiel de la firme. On a supposé que l’IICAP des firmes est égal à 10 ans et ceci en se référant à Mauboussin et Johnson (1997). Le taux de rendement interne moyen attendu sur les nouveaux investissements stratégiques (R), qui sera un intrant dans l’équation 2, est une inconnue mathématique qui est obtenue en résolvant l’équation (1) après avoir entré les autres valeurs, c'est-à-dire la valeur intrinsèque (VI), le bénéfice d’exploitation net après impôts (BENAI), le coût moyen pondéré du capital (WACC) et le montant à être investi chaque année dans les nouveaux projets (I). Dans l’équation (2), on utilise la valeur marchande de la firme (VM), toute chose égale par ailleurs, pour calculer la MICAP de la firme. BENAI I ( R − WACC ) IICAP + WACC WACC (1 + WACC ) BENAI I ( R − WACC ) MICAP VM = + WACC WACC (1 + WACC ) VI = (1) (2) Dans ces deux équations, on suppose que la différence entre la valeur intrinsèque et la valeur marchande de la firme est due à la différence entre l’IICAP et de la MICAP. Enfin, il convient de signaler que notre méthode de calcul du MICAP se base essentiellement sur l’hypothèse de la convergence du 257 rendement marginal du capital investi vers le coût marginal du capital. Cette hypothèse constitue l’élément de base de la définition de la durée de l’avantage concurrentiel. Rappelons, que cette dernière correspond à l’année anticipée de la disparition de l’avantage concurrentiel de la firme, de telle sorte qu’à cette date le rendement marginal sur le capital est égal au coût marginal du capital et la valeur économique ajoutée (VÉA) de la firme atteint son maximum. Bien que la convergence du rendement marginal sur le capital vers le coût marginal du capital puisse s’effectuer de plusieurs façons, on a choisi la convergence linéaire afin de simplifier les calculs. 3.2. Hypothèses de recherche Une firme qui a une MICAP élevée est une firme qui dispose d’un avantage concurrentiel durable. C’est l’anticipation du marché quant à l’avantage concurrentiel futur des firmes qui explique en bonne partie les écarts des MICAP des différents titres. Il faut bien comprendre que la MICAP est une mesure de la compétitivité d’une firme telle qu’anticipée par le marché. Ce qui nous conduit à l’interrogation la plus importante : Comment la MICAP, qui n’est autre qu’une anticipation du marché quant à la durée de l’avantage concurrentiel d’une firme, pourrait être un bon indicateur de la performance courante et future et nous servir dans la construction de portefeuilles? Dans ce qui suit, on va répondre à cette question à travers les trois volets suivants : (1) D’abord, il convient de signaler que plusieurs autres indicateurs issus du marché ont été utilisés, depuis longtemps, pour la sélection des titres ayant le meilleur potentiel de rendement. Par exemple, le ratio cours-bénéfice, le momentum, le market-to-book, le q de Tobin …etc. L’avantage du MICAP réside dans le fait qu’il utilise plusieurs données à la fois. Pris isolément, la valeur marchande d’un titre, son coût de capital, son BENAI, son taux de rendement implicite,…etc., ne signifient pas grand-chose. Cependant, réunis ensemble, la MICAP donne une meilleure image, une synthèse de la situation. (2) Ensuite, la MICAP est fortement liée au concept de l’avantage concurrentiel durable. En effet, une firme qui possède un avantage concurrentiel durable présente normalement une CAP longue et vice versa. L’avantage concurrentiel durable est devenu pour les praticiens une question centrale, non seulement pour comprendre et expliquer l’hétérogénéité des performances des firmes, mais aussi pour prédire leurs perspectives de croissance. Selon Besanko, Dranove et Shanley (1996), un avantage concurrentiel permet à la firme de réaliser des performances supérieures à la moyenne de son industrie. C’est dans un tel cadre, que plusieurs auteurs, comme Mueller (1986), Geroski et Jacquemin (1988), Schohl (1990), Droucopoulos et Lianos (1993), Goddard et Wilson (1996) ont constaté empiriquement que l’avantage concurrentiel est un bon indicateur de la performance future de la firme. Bien qu’il soit difficile de quantifier parfaitement l’avantage concurrentiel d’une firme, on peut dire que la MICAP est un bon indicateur de celui-ci. Dès lors, plus la MICAP d’une action est élevée, plus les anticipations des investisseurs face à ce titre sont favorables. Ainsi, l’action sera généralement très prisée par le marché. (3) Enfin, on peut avancer l’hypothèse de la circulation bidirectionnelle de l’information et ses conséquences sur les décisions des dirigeants des firmes (Sunder (1989), Bommel (1997), Dow et Gorton (1997), Dye et Sridhar (2002), Luo (2005), etc.). Selon cette hypothèse, l’information circule dans les deux sens, de la firme vers le marché des capitaux et de celui-ci vers la firme. C’est à ce dernier sens qu’on s’intéresse le plus pour expliquer la possibilité qu’un indicateur issu du marché pourrait être un bon indicateur des rendements potentiels des titres. Le modèle de la CAP de Rappaport et Mauboussin (2001) repose largement sur l’hypothèse de la circulation bidirectionnelle de l’information. D’après Mills et Dahlhoff (2003), il y a quatre étapes dans le modèle de la CAP de Rappaport et Mauboussin (2001): (1) calculer la MICAP à partir de la valeur marchande courante de l’action, (2) comparer la MICAP avec l’IICAP, (3) analyser soigneusement tout écart positif ou négatif et (4) prendre, selon le résultat de cette analyse, les décisions adéquates pour maintenir ou améliorer cette durée de l’avantage concurrentiel. 258 C’est à ce moment là que l’information issue du marché aura un impact, d’abord sur la prise de décisions par les dirigeants des firmes et par la suite sur les rendements potentiels des titres financiers. La logique de Mills et Dahlhoff (2003) repose sur le fait que dans un marché efficient, le niveau des cours des titres reflète les anticipations conjuguées des investisseurs quant à la capacité des firmes de saisir un avantage concurrentiel et de le conserver pendant une période de temps bien déterminée (CAP). Autrement dit, le marché s’attend à ce que les firmes à MICAP élevé réalisent des performances économiques supérieures. Toutefois, il ne faut pas oublier que les investisseurs récompensent les firmes à l’avance. En effet, si les attentes des investisseurs ne se confirment pas dans la réalité, la détérioration de la performance au marché des titres concernés sera certaine. C’est la tâche des dirigeants de satisfaire les attentes du marché, ils doivent non seulement accroître la compétitivité de leur firme, mais, ils doivent aussi, mettre en considération les informations issues du marché lors de la prise de décisions. Ainsi, les hypothèses à tester dans ce travail sont les suivantes : H 1 : Les actions à forte variation positive du MICAP auront un meilleur rendement. H 2 : La variation du MICAP est une variable pertinente en termes de contenu informationnel pour expliquer la performance financière courante de la firme. H 3 : La variation du MICAP est une variable pertinente en termes de capacité prédictive quant à la valeur boursière. 3.3. L’échantillon et la période d’étude Notre étude concerne la période 2000 à 2004. Après avoir exclus, comme dans la plupart des études empiriques, pour conserver l’homogénéité de l’échantillon, les firmes des secteurs d’activités Finance, Assurance et l’Immobilier (2-Digit SIC: 60-67) de la base de données initiale de Stern & Stewart (2005), on a sélectionné un échantillon aléatoire de 80 firmes. Le tableau 1 classe les firmes de l’échantillon par secteur d’activité sur la base du code SIC 2-Digit. Les bases de données de Stern & Stewart (2005), Compustat, Datastream et StockPointer ont été utilisées pour extraire les données nécessaires à l’étude empirique. La taille de l’échantillon est limitée en raison (1) de l’importance du traitement qu’il faut apporter à chaque firme pour générer les données, (2) de la méthodologie de la mesure du MICAP qui exige au départ, comme il se doit, d’avoir des firmes qui ont un rendement marginal sur leur capital supérieur à leur coût marginal du capital. 3.4. Méthodologie de recherche Pour tester l’hypothèse 1 on a poursuivi les étapes suivantes : -1ère étape : Pour chacune des années 2000 à 2004, la MICAP de chacune des 80 firmes a été calculée, selon la méthode présentée précédemment. -2ème étape : Les firmes sont classées par ordre croissant non pas de leur MICAP, mais de leur variation de MICAP. La justification est la suivante : En classant les firmes par ordre de MICAP, les firmes qui appartiennent aux secteurs à faible croissance ne feront jamais partie des portefeuilles à MICAP élevé. Or, parmi ces firmes il y en a certainement plusieurs dont les actions constituent un bon achat. Ainsi, en sélectionnant les firmes sur la base de la variation du MICAP, les portefeuilles à fortes et faibles variations du MICAP seront constitués des firmes appartenant à des industries de toutes catégories. -3ème étape : À chaque fin d’années 2001, 2002, 2003 et 2004, cinq portefeuilles sont formés à partir des 80 firmes de notre échantillon de la manière suivante : le premier portefeuille contient les 16 firmes ayant les plus faibles variations du MICAP et le cinquième portefeuille, les 16 firmes ayant les plus fortes variations positives du MICAP. - 4ème étape : Pour chacune des années 2002 à 2005, le rendement annuel de chaque firme de notre échantillon a été calculé, comme suit : 259 Ri ,t = Pi ,t − P i ,t −1 + Di ,t Pi ,t −1 (3) Avec, - Ri ,t : Le rendement annuel du titre i pour l’année t. - Pi ,t : Le prix de fermeture à la dernière journée de transactions de décembre du titre i pour l’année t. - Di ,t : Dividende annuel du titre i pour l’année t. Nous avons considéré trois stratégies d’investissement. La première stratégie est la stratégie «Achat-détention». Cette stratégie consiste à acheter un des 5 portefeuilles (sélectionné sur la base de la ∆MICAP) à une année X et à le conserver pendant Y année(s). Étant donné que notre étude s’étend sur une période de 4 années (2002 – 2005), on peut dénombrer sur cette période 10 horizons de détention. Nous avons calculé le rendement annuel moyen qu’un investisseur aurait pu réaliser en appliquant cette stratégie sur chacun des 10 horizons envisageables. La deuxième stratégie est la stratégie «Ré-sélection chaque année». Cette stratégie consiste, par exemple à acheter le portefeuille 5 constitué en décembre 2001, à le conserver un an, puis à l’échanger en décembre 2002 contre le nouveau portefeuille à plus forte variation du MICAP constitué en décembre 2002, etc. Par conséquent, les firmes composant le portefeuille 5 en décembre 2002 ne sont pas nécessairement les mêmes que celles qui constituent le portefeuille 5 en décembre 2001. Encore une fois, nous avons calculé le rendement annuel moyen qu’un investisseur aurait pu réaliser en appliquant cette stratégie sur chacun des 10 horizons envisageables. La troisième stratégie consiste simplement à «l’achat de l’indice S&P 500». Nous avons calculé le rendement réalisé par l’indice sur chacun des 10 horizons envisageables durant la période 2002 à 2005. Par la suite, nous ferons références aux stratégies «Achat-Détention» et «Ré-sélection chaque année» par l’appellation stratégie 1 et 2, respectivement. Pour tester l’hypothèse 2, on a procédé comme suit : En premier lieu, on a utilisé la méthodologie de rangs pour examiner la relation qui peut exister entre la ∆MICAP et certaines mesures de la performance courante de la firme. L’avantage de cette méthode réside dans le fait qu’elle permet d’analyser la corrélation de ∆MICAP avec les autres indicateurs de performance sans passer par des tests statistiques. Pour cela, on a sélectionné les huit indicateurs de performance suivants : - ROC : Le rendement sur le capital investi. - EVAC : La valeur économique ajoutée normée par le capital investi au début. - MVAC : La valeur marchande ajoutée normée par le capital investi au début. - ∆EVAC : La variation de la valeur économique ajoutée normée par le capital investi au début. - ∆MVAC : La variation de la valeur marchande ajoutée normée par le capital investi au début. - ROA : Le rendement sur l’actif. - ROE : Le rendement sur les fonds propres. - ∆VENTES en % : La variation des ventes en pourcentage. En second lieu, on a examiné les coefficients de corrélation de Pearson entre la ∆MICAP et les indicateurs de performance sélectionnés, en tenant compte des données groupées («pooled data»). 260 Pour tester l’hypothèse 3, on a poursuivi les étapes suivantes : - 1ère étape : D’abord, on a classé les firmes par ordre croissant de leur ∆MICAP. Ensuite, on a construit, pour chaque année, cinq portefeuilles : le 1er portefeuille contient les 16 firmes ayant les plus faibles ∆MICAP, il s’agit du 1er quintile, soit 20% de l’échantillon, le 5ème portefeuille (5ème quintile) contient les 16 firmes ayant les plus fortes ∆MICAP. Le 2ème portefeuille, le 3ème portefeuille et le 4ème sont formés des 16 firmes chacun ayant des ∆MICAP intermédiaires. Enfin, on a examiné pour chaque portefeuille, le rendement annuel futur moyen, le bêta moyen, la taille moyenne mesurée par la valeur marchande de fonds propres, le ratio valeur marchande / valeur comptable (VM/VC) et le ratio Bénéfices / Cours. - 2ème étape : Afin d’examiner l’effet joint du ΔMICAP avec les autres variables (bêta, taille, VM / VC et le ratio Bénéfice / Cours), on a utilisé l’approche bidimensionnelle des variations dans les rendements annuels futurs moyens. Cette méthode permet de vérifier si les firmes à ΔMICAP élevée réalisent des performances supérieures par rapport aux firmes à ΔMICAP faible, après avoir contrôlé l’effet des autres variables. Par exemple, pour la ΔMICAP et le bêta, on a commencé par classer (par ordre croissant) les firmes en quartiles basés sur leur bêta .Ensuite, chaque quartile a été divisé, à son tour, en 4 portefeuilles en se basant sur l’ordre croissant des ΔMICAP. Enfin, on a examiné les rendements annuels futurs moyens des portefeuilles formés selon bêta et ΔMICAP. - 3ème étape : D’abord, on a examiné les coefficients de corrélation de Pearson entre la variable dépendante (le rendement annuel futur) et les variables explicatives (la ΔMICAP et les variables de contrôle) en tenant compte des données groupées. Par la suite, on a effectué les régressions sur données de panel des modèles suivants : Modèle 1 : RENDAFt +1 = α 0t + α 1t BETAi ,t + ε i ,t Modèle 2 : RENDAFt +1 = α 0t + α 1t ΔMICAPi ,t + ε i ,t Modèle 3 : RENDAFt +1 = α 0t + α 1t BETAi ,t + β 2 t ΔMICAPi ,t + ε i ,t Modèle 4 : RENDAFt +1 = α 0t + α 1t Ln(VM ) i ,t + ε i ,t Modèle 5 : RENDAFt +1 = α 0t + α 1t (VM / VC )i ,t + ε i ,t Modèle 6 : RENDAFt +1 = α 0t + α 1t ΔMICAPi ,t + α 2t Ln(VM ) i ,t + ε i ,t Modèle 7 : RENDAFt +1 = α 0t + α 1t ΔMICAPi ,t + α 2t (VM / VC )i ,t + ε i ,t Modèle 8 : RENDAFt +1 = α 0t + α 1t Ln(VM ) i ,t + α 2t (VM / VC )i ,t + ε i ,t Modèle 9 : RENDAFt +1 = α 0t + α 1t ΔMICAPi ,t + α 2t Ln(VM ) i ,t + α 3t (VM / VC )i ,t + ε i ,t Modèle 10 : RENDAFt +1 = α 0t + α 1t (B / C )i ,t + ε i ,t Modèle 11 : RENDAFt +1 = α 0t + α 1t ΔMICAPi ,t + α 2t (B / C )i ,t + ε i ,t Modèle 12 : RENDAFt +1 = α 0t + α 1t ΔMICAPi ,t + α 2 t Ln(VM ) i ,t + α 3t (B / C )i ,t + ε i ,t Modèle 13 : RENDAFt +1 = α 0t + α 1t BETAi ,t + α 2t ΔMICAPi ,t + α 3t Ln(VM ) i ,t + α 4t (VM / VC )i ,t + α 5t (B / C )i ,t + ε i ,t Avec, - RENDAF : Le rendement annuel futur, il est mesuré par le rendement total d’une année. - ΔMICAP : La variation du MICAP (i.e. ΔMICAPt = MICAPt − MICAPt −1 ) - VM/VC : Le ratio valeur marchande des fonds propres / valeur comptable des fonds propres - B/C : Le ratio bénéfice/cours. - ln(VM) : Le log népérien de la valeur marchande des fonds propres - BETA : Le risque systématique. 261 4. Résultats et interprétations 4.1. Résultats des tests de l’hypothèse 1 : Le tableau 2 rapporte les résultats de la stratégie 1, c'est-à-dire la stratégie «Achat-détention». Il indique les rendements annuels moyens des cinq portefeuilles pour les 10 horizons possibles. Le tableau 3 présente les résultats de la stratégie 2, c'est-à-dire la stratégie «Ré-sélection chaque année». Il indique les rendements annuels moyens des cinq portefeuilles pour les 10 horizons envisageables. Le tableau 4 présente les tests de différences entre les rendements de ces cinq portefeuilles (données groupées). Les tests de différence de la moyenne (t-test) et de la médiane (Wilcoxon-test) sont rapportés, respectivement, aux colonnes 2 et 3 du tableau. Ces tests montrent une différence significative de la moyenne et de la médiane des rendements du portefeuille 1 versus les rendements des portefeuilles 4 et 5, des rendements du portefeuille 2 versus les rendements des portefeuilles 4 et 5, et les rendements du portefeuille 3 versus les rendements du portefeuille 5. À partir des résultats obtenus pour les trois stratégies d’investissement, c'est-à-dire les rendements annuels moyens des cinq portefeuilles pour les 10 horizons possibles, on peut répondre aux questions suivantes : (1) Question 1 : Est-ce que l’application de la stratégie 1 ou de la stratégie 2 par l’achat de portefeuilles à forte ΔMICAP génère un rendement supérieur à ce qui aurait été réalisé à l’aide de portefeuilles à faible ΔMICAP? (2) Question 2 : Est-ce que la performance générée à l’aide des portefeuilles à forte ΔMICAP selon la stratégie 2 a été supérieure à celle obtenue par la stratégie 1 ? (3) Question 3 : Est-ce que la performance réalisée à l’aide des portefeuilles à forte ΔMICAP (que ce soit par la stratégie 1 ou la stratégie 2) a été supérieure à celle de l’indice S & P 500? D’abord, pour répondre à la question 1, on utilise un test simple qui consiste à vérifier sur combien d’horizons la performance des portefeuilles à plus fortes ΔMICAP a été supérieure à celle des portefeuilles à plus faibles ΔMICAP et ce, pour chacune des deux stratégies. Les tableaux 5 et 6 présentent les résultats de ce test pour la stratégie 1 et la stratégie 2, respectivement. Les résultats de ce premier test montrent que la performance des portefeuilles à fortes ΔMICAP tend à surpasser la performance des portefeuilles à faibles ΔMICAP et ceci, pour les deux stratégies. Par exemple, selon la stratégie 1, la performance des portefeuilles à plus fortes ΔMICAP (portefeuille 5) a surpassé la performance des portefeuilles à plus faibles ΔMICAP (portefeuille 1) sur 9 des 10 horizons envisageables durant la période 2002-2005. En outre, ce résultat est meilleur pour la stratégie 2 puisqu’on a trouvé que pour tous les horizons envisageables (10/10), la performance des portefeuilles 5 a dépassé celle des portefeuilles 1. D’une manière générale, on remarque que les résultats de la stratégie 2 sont plus intéressants que ceux de la stratégie 1. En effet, selon la stratégie 2 la performance des portefeuilles de rang supérieur quant à la ΔMICAP a surpassé la performance des portefeuilles de rang inférieur, 95 fois sur 100 cas possibles. Alors que pour la stratégie 1, le même résultat a été vérifié 90 fois sur les 100 cas possibles. Ces résultats supportent l’hypothèse 1 que nous avons énoncée précédemment : Les actions à forte variation positive du MICAP auront un meilleur rendement. Ensuite, pour répondre à la question 2, on a opposé les portefeuilles 3 à 5 (puisque ce sont les portefeuilles qui ont le mieux performé) de la stratégie 1 et la stratégie 2 sur chacun des 6 horizons envisageables durant la période 2002-2005 (étant donné que les quatre horizons d’une seule année sont identiques pour les deux stratégies). Dans le tableau 7, on présente les résultats de cette comparaison. À partir de ce test, on peut dire que la stratégie 2 a été plus performante que la stratégie 1 sur la majeure partie des horizons étudiés. Ce résultat est particulièrement vrai pour les portefeuilles 4 et 5. En outre, en examinant, les rendements moyens des portefeuilles 1 à 5 sur les 10 horizons possibles pour les deux stratégies, on a constaté que la stratégie 2 a été plus performante que la stratégie 1 pour les portefeuilles 4 et 5, il s’agit des portefeuilles à forte ΔMICAP. Quant à elle, la stratégie 1 a été plus performante que la stratégie 2 pour les portefeuilles 1 et 2, c'est-à-dire pour les portefeuilles à faible ΔMICAP. Pour le portefeuille 3 la performance moyenne de deux stratégies a été presque identique. À la lumière de ces 262 résultats, on peut dire, ceteris paribus, qu’un investisseur aurait intérêt à appliquer la stratégie 2 en achetant les portefeuilles à forte ΔMICAP, c'est-à-dire les portefeuilles 4 et 5. Enfin, pour répondre à la question 3, on a opposé les stratégies 1 et 2 à la troisième stratégie (achat du S&P 500). En effet, on a comparé la performance des portefeuilles 1 à 5 sélectionnés sur la base de la première et la deuxième stratégie avec la performance du S&P 500 sur les 10 horizons considérés. Le tableau 8 présente les résultats de cette comparaison. D’après ces résultats, on remarque que la performance des portefeuilles à forte variation du MICAP, pour les stratégies 1 et 2, a surpassé celle du S&P 500 de façon assez systématique. Ceci est particulièrement vrai dans le cas des portefeuilles 4 et 5 qui ont surpassé la performance de l’indice 10 fois sur les 10 horizons considérés. 4.2. Résultats des tests de l’hypothèse 2 : Le tableau 9 présente les résultats synthétiques de la méthodologie de rangs. Dans une première étape, on a divisé les 80 firmes classées selon l’ordre décroissant de leur ΔMICAP moyenne en quatre quartiles. Dans une deuxième étape, on a calculé le pourcentage des firmes, classées selon l’ordre décroissant des autres indicateurs de performance et qui appartiennent au même quartile issu du classement selon la ΔMICAP moyenne. Par exemple, pour le rendement sur le capital investi (ROC), le 20% de la deuxième colonne signifie que 4 firmes parmi les 20 firmes du 1er quartile du classement des firmes en ordre décroissant de leur ROC moyen, appartiennent au 1er quartile issu du classement des firmes selon l’ordre décroissant de leur ΔMICAP moyenne. Les résultats du tableau 9 montrent que la ∆MVAC est la plus reliée à la ∆MICAP. En effet, le pourcentage des firmes, classées selon l’ordre décroissant de leur ∆MVAC moyenne, et qui appartiennent aux mêmes quartiles issus du classement des firmes, selon l’ordre décroissant de leur ∆MICAP moyenne, est systématiquement plus élevé par rapport aux pourcentages des autres indicateurs de performance. Un autre indicateur de performance est, lui aussi, relié à la ∆MICAP, il s’agit de la MVAC. Un examen minutieux des résultats du tableau 9 montre qu’il existe une relation entre la ∆MICAP et la ∆EVAC. Toutefois, cette relation semble être moins importante que celles de la ∆MICAP avec la ∆MVAC, ou encore de la ∆MICAP avec la MVAC. Pour les autres indicateurs de performance, les résultats du tableau 9 ne permettent pas de tirer de conclusions quant à l’existence d’une relation entre ces indicateurs et la ∆MICAP. L’utilisation d’un test statistique à savoir le coefficient de corrélation de Pearson s’avère, dans un tel cas, indispensable. Pour cela, on a déterminé les coefficients de corrélation de la ∆MICAP et les indicateurs de performance sélectionnés (données groupées). Les résultats (non présentés ici en raison des contraintes de pages) montrent que la ∆MICAP est corrélée négativement et significativement au seuil de 1% avec le ROC. Le coefficient de corrélation est égal à -15,7%. Une association négative et significative au seuil de 1% est aussi trouvée entre la ∆MICAP et l’EVAC, avec un coefficient de corrélation de 17,6%. Les résultats indiquent des associations négatives mais non significatives entre la ∆MICAP et les autres indicateurs de performance (MVAC, ∆EVAC, ROA, ROE et ∆VENTES en %). Par contre, on a trouvé une relation positive et significative au seuil de 1% entre la ∆MICAP et la ∆MVAC, soit un coefficient de corrélation de 15,4%. Un tel résultat signifie que la ∆MICAP présente un contenu informationnel important pour expliquer la variation annuelle de la performance au marché de la firme, mesurée par la ∆MVAC. Ces résultats permettent d’accepter l’hypothèse 2. En effet, la relation positive et significative entre la ∆MICAP et la ∆MVAC et les relations négatives et significatives entre la ∆MICAP et le ROC, d’une part, et entre la ∆MICAP et l’EVAC, d’autre part, peuvent être expliquées par le fait que la compétitivité d’une firme présente deux dimensions (Porter (1985)) : Une première dimension d’ordre quantitatif. Dans un tel cas, la compétitivité est assimilée à la rentabilité. Et une deuxième dimension, d’ordre qualitatif, qui exprime une capacité ou encore une potentialité de croissance. Si on considère que la ∆MVAC est une mesure approximative des potentialités de croissance des firmes, on peut dire que c’est la deuxième dimension de la compétitivité qui l’emporte par rapport à la première dimension. Autrement dit, le marché apprécie la durée de l’avantage concurrentiel de la firme en donnant plus d’importance à ses 263 potentialités de croissance, plutôt qu’à la rentabilité de son activité courante. Ainsi, d’après les résultats de la méthode de rangs et les coefficients de corrélation de Pearson on peut dire que la ∆MICAP est une variable pertinente en termes de contenu informationnel pour expliquer la performance financière courante de la firme. En guise de conclusion, l’hypothèse 2 est acceptée. 4.3. Résultats des tests de l’hypothèse 3 : Le tableau 10 présente les caractéristiques des cinq portefeuilles constitués selon la ∆MICAP moyenne. On remarque que la ∆MICAP moyenne est positivement reliée au rendement annuel futur moyen. Ce dernier est une fonction croissante monotone de la ∆MICAP. En effet, le portefeuille de rang supérieur affiche toujours le rendement annuel futur moyen le plus élevé. D’après les résultats du tableau 10, il semble que la relation positive entre le rendement annuel futur moyen et la ΔMICAP moyenne est indépendante de l’effet des autres variables reconnues dans la littérature financière pour avoir un pouvoir explicatif et prédictif important sur les rendements potentiels des firmes (le bêta, le ratio VM/VC, le ratio B/C et la taille). Par exemple, les résultats de la dernière colonne du tableau 10 montrent que la taille moyenne (i.e. Ln(VM) moyen) décroît systématiquement entre le portefeuille 1 et le portefeuille 3, en passant de 9,230 à 8,440. Mais, ceci n’implique pas l’existence d’une relation négative entre la ΔMICAP moyenne et la taille moyenne, puisque cette dernière recommence à croître entre le portefeuille 3 et le portefeuille 5, en passant de 8,440 à 8,683. Pour mieux examiner l’effet joint de la ΔMICAP et les variables de contrôle, on a utilisé l’approche bidimensionnelle. Celle-ci, consiste à analyser la relation qui peut exister entre le rendement annuel futur moyen et la ΔMICAP moyenne après avoir contrôlé l’effet des autres variables, à savoir, le bêta moyen, le ratio VM/VC moyen, le ratio B/C moyen et la taille moyenne. Les résultats de ces études bidimensionnelles, présentés dans les tableaux 11, 12, 13 et 14, ont indiqué que la relation positive entre le rendement annuel futur moyen et la ΔMICAP moyenne persiste même après avoir contrôlé l’effet des autres variables. Par exemple, le tableau 14 montre qu’après avoir contrôlé l’effet du ratio B/C, la relation positive entre les rendements annuels futurs moyens et la ΔMICAP moyenne est robuste. En effet, indépendamment du niveau du ratio B/C dans les portefeuilles constitués, on remarque que le 4ème quartile présente toujours le rendement annuel futur moyen le plus élevé par rapport à ceux des quartiles de rang inférieur. Ceci est vrai aussi pour le 3ème quartile par rapport au 1er quartile et par rapport au 2ème quartile. À titre d’exemple, pour le portefeuille à faible-B/C et le portefeuille à B/C-3, le rendement annuel futur moyen est une fonction croissante monotone de la ΔMICAP moyenne. Pour tester la robustesse de cette relation positive entre la ΔMICAP moyenne et les rendements annuels futurs moyens, on a examiné, en premier lieu, les coefficients de corrélation de la variable dépendante (RENDAF) et les variables explicatives (la ΔMICAP et les variables de contrôle) en tenant compte des données groupées (résultats non présentés ici faute d’espace). Les principaux résultats ont indiqué que le RENDAF est corrélé positivement et significativement au seuil de 1% avec la ΔMICAP. Le coefficient de corrélation est égal à 26,6%. Une association positive et significative au seuil de 1% est aussi trouvée entre le RENDAF et le B/C. Par contre, on a constaté que le RENDAF est corrélé négativement et significativement au seuil de 1% avec la taille (ln(VM)). On a remarqué aussi une corrélation positive mais non significative entre le RENDAF et le BETA, et une corrélation négative et significative au seuil de 10% entre le RENDAF et le VM/VC. Les résultats ont indiqué une association négative mais non significative entre la ΔMICAP et le VM/VC, d’une part, et entre la ΔMICAP et le B/C, de l’autre part, avec des coefficients de corrélation de -9,0 % et -3,4% respectivement. On a constaté aussi une corrélation positive mais non significative entre la ΔMICAP et le BETA, avec un coefficient de corrélation de 7,65%. Par contre, la ΔMICAP est corrélée négativement et significativement au seuil de 10% avec la taille. 264 En second lieu, on a régressé sur données de panel les modèles 1 à 13. Le tableau 15 rapporte les résultats de ces régressions. Pour le modèle 1, les résultats montrent que le bêta présente un pouvoir prédictif faible des rendements annuels futurs. En effet, le R 2 ajusté obtenu est très faible, soit 0,31%. Ce résultat ne rejette pas totalement le MÉDAF, mais signifie que l’exposition au risque de marché, représentée par le coefficient bêta dans le MÉDAF, n’est pas le seul facteur explicatif des rentabilités boursières. Lorsqu’on a régressé la ∆MICAP, seule, sur les rendements annuels futurs (modèle 2), on a trouvé un coefficient positif et significatif au seuil de 1%. Le R 2 ajusté obtenu est nettement plus élevé que celui du modèle 1, soit 6,81%. Dans le modèle 3, on a régressé le bêta et la ∆MICAP ensemble sur les rendements annuels futurs. On a trouvé des coefficients positifs et significatifs au seuil de 5% pour le bêta et au seuil de 1% pour la ∆MICAP, le R 2 ajusté a enregistré une faible amélioration par rapport au modèle 2, en passant de 6,81% à 6,86%. Les résultats des modèles 1, 2 et 3 montrent que la ∆MICAP est un facteur explicatif des rendements boursiers futurs nettement meilleur que le bêta. Pour estimer conjointement le pouvoir explicatif de la taille et la ∆MICAP, on a régressé dans le modèle 6, ces deux variables sur les rendements annuels futurs. Les résultats montrent que le coefficient de la taille est négatif et significatif au seuil de 1%, alors que le coefficient de la ∆MICAP est positif et significatif au seuil de 1%. En plus, le R 2 ajusté obtenu est nettement plus élevé, soit 9,19%. La régression du ratio VM/VC et la ∆MICAP sur les rendements annuels futurs (modèle 7) donne un R 2 ajusté égal à 7,08%. Toutefois, le coefficient du ratio VM/VC est toujours négatif et non significatif, alors que le coefficient du ∆MICAP a conservé son signe positif et sa signification au seuil de 1%. Dans le modèle 9, on a regroupé les trois variables explicatives ensemble, c'est-à-dire, la ∆MICAP, la taille et le ratio VM/VC. Les résultats montrent qu’il n’y a pas des changements au niveau des signes et de la signification des coefficients de ces trois variables. Ce qui signifie que les deux premières variables (la ∆MICAP et la taille) sont toujours des facteurs explicatifs des rendements annuels futurs. Alors que le ratio VM/VC ne semble pas avoir un pouvoir prédictif des rendements espérés. Le R 2 ajusté obtenu pour le modèle 9, est à l’ordre de 9,43. L’introduction de la ∆MICAP et le ratio B/C dans un même modèle de régression (modèle 11) permet d’obtenir un R 2 ajusté important soit 11,33%. Les coefficients de deux variables sont toujours positifs et significatifs au seuil de 1%. Dans le modèle 12, on a régressé la taille, le ratio B/C et la ∆MICAP sur les rendements annuels futurs. Les résultats montrent que les coefficients des trois variables affichent les signes attendus et gardent leur signification au seuil de 1%. En outre, le R 2 ajusté obtenu est le plus élevé jusqu’à maintenant, soit 12,51%.Enfin, on a régressé le modèle global (modèle 13), qui englobe les cinq variables explicatives. On a obtenu un R 2 ajusté de 13,14%, c'est-à-dire légèrement supérieur à celui du modèle 12, soit 12,51%, ainsi qu’à celui du modèle 11, soit 11,33%. En outre, on sait que le modèle 12 englobe trois variables explicatives et le modèle 11 englobe uniquement deux variables explicatives. Ainsi, on peut dire que le R 2 ajusté élevé du modèle global peut être dû au nombre élevé des variables explicatives, puisque économétriquement le R 2 augmente avec le nombre de variables explicatives dans le modèle. Dans ce modèle global, tous les coefficients affichent les signes attendus, mais au niveau de la signification on remarque que seul le coefficient du ratio VM/VC n’est pas significatif. D’après les résultats de ces 13 modèles de régression sur des données de panel, on peut conclure que l’hypothèse 3 est acceptée. En effet, la ∆MICAP est une variable pertinente en termes de capacité prédictive quant au rendement boursier futur. Même après avoir contrôlé pour l’effet des autres variables comme le bêta, le ratio VM/VC, le ratio B/C et la taille, on a constaté que la ∆MICAP conserve toujours sa capacité prédictive et explicative des rendements annuels futurs. CONCLUSION Porter (1985), dans son modèle de l’avantage concurrentiel, affirme qu’il ne faut pas confondre avantage concurrentiel durable et position dominante sur le marché. Seul le premier est garant d'une performance supérieure et durable dans une industrie. En effet, un avantage concurrentiel durable crée des 265 barrières puissantes autour de l’activité de la firme, ce qui lui permet, non seulement, de faire face avec succès aux forces de la concurrence, mais aussi de creuser l’écart sur ses concurrents et de résister aux périodes économiques difficiles dans son secteur. Dans les industries fortement concurrentielles, un avantage concurrentiel durable n’est pas facile à repérer par les investisseurs. Le problème réside dans la difficulté de mesurer parfaitement un avantage concurrentiel durable et de tenir compte de son aspect qualitatif. Contrairement aux autres mesures de performance, un avantage concurrentiel durable ne peut pas être réduit directement à une formule ou un rapport. Dans cette recherche, on a quantifié l’avantage concurrentiel durable de la firme dans le cadre de son équilibre économique concurrentiel, via le concept de la durée de l’avantage concurrentiel. Il s’agit d’un concept fort intéressant dans le processus de la création de valeur (Rappaport (1986)), mais il est souvent négligé par les investisseurs. Les raisons de cette ignorance sont sans doute dues à la difficulté de l’évaluer. Pour plusieurs auteurs, l’explication la plus pertinente de l’hétérogénéité des performances des firmes est fondée sur le concept de la durée de l’avantage concurrentiel. Notre logique dans ce travail s’est basé sur les quatre points suivants : (1) Dans un marché financier efficient, les cours des titres reflètent toutes les informations disponibles, y compris les avantages concurrentiels durables et les anticipations conjuguées des investisseurs quant aux perspectives futures des firmes dans le cadre de leur équilibre économique concurrentiel, (2) l’avantage concurrentiel durable est considéré par plusieurs auteurs comme étant le meilleur indicateur de la performance future de la firme, (3) la clé de succès à long terme d’une firme est largement déterminée par sa capacité de saisir un avantage concurrentiel et de le conserver, autrement dit d’avoir une durée de l’avantage concurrentiel longue et de la maintenir, (4) l’hypothèse de la circulation bidirectionnelle de l’information et ses conséquences sur les décisions des dirigeants des firmes. Ainsi, notre objectif dans cette recherche était de calculer la durée de l’avantage concurrentiel impliquée par le marché (MICAP) et d’analyser son contenu informationnel et sa capacité prédictive quant aux rendements boursiers. Mais face à l’impossibilité de déterminer une MICAP relative à l’industrie, ainsi que sa standardisation, on a étudié cette variable en termes de sa variation annuelle plutôt qu’en termes de son niveau brut. À notre connaissance, aucune étude ne s’est proposé d’analyser empiriquement la pertinence de la ΔMICAP comme indicateur de la performance courante et future de la firme. En utilisant la formule de Modigliani et Miller (1961) et en se basant sur l’hypothèse de la convergence du rendement marginal du capital investi vers le coût marginal du capital, donc de l’existence des équilibres concurrentiels, on a calculé les ΔMICAP d’un échantillon de 80 firmes appartenant à la base de données Stern et Stewart (2005) sur la période 2000-2004. Sur le plan méthodologique, on a choisi, en premier lieu, d’analyser trois stratégies d’investissement : La stratégie «Achat-détention», la stratégie «Ré-sélection chaque année» et la stratégie d’achat de l’indice S&P 500. Pour cela, on a classé les firmes selon l’ordre croissant de leurs ΔMICAP et on a constitué cinq portefeuilles à la fin de chaque année : 2001, 2002, 2003 et 2004. On a comparé les rendements de ces portefeuilles par l’intermédiaire de plusieurs tests. Les résultats ont montré que les actions à ΔMICAP élevée ont un meilleur rendement que les actions à ΔMICAP faible et que la performance des portefeuilles à ΔMICAP élevée a surpassé celle de l’indice S&P 500 de façon assez systématique. En deuxième lieu, on a utilisé la méthodologie de rangs et les coefficients de corrélation pour analyser la relation de la ΔMICAP avec les indicateurs de performance sélectionnés. Les résultats de cette analyse rapportent l’existence d’une association positive et significative entre la ΔMICAP et la ΔMVAC. Les résultats indiquent aussi des relations négatives et significatives entre la ΔMICAP et certains indicateurs de performance comme le ROC et l’EVAC. En liant ces résultats avec l’existence des deux dimensions de la compétitivité, une dimension quantitative et une dimension qualitative, il semble que la ΔMICAP présente un contenu informationnel quant à la dimension qualitative, à savoir les potentialités de croissance de la firme (ΔMVAC). En troisième lieu, on a analysé les caractéristiques des cinq portefeuilles constitués selon la ∆MICAP moyenne et on a effectué des analyses bidimensionnelles des variations dans les RENDAF moyens afin d’examiner l’effet joint de la ΔMICAP avec les autres variables de contrôle : BETA, VM/VC, B/C et ln(VM). Pour tester la capacité prédictive de la ΔMICAP, 266 on a examiné les coefficients de corrélation de la variable dépendante (le RENDAF) et les variables explicatives (la ΔMICAP et les variables de contrôle) et on a régressé sur données de panel treize modèles. Les résultats ont montré que la ΔMICAP présente un pouvoir prédictif important quant à la création de valeur boursière, même en présence des variables de contrôle. Brefs, tels sont les principaux résultats de ce travail. Certes les résultats seraient plus généralisables si nous avions pu inclure toutes les firmes non financières de la base de données Stern et Stewart (2005). Toutefois, l’indisponibilité de certaines données et la quantité considérable de temps nécessaire pour calculer les MICAP de chaque firme, nous ont empêchés de réaliser un tel objectif. L’application de cette étude dans le contexte canadien et l’analyse du phénomène de la persistance des rendements économiques supérieurs pour les firmes ayant des ΔMICAP élevées pourraient être l’objet de recherches futures. Tableau 1: Classement des firmes de l’échantillon sur la base du code SIC 2-Digit 2-Digit SIC Secteur d’activité Nombre de firmes 15 - 17 Construction 4 20 – 39 Industriel 37 40 - 49 Transport, Communications et 18 Équipement Publics 50 - 51 Commerce de Gros 3 52 - 59 Commerce de Détail 14 70 - 89 Service 4 Total 80 Pourcentage (%) 5 46,25 22,5 3,75 17,5 5 100 Tableau 2 : Les rendements annuels moyens des 5 portefeuilles sur les 10 horizons possibles pour la stratégie 1 Période de détention Fin 2001-Fin 2002 Fin 2001-Fin 2003 Fin 2001-Fin 2004 Fin 2001- Fin 2005 Fin 2002-Fin 2003 Fin 2002-Fin 2004 Fin 2002-Fin 2005 Fin 2003-Fin 2004 Fin 2003-Fin 2005 Fin 2004-Fin 2005 Portefeuille1 (faible- ΔMICAP) -0,0238 0,0864 0,0691 0,0319 0,2101 0,1736 0,1427 0,1290 0,1164 -0,0561 Portefeuille2 Portefeuille3 Portefeuille4 -0,1021 0,0964 0,0468 0,0075 0,2498 0,2002 0,1749 0,1410 0,0995 0,0397 -0,0706 0,1126 0,0669 0,0248 0,2982 0,2458 0,2033 0,1824 0,1283 0,0900 -0,0164 0,1235 0,1038 0,0509 0,3778 0,2801 0,2458 0,1990 0,1326 0,1095 Portefeuille5 (forte- ΔMICAP) 0,0491 0,2079 0,1418 0,0996 0,4181 0,3089 0,2340 0,2325 0,1143 0,1344 Tableau 3 : Les rendements annuels moyens des 5 portefeuilles sur les 10 horizons possibles pour la stratégie 2 Période de détention Fin 2001-Fin 2002 Fin 2001-Fin 2003 Fin 2001-Fin 2004 Fin 2001-Fin 2005 Fin 2002-Fin 2003 Fin 2002-Fin 2004 Fin 2002-Fin 2005 Fin 2003-Fin 2004 Fin 2003-Fin 2005 Fin 2004-Fin 2005 Portefeuille1 (faible- ΔMICAP) -0,0238 0,0869 0,0620 0,0155 0,2101 0,1688 0,1011 0,1290 0,0323 -0,0561 Portefeuille2 Portefeuille3 Portefeuille4 -0,1021 0,0593 0,0277 0,0041 0,2498 0,1942 0,1577 0,1410 0,0892 0,0397 -0,0706 0,0984 0,0647 0,0433 0,2982 0,2390 0,2058 0,1824 0,1353 0,0900 -0,0164 0,1641 0,1114 0,0900 0,3778 0,2853 0,2525 0,1990 0,1534 0,1095 267 Portefeuille5 (forte- ΔMICAP) 0,0491 0,2197 0,1640 0,1401 0,4181 0,3220 0,2860 0,2325 0,1824 0,1344 Tableau 4: Tests de différence sur les rendements des cinq portefeuilles (données groupées) t-test : p-value de Wilcoxon-test: p-value différence de moyenne de différence de médiane Portefeuille 1 vs Portefeuille 2 0,6472 Portefeuille 1 vs Portefeuille 3 0,1640 Portefeuille 1 vs Portefeuille 4 0,0061*** Portefeuille 1 vs Portefeuille 5 0,0004*** Portefeuille 2 vs Portefeuille 3 0,3451 Portefeuille 2 vs Portefeuille 4 0,0323** Portefeuille 2 vs Portefeuille 5 0,0032*** Portefeuille 3 vs Portefeuille 4 0,3421 Portefeuille 3 vs Portefeuille 5 0,0770* Portefeuille 4 vs Portefeuille 5 0,3210 (***), (**), (*) Statistiquement significatifs, respectivement, à 1%, 5% et 10%. 0,6872 0,4443 0,0152** 0,0028*** 0,7369 0,0551* 0,0121** 0,1591 0,0326** 0,4703 Tableau 5: Comparaison de la performance des portefeuilles de la stratégie 1 (Achat-détention) Portefeuille ( j ) Portefeuille 5 Portefeuille 4 Portefeuille3 Portefeuille2 Portefeuille1 - 8/10 - 9/10 10/10 - 10/10 10/10 10/10 - 9/10 10/10 8/10 6/10 - Portefeuille ( i ) Portefeuille 5 Portefeuille 4 Portefeuille 3 Portefeuille 2 Portefeuille 1 Tableau 6 : Comparaison de la performance des portefeuilles de la stratégie 2 (Ré-sélection chaque année) Portefeuille (j) Portefeuille 5 Portefeuille 4 Portefeuille3 Portefeuille2 Portefeuille1 - 10/10 - 10/10 10/10 - 10/10 10/10 10/10 - 10/10 10/10 9/10 6/10 - Portefeuille (i) Portefeuille 5 Portefeuille 4 Portefeuille 3 Portefeuille 2 Portefeuille 1 Tableau 7: La comparaison de la performance des portefeuilles constitués selon la stratégie 1 et 2 Portefeuille (j) de la stratégie 1 Portefeuille (i) de la stratégie 2 Portefeuille 5 Portefeuille 4 Portefeuille 3 Portefeuille 5 Portefeuille 4 Portefeuille 3 6/6 2/6 1/6 6/6 6/6 1/6 6/6 6/6 3/6 Tableau 8: La comparaison de la performance des portefeuilles de la stratégie 1 et 2 avec l’indice S & P 500 Stratégie 1 5/10 7/10 9/10 10/10 10/10 Portefeuilles «x» vs S & P 500 1 2 3 4 5 268 Stratégie 2 4/10 5/10 10/10 10/10 10/10 Tableau 9: Résultats synthétiques de la méthodologie de rangs 1er quartile 2ème quartile 3ème quartile 4ème quartile 1er et 2ème quartiles 1er, 2ème et 3ème quartiles ROC EVAC MVAC ∆EVAC ∆MVAC ROA ROE ∆VENTES 20% 15% 20% 5% 47,5% 68,33% 20% 15% 30% 10% 47,5% 71,66% 30% 10% 30% 20% 52,5% 73,33% 30% 20% 15% 20% 42,5% 71,66% 35% 35% 30% 40% 62,5% 80% 25% 10% 25% 10% 47,5% 68,33% 25% 20% 10% 20% 45% 71,66% 5% 40% 5% 20% 35% 70% B/C moyen 0,055 0,062 0,063 0,055 0,053 0,058 Ln(VM) moyen 9,230 9,027 8,440 8,550 8,683 8,786 Tableau 10: Caractéristiques des portefeuilles constitués selon la ΔMICAP moyenne ΔMICAP RENDAF moyen VM/VC β moyen moyenne en % moyen -19,550 6,635 3,368 0,537 Faible-ΔMICAP -3,065 9,662 5,187 0,665 ΔMICAP-2 0,083 12,454 3,230 0,571 ΔMICAP-3 3,180 17,585 5,450 0,658 ΔMICAP-4 20,656 21,840 1,460 0,754 Forte-ΔMICAP 0,261 13,635 3,739 0,637 Moyenne RENDAF est le rendement annuel futur, (i.e. RENDAF (t+1)), la ΔMICAP est la variation du MICAP, VM/VC est le ratio valeur marchande des fonds propres par rapport à la valeur comptable des fonds propres, le B/C est le ratio bénéfice/cours, ln(VM) est le log népérien de la valeur marchande des fonds propres en M$ et BETA est le risque systématique. Tableau 11 : Rendements annuels moyens des portefeuilles constitués selon Bêta et ΔMICAP Faible-ΔMICAP ΔMICAP-2 ΔMICAP-3 Forte-ΔMICAP Total Faible- Bêta Bêta -2 Bêta -3 Forte - Bêta Total 0,093 0,092 0,105 0,152 0,110 0,024 0,098 0,134 0,158 0,104 0,107 0,081 0,191 0,248 0,157 0,108 0,108 0,213 0,264 0,173 0,083 0,095 0,161 0,206 Tableau 12 : Rendements annuels moyens des portefeuilles constitués selon VM/VC et ΔMICAP Faible-ΔMICAP ΔMICAP-2 ΔMICAP-3 Forte-ΔMICAP Total Faible- VM/VC VM/VC -2 VM/VC -3 Forte – VM/VC Total 0,092 0,170 0,207 0,195 0,166 0,146 0,130 0,185 0,184 0,161 0,035 0,021 0,180 0,300 0,134 0,008 0,056 0,094 0,142 0,075 0,070 0,094 0,166 0,205 Tableau 13 : Rendements annuels moyens des portefeuilles constitués selon Taille et ΔMICAP Faible-ΔMICAP ΔMICAP-2 ΔMICAP-3 Forte-ΔMICAP Total Faible- ln(VM) Ln(VM) -2 Ln(VM) -3 Forte – ln(VM) Total 0,090 0,190 0,174 0,253 0,177 0,178 0,131 0,169 0,240 0,179 0,049 0,047 0,122 0,234 0,113 0,060 0,010 0,127 0,103 0,075 0,094 0,095 0,148 0,208 Tableau 14 : Rendements annuels moyens des portefeuilles constitués selon B/C et ΔMICAP Faible-ΔMICAP ΔMICAP-2 ΔMICAP-3 Forte-ΔMICAP Total Faible- B/C B/C -2 B/C -3 Forte – B/C Total 0,029 0,097 0,145 0,185 0,114 0,100 0,028 0,218 0,270 0,154 0,021 0,090 0,120 0,155 0,097 0,172 0,139 0,181 0,228 0,180 0,081 0,088 0,166 0,210 269 Tableau 15 : Les régressions sur données de panel (*) Modèles Constante β ΔMICAP Ln(VM) 1 2 3 4 5 6 7 8 9 10 11 12 13 -0,0335 0,1354*** -0,0123 3,7809*** 0,1434*** 3,5356*** 0,1404*** 3,7481*** 3,5228*** -0,0646 -0,0821 3,0144*** 2,7317*** VM/VC B/C 0,2666** 0,2319** 0,0034*** 0,0037*** -0,4148*** -0,0019 0,0028*** 0,0034*** 0,0028*** 0,2169** 0,0037*** 0,0030*** 0,0029*** -0,3869*** -0,4107*** -0,3854*** -0,3412*** -0,3265*** -0,0013 -0,0007 -0,0003 0,0000 3,4777*** 3,7626*** 2,0681** 2,3310*** R 2 − ajusté 0,31% 6,81% 6,86% 3,35% 0,66% 9,19% 7,08% 3,93% 9,43% 4,07% 11,33% 12,51% 13,14% F-test 5,22** 19,24*** 11,85*** 53,13*** 1,81 35,58*** 10,08*** 26,64*** 23,64*** 14,23*** 19,37*** 26,02*** 16,64*** RENDAF est le rendement annuel futur, ΔMICAP est la variation du MICAP, VM/VC est le ratio valeur marchande des fonds propres par rapport à la valeur comptable des fonds propres, le B/C est le ratio bénéfice/cours, ln(VM) est le log népérien de la valeur marchande des fonds propres en M$ et BETA est le risque systématique. (***), (**), (*) Statistiquement significatifs, respectivement, à 1%, 5% et 10%. (*) Le test de Hausman révèle que le modèle à effets fixes est supérieur au modèle à effets aléatoires. Les coefficients estimés sont donc ceux acquis avec le premier modèle. En plus, nous avons inclus des variables binaires représentant chacune une année de la période d’étude afin d’éliminer l’effet temps, pouvant exister le cas échéant. BIBLIOGRAPHIE 1- Barney, J. (1991). Firm resources and sustained competitive advantage. Journal of Management, 17, 99-120. 2- Besanko, D., Dranove, D. & Shanley, M. (1996). The economics of strategy. John Wiley & Sons, Inc., New York. 3- Bommel, J. (1997). Messages from market to management: The case of IPO’s. Working paper, INSEAD. 4- Boston Consulting Group. (2000). New Perspectives on value creation. BCG report. 5- Boston Consulting Group. (2001). Dealing with investors’ expectations. BCG report. 6- Boston Consulting Group. (2003). The Value Creators: A study of the world’s top performers. BCG report. 7- Damodaran, A. (2006). Finance d’entreprise: Théorie et pratique. Traduit par Hirigoyen, G., De Boeck Université. 8- Dow, J. & Gorton, G. (1997). Stock market efficiency and economic efficiency: Is there a connection? Journal of Finance, 52 (3), 1087-1129. 9- Droucopoulos, V. & Lianos, T.P. (1993). The persistence of profits in the Greek manufacturing industry. Internat. Rev. Appl. Econom, 7, 163-176. 10- Dye, R.A. & Sridhar, S. (2002). Resource allocation effects of price reactions to disclosures. Contemporary Accounting Research, 19 (3), 385-410. 11- Fama, E. (1976). Foundations of finance. New York: Basic Books. 12- Geroski, P.A. & Jacquemin, A. (1988). The persistence of profits: A European comparison. The Econom. J.: The J. Royal econom. Society, 98, 375-389. 13- Goddard, J.A. & Wilson, J.O.S. (1996). Persistence of profits for UK manufacturing and service sector firms. Service Industries J., 16. 14- Hamel, G. & Prahalad, C.K. (1994). Competing for the future. Boston, Harvard Business School Press. 15- Jahnke, G., Klaffke, S. & Oppenheimer, H. (1987). Price-Earnings Ratios and security performance. Journal of Portfolio Management, Fall, 39-46 270 16- Luo, Y. (2005). Do insiders learn from outsiders? Evidence from mergers and acquisitions. The Journal of Finance, LX (4). 17- Mauboussin, M. (1997). Thoughts on valuation. Credit Suisse Equity Research, 21. 18- Mauboussin M.: (1998). Why strategy matters? Credit Suisse Equity Research, 29. 19- Mauboussin, M. & Bartholdson, K. (2002). Measuring the moat:. Assessing the magnitude and sustainability of value creation. Credit Suisse Equity Research, 16. 20- Mauboussin, M. & Schay, A. (2001). Thoughts on valuation II. Credit Suisse Equity Research, 31. 21- Mauboussin, M. & Johnson, P. (1997). Competitive advantage period CAP: The neglected value driver. Financial Management, 26(2), 67-74. 22- Mc-Gahan, A. (1999). Competition, strategy and business. California Management Review, 41(3). 23- Mills, R.W. & Dahlhoff, J.U. (2003). Competitive advantage period (CAP) analysis: Looking at share price and firm value from a different perspective. Henley Centre for Value Improvement, Henley Discussion Paper, 3. 24- Mills, R. (1997). Competitive advantage period. Manager Update, 9(1). 25- Mills, R. (1994). Finance, strategy and strategic value analysis. Lechlade: Mars Business Associates. 26- Mills, R. (1998). The dynamics of shareholder value. Lechlade: Mars Business Associates. 27- Modigliani, F. & Miller, M.H. (1961). Dividend policy, growth, and the valuation of shares. Journal of Business, 34, 411-433. 28- Mueller, D.C. (1986). Profits in the long run. Cambridge University Press, Cambridge. 29- Porter, M.E. (1980). Competitive strategy techniques for analyzing industries and competitors. New York, NY, Free Press. 30- Porter, M.E. (1985). Competitive advantage: Creating and sustaining superior performance. New York, Free Press. 31- Rappaport, A. (1998). Creating shareholders value: A guide for managers and investors. New York: Free Press. 32- Rappaport, A. (1986). Creating shareholder value: The new standard for business performance. The Free Press. 33- Rappaport, A. & Mauboussin, M.J. (2001). Expectations investing: Reading stock prices for better returns. Harvard Business School Press. 34- Rappaport, A. (1987). Stock market signals to managers. Harvard Business Review, Nov–Dec, 57-62. 35- Schohl, F. (1990). Persistence of profits in the long run: A critical extension of some recent findings. Internat. J. Indust. Organ, 8, 385-404. 36- Sinha, R. (2006). Corporate governance and shareholder value analysis. Global Business Review, 7(1). 37- Stewart, G.B. (1991). The quest for value: A guide for senior managers. Harper Business. 38- Sunder, S. (1989). Proof that in an efficient market, event studies can provide no systematic guidance for revision of accounting standards and disclosure policy for the purpose of maximizing shareholder wealth. Contemporary Accounting Research, 5, (2), 452-460. 39- Tobin, J. (1969). A general equilibrium approach to monetary theory. Journal of Money, Credit and Banking,, 1, 15-29. 40- Tseng, K.C. (1988). Low price, price-earnings ratio, market value, and abnormal returns. Financial Review, 23(3), 333-343. 41- Wiggins, R. & Ruefli, T.W. (2002). Sustained competitive advantage: Temporal dynamics and the Incidence and persistence of superior economic performance. Organization Science, 13(1), 81-105 42- Wigmore, B.A. (1991). How can we explain the growth of the S&P 500 in the 1980s? Goldman Sachs Internal Research. 43- Yook, K.C. & McCabe, G.M. (2001). MVA and the cross-section of expected stock returns. Journal of Portfolio Management, 27(3), 75-87. 271 ASAC 2008 Halifax, Nova Scotia Marianne Schovsbo MBA Student Alex Ng University of Northern British Columbia LIFE SCIENCE VENTURE CAPITALISTS: AN EXPLORATORY STUDY Nineteen venture capitalists are interviewed to understand from their experience - selection, evaluation, and risks factors involved in deciding on life science investments. Results show large and varying support for the theory on venture capitalists’ role in reducing costs of information asymmetry in investing, as well as, national differences between Canadian and US venture capitalists. The development of life science technologies, particularly biotechnology and medical devices, are probably some of the most intricate, lengthy and complex processes that present themselves today within our global economies and financial systems (Baeyens et al., 2006; Baum and Silverman 2004; DiMasi et al., 2003; Fetterhoff and Voelkel, 2006; Shepard et al., 2003). Biotechnology and medical devices reach into a large and diverse marketplace touching all fields of human knowledge, including pharmaceuticals, food, fuel, as well as our waste product processes. In a sense these technologies encompass some of the most pioneering and valuable creations that people have had the courage to invent (DiMasi et al., 2003; Rousu et al., 2004). In fact, thirty-five percent (35%), or 95 out of 256, of all new therapeutic products that are approved in the past ten years come directly from the life science field (Edwards, et al., 2003). This is an amazing feat for a life science industry that did not truly exist before its conception in the late 1970’s (Amir-Aslani and Negassi 2006; Bains 2004; Edwards et al., 2003). Life science technologies, however, are some of the most comprehensive and difficult to evaluate as investment opportunities (Baeyens et al., 2006; Baum and Silverman 2004; DiMasi et al., 2003) . If Venture Capitalists (VC) can only invest in a specific number of portfolio investments, it is crucial that the investment opportunities chosen are those that are also the most likely to succeed. Investor perception risks due to the lack of qualification, venture evaluation, technology comprehension, information asymmetry or overconfidence, are factors that limit Venture Capitalists from choosing investments that create wealth (Coombs et al., 2006; Franke et al., 2006; Shepherd et al., 2003). With the tremendous amount of capital needed, along with the development risks involved, it is perhaps somewhat surprising that these life science technologies get funding at all, by venture capitalists. The Canadian venture capital industry experience is not greatly studied. The limited extant literature demonstrates empirical study of venture capital firms through primary survey data (Thornhill and Amit, 2001) or secondary survey data (Amit et al, 1998; Cumming and MacIntosh, 2003). The purpose of this study is to contribute to the Canadian VC literature, through an exploratory and qualitative approach, yielding deeper answers to the theory on the role of venture capitalists (Amit et al., 1998) and on differences between Canadian and US VC firms (Cumming & MacIntosh, 2003). To date, no Canadian research is dedicated into the understanding of risk perception and venture evaluation among Venture Capitalists who specialize in life science investment opportunities. A qualitative approach is taken in this study to understand venture capital managers whose tacit knowledge about a very complex 272 subject, such as investing in life sciences, can be revealed in ways that firm level survey data cannot. Thus, in-depth interviews provide data that is not limited for the purposes of singular, highly defined hypothesis testing, and more importantly, provide a rich basis of understanding venture capital manager perspectives, and decision-making processes. Often, this exploration can benefit future VC studies to consider novel factors to explain issues. Risks associated with lack of life science comprehension and perception biases may be large contributors to whether a life science venture reaches a successful exit or even gets funded. Understanding the risk perceptions combined with the comprehension and aversion strategies used by Venture Capitalists that specialize in life science investments, is therefore important. This is a study in which nineteen semi-formal interviews are conducted with Venture Capitalists located on the West Coast of Canada and United States. Differences between West Coast Canadian and United States venture capitalists are also found. This paper is organized as follows: 2) Background on venture capitalists, 3) Literature Review, 4) Methodology, 5) Results and Discussion and 6) Conclusion. 2. Background on Venture Capitalists in Life Sciences Venture capital is an important source of early stage investment capital, especially in the United States where approximately thirty-five percent (35%) of all venture capital investments, whether in life science, high tech, etc., are dispersed to seed and start up fledging firms (Fried and Hisrich, 1994; Manigart et al., 2000; National Venture Capital Association). As selectors and investors of new ventures, venture capitalists are often portrayed, within the business world, as having superhuman characteristics for picking the right projects (Baeyens et al., 2006; Baum and Silverman 2004; Bishop and Nixon, 2005). They are considered to be highly qualified selectors and investors of high risk projects, are described as reducers of information asymmetries, said to be “coaches” and “monitors” after investments are made, and are depicted as having an instinctive capability to act within complex and uncertain environments (Amit et al., 1998; Baeyens et al., 2006; Baum and Silverman 2004; Bishop and Nixon, 2005; Shepard et al., 2003). However, venture capital investment opportunities, generally, have a low success rate where only one in two hundred projects becoming highly profitable, with the majority of projects only just making a return on investment, and where approximately fifteen to twenty percent (15-20%) fail outright (Amit et al., 1998; Berlin 1998; Sahlman, 1990; Zacharakis and Meyer, 2000). Life science technologies, however, are considered even higher venture risks (Baeyens et al., 2006; Danzon et al., 2005; DiMasi et al., 2003). The initial failure rate during the discovery and developmental stages can be enormous, with postulations that only one in 10,000 new life science entities succeed to the market as new therapeutics (DiMasi et al., 2003; Evans and Varaiya, 2003; Fetterhoff and Voelkel, 2006; Pharma Ventures 2005). 3. Literature Review of Canadian Venture Capital Amit et als’(1998) seminal paper proposes a clear theory on venture capitalists whose “raison d’être” is their ability to reduce the cost of information asymmetries. Information asymmetry especially exists between what the venture owner knows versus what the venture capitalists could know about venture success. Their theoretical analysis yields four empirical predictions: 1) There are strong industry effects because venture capitalists can operate better than ordinary investors in selecting investments in industries, such as life sciences, where there is high information asymmetry; 2) Venture capitalists prefer projects where monitoring and selection costs are low or information asymmetry costs are lower; 3) Venture capital exits will tend to be sales to informed investors, and exits that lead to IPO’s tend to be higher quality ventures; 4) There would be a negative relationship between the venture capital invested in a venture firm and its performance. Their empirical analysis of Venture Canada Association’s survey 273 data on 387 Canadian venture firm companies during 1987-1994 yield supportive results. Amit et al. (1998) find that when examining Venture Capital firms within Canada, they discover that Canadian venture capitalists are large investors of the technology industries, such as life science, but at later stages of development, rather than at the earlier entrepreneurial stages. They find that Canadian venture capitalists exit their portfolio firms through insider sales and management buy-outs rather than mergers and acquisitions and IPOs -- even during the period of the early 1990’s where the markets are rather favorable to IPOs. Cumming and MacIntosh (2003) examine cross-country comparison, US and Canada, of full and partial venture capital exits. They seek to examine what factors determine the choice of how venture capital firms exit from their ventures. Empirical results on Canadian and US venture capital association survey data during 1992 to 1995 offers support for their central hypothesis; that the greater the degree of information asymmetry between the selling VC and the buyer, the greater the likelihood of a partial exit to signal quality. Of particular interest, their study finds differences between Canadian and US venture capital firms highlighting the impact of legal and institutional factors on exits across these countries. Other studies on the Canadian venture capital experience examine a special type of tax-driven venture capital vehicle, known as the “Labour Sponsored Venture Capital Corporation” (Cumming and MacIntosh, 2006). Analysis of firm level data from 1977-2001 show empirical support for their hypothesis that this special type of venture capital indeed does “crowd out” other types of VC funds because of its tax advantages. Cumming (2005) examines forms of financing in 12,363 Canadian and US venture capital firms towards entrepreneurial firms. His empirical results show that agency costs, capital gains taxation, learning and market conditions affect financing contracts made with venture firms. 4. Methodology 4.1 Semi-Structured Interview Method and Questionnaire The semi-structured interview technique is chosen for this study. This method is chosen because recent literature shows that Venture Capitalists are not likely to return mail questionnaires before direct and personal contact is made. (Baeyens et al, 2006). Face to face and telephone interviews are shown to be effective because they establish a relationship with the Venture Capitalists (Baeyens et al, 2006). Furthermore, in the semi-structured interview technique, the researcher is able to exercise his or her inventiveness with respect to following up a respondent’s answer to a question as well as the ability to clarify questions that the respondent may not understand (Hair et al., 2003). The semi-structured interview process, however, is structured to ensure that cohesiveness and comparability can be made between respondents. This approach is noted to give unexpected and insightful information that can augment the findings of the research (Hair et al., 2003). The interviewing period took place between the end of February to mid July, 2007. The interview is conducted either face to face or over the telephone with a brief introduction of the research that is to be conducted and why. It is aimed that the interview last between fifteen and twenty minutes, but some of the interviews expanded a much longer time period, with the longest lasting almost two hours. Many of the Venture Capital respondents spent extra time discussing their positions and opinions. The questionnaire is designed and structured in a three part sequence: 1) Opening questions asking specific background information with respect to the person being interviewed; 2) A middle section that directs questions specifically towards the research objectives, such as Venture Capital preferred investment criteria; 3) A final section which asks open general information questions such as “What are the preferred Venture Capital exit routes?”, and “What are the general costs of life science projects”. Appendix A (available upon request) shows the established questionnaire, used with all interviews. The 274 questionnaire questions are a mix of Likert Scale and open ended opinion questions. All respondents’ comments are presented in tables and are available upon request. 4.2 Participant Sample Description Ten (10) Venture Capitalists on the West Coast of Canada from the Vancouver region; and nine (9) Venture Capitalists on the West Coast of United States, six (6) from the Seattle region and three (3) from California, are contacted and successfully interviewed. Of the ten (10) Canadian respondents, two (2) are from the same company and of the nine (9) United States respondents, two (2) are from the same company. Of the respondents from West Coast Canada, three (3) of the ten (10) Venture Capitalists interviewed are female and the rest are male; in contrast, all respondents are male from the West Coast United States. Most of the Venture Capitalists interviewed are Senior Investors and Managers including Principals, Vice Presidents, and Partners. 5. Results and Discussion 5.1 On the Theory of Venture Capitalists 5.1a) Strong industry effects in venture capitalist firms. Are there strong industry effects with respect to life science investments in the venture capital industry as theorized by Amit et al. (1998)? A series of six questions are asked with respect to screening life science investment opportunities and the respondent’s strategy. The first question of this series asked whether there is a focus on specific life science industries, such as drug development versus device. The United States respondents (8/9) are more inclined to strongly agree and agree than the Canadian respondents (7/10), but there are no significances between the two groups. Combining the two study groups, there is a strong trend of agreement against disagreement with fifteen (15) of the respondents both strongly agreeing/agreeing and only four (4) of the Venture Capital respondents both disagreeing/strongly disagreeing (Figure 1). Figure 1. Percentage of Venture Capitalists who agree that they focused on Specific Life Science Industries when Screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) 60% Percentage of VCs 50% 40% West Coast Canada: Vancouver 30% West Coast United States: Seattle and California 20% 10% 0% Strongly Agree Agree Neither Disagree Strongly Disagree The second question of the series asked whether there is a focus on specific life science technologies such as single agent small molecules, biologics, nanotechnology etc., when screening investment opportunities. It is found that 70% of the United States respondents agree/strongly agree compared to 20% of the Canadian respondents (Figure 2). 275 Figure 2. Percentage of Venture Capitalists who agree that they focused on Specific Life Science Technologies when Screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) 60% 50% W est C oast C anada: V anc ouv er W e s t C o a s t U n it e d S t a t e s : S e a t t le s C40% V f o e g 30% ta n e rc 2 0 % e P 10% 0% S tr o n g l y A g r e e A g ree N e i th e r D is a g re e S tr o n g l y D is a g re e The third question of the series asked whether there is a focus on specific disease indications, such as diabetes, cancer, or heart disease, when screening investment opportunities (Figure 3). It is found that twelve (12) of all the respondents disagree with this strategy. Three (3) of the ten (10) respondents from Canada, however, did agree that there is a specific disease indication focus when screening investment proposals. Figure 3. Percentage of Venture Capitalists who agree that they focused on Specific Disease Indications when screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) 60% Percent of VCs 50% 40% West Coast Canada: Vancouver 30% West Coast United States: Seattle and California 20% 10% 0% Strongly Agree Agree Neither Disagree Strongly Disagree The fourth question of the series asked whether there is a focus on preferred stage of technology development when screening investment opportunities (Figure 4). This question received a wide spread of responses with eleven (11) of the total nineteen (19) respondents either strongly agreeing or agreeing. Figure 4. Percentage of Venture Capitalists who agree that they focused on a Preferred Stage of Technology when Screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) 45% 40% Percent of VCs 35% 30% 25% Wes t Coas t Canada: Vancouver 20% Wes t Coas t United States : Seattle and California 15% 10% 5% 0% Strongly Agree Agree Neither Disagree Strongly Disagree The fifth question of the series asked whether there is a focus on geographical location of the investment opportunity when screening investment opportunities (Figure 5). It is found that fourteen (14) 276 out of all respondents either strongly agree/ agree that geographical location of the investment opportunity is a focus. Figure 5. Percentage of Venture Capitalists who agree that they focused on a Preferred Geographic Location when Screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) 90% 80% 70% W e s t C o as t C an ad a: V an c o u v e r s60% C V f 50% o t n4 0 % e rc e P3 0 % W e s t C o a s t U n it e d S t a t e s : S e a t t le a n d C a lif o r n ia 20% 10% 0% S tr o n g l y A g r e e A g re e N e i th e r D is a g re e S tr o n g l y D i s a g r e e The sixth and final question of the series asked whether there is a focus on minimum or maximum investment required to exit when screening investment opportunities (Figure 6). It is found that 90% of United States respondents agree or strongly agree compared to 50% the Canadian respondents. Comments from five of the Canadian respondents addressed that minimum and maximum investment is not a focus because it depended on the fund rather than the technology being assessed. Figure 6. Percentage of Venture Capitalists who agree that they focused on the Minimum or Maximum Investment Required to Exit when Screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) 80% 70% 60% s C V 50% f o 40% t n e 30% rc e P 20% W est Co ast Can ad a: Van co uver W est Co ast Un ited States: Seattle an d Califo rn ia 10% 0% Strongly Agree Agree Neither Disagree Strongly Disagree 5.1b) Discussion: strong industry effects in VC firms. There are strong industry effects because venture capitalists can operate better than ordinary investors in selecting investments in industries, such as life sciences, where there is high information asymmetry (Amit et al., 1998). These findings on specific life science investment strategies are supportive of Amit et als’ (1998) theoretical prediction. That is, there are indeed very specific life science investment strategies and criteria that Venture Capitalists use and seek. Fried and Hisrich’s (1994) model of Venture Capital investment decision making process with inputs for Baeyens et al., (2005) seems to be very much so, the model of choice, with specific focus on life science technologies targeted, preferred stages of development chosen, geographical location and minimum, maximum investment required to exit. 5.1c) VC’s preferences for low monitoring and selection costs. When asked what the responsibilities of their positions are with respect to investment selection and evaluation, all respondents replied that they are involved at all levels of investment selection process within the firms. A typical answer is that he/she is involved in initial calls and opportunity searches, investment proposal reviews, 277 evaluation, investment decisions/negotiations and board representation. All of the respondents spent time in active monitoring roles within their selected portfolio companies, either observing board practices, active on the board, networking, presenting achievements of their portfolios companies and in general, assisting portfolio companies in any manner that they could. The respondents are asked what main investment strategies are used when screening life science opportunities (Figure 7). It is important to note that for this question, the respondents are able to identify more than one strategy. Of the five identified categories, management, return on investment, location, stage of development and the technology itself, three categories: management, stage and technology, are mentioned by eighteen (18) of the nineteen (19) respondents, with only one respondent mentioning return and none of the respondents mentioning location. Figure 7. Main Factors Considered by Venture Capitalists when Screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) W e st C o a st C a n a d a : Va n co u ve r W e s t C o a s t U n ite d S ta te s : S e a ttle a n d C a lif o rn ia 100% 90% 80% s C V f o t n e c r e P 70% 60% 50% 40% 30% 20% 10% 0% M an ag em en t R e tu rn L o c a ti o n S ta g e T e c h n o lo g y The respondents are next asked four questions with respect to criteria used when screening investment opportunities. These included financial, potential market size, intellectual property, and management team criteria. Firstly, the respondents are asked whether financial criteria are important to estimate potential proposal worth, (Figure 8). Almost 100% of the respondents (18/19) replied that they either strongly agree or agree that the financial criteria are important. The respondents are then asked if potential market size criteria are important when screening investment opportunities. Here again 100% of the respondents either strongly agree or agree. Figure 8. Percentage of Venture Capitalists who agree that Financial Criteria are important to estimate potential value of Investment Opportunity. West Coast Canada (n=10) and West Coast United States (n= 9) 100% 90% P ercent of V C s 80% 70% Wes t Coas t Canada: Vancouver 60% 50% Wes t Coas t United States : Seattle and California 40% 30% 20% 10% 0% Str ongly Agr e e Agr e e Ne ithe r Dis agr e e Str ongly Dis agr e e The respondents are next asked whether intellectual property is important when screening investment opportunities (Figure 9). In this case fifteen (15) of the respondents strongly agree, with three (3) respondents simply agreeing. One Venture Capitalist from the United States neither agree nor disagree. 278 Figure 9. Percentage of Venture Capitalists who agree that Intellectual Property Status is important when screening Investment Opportunity. West Coast Canada (n=10) and West Coast United States (n= 9) 100% 90% Percent of VCs 80% 70% Wes t Coas t Canada: Vancouver 60% Wes t Coas t United States : Seattle and California 50% 40% 30% 20% 10% 0% Strongly Agree Agree Neither Disagree Strongly Disagree Respondents are finally asked whether management team and skill is important when screening investment opportunities (Figure 10). Here 100% of the United States respondents strongly agree. Canadian respondents also strongly agree, but two (2) of the respondents simply agree. Figure 10. Percentage of Venture Capitalists who agree that Management Team and Their Skills are Important when screening Investment Opportunities. West Coast Canada (n=10) and West Coast United States (n= 9) 120% 100% Percent VCs 80% West Coast Canada: Vancouver 60% West Coast United States: Seattle and California 40% 20% 0% Strongly Agree Agree Neither Disagree Strongly Disagre e Respondents are asked whether it is important to exclude investment opportunities that are scientifically difficult to understand. Canadian respondents are spilt in opinion, with 50% agreeing and 50% neither agreeing nor disagreeing or simply disagreeing. United States respondents are also split on this question, but tended to strongly agree, 33%, rather than to simply agree (Figure 11). Figure 11. Percentage of Venture Capitalists who agree that it is important to exclude Investment Opportunities that are Scientifically Difficult to Understand. West Coast Canada (n=10) and West Coast United States (n= 9) 100% 90% West Coast Canada: Vancouver Percent VCs 80% 70% West Coast United States: Seattle and California 60% 50% 40% 30% 20% 10% 0% Strongly Agree Agree Neither Disagree Strongly Disagree Respondents are asked about investment choices and direct selection preferences. Almost 100% of all the respondents either strongly agree or agree that past investment choices influenced future investment choices (Figure 12). When respondents are asked whether they tended to select investment 279 opportunities that are related to their educational/work background, 63% of the respondents (12/19) either strongly agree or agree (Figure 13). Figure 12. Percentage of Venture Capitalists who agree that Past Investment Choices Influenced Future Investment Choices. West Coast Canada (n=10) and West Coast United States (n= 9) 80% 70% Percent VCs 60% West Coast Canada: Vancouver 50% West Coast United States: Seattle and California 40% 30% 20% 10% 0% Strongly Agree Agree Ne ithe r Disagre e Strongly Disa gree Percent VCs Figure 13. Percentage of Venture Capitalists who agree that they tended to select Life Science Investment Opportunities related to their educational background over opportunities that are not related. West Coast Canada (n=10) and West Coast United States (n= 9) West Coast Canada: Vancouver 50% 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% West Coast United States: Seattle and California Strongly Agree Agree Neither Disagree Strongly Disagree When asked about whether they tended to choose/select investment opportunities that are similar to past successes over those proposals that are not, 60% of all respondents either strongly agree or agree, and 30% neither agree nor disagree (Figure 14). Interestingly 75% of the United States respondents either strongly agree or agree, whereas 40% of the Canadian respondents simply agree. Also, 40% of the Canadian respondents neither agree nor disagree. Three (3) of the Canadian respondents mentioned that they chose this category because it is easy to become “biased” and complacent with respect to this type of decision making and that it is important to “keep one’s eye on the ball to changing variables”. Figure 14. Percentage of Venture Capitalists who agree that they tended to select Life Science Investment Opportunities similar to past successes over those which are not. West Coast Canada (n=10) and West Coast United States (n= 9) 60% West Coast Canada: Vancouver Percent VCs 50% West Coast United States: Seattle and California 40% 30% 20% 10% 0% Strongly Agree Agree Neither Disagree Strongly Disagree Respondents are also asked whether Venture Capital experience/expertise is positively linked to successful investment selection (Figure 15). Of all the respondents, 90% either strongly agree or agree. United States respondents strongly agree compared to Canadian respondents who simply agree. 280 Figure 15. Percentage of Venture Capitalists who agree that Venture Capital expertise/experience is positively linked to successful investment selection. West Coast Canada (n=10) and West Coast United States (n= 9) 80% W e st C oa st C a n a da : V an co u ver 70% 60% W e s t C o a s t U n ite d S ta te s : S e a ttle a n d C a li f o rn ia s C5 0 % V t 40% n e rc 3 0 % e P 20% 10% 0% S tro n g l y A g re e A g ree N e i th e r D isag ree S tro n g l y D i s a g re e Respondents are asked whether they would or would not invest in a difficult to understand life science investment opportunity. The majority (90%) of United States respondents strongly agree/agree compared to Canadian respondents, who agree 60% of the time. Of note, 30% of the Canadian respondents disagree (Figure 16). Figure 16. Percentage of Venture Capitalists who would not invest in a difficult to understand Life Science Investment Opportunity. West Coast Canada (n=10) and West Coast United States (n= 9) W e s t C o a s t C a n a d a : V a n c o u ve r 80% W e s t C o a s t U n i t e d S t a te s : S e a tt l e a n d C a lif o rn ia 70% 6 s C5 V t n4 e c r e3 P 2 0% 0% 0% 0% 0% 10% 0% S tr o n g l y A g re e A g re e N e i th e r D is a g re e S tr o n g l y D is a g re e 5.1d) Discussion: venture capitalists and monitoring and selection costs. Do venture capitalists prefer projects where monitoring and selection costs are low or information asymmetry costs are lower as theorized by Amit et al. (1998)? These findings show support for Amit et al’s (1998) view that low monitoring costs influences selection of venture firms. Geographical location, the nearness of the actual development of the investment opportunity, is found to be important. As one United States respondent mentioned, he is not concerned where a technology or investment opportunity came from, but rather that the technology or investment could be established and developed near him, making it logistically easier for him to monitor. This brings into account the importance of monitoring by the Venture Capitalists. All of those interviewed are active monitors of their investment portfolios, whether indirectly or directly as board members. As discussed earlier in the literature section, Venture Capital monitoring is directly linked with the successful development of the investment, in particularly within the life science industry where the development timelines are both intricate and lengthy. Therefore selection of investment opportunities that logistically favour the ease of monitor is not surprising. These findings show further support for Amit et als’ (1998) prediction that low selection costs to venture capitalists are important as it influences their selection behaviour. Results show that venture capitalists deal with the costs of information asymmetry, that being the risk of not knowing about an investment risk as much as the venture owner, by selecting investments that align with their personal education, past experience and successes and avoiding investments which are difficult to understand. Indeed, expertise and experience are recognized as key to investment success in the VC industry. 281 A key tenet of Amit et als’ (1998) theory is that venture capitalists exist because they can reduce the cost of information asymmetry in investments. These findings show that venture capitalists mitigate this cost by relying heavily on management team ability. In fact, from the intense focus that the respondents gave to the selection of trusted management teams it would seem that venture capitalists may perceive the cost of information asymmetry risk as the highest possible risk with respect to life science investments. This type of adverse selection may in turn increase unforeseen risks, or perhaps worse, screen out novel life science technologies that may have huge impacts on unmet medical needs and lucrative investment benefits. 5.1e) The nature of VC capital exits. What are venture capitalists’ preferences for exiting their project? Respondents are asked what they considered as the most desirable exit route for life science investments (Figure 17). All respondents, 100%, replied that M&A is, in the recent past, the most desirable route of choice. All respondents also comment that exit routes depended on market conditions, highlighted by four (4) of the United States respondents who explain that because M&A and IPO’s depend on market conditions and interest rates, exit routes are continuously changing. Four (4) of the 40% of the Canadian respondents considered IPO’s as desirable exit routes, but only 10% of the United States respondents find this route desirable. With respect to IPO’s, two (2) of the United States respondents comment that they consider them more as a refinancing option than an exit route. General comments are directed towards the stock market volatility in technology stocks, the length of time it took for an IPO financing period, the paying off of underwriters and generally the long waiting game of getting one’s money back out of the market. As one of the United States respondents mention: “IPO’s are one big headache right now, because of the market swings and the uncertainty. A lot of the bigger pharmaceutical companies are currently searching for technology to augment their pipelines and therefore there are still lots of opportunities for M & A's” Figure 17. Most desirable Exit Routes for Life Science Investments. West Coast Canada (n=10) and West Coast United States (n= 9) 120% West Coast Canada: Vancouver Percent VCs 100% West Coast United States: Seattle and California 80% 60% 40% 20% 0% Earn Out IPO M&A Depends on Market Other The respondents are also asked how long they expected to wait in order for an investment to return capital (Figure 18). The majority of the respondents (90%) gave responses within two categories: 3-5 years and 5-7 years, with four of these respondents replying 3-7 years. Interestingly, 50% of Canadian respondents, after picking a specific time range for investment, are willing to wait longer than 7 years. This is not the case with the United States respondents. 282 Figure 18. Venture Capitalist’s Expectations for Time Period between Initial Investment to Capital Return. West Coast Canada (n=10) and West Coast United States (n= 9) 80% 70% 60% s C5 0 % V t n4 0 % e c r 30% e P 20% W e st C o a st C a n a d a : V a n co u ve r W e s t C o a s t U n i te d S ta te s : S e a ttl e a n d C a lif o rn ia 10% 0% 1 -3 Y e a r s 3 -5 Y e a r s 5 -7 Y e a r s Longer 5.1f) Discussion: the nature of VC capital exits. Findings support Amit et als’ (1998) view that exits will tend to be sales to informed investors, and exits that lead to IPO’s tend to be higher quality ventures. Patience and tolerance, however, are two important characteristics required in life science technology development and investment. Years are required before an investment may see a return. While periods ranging from three (3) to seven (7) years are the ideal time ranges sought by investors, this does not reflect the true timelines of life science technology development already mentioned, in particular within the therapeutics arena where fifteen (15) years is not an unusual development time span. How do the venture capitalists move the technologies that they invest in, into the next stages of development? It is assumed that as soon as the life science technology is developed to a certain phase, exit strategies are pursued that would allow that technology to go to the next level of development. From the interviews, it is found that M&A is, in the recent past, the exit route of choice. This is concurred with by both the Canadian and the United States respondents. IPO’s are considered an exit route in the late 1990s but are no longer considered as such by most of the United States respondents. The United States respondents are particularly against the IPO method as an exit route, exclaiming that such routes are “…a pain in the butt…” and “…only a way of driving a person insane…” However, two respondents agree that IPO is a good refinancing option reserved for high quality ventures which concurs with Amit et al. (1998). 5.1g) VC’s views on venture firm performance. What are venture capitalists’ views on determinants of venture firm performance or success? Respondents are asked what they think are some of the key challenges venture capitalists have in terms of selecting life science opportunities to invest in. The responses are compiled into six main categories: financial efficiency challenges, management team challenges, technological evaluation challenges, opportunities availability challenges, future trends and risk challenges and finally deal processing/negotiation challenges (Figure 19). Figure 19. Key Challenges in selection of Life Science Investment Opportunities by Venture Capitalists. West Coast Canada (n=10) and West Coast United States (n= 9) West Coast Canada: Vancouver 80% 70% West Coast United States: Seattle and California Percent VCs 60% 50% 40% 30% 20% 10% 0% Financial Management Tech Evaluation Opportunities Available Future Trends/Risk 283 Deal Processing Of note, the main challenge is management teams (12 responses), technological evaluation (9 responses), opportunities availability and future trends and risks (7 responses each). Canadian respondents (4 responses) are also concerned with deal processing/negotiation challenges. With respect to opportunity availability challenges, these are more pronounced with the Canadian respondents than the United States. The United States respondents are more concerned about future trends and unforeseen risks than the Canadian respondents. 5.1h) Discussion: VC’s views on venture firm performance. Amit et al. (1998) theorizes that there would be a negative relationship between the venture capital invested in a venture firm and its performance. This study’s findings differ from Amit et al. (1998) in that venture capitalists stress the importance of management in terms of their greatest challenges and screening criterion with respect to performance. There are four specific screening criteria discussed, including financial criteria, market size, intellectual property and management. While all these screening criteria are agree to be important, management team is the overwhelming response choice by both Canadian and United States VC’s as being the number one aspect that is sought after. Management experience and the ability to work with selected management are also found as being the number one challenge with respect to having successful opportunity development. Management characteristics are almost more important than the actual technology itself. As one United States respondent comments: “It is easier to get to an opportunity exit with an experienced management team than with a management team that has little to no experience. I’d prefer to work with a poor technology that is being developed by an experienced management team any day over an incredible stellar technology that has an inexperienced management team. I have seen many stellar technologies fail because of inexperienced management teams.” All in all, findings imply that, financials as Amit et al. (1998) see it - the importance of the amount of venture capital invested, actually plays an overstated and small role in venture performance. Technology is the strong second strategy choice of the United States respondents, but surprisingly not the overall focus of the Canadian respondents who responded that stage of the technology is more important. One of the United States respondents mentions that his most pressing difficulty is finding interesting and novel technologies that could be nurtured. 5.2 Venture Capitalist Differences between Canadians and Americans When comparing total years of previous life science experience, Canadian Venture Capital respondents have a total of 24 years combined experience, an average of 2.4 years and a median of 1 year compared to United States respondents who had a total of 63.5 years combined experience, an average of 7 years and a median of 5 years (Figure 20). United States VC’s clearly appear to have more experience. Figure 20. Previous Experience in a Life Sciences Company for West Coast Canadian (n=10) and West Coast United States (n= 9) Venture Capitalists 70 60 Years 50 West Coast Canada: Vancouver 40 West Coast United States: Seattle and California 30 20 10 0 Tota l Yea rs of Ex perie nce in Life Scie nce Ave rage pe r VC 284 When asked what their level of completed education is, 100% of respondents from both countries have completed High School and have a University undergrad degree. Aside from their Undergrad degrees, 20% of the Canadian Venture Capitalists have technical or professional degrees such as in engineering, law or medicine, 20% have Master degrees, 10% have PhDs and 70% have MBA degrees. With respect to the United States Venture Capitalists, 33% have technical or professional degrees, 22% have Master degrees, 44% have PhDs and 44% have MBA degrees. The major difference between the two countries is the PhD and MBA categories where there are more Venture Capitalists from the United States who have PhDs and more Venture Capitalists from Canada that have MBA degrees status. It is interesting to note that of all 19 respondents combined from both countries, 11 of the respondents had MBA degrees, that being 58% of the sample size. United State VC’s appear to have more advanced educational training than Canadian VC’s. The respondents are asked whether it is important to exclude investment opportunities that have unclear regulatory guidelines and risks. Interestingly 67% of the United States Venture Capital respondents either strongly agree or agree compared to 30% of the Canadian respondents and 40% of the Canadian correspondents disagree (Figure 21). Figure 21. Percentage of Venture Capitalists who agree that it is important to reject Investment Opportunities that have Unclear Regulatory Guidelines and Risks. West Coast Canada (n=10) and West Coast United States (n= 9) 45% W e st C o a st C a n a d a : V a n co u ve r 40% 35% s C3 0 % V f 25% o t 20% n e rc 1 5 % e P10% W e s t C o a s t U n i te d S ta te s : S e a ttl e a n d C a lif o rn ia 5% 0% S tr o n g l y A g re e A g re e N e i th e r D is a g re e S tr o n g l y D is a g re e The respondents are asked whether it is important to exclude investment opportunities that carry negative public opinion such as foetus stem cell research, genetically modified organisms etc (Figure 22). Captivatingly, the Canadian respondents (7/10) agree that it is important to exclude these types of investment opportunities compared to the United States respondents (6/9). Figure 22. Percentage of Venture Capitalists who agree that it is important to exclude Investment Opportunities that carry Negative Public Opinion. West Coast Canada (n=10) and West Coast United States (n= 9) W e s t C o a s t C a n a d a : V a n co u v e r 80% W e s t C o a s t U n ite d S ta te s: S e a ttle a n d C a lif o rn ia 70% 60% s C50% V t 40% n e rc 30% e P 20% 10% 0% S tro n g l y A g ree A g ree N eith er D isag ree 285 S tro n g ly D isag ree With respect to this question, the comments given by the respondents are quite different between the two countries, where three (3) of the Canadian respondents mention that Venture Capitalists are risk adverse, do not want to be tainted by bad news, or simply that because their funds are supported by the Canadian government, they have an ethical responsibility of not being involved with investment opportunities which are perceived as unethical by the public. Comments from two of the United States respondents reflect the emphasis on attempts to first overcome any public negativity and then deciding whether to move forwards, especially if the investment opportunity had good potential for great returns. One of the Canadian respondents concurs with this sentiment. However, one United States respondent also mentions that even though, generally, he would not consider this an issue, if the investment opportunity is “too hot to handle” it may be difficult to exit such an opportunity. 5.2 Discussion of Canadian and American venture capitalist differences. Cumming and MacIntosh (2003) find differences between Canadian and US venture capital firms highlighting the impact of legal and institutional factors on exits across these countries. More specifically, “more stringent escrow and hold period requirements in Canada, for example, make it more likely that IPO’s will be effected as partial exits in Canada.” This study is a comparative study that interviews venture capitalists from the Canadian and the United States West Coast. Whereas firm level specific portfolio differences and stage of technology are not researched, selection behaviors are. There are some interesting differences that imply that investment behavior and risks are perceived quite differently between the two countries with respect to life science investments. United States respondents are less apt to agree that public opinion mattered in investment decisions than the Canadian respondents are. Where this most likely reflects the investment climate and portfolios that are held by the Venture Capitalists, it may also be a reflection on different value systems of the two countries. Furthermore, risks such as regulatory risks are perceived more strongly by the United States respondents than the Canadian respondents, reflecting that perhaps governmental risks and legal liability are more strongly linked to choice in the United States than in Canada. This observation would also support the Manigart et al, (2000) discussion on how culture, legal and governmental systems affect the behaviors of investors within a particular country. Conclusion If venture capitalists can only invest in a specific number of portfolio investments, it is crucial that the investment opportunities chosen are those that are most likely to succeed. While the Canadian experience of venture capitalism is examined empirically, this qualitative study examines the subjective experiences of venture capitalist managers which to date is yet to be investigated. This is a study in which nineteen semi-formal interviews are conducted with venture capitalists located on the West Coast of Canada and United States. These findings show large and varying support for Amit et als’ (1998) theory on venture capitalists; yet findings show several more insights. The most important of which is the need for VC’s to develop trust with their venture managers as this is the crux (their key challenge and criterion) by which they can deal with their greatest source of risk, asymmetric information, in selecting investments. Canadian Venture Capitalists tend to present themselves as cautious investors, concerned with ethics and moral issues as well as selectors of criteria and techniques. The United States investors, on the other hand, are more concerned with choosing investments for the sponsorship of important innovation to create wealth, searching internationally if required. The first and intentional limitation of this study is the geographical scope: that being West Coast Canada and United States. Further study into the direct linkages of education and experience levels with investment success rate should be pursued. 286 References 1. Amir-Aslani, Arsia and Syoum Negassi. (2006). Is technology integration the solution to biotechnology’s low research and development productivity? Technovation. Volume 26, 573-582. 2. Amit, Raphael, James Brander and Christoph Zott. (1998). Why do venture capital firms exist? Theory and Canadian evidence. Journal of Business Venturing, Volume 13, 441-466. 3. Baeyens, Kathleen, Tom Vanacker and Sophie Manigart. (2006). Venture capitalists’ selection process: the case of biotechnology proposals. International Journal of Technology Management, Volume 34, 28-46. 4. Bains, W. (2004). Failure rates in drug discovery and development: will we ever get any better? Drug Discovery World, Fall, 9-18. 5. Baum, Joel A.C. and Brian S. Silverman. (2004). Picking winners or building them? Alliance, intellectual, and human capital as selection criteria in venture financing and performance of biotechnology startups. Journal of Business Venturing, Volume 19, 411-436. 6. Bishop, Karen and Robert D. Nixon. (2006). Venture Opportunity Evaluations: Comparisons between venture capitalists and inexperienced pre-nascent entrepreneurs. Journal of Developmental Entrepreneurship, Volume 11, 19-33 7. Cumming, Douglas J. (2005). Agency costs, institutions, learning and taxation in venture capital contracting. Journal of Business Venturing, Volume 20(5), 573-622. 8. Cumming, Douglas J. and Jeffrey G. MacIntosh. (2006). Crowding out private equity: Canadian evidence. Journal of Business Venturing, Volume 21(5), 569-609. 9. Cumming, Douglas J. and Jeffrey G. MacIntosh. (2006). A cross-country comparison of full and partial venture capital exits. Journal of Banking and Finance, Volume 27(3), 511-548. 10. Danzon, Patricia M. (2005). Productivity in pharmaceutical biotechnology R&D: the role of experience and alliances. Journal of Health Economics, Volume 24, 317-339. 11. DiMasi, Joseph A., Ronald E. Hansen and Henry G. Grabowski. (2003). The price of innovation: new estimates of drug development costs. Journal of Health Economics, Volume 22, 151-185. 12. Edwards, M.G., Murray F. and R. Yu (2003). Value Creation and sharing among universities, biotechnology and Pharma. Nature Biotechnology. Vol. 21, 618-624. 13. Evans, Anne G. and Nikhil P. Varaiya. (2003). Anne Evans: Assessment of a Biotechnology Market Opportunity, Entrepreneurship: Theory and Practice. Fall, 87-105. 14. Fetterhoff, Terry J. and Dirk Voelkel. (2006). Managing open innovation in biotechnology. Research Technology Management, Volume 49, 14-18. 15. Fried, Vance H. and Robert D. Hisrich. (1994). Toward a Model of Venture Capital Investment Decision Making. Financial Management, Volume 23, No. 3, 28-37. 16. Hair, Joseph F Jr., Arthur H. Money, Philip Samouel and Barry Babin. (2003). Essentials of Business Research Methods. John Wiley & Sons; Leyh Publishing, LLC. 17. Manigart, Sophie, K. DeWaele, M. Wright, K. Robbie, P. Desbriere, H.J. Sapienza and A. Beekman. (2000). Venture capitalists, investment appraisal and accounting information: a comparative study of the USA, UK, France, Belgium and Holland. European Financial Management, Volume 6, No. 3., 389-403. 18. Pharma Ventures. (2005). Pharmaceutical Deal Structures: The Essential Manual for Deal Makers. Pharma Ventures Ltd. Magdalen Centre. Oxford Science Park, Oxford OX4 4GA, UK. 287 19. Rousu, Mathew C. (2004). Estimating the public value of conflicting information: the case of genetically modified foods. Land Economics, Volume 80, 125-135. 20. Sahlman, William A. (1990). The Structure and Governance of Venture-Capital Organizations. Journal of Financial Economics, Volume 27, 473-521. 21. Zacharakis, Andrew L. and Meyer, G. Dale. (2000). The potential of actuarial decision models: can they improve the venture capital investment decision? Journal of Business Venturing, Volume 15, 323-346. 288 ASAC 2008 Halifax, Nova Scotia In Sunwoo Bachelor of Commerce Concordia University – John Molson School of Business MARKET VALUATION OF CORPORATE CHARITABLE DONATIONS This study investigates the relationship between corporate charitable donations and stock market valuations Introduction Over the past decade, Corporate Social Responsibility (CSR) has become a major concern for firms in appeasing the skepticism of stakeholder groups including shareholders, consumers, and organizational members at various levels (Donaldson and Preston, 1995; Kapstein, 2001). Significant changes have occurred in the way in which firms would like to be regarded, as evidenced by the increased reporting of charitable activities in annual reports. The literature on CSR has treated many issues such as the determinants thereof, within stakeholder (Roberts, 1992) and financial (Waddock and Graves, 1997) models. A recent UK study by Brammer and Millington (2004) focuses on the determinants of one aspect of CSR, charitable donations, rather than investigating a composite measure of CSR which bundles its different aspects (e.g. employee relations, environmental policy, etc.). Along the same focus on charitable donations, other angles addressed in the US and Canada include: the relationship between charitable giving and corporate governance (e.g. Bartkus et al., 2001; Coffey and Wang, 1998); the motivation behind the allocation of corporate donations among various categories of recipients (e.g. Leclair and Gordon, 2000); and the impact of tax changes on corporate charitable giving (e.g. Schwartz, 1968). In becoming an increasingly popular topic, corporate charitable giving has also become the subject of criticism by observers who believe that it, as well as that of CSR activities in general, is not only disingenuous, but is in fact damaging. In a special report by The Economist on CSR, Clive Crook (2005) separates the positive effects of charitable donations into increased social welfare and an increase in firm profits. According to Crook, few companies achieve both effects through their donations, and many fail to achieve either. While the article does not provide statistical support for its conclusions, its criticism nonetheless introduces questions that may deserve more attention (in order to judge the full effect of CSR activities such as charitable donations). Given that such criticism of a fundamentally positive activity exists, it is of interest to determine how capital markets respond to charitable donations. Though analyzing charitable donations has the advantage of dealing with an easily quantifiable activity, the disadvantage is the lack of publicly available data. Aggregate corporate giving is publicly available, but firms are bound to disclose neither the timing nor the amount of their donations. This may explain the disproportionate relationship between interest in charitable activities and existing in-depth analyses thereof. The data used in this paper is a collection of the annual surveys conducted by The Chronicle of Philanthropy, which collects information on a large number of Fortune 500 companies in terms of cash as well as cash and product giving. This paper investigates corporate charitable giving in the following ways. 289 First, we measure the effect of corporate charitable contributions on market valuation while allowing for the effects of firm characteristics on both the level of charitable donations and market valuation itself. Second we explore possible industry-specific trends in the impact of charitable donations on market valuation. To this end we identify sectors which may be seen especially sensitive to stakeholder perception as outlined by Brammer and Millington (2004); and Gordon and Leclair (2000). Third, we also study the amount of donations in excess of the expected amount to determine whether it has a significant effect on market valuation motivated by the need to signal positive CSR behavior. Finally, we analyze whether market valuation of CSR behavior has changed within recent years. The following analysis consists of four sections. The first section introduces the stakeholder model of charitable donations. The section ends with the hypotheses tested in this study. The derivation of the sample is discussed in the second section. The empirical tests conducted and a brief statement of results is presented in the third section. A further discussion of results and their implications, as well as suggestions for further study are outlined in the concluding section. Corporate Charitable Donations: The Stakeholder Model The Stakeholder Model used by Brammer and Millington (2004)(figure1.1) attempts to explain the relationships and variables relevant in determining the level of charitable donations. According to stakeholder theory, the firm faces the task of managing the impact of competing or conflicting stakeholder demands on the overall objectives of the company (Freeman, 1984). The model describes the fact that varying attitudes are filtered through firm-specific characteristics before resulting in actual charitable activities (Brammer and Millington, 2004) and it posits that an expected level of charitable donations can be predicted. This dynamic of stakeholder demands being filtered through firm-characteristics, and resulting in a level of charitable donations is essentially the focus of Brammer and Millington (2004). Their model also implies that there should be a feed-back effect of charitable donations that comes back full circle to influence the stakeholders. Figure 1.1. The stakeholder model 290 The focus of this paper is to determine, within the schemata of the stakeholder model, what subsequent effect the resulting response (to stakeholder pressures concerning charitable donations) has on stock market valuation. In particular, we attempt to identify distinct motivations for charitable donations. Conceptually, we hypothesize the following progression of actions: the firm finds itself within a constellation of competing and/or conflicting demands concerning charitable donations (or other CSR activities) and produces a response as a function of its firm-specific characteristics which, along with the respective stakeholder pressures, dictate the specific motivations driving charitable giving. These characteristics include financial measures such as profit and advertising expenditures which may deter or favor charitable donations, as well as industry classifications which will dictate the relative power, urgency, and legitimacy of specific stakeholder group demands (Mitchell et al., 1997). This study predicts that there are three possible motivations behind corporate charitable donations. First, there is the marketing benefit of visibility in the media, which is expected to increase sales revenues. Second, donations may help to mitigate negative environmental or social impacts on stakeholders. Thirdly, charitable activity may be used as a signal of being a good corporate citizen. However, while in aggregate we expect a positive relationship between donations and market valuation, we also expect to observe variation in this relationship across industries, because of the difference in motivations prevalent in certain sectors, and across time, as CSR activities have gained increasing public attention in recent years. Philanthropic activity by a pharmaceutical firm might, for example, increase brand image and cause an increase in sales, while the same action by an industrial manufacturer may not produce the same outcome. Instead, the response, if any, may go towards mitigating the negative environmental effects its operations might have on the public (Brammer and Millington, 2004). Furthermore, stakeholder perception, and the resulting market valuation of a CSR signal may have fundamentally changed in recent years. To differentiate between the three suggested motivations we (1) proxy for negative social and environmental effects using SECOST (paper products, petroleum, tobacco, and alcohol sectors); (2) split the donations into two parts- a part due to potential business benefits (e.g. increased sales) and an excess part; (3)and split the sample into two sub samples consisting of donations from 1997 to 2000 and 20012003. Step (1) seeks to differentiate between the marketing motivation and the need to mitigate negative social or environmental effects. Step (2) is taken because if the market values corporate donations due to their marketing effect, then we would expect to find a positive relationship only between the predicted (expected) level of donations and market valuation. If, on the other hand, donations are viewed by the market as a signal of good corporate citizenship, we expect the positive relationship to be largely due to the excess donations (i.e. donations over and above those predicted by the business fundamentals). This last conclusion is based on the fact that the unexpected level of donations given cannot be explained by “business fundamentals” such as quantitative attributes (e.g. marketing expense, ROA, etc.) and must thus be motivated by the need to engage in philanthropic activity as a signal of good corporate citizenship. Step(3) is taken to gauge whether recent industry and media focus on CSR behavior may have increased the intrinsic value of such activities, we split the sample into two sub samples. Since the media focus regarding CSR began to take off in 2001 we use that year to split the sample. We expect to find a stronger relationship between excess donations and market valuation in the later sub sample under the assumption that the recent media focus of CSR has led to an increase in perceived value of such activities. This paper hypothesizes that the interrelationships described above and their effects are interpreted and priced by the market. Based on the above discussion, we now summarize the hypotheses. Presentation of Hypotheses. H1: Market valuation will be positively related to charitable donations. 291 H2: H3: H4: Market valuation will be differently related to charitable donations for firms that operate in sectors that impose a social or environmental cost on its stakeholders. If charitable donations are motivated largely by business fundamentals, predicted (expected) charitable donations will have a substantial effect on market valuation. If charitable donations are used to signal good corporate citizenship1, excess (residual term) charitable donations will have a substantial effect on market valuation. As the recent media focus on CSR may have increased the intrinsic value of corporate charitable donations, we expect to observe a stronger relationship between donations and market valuation in the later years of our sample. Data and Sample Selection The data used in this study comes from several sources. The annual data on cash and product giving is obtained from The Chronicle of Philanthropy’s annual studies in which 150 largest Fortune 500 firms are asked to report their level of donations. The 1999 values, however, represent estimates made by companies in 1998 about their planned future donations. All other values represent actual giving. The dataset is comprised of all the firms that chose to report. Compustat is used to obtain the various data items that are used to construct the remaining dependent variables with the exception of the percentage owned by institutional shareholders, which is from the Compact D database, as well as Wynan’s International Preferred Yield Index and the implicit price deflator for GDP for nonresidential fixed investments which are taken from public sources. We excluded financial firms because they are generally assumed to operate differently than other firms. Table 1 shows how we arrive at our final sample. Table 1 Summary of Dataset Breadth The initial sample was obtained from the Chronicle of Philanthropy 1997 1998 1999 2000 2001 2002 Initial sample 89 89 68 75 84 98 2003 Total 89 592 Financial firms 28 28 28 28 28 28 28 196 Compustat n.a. 9 6 5 4 2 2 0 28 Final Dataset 52 55 35 43 54 68 61 368 1 It is important to note that we distinguish between motivation to mitigate negative social or environmental impacts and the motivation to signal good corporate citizenship. We control for this difference by including the sensitive industry variable in the expected (predicted) amount of charitable donations and equating CSR motivated donations to the excess (residual term) amount of charitable donations. 292 Empirical Tests and Results Tobin’s Q and the Modified Lerner Recursive Algorithm To investigate whether charitable donations have a significant association with firm value, we estimate least-squares regressions, using Tobin’s Q as the dependent variable and charitable donations (as a percentage of profits) as one of many explanatory variables. The dependent variable, measured for each company at the close of each fiscal year ending in calendar 1997 through 2003, is defined as Tobin’s Q = Market Value of Assets Replacement Cost of Assets (1) Tobin’s Q may be measured in a variety of ways. We use a modified version of Tobin’s Q used by Perfect and Wiles (1994). The qpw method has the advantages that (1) all of the necessary data is publicly available, except for a price deflator, and a preferred stock yield index, (2) asset replacement cost data, which is not reliably reported, is not necessary, and (3) the measure can easily be computed using a small set of algorithms for a large number of firms (Perfect and Wiles, 1994). In particular, we calculate qpw as: qpw = COMVAL + PREFVAL + PWBOND + STDEBT PWRC (2) where COMVAL is the year-end value of the firm’s common stock(the product of Compustat item 25 and 199); PREFVAL is the estimated year-end market value of the firm’s preferred stock (the quotient of Compustat data item 19 and Wynan’s International Preferred Yield Index); PWBOND is the market value of the firm’s long-term debt (Compustat item 9); STDEBT is the year-end book value of the firm’s shortterm debt with a maturity less than one year (Compustat item 44); and PWRC is estimated replacement cost using the modified LR technique. The modified LR technique for determining PWRC is based on the following equation of the original LR technique: RNPLR,t = RNPLR,t-1 [(1+Фt)/ (1+δt)(1+θt)] + It, for t>0; (3) where RNPLR,t is the LR estimated value for net plant replacement cost in year t; RNPLR,0 is HNP0; Фt is the growth of capital prices in year t; δt is the real depreciation rate in year t; is the cost-reducing rate of technological progress; and θt is the investment in new plant in year t. The growth of capital goods prices in year t, Фt, is estimated by the Gross National Product deflator for nonresidential fixed investment. The real depreciation rate in year t is estimated by: δt = DEP t HNP t-1; (4) where DEPt = book depreciation in year t. The Modified LR technique, however, makes the following simplifying assumptions: that firm replacement cost estimates are not available; and that LR’s rate of cost-reducing technical progress, θt, is zero. The result is the following Perfect and Wiles (1994) model: RNPPW, t = RNPPW,t-1 [(1+Фt)/ (1+δt)] + It, 293 for t>0; (5) and RNPPW, t0 = HNPPW,t0 (6) We compute the value of PWRC in the manner described above with the exception that the Implicit Price Deflator for GDP (nonresidential fixed investments) in constant year 2000 dollars is used instead of the Gross National Product deflator for nonresidential fixed investment. We compile the constituents of the denominator in equation (2) (using Compustat tape items) as follows: HNPPW,to is equated to gross value of property, plant, and equipment (#7); DEP t is equated to depreciation and amortization (#14), and It is equated to capital expenditures on property, plant, and equipment (#30). OLS Regression Models: Basic Model To test the hypothesis (1) developed in section 2, we employ the following OLS model: ltq = β0 + β1 CHDONPER + β2 ROA + β3 FIO + β4 FS + β5 LEVER + β6 YEAR1997 + β7 YEAR1998 + β8 YEAR1999 + β9 YEAR2000 + β10 YEAR2001 + β11 YEAR2002 + ε (7) where ltq is the log of one plus qpw. CHDONPER is charitable donations, defined as total cash and product giving by a firm divided by the operating income before depreciation. We expect the coefficient of CHDONPER to be positive as suggested in hypothesis (1) above. A large number of studies has examined the determinants of firm-valuation, and as a result there are a proportionate amount of variables that might be considered as control variables in the present study. We have selected the most widely used control variables. In particular, we use ROA, future investment opportunities (FIO), firm size (FSIZE), and leverage (LEVER) (e.g. Yermack, 1996; Smith and Watts, 1992; Brammer and Millington, 2004). All of the control variables listed here are derived from Compustat tape data items as follows: ROA is operating income before depreciation (#13) divided by total assets (#6); FIO is a ratio of property, plant, equipment capital expenditures (#30) to net sales (#12); FSIZE is the log of COMVAL plus PREFVAL (derived as described previously); and LEVER is the ratio of total debt outstanding to total assets (#6). Panel A of Table 2 summarizes our measurement of charitable donations on a yearly basis. We note that both the mean CHDONPER and CHDON increase significantly over the sample period, while their respective medians also increase, but at a slower rate. Panel B of Table 2 summarizes the independent variables. The year dummies (YEAR1997, YEAR1998, YEAR1999, YEAR2000, YEAR2001, and YEAR2002) attempt to control for the influence of overall bullish or bearish attitudes in the market on valuations. Regression 1 in Table 3 reports the results with ltq as the dependent variable. We find that firms with a higher proportion of giving to profits also have higher market valuations. This relationship is statistically significant at the 1% level and is of marginal economic significance as there is an 8.51% change in ltq per standard deviation change, an increase of $0.01722 per dollar of operating income before depreciation, in CHDONPER. OLS Regression Models: Expanded Model To test hypothesis (2) we use the following regression models: ltq = β0 + β1 CHDONPER + β2 ROA + β3 FIO + β4 FS + β5 LEVER + β6 YEAR1997 + β7 YEAR1998 + β8 YEAR1999 + β9 YEAR2000 + β10 YEAR2001 + β11 YEAR2002 + β12 SENSITIVE + ε 294 (8) ltq = β0 + β1 CHDONPER + β2 ROA + β3 FIO + β4 FS + β5 LEVER + β6 YEAR1997 + β7 YEAR1998 + β8 YEAR1999 + β9 YEAR2000 + β10 YEAR2001 + β11 YEAR2002 + β12 SENSITIVE + β13 MKTEFFECT + ε (9) Table 2 Summary Statistics of Main Variables The sample consists of 368 cash and product donations between 1997 and 2003. In Panel A, CHDONPER is the ratio of charitable donations to total operating income before depreciation. CHDON is the total dollar value (in millions) of cash and product donations. In Panel B, ROA is operating income before depreciation (Compustat data item #13) divided by total assets (#6); FIO is a ratio of property, plant, equipment capital expenditures (#30) to net sales (#12); FSIZE is the log of COMVAL plus PREFVAL (derived as described previously); LEVER is the ratio of total debt outstanding to total assets (#6); SECOST is a dummy variable that takes on a value of one if the firm is in the petroleum, tobacco, paper products, or alcohol industry; MITIGATE is the interactive variable that is the product of SECOST and CHDONPER; PERCENT is defined as the proportion of total shares held by institutional investors; and MARKETING is defined as the ratio of total advertising expenses (# 6) to total operating income before depreciation (13). The figures below represent mean values and median values in parentheses. Panel A: summary statistics for CHDONPER and CHDON per year Year CHDONPER 1997 1998 1999 2000 2001 2002 2003 Total 0.7404 0.7659 1.2552 0.9096 1.1756 1.1179 1.2312 1.0203 (0.6264) (0.5590) $32.500 $37.700 $37.400 $61.400 $66.600 $67.400 $84.300 ($18.100) ($21.800) ($16.700) ($26.600) ($25.900) ($22.900) ($25.000) $55.900 ($21.800) (0.4563) (0.4883) (0.4219) (0.5976) (0.7436) (0.6662) CHDON Panel B: summary statistics for dependent and independent variables Ltq CHDOPERC ROA FIO FSIZE LEVER SECOST MITIGATE PERCENT MARKETING Min 0.1627 0.0003 -0.2077 0.0017 8.9685 0.0000 0.0000 0.0000 0.0995 0.00002 Mean 1.0325 0.0102 0.1531 0.0747 10.3995 0.2393 0.1748 0.0007 0.5738 0.0358 295 Median 0.8490 0.0056 0.1477 0.0565 10.3671 0.2352 0.0000 0.0000 0.5693 0.0269 Max Std. dev. 4.8837 0.7646 0.2484 0.0172 0.4306 0.0749 0.3623 0.0635 11.7062 0.5468 0.6645 0.1352 1.000 0.3800 0.0161 0.0022 0.9569 0.1311 0.1342 0.02862 Table 3 The Relationship between Charitable Donations and Market Valuation The estimation is done using ordinary least-squares regressions with robust standard errors. The dependent variable is market valuation, ltq, defined as the log of one plus qIN. ROA is operating income before depreciation (#13) divided by total assets (#6); FIO is a ratio of property, plant, equipment capital expenditures (#30) to net sales (#12); FSIZE is the log of COMVAL plus PREFVAL; and LEVER is the ratio of total debt outstanding (#9 plus #44) to total assets (#6). SECOST is a dummy variable that takes a value of one if the firm is produces paper products, tobacco, alcohol, or petroleum. MITIGATE is an interactive dummy defined as the product of SECOST and CHDONPER which attempts to measure if charitable donations by sensitive sector firms have a different effect on market valuation. ***, **, and * denote significance at the 1, 5, and 10 percent level, respectively. 1 2 3 4 Constant -5.281 -4.996 -5.126 -4.987 (9.24)** (8.75)*** (8.95)*** (8.72)*** CHDONPER 6.787 5.200 6.042 5.111 (4.21)** (3.42)*** (3.80)*** (3.38)*** ROA 0.674 0.881 0.902 0.851 (1.48) (1.97)** (1.99)** (1.90)* FIO -4.299 -4.488 -4.430 -4.485 (8.14)** (8.62)*** (8.39)*** (8.61)*** FS 0.619 0.599 0.607 0.599 (10.72)** (10.60)*** (10.55)*** (10.62)*** LEVER -0.955 -1.026 -0.984 -1.031 (3.73)** (4.11)*** (3.91)*** (4.13)*** YEAR1997 0.204 0.151 0.173 0.150 (2.38)* (1.85)* (2.07)** (1.83)* YEAR1998 0.201 0.159 0.180 0.157 (2.39)* (1.95)* (2.18)** (1.92)* YEAR1999 0.290 0.240 0.257 0.240 (3.49)** (2.88)*** (3.10)*** (2.88)*** YEAR2000 0.245 0.254 0.244 0.256 (2.57)* (2.80)*** (2.62)*** (2.84)*** YEAR2001 0.181 0.193 0.188 0.194 (2.00)* (2.23)** (2.12)** (2.23)** YEAR2002 0.117 0.112 0.123 0.109 (1.47) (1.49) (1.57) (1.46) SECOST -0.425 -0.512 (5.36)*** (3.22)*** MITIGATE -47.953 15.382 (4.34)*** (0.94) Observations R-squared 362 0.53 362 0.57 362 0.55 296 362 0.57 SECOST is a dummy variable that takes on a value of one if the firm is in an industry that imposes social or environmental costs on the public and thus has a vested interest to reduce the risk of adverse reactions from stakeholder groups (Brammer and Millington, 2004). The final set includes the following industries: Paper products, Petroleum, Tobacco, Cigarettes, and Alcohol2. An interactive dummy, MITIGATE, attempts to measure whether there is a significantly different effect of charitable donations made by these “sensitive” firms on market valuation than that of charitable donations made by firms in other sectors. We use the dummy is first regressed in regression 2 to test if there was a general bullish or bearish trend among sensitive industries. Then, allowing for the differences in both the intercept and the slope, we use both SECOST and MITIGATE in regression 4 to determine whether charitable donations made by sensitive industries have a different impact on market valuation (hypothesis 2). Regressions 2,3, and 4 in Table 3 show that while firms in SECOST industries have lower valuations (with statistical significance at the 1% level and economic significance as shown by a coefficient to dependent variable mean proportion of 50%), there is no statistically significant difference in the valuation of charitable donations by sensitive industry firms and the donations of firms in other industry sectors (as MITIGATE is found to be statistically insignificant). OLS Regression Models: Two-Stage Regression Model Hypothesis (3) aims to take into account that, given the stakeholder model and its influences, the level of a firm’s charitable donations can to some extent be predicted. In the first stage of this two-stage regression model, we attempt to estimate the level of charitable donations as a proportion of profits by regressing it against several firm variables. CHDONPER = β0 + β1 ROA + β2 FIO + β3 FS + β4 LEVER + β5 YEAR1997 + β6 YEAR1998 + β7 YEAR1999 + β8 YEAR2000 + β9 YEAR2001 + β10 YEAR2002 + β11 PERCENT+ β12 MARKETING + β13 SECOST+ ε (10) The independent variables ROA, FSIZE, FIO, and LEVERAGE are used once more as past literature suggests that these variables, which are used as control variables in equation (6), may influence charitable giving. In addition to these we add a variable called MARKETING, defined as the ratio of total advertising expenses (Compustat item # 6) to total operating income before depreciation (Compustat item # 13). In their study, Gordon and Leclair (2000) find that there is a statistically significant relationship between the level of advertising expenses and the level of charitable giving. Given that charitable donations may indeed be part of an overall marketing effort, it follows that the level of advertising that a firm engages in may contribute to determining its overall level of donations. We therefore expect a positive relationship between the level of advertising expenses and charitable donations. Furthermore, the independent variable PERCENT, representing the percentage of stock owned by institutional shareholders, is also incorporated. This variable is used as a proxy for the existence of 2 The existing literature also states that consumer-focused sectors are more likely to be strongly influenced by public perception, but as the entire sample of firms used in this study is composed of Fortune 500 firms, presumably all being under a high level of public scrutiny, we omit a “consumerfocus” dummy 297 active monitors. The reasoning behind this last variable is that this specific financial stakeholder group will exert a material influence on the level of donations that will be proportionate to its magnitude (as a percent of total share ownership). Regression 1 in Table 4 shows that both PERCENT and MARKETING are found to be statistically significant at the 5% and 10% level, respectively, and economically significant. A one standard deviation change in PERCENT and MARKETING will bring about a 47.6% and 24.4% change, respectively, in CHDONPER. SECOST is also included as an independent variable and is found to statistically significant at the 5% level and economically significant (in that the coefficient of SECOST is of a 78% magnitude in comparison to the dependent variable mean). Essentially, by incorporating all of the above independent variables, we attempt to determine the level of charitable donations that business fundamentals would dictate. That is to say that we would expect a firm with a certain set of characteristics pertaining to industry and several financial measures to donate accordingly. It follows that the difference between actual donations, and those predicted by equation (10), that are captured in the residual error term, would be due to motivations unrelated to the aforementioned “business fundamentals”. Assuming that charitable donations not prompted by business fundamental rationale are due CSR motivations, we expect excess donations (XCHDONPER) to have a substantial impact on market valuation as stated in the second part of hypothesis (3), which we test in the following regression: ltq = β0 + β1 XCHDONPER + β2 ROA + β3 FIO + β4 FS + β5 LEVER + β6 YEAR1997 + β7 YEAR1998 + β8 YEAR1999 + β9 YEAR2000 + β10 YEAR2001 + β11 YEAR2002 + ε (11) Regression 2 in Table 4 shows that firms giving more charitable donations than expected have higher market valuations. This relationship is statistically significant at the 1% level and is of substantial economic significance: a one standard deviation change in XCHDONPER brings about a 10.4% change in ltq. In the final stage of analysis we regress excess donations and expected donations on the dependent variable over two different time periods. As the year 2001 marks a sharp increase in media focus on CSR, we use this year to separate our sample. Regression 1 and 2 in Table 5 show the relationship between excess donations and expected donations to market valuation, respectively, for the time period from 2000 to 2003; and regressions 3 and 4 show the same relationships for the time period from 1997 to 2000. We find that in the earlier sample, expected donations were statistically significantly related to market valuation at the 10% level with economic significance. In the latter sub sample, however, expected charitable donations lose their statistical significance, and excess charitable donations is found to be very strongly related to market valuation (with statistical significance at the 1% level). This relationship is found to be of economic significance. This suggests that recent media focus has shifted the perceived market value of charitable donations away from the portion due to business fundamentals towards CSR behavior as predicted in hypothesis (4). Due to the large number of missing MARKETING variable entries, we run a robustness test3 as shown in regressions 3 and 4 and find that our results hold. 3 missing advertising expense values are equated to “0” 298 Table 4 The Relationship Between Excess Charitable Donations and Market Valuation The estimation is done using ordinary least-squares regressions with robust standard errors. In regression 1 the dependent variable is CHDONPER, defined as the ratio of charitable donations to total operating income before depreciation. ROA is operating income before depreciation (#13) divided by total assets (#6); FIO is a ratio of property, plant, equipment capital expenditures (#30) to net sales (#12); FSIZE is the log of COMVAL plus PREFVAL (derived as previously); and LEVER is the ratio of total debt outstanding to total assets (#6). In regression 2 the dependent variable is market valuation, ltq, defined as the log of one plus qpw. CHDONPER is replaced by XCHDONPER, defined as the residual term in regression [1], the unexplained portion of CHDONPER; and EXPCHDONPER is defined as the explained portion of CHDONPER. ). SECOST is a dummy variable that takes a value of one if the firm is produces paper products, tobacco, alcohol, or petroleum. MITIGATE is an interactive dummy defined as the product of SECOST and CHDONPER which attempts to measure if charitable donations by sensitive sector firms have a different effect on market valuation. PERCENT is defined as the proportion of total shares held by institutional investors, and MARKETING is defined as the ratio of total advertising expenses (# 6) to total operating income before depreciation (13) (MARKETING1 is used for the robustness test to allow for a larger sample size by equating missing advertising expense values to “0”). ***, **, and * denote significance at the 1, 5, and 10 percent level, respectively. Dependent variable Constant ROA FIO FS LEVER YEAR1997 YEAR1998 YEAR1999 YEAR2000 YEAR2001 YEAR2002 PERCENT MARKETING (1) CHDONPER (2) ltq (3) ltq (4) CHDONPER (5) ltq (6) ltq -0.083 (2.36)* -0.009 (0.43) -0.018 (0.86) 0.007 (2.31)* 0.015 (1.89) -0.007 (1.75) -0.006 (1.57) -0.006 (1.70) -0.003 (0.85) 0.001 (0.23) -0.001 (0.19) 0.037 (3.58)** 0.087 (2.18)* -5.585 (6.58)*** 1.242 (2.91)*** -5.442 (10.17)*** 0.676 (8.15)*** -1.773 (5.86)*** 0.061 (0.63) 0.110 (1.14) 0.135 (1.15) 0.268 (2.39)** 0.226 (2.31)** 0.155 (1.61) -5.618 (6.33)*** 0.982 (2.14)** -5.058 (7.04)*** 0.666 (7.68)*** -1.917 (6.19)*** 0.144 (1.25) 0.187 (1.63) 0.221 (1.64) 0.294 (2.65)*** 0.209 (1.97)* 0.156 (1.58) -0.014 (0.96) 0.014 (0.96) -0.018 (1.34) 0.002 (1.48) 0.002 (0.42) -0.004 (1.54) -0.005 (1.62) -0.005 (1.91)* -0.003 (1.07) 0.001 (0.18) -0.001 (0.20) 0.005 (1.23) -5.585 (6.58)*** 1.242 (2.91)*** -5.442 (10.17)*** 0.676 (8.15)*** -1.773 (5.86)*** 0.061 (0.63) 0.110 (1.14) 0.135 (1.15) 0.268 (2.39)** 0.226 (2.31)** 0.155 (1.61) -5.618 (6.33)*** 0.982 (2.14)** -5.058 (7.04)*** 0.666 (7.68)*** -1.917 (6.19)*** 0.144 (1.25) 0.187 (1.63) 0.221 (1.64) 0.294 (2.65)*** 0.209 (1.97)* 0.156 (1.58) -0.066 -0.036 MARKETING1 SECOST -0.008 -0.066 -0.036 299 0.079 (2.30)** -0.006 (3.50)** MITIGATE XCHDONPER (0.37) -38.899 (1.91)* 6.925 (4.17)*** EXPCHDONPER Observations R-squared 179 0.16 176 0.74 (0.18) -31.345 (1.51) 11.814 (1.33) 176 0.72 300 (4.62)*** 353 0.11 (0.37) -38.899 (1.91)* 6.925 (4.17)*** 176 0.74 (0.18) -31.345 (1.51) 11.814 (1.33) 176 0.72 Table 5 The Relationship Between Excess Charitable Donations and Market Valuation over Time The estimation is done using ordinary least-squares regressions with robust standard errors. In regression 1 the dependent variable is CHDONPER, defined as the ratio of charitable donations to total operating income before depreciation. ROA is operating income before depreciation (#13) divided by total assets (#6); FIO is a ratio of property, plant, equipment capital expenditures (#30) to net sales (#12); FSIZE is the log of COMVAL plus PREFVAL (derived as previously); and LEVER is the ratio of total debt outstanding to total assets (#6). In regression 2 the dependent variable is market valuation, defined as the log of one plus qIN (an altered version of the modified Lerner measure). CHDONPER is replaced by XCHDONPER, defined as the residual term in regression [1], the unexplained portion of CHDONPER; and EXPCHDONPER is defined as the explained portion of CHDONPER. ***, **, and * denote significance at the 1, 5, and 10 percent level, respectively. Constant ROA FIO FS LEVER 2001-2003 1 2001-2003 2 1997 – 2000 3 1997 - 2000 4 -5.617 (4.93)*** 0.406 (0.59) -5.820 (6.31)*** 0.690 (6.04)*** -1.632 (4.16)*** -5.527 (4.51)*** 0.173 (0.24) -5.458 (4.70)*** 0.671 (5.33)*** -1.654 (4.18)*** -6.487 (4.86)*** 1.734 (3.09)*** -5.098 (7.30)*** 0.760 (5.84)*** -2.226 (4.32)*** 0.000 -6.810 (5.45)*** 0.864 (1.33) -4.172 (4.07)*** 0.787 (6.66)*** -2.793 (6.85)*** 0.000 (0.00) 0.042 (0.00) 0.020 (0.41) 0.075 (0.20) 0.067 (0.60) 0.225 (0.57) 0.097 (1.66) (0.52) -0.288 (1.11) -15.048 (0.52) 0.705 (0.10) -0.181 (0.67) -1.481 (0.07) year== 1997.0000 year== 1998.0000 year== 1999.0000 year== 2000.0000 year== 2001.0000 year== 2002.0000 SECOST MITIGATE XCHDONPER 0.241 0.227 (2.35)** 0.161 (2.04)** 0.160 (1.69)* -0.061 (0.37) -32.156 (1.58) 6.637 (4.44)*** (1.63) -0.056 (0.30) -25.197 (1.27) 301 EXPCHDONPER Observations R-squared 8.709 (0.78) 99 0.70 27.113 (1.77)* 99 0.68 77 0.79 302 77 0.80 Summary, Interpretation of Results, and Conclusion Corporate Social Responsibility has in recent years received much media attention, and with it, a proportionate amount of skepticism. This paper has investigated the effect of charitable donations, one of the more objectively quantifiable aspects of CSR activity, on market valuation over the period of 19972003 within the framework of stakeholder theory. In a broad sense, this study has separated the motivations behind charitable donation into those driven by business fundamentals and those driven by a desire to signal CSR. We divide the business fundamentals motivation further into a marketing motivation and the desire to mitigate negative social and environmental impacts on stakeholders. Having defined three distinct motivations we empirically study how the market values the charitable donations in which they result. Our main findings are as follows. First, firms that engage in charitable donations, on the whole, have higher market values. Second, though firms in sectors incurring social or environmental costs tend to have lower market valuations, their charitable donations are not priced differently than those of firms in other sectors. Third, the amount of unexpected charitable donations given results in a higher market valuation while the expected amount does not. Fourth, the higher market valuation of the unexpected portion of charitable giving is a phenomenon that is found to be prevalent in recent years, coinciding with the increased media focus on CSR. Finally, percent ownership by institutional shareholders, level of advertising, and whether the firm imposes social or environmental costs are all found to be statistically and economically relevant to the level of charitable donations. The former of these variables has not been discussed in the relevant literature and may deserve further investigation. Our findings are consistent with hypothesis (1), (3), and (4). We have found empirical evidence that charitable donations are motivated by two distinct rationales: business fundamentals, and nonbusiness fundamentals (which we presume to be CSR motivated). We infer from the relative strength of the relationship between unexpected donations and market valuation over time that stakeholders, and the resulting market perception have become more scrutinizing. As CSR activities have gained increased media exposure and proportionate amounts of criticism, it seems that markets have learned to distinguish between genuine CSR activity and business fundamentals-driven activity thinly veiled under the guise of CSR activity. 303 References Bartkus, B. R., Morris, S. A., Seifert, B. 2001. “Stronger governance mechanisms: do they influence corporate philanthropy?” Paper presented at the Academy of Management, Washington DC. Brammer, S., Millington, A. (2004). The development of corporate charitable contribution in the UK: A Stakeholder Analysis. Journal of Management Studies, 41, 1411-1434 Coffey, B.S., Wang, J. (1998). Corporate stakeholders and corporate finance. Financial Management, 16, 5-14. Crook, C. (2005). A survey of corporate social responsibility: the good company. The Economist, January 22nd –28th, p. 3-22. Donaldson, T., Preston, L. E. (1995). The stakeholder theory of the corporation: concepts, evidence and implications. Academy of Management Review, 20, 65-91. Freeman, R.A. (1984). Strategic management: a stakeholder approach. Englewood Cliffs, NJ: Prentice Hall. Kapstein, E. B. (2001). The corporate ethics crusade. Foreign Affairs, 80, 105-19. Leclair, M., Gordon, K. (2000). Corporate support for artistic and cultural activities: what determines the distribution of corporate giving? Journal of Cultural Economics, 24, 225 -241. Perfect, S., Wiles, K. (1994). Alternative construction of Tobin’s Q: an empirical comparison. Journal of Empirical Finance, 1, 313-341. Roberts, R.W. (1992). Determinants of corporate social responsibility disclosure: an application of stakeholder theory. Accounting Organizations and Society , 17, 595-612. Schwartz, R.A. (1968). Corporate philanthropic contributions. Journal of Finance, 23, 479 – 497 Waddock, S.A., Graves, S.B.(1997). The corporate social performance-financial performance link. Strategic Management Journal, 18, 303-19. Yermack, D. (1996). Higher market valuation of companies with a small board of directors. Journal of Financial Economics, 40, 185-211. 304 ASAC 2008 Halifax, Nova Scotia Oumar Sy Dalhousie University Vihang Errunza Desautels Faculty of Management McGill University DOES DIVERSIFICATION EXPLAIN MARKET ANOMALIES?1 Empirical evidence suggests that stockholders do not diversify their portfolios to any significant level, and they are compensated for bearing idiosyncratic risk. In this paper, we breakdown the abnormal returns of size, value, long-term-reversal, momentum, and short-term contrarian premiums into diversification and net selectivity components to find that the anomalies in the U.S. and the international capital markets vanish when the effects of foregone diversification are taken into account. Introduction The Capital Asset Pricing Model (CAPM) of Sharpe (1964), Lintner (1965), and Mossin (1966) has long served as a paradigm for pricing securities. However, the ability of the model to explain the behavior of expected returns has been questioned with the discovery of several so-called market anomalies, i.e., investment strategies that have historically earned high abnormal returns. The most important anomalies reported in the literature include the size effect (Banz (1981)), the value effect (Rosenberg et al. (1985)), the long-term reversal effect (De Bondt and Thaler (1985)), the momentum effect (Jegadeesh and Titman (1993)), and the short-term contrarian effect (Jegadeesh (1990)). Although there have been a number of attempts to explain the presence of market anomalies (see the reviews of Campbell (2000) and Schwert (2003)), no study has explored the potential effects of foregone diversification. Hence, the main objective of this paper is to investigate whether the anomalies reflect the cost of holding undiversified portfolios. This investigation is motivated by two welldocumented stylized facts: the reluctance of stockholders to diversify their portfolios despite the documented benefits of diversification and the pricing of idiosyncratic risk. According to Kelly (1995), the median stockholder in the U.S. owns a single publicly traded stock and less than one-third of the stockholders have more than ten stocks.2 If a large majority of investors hold undiversified portfolios, it can be reasonably presumed that this behavior will ultimately influence asset prices. Indeed, Goetzmann and Kumar (2003) investigate this issue and find that the diversification decisions of individual investors do influence asset prices. They argue, 1 We are grateful to Najah Attig, Omrane Guedhami, John Rumsey, and Sergei Sarkissian for comments. We also thank seminar participants at Dalhousie University and University of South Carolina for discussions and comments. Errunza acknowledges financial support from the Bank of Montreal Chair at McGill University, IFM2, and SSHRC, while Sy acknowledges financial support from FCAR, IFM2, and McGill Finance Research Centre. 2 See similar evidence in Blume and Friend (1975), Barber and Odean (2000), and Polkovnichenko (2005). 305 If individual investors systematically hold less than fully diversified portfolios, they are likely to demand compensation for the idiosyncratic risk in their equity portfolios. If this sensitivity is widespread, their portfolio decisions are likely to generate pervasive forces that can influence returns. Consequently, ceteris paribus, stocks with a less diversified individual investor clientele are likely to earn higher expected returns. (p. 4) If bearing diversifiable risk is widespread and rewarded, 3 then the commonly used measures of performance to generate the anomalies can be questioned because most overlook idiosyncratic risk. In other words, if idiosyncratic risk is priced, it will be natural to observe reliable abnormal returns on less than perfectly diversified portfolios. The abnormal returns will then capture the reward-to-theidiosyncratic-risk inherent in the portfolios (the effects of foregone diversification) in addition to the manager’s potential ability to pick stocks. To investigate this hypothesis, we develop a generalized multifactor framework to decompose the abnormal return of a portfolio into a component due to foregone diversification (Diversification) and a component attributable to skill (Net Selectivity). We then recognize the impact of idiosyncratic risk when the investment portfolio is not well diversified. Diversification measures the reward to the idiosyncratic risk endured once the investor decides to concentrate on few securities. Net Selectivity is the abnormal return purged of the effects of foregone diversification. We investigate 25 key portfolios that represent the size premiums within value quintiles, and the value, long-term reversal, momentum, and short-term contrarian premiums within size quintiles. We find that Diversification is the most significant component of abnormal performance, averaging more than two-third (72.78%) of the abnormal returns across the 25 premiums in the United States. The importance of foregone diversification is even more robust when we consider the 25 premiums computed using international stocks. More importantly, we find that the abnormal returns for all the premiums become insignificant or negative once the effects of foregone diversification are taken into consideration. Overall, our results suggest that the size, value, long-term reversal, momentum, and short-term contrarian effects persist mainly because arbitrage is impeded by the cost of foregone diversification. This research is related to a number of studies that attempt to explain the anomalies. The closest among these is probably Fama and French (1996). In that paper, the authors show that most of the anomalies vanish in their three-factor model (Fama and French (1993)). Our present study contributes to this literature by decomposing the abnormal returns related to the anomalies into components attributable to Diversification and Net Selectivity. For 12 out of the 25 premiums investigated in the U.S., the threefactor model augmented with the momentum factor (Carhart (1997)) generates reliably positive abnormal returns at the 5% level of statistical significance. These significant abnormal returns vanish, however, when they are purged of the effects of foregone diversification. This study is particularly related to an emerging literature on the relation between the anomalies and costly arbitrage. For instance, Ali, Hwang, and Trombley (2003) show that the value effect exists only for high idiosyncratic risk stocks, while Fu (2005) finds a similar result for the size effect.4 Not only the current study expands this literature to investigate a large set of anomalies, but—rather than looking at the effects for different idiosyncratic risk groups—it uses a more direct approach that consists in removing the effect of idiosyncratic risk on the abnormal returns observed on the anomalies. To that end, we derive a multifactor framework that can be seen as a generalization of Fama’s (1972) breakdown of 3 See Lintner (1965), Douglas (1969), Tinic and West (1986), Merton (1987), Lehmann (1990), Malkiel and Xu (2002), Goyal and Santa-Clara (2003), Jiang and Lee (2004), Fu (2005), and Spiegel and Wang (2005) for theoretical and empirical evidence that idiosyncratic risk is priced. 4 See Pontiff (1996) for a comprehensive review of this literature. 306 investment performance. Our research is also unique in the sense that it is the first to appraise the role of idiosyncratic risk on the abnormal returns observed on the anomalies. The paper is also related to Shleifer and Vishny (1997). According to the authors, In reality, arbitrage resources are heavily concentrated in the hands of a few investors that are highly specialized in trading a few assets, and are far from diversified. As a result, these investors care about total risk, and not just systematic risk. Since the equilibrium excess returns are determined by the trading strategies of these investors, looking for systematic risk as the only potential determinant of pricing is inappropriate. Idiosyncratic risk as well deters arbitrageurs, whether it is fundamental or noise trader idiosyncratic risk. (p. 52)5 Under Shleifer-Vishny’s framework, there can be an anomaly without it being eliminated by the arbitrageurs. The presence of idiosyncratic risk could impair the short-term performance of the arbitrage portfolio and so stimulate a premature withdrawal of the funds under management. This risky prospect will ultimately deter arbitrage, and therefore permits the persistence of a seemingly profitable anomaly. Our main result is that the documented abnormal returns on the reported anomalies are not large enough to compensate for the idiosyncratic risk that investors must bear in order to profit from the mispricing in returns. The paper is organized as follows. Section II investigates the idiosyncratic risk of U.S. stocks. Section III presents our generalized multifactor breakdown of investment performance. Section IV presents the anomalous premiums to be explained and investigates the role of foregone diversification for the U.S. and the international capital markets. Section V concludes. The Relative Importance of Idiosyncratic Risk A necessary condition for our hypothesis—that the anomalies reflect foregone diversification—to hold is that the idiosyncratic component of returns be nontrivial. Campbell et al. (2001) investigate the behavior of U.S. stocks to find that there is not only a noticeable increase in firm-level idiosyncratic volatility over 1962-1996, but also a decrease in the correlations between stocks, suggesting an enhancement of the potential benefits of portfolio diversification. Despite this, most stockholders continue to hold concentrated portfolios. In a recent study, Goetzmann and Kumar (2003) re-examine the diversification behavior of U.S. stockholders in the 1990s to find that the median number of stocks in a portfolio is three, a small change from the previous evidence. In this section, we examine the relative importance of the idiosyncratic risk of U.S. stocks over the more recent decade. We proceed as follows. First, we use all the NYSE, AMEX, and NASDAQ stocks that have complete data over the period January 1995 to December 2004 (2,448 stocks) to randomly draw 500 portfolios each of n stocks (n = 1 to 200). For each portfolio, we compute the returns by equally weighting the returns of the n member stocks. Second, we decompose the monthly portfolio returns (into systematic and idiosyncratic components) by running either a CAPM-based regression: r pt = a p + b pM rMt + e pt , (1) or a four-factor regression: 5 In a very interesting paper, Pontiff (1996) discuss why idiosyncratic risk is the single largest cost faced by arbitrageurs, and that it matters even if when arbitrageurs have access to a large number of projects. 307 Table 1 Idiosyncratic Risk as a Function of the Number of Stocks in the Portfolio, 1995:1–2004:12 Total risk Panel A. CAPM Idiosyncratic risk Total risk Panel B. Four-factor model Idiosyncratic risk Ratio Ratio 1 13.66 12.74 93.75 13.66 12.18 89.29 2 10.55 9.41 89.43 10.55 9.01 85.47 3 9.26 7.92 85.68 9.26 7.56 81.77 4 8.41 6.95 82.79 8.41 6.63 78.92 5 7.94 6.35 80.09 7.94 6.05 76.35 6 7.47 5.79 77.49 7.47 5.50 73.75 7 7.14 5.37 75.28 7.14 5.10 71.64 8 6.87 5.01 72.92 6.87 4.76 69.43 9 6.71 4.77 71.19 6.71 4.54 67.80 10 6.53 4.52 69.29 6.53 4.30 66.07 15 6.00 3.75 62.41 6.00 3.56 59.51 20 5.69 3.24 56.90 5.69 3.08 54.17 30 5.38 2.66 49.39 5.38 2.53 47.12 50 5.10 2.05 40.33 5.10 1.96 38.50 100 4.89 1.46 29.92 4.89 1.40 28.58 150 4.80 1.20 24.97 4.80 1.14 23.84 200 4.76 1.04 21.78 4.76 0.99 20.80 Number of stocks in portfolio The table illustrates the behavior of idiosyncratic risk as a function of the number of stocks in the portfolio. The simulation proceeds as follows. We use all the NYSE, AMEX, and NASDAQ stocks that have complete data between January 1995 and December 2004 (2,448 stocks) to randomly draw 500 portfolios of n stocks (n = 1 to 200). For each equally-weighted portfolio, we compute the total risk as the standard deviation of the returns, the idiosyncratic risk as the standard deviation of the residuals from equation (1) (for the CAPM) or equation (2) (for the four-factor model), and the ratio of the idiosyncratic risk to the total risk (ratio). The table shows the mean of the three variables across the 500 portfolios. Panel A shows the results obtained when the idiosyncratic risk is computed from a CAPM regression, while panel B shows the four-factor model results. All returns are in percent per month. The data are from the Center for Research in Security Prices (CRSP). The data for the factors are from Professor Ken French’s data library. 308 r pt = a p + b pM r Mt + b pS SMBt + b pH HMLt + b pW WMLt + e pt , (2) where r p is the return on the portfolio in excess of the one-month Treasury bill rate, rM is the excess return on the value-weighted market portfolio of NYSE, AMEX, and NASDAQ stocks, SMB is the size factor (the difference between the returns to portfolios of small and large stocks of similar book-to-market (B/M) ratio), HML is the value factor (the difference between the returns to portfolios of high and low B/M stocks of similar size), WML is the momentum factor (the difference between the returns to portfolios of prior t - 12 to t - 2 months composed of winning and losing stocks of similar size), and e p is the error term. The intercept a p measures the portfolio abnormal return while the b s measure the systematic exposures to market, size, value, and momentum risks. Finally, we estimate the idiosyncratic risk by taking the standard deviation of the error term. Table 1 illustrates the behavior of idiosyncratic risk as a function of the number of stocks in the portfolio. We report the means (across the 500 portfolios) of (i) the standard deviations of the portfolio excess returns (total risk), (ii) the standard deviations of the error terms (idiosyncratic risk), and (iii) the ratios of idiosyncratic risk to total risk. The standard deviation of a typical stock is about 13.66% per month, which is more than twice as high as the standard deviation of the U.S. market excess returns during the same period (4.65% per month). As expected, idiosyncratic risk represents the most important source of risk for U.S. stocks, averaging about 90% of the standard deviation of a typical stock. The importance of idiosyncratic risk does not seem to rest on the asset-pricing model used because the results change little when the four-factor model is used instead of the CAPM. The conventional wisdom in the finance literature is that diversification can be quickly achieved by keeping roughly 20 stocks in the portfolio (Bloomfield et al. (1977)). Such a level of diversification would have left more than half of the portfolio standard deviation unexplained by systematic risk over the period 1995-2004. The most striking result from the table is probably the fact that idiosyncratic risk does not seem to disappear even when the simulated portfolio is designed to have a large number of stocks. Consistent with Statman’s (2004) finding that more than 300 stocks are needed to optimize the level of diversification of a typical portfolio, the ratio of portfolio idiosyncratic risk to portfolio standard deviation is on average 21% when there are 200 stocks in the portfolio.6 The significance of idiosyncratic risk is important for understanding the potential role of foregone diversification in the reported anomalies, because most of the portfolios used to generate the anomalies do not include a large number of stocks. For instance, only five of the 25 Fama-French size-B/M portfolios have more than 200 stocks in December 2004, and none of the portfolios have 200 stocks in July 1963. Generalized Multifactor Decomposition of Investment Performance The use of a benchmark is central to portfolio performance evaluation as it enables the excess return achieved by an investment portfolio over a period to be broken down into two components: the required return given the level of risk and an abnormal return. Usually, the portfolio excess return is compared to a benchmark that has the same level of systematic risk, R ( b p ) , resulting in the measure of abnormal return (usually referred to as Selectivity = E[ r pt ] - R ( b p ) ). The main problem with Selectivity is that it assumes 6 In a recent paper, Bennett and Sias (2007) estimate that an arbitrageur would need 31,294 stocks to form an “approximately well-diversified” portfolio over 2000-2004! 309 that the investment portfolio is well diversified, and hence overlooks the cost of foregone diversification. We overcome this drawback by purging the effects of foregone diversification from Selectivity: Net Selectivity = Selectivity – Diversification. (3) Diversification = R ( s p ) - R ( b p ) , where R ( s p ) is the diversified benchmark with the same level of total risk as portfolio p , appraises the cost of foregone diversification. Assume the following multifactor asset-pricing model: 7 E[ r pt ] = å K k= 1 b pk E[ rkt ] , (4) where E[.] denotes the expectations operator, rk is the return on factor k ( k = 1 to K ), and b pk is the beta associated with factor k . From equation (4), it is clear that the benchmark that has the same level of K systematic risk as portfolio p (with regard to all the risk factors) is R ( b p ) = å k = 1 b pk E[ rkt ] . The benchmark with the same total risk as portfolio p can be seen as a portfolio of the factors R( s p ) = å K k= 1 v pk E[ rkt ] for which the weights (the v s) are chosen such that Var [ R( s p )] = Var [ r pt ] = s 2p . Since the benchmark variance is Var [ R( s p )] = å K k= 1 å K l=1 v pk v pl s kl , where s kl is the covariance between factors k and l , it follows that there are an infinite number of ways to choose the v s to match the portfolio total risk. A practical solution is obtained by assuming an equal weight for each factor ( v pk = v p , " k ). Solving for v p and substituting yields the explicit expression of the multifactor benchmark that matches the total risk of portfolio p , æ çç ç R ( s p ) = s p ççç çç ççè K å E[r ] å å s k= 1 kt K K k= 1 l=1 kl ö ÷ ÷ ÷ ÷ ÷ . ÷ ÷ ÷ ÷ ÷ ÷ ø (5) Table 2 shows the explicit components of investment performance for the multifactor-asset pricing model and two special cases: the CAPM and the four-factor model. For the sake of simplicity, we discuss the decomposition of performance for the CAPM. For this model, the usual benchmark against which the portfolio return is compared is R ( b p ) = b pM E[ rMt ] , which is the return on the combination of the market and the riskless portfolios that matches the systematic risk of portfolio p (i.e., the benchmark in the Security Market Line (SML)). This yields Jensen’s (1968) measure of selectivity, a p = E[ r pt ] - b pM E[ rMt ] . However, if the portfolio is not well diversified, the Selectivity will be problematic because R ( b p ) understates the (idiosyncratic) risk faced. The solution to this problem is to use the benchmark that has the same total risk as the portfolio (i.e., the benchmark in the Capital Market Line (CML)), R( s p ) = s p ( E[ r Mt ] / s M ) , resulting in the Net Selectivity = E[ r pt ] - s p ( E[ rMt ]/ s M ) . 7 The multifactor asset-pricing model can be seen as either the ICAPM of Merton (1973) or the APT of Ross (1976). 310 Table 2 Multifactor Breakdown of Investment Performance Components Multifactor pricing model R( b p ) = Benchmark (systematic risk) æ çç ç R ( s p ) = s p ççç çç çç è Benchmark (total risk) æ çç ç E[ r pt ] - s p ççç çç çç è Net Selectivity å æ ç K ç çç çç k = 1ç çç çè k= 1 å å å k= 1 E[ rkt ] K k= 1 K å k= 1 å å å k= 1 E[ rkt ] K k= 1 sp å å k= 1 K l=1 l=1 s kl ö÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ø s kl K l=1 R( b p ) = s kl ö ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ø ö ÷ ÷ ÷ ÷ - b pk ÷ E[ rkt ] ÷ ÷ ÷ ÷ ÷ ÷ ø æ çç ç R ( s p ) = s p ççç çç çç è æE[ r ] ÷ ö ÷ R( s p ) = s p çç Mt ÷ çè s M ÷ ø æ çç ç E[ r pt ] - s p ççç çç çç è M M 4 k= 1 b pk E[ rkt ], 4 å å å k= 1 a p = E[ r pt ] - æE[ r ]ö ÷ E[ r pt ] - s p çç Mt ÷ ÷ ÷ çè s ø æs p ö çç - b ÷ ÷ pM ÷E[ rkt ] ççès ÷ ø å k = M , SMB , HML , WML a p = E[ r pt ] - b pM E[ rMt ] b pk E[ rkt ] K K K Four-Factor model R ( b p ) = b pM E[ rMt ] b pk E[ rkt ] K a p = E[ r pt ] - Selectivity Diversification å K CAPM å æ ç 4 ç çç çç k = 1ç çç çè E[ rkt ] 4 4 k= 1 l=1 4 å k= 1 b pk E[ rkt ] 4 å å å k= 1 E[ rkt ] 4 4 k= 1 l=1 sp K K k= 1 l=1 å å s kl ö÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ø s kl s kl ö ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ø ö ÷ ÷ ÷ ÷ E[ rkt ] - b pk ÷ ÷ ÷ ÷ ÷ ÷ ÷ ø The table summarizes the components of investment performance for the multifactor pricing model and two special cases: the CAPM and the four-factor model. The construction of the components is described in detail in Section III. 311 Table 3 The Premiums to be Explained, 1963:7–2004:12 Panel A. Size and B/M portfolios Low 2 3 4 Panel B. Size and long-term reversal portfolios High H-L t(H-L) Loser 2 3 4 Winner L-W t(L-W) Small 0.73 1.31 1.36 1.57 1.67 0.60 4.17 1.60 1.43 1.49 1.30 0.91 0.41 4.40 2 3 0.89 0.90 1.15 1.22 1.40 1.22 1.46 1.35 1.55 1.52 0.49 0.37 3.64 2.70 1.62 1.53 1.35 1.37 1.44 1.28 1.41 1.27 1.09 1.13 0.24 0.25 2.45 2.35 4 Big 1.01 0.90 0.99 0.97 1.22 0.97 1.34 1.06 1.37 1.06 0.36 2.77 1.23 1.03 1.20 0.95 1.01 0.83 1.80 1.00 1.19 1.12 0.20 0.13 1.42 1.29 0.32 2.29 -0.14 -0.76 0.25 1.43 0.28 1.88 0.31 2.21 0.39 2.61 0.25 1.44 0.23 1.71 0.34 2.64 0.28 2.10 0.08 0.58 Loser 2 3 4 Winner W-L Small 2 0.35 0.41 1.17 1.04 1.44 1.27 1.57 1.51 2.01 1.81 3 4 0.59 0.59 0.98 0.99 1.14 1.03 1.25 1.23 Big 0.63 0.89 0.77 -0.23 -1.26 0.16 1.11 0.45 3.51 S-B t(S-B) Panel C. Size and momentum portfolios S-B t(S-B) Panel D. Size and short-term contrarian portfolios t(W-L) Loser 2 3 4 Winner L-W t(L-W) 1.03 0.94 7.45 6.60 1.76 1.63 1.32 1.55 1.25 1.27 1.12 1.09 0.42 0.77 0.77 0.66 6.51 5.38 1.75 1.61 0.71 0.63 4.43 3.66 1.64 1.54 1.43 1.41 1.25 1.17 1.01 0.94 0.71 0.68 0.68 0.67 5.59 4.97 0.98 1.29 0.37 1.96 1.11 1.06 0.95 0.88 0.72 0.29 2.00 0.44 3.20 0.47 3.11 0.37 2.18 0.20 1.43 0.20 1.44 0.20 1.41 -0.10 -0.69 At the end of each June, portfolios are formed as the intersections of independent sorts of NYSE, AMEX, and NASDAQ stocks into five size groups, five book-to-market (B/M) groups, five prior t - 60 to t - 13 months return groups, five prior t - 12 to t - 2 months return groups, and five prior-month-return groups. All quintile breakpoints are computed using solely NYSE stocks. For each B/M group, a size premium (S-B) is formed as the difference between the average returns on the two smallest and two biggest size portfolios of the B/M quintile. Twenty extra premiums are formed from the intersections with the size quintile groups. For each size group, the value premium (H-L) is the difference between the average returns on the two highest and the two lowest B/M portfolios of the size quintile, the long-term reversal premium (L-W) is the difference between the average returns on the two lowest and the two highest prior (13-60 month) return portfolios of the size quintile, the momentum premium (W-L) is the difference between the average returns on the two highest and the two lowest prior (2-12 month) return portfolios of the size quintile, and the short-term contrarian premium (L-W) is the difference between the average returns on the two lowest and the two highest prior-month-return portfolios of the size quintile. The table reports the average return of the portfolios and the premiums with the t-statistics that the premium is zero. The data for the portfolios are from Professor Ken French’s data library. The sample period is July 1963 to December 2004 (498 observations). All returns are in percent per month. In boldface are 25 premiums to be explained. 312 The Net Selectivity is proportional to the difference of Sharpe ratios between the portfolio and the market, Net Selectivity = s p {( E[ r pt ]/ s p ) - ( E[ rMt ]/ s M )}, and so measures whether the portfolio beats the market’s Sharpe ratio.8 Furthermore, the Net Selectivity will give the same ranking as Modigliani and Modigliani’s (1997) RAP (Risk-Adjusted Performance), Net Selectivity = ( s p / s M )( RAPp - RAPM ) , and Statman’s (2000) eSDAR (Excess Standard-Deviation-Adjusted Return), Net Selectivity = ( s p / s M )eSDAR p . Diversification = {( s p / s M ) - b pM }E[ rMt ] , the difference between the CML and SML benchmarks, measures the cost of foregone diversification. Since the market beta can always be written as b pM = r pM s p / s M , where r pM is the correlation between the portfolio and the market, it follows that Diversification = (1 - r pM )s p ( E[ rMt ]/ s M ) . The latter expression shows that Diversification is trivial if the portfolio is well diversified.9 It also shows—as any correlation coefficient is lower than one—that Diversification will be positive as long as there is positive risk-return tradeoff. Therefore, as a measure of abnormal return, Selectivity will tend to overstate the manager’s performance when the portfolio is not well diversified. Diversification and Market Anomalies A. The Premiums to be Explained This study investigates the size, value, long-term reversal, momentum, and short-term contrarian premiums obtained from the intersections of independent sorts of stocks into five size (market capitalization) groups, five B/M groups, five prior t - 60 to t - 13 months return groups, five prior t - 12 to t - 2 months return groups, and five prior-month-return groups. We first focus on the NYSE, AMEX, and NASDAQ stocks and then examine the robustness of our results with international stocks. The premiums are composed from the returns differentials between extreme quintile portfolios within a particular group, using monthly data from July 1963 to December 2004 (498 observations).10 The data are from Professor Ken French’s data library.11 Table 3 shows the average returns of the various characteristic-sorted portfolios and the resulting premiums. Panel A shows the results for the size- and B/M-sorted portfolios. For each of the B/M quintiles, there is a size premium, which is the difference of average returns on the two smallest and the two biggest size portfolios. The average size premium is significant at the 5% level in the two quintiles of highest B/M ratio. Similarly, there is a value premium for each size quintile, which is the difference of average returns on the two highest and the two lowest B/M portfolios. Consistent with the reported value effect, four of the five value premiums are significant on average at the 5% level. 8 It can be shown that the CAPM implies that no investment strategy beat the market’s Sharpe ratio. Indeed, as any correlation coefficient is lower than one, dividing both sides of the CAPM by s p and solving yields E[ r pt ]/ s p = r pM ( E[ rMt ]/ s M ) £ E[ rMt ]/ s M . 9 Indeed, since from the CAPM-based regression (equation (1)), the idiosyncratic variance can be written as 2 2 s e2 = s p2 - b pM s M2 = s p2 (1 - r pM ) , it follows that for a diversified risky portfolio s e2 = 0 and b pM > 0 , so that r pM = 1 . This shows that Diversification = (1 - r pM )s p ( E[ r Mt ] / s M ) = 0 and Selectivity = Net Selectivity for any well-diversified risky portfolio. 10 The beginning of the sample period is set to coincide with the one used by Fama and French (1993). Another argument for using the post-1963 period is that the average number of securities in portfolio is significantly higher than in the pre-1963 period. We can report, however, that using the pre-1963 data has no effect on evidence presented in this paper. 11 The data are available at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. 313 Panel B of Table 3 shows that long-term reversal effect is not trivial; the difference of average returns on the two lowest and two highest prior t - 60 to t - 13 months return portfolios is significant at the 5% level in all but one size quintile. The momentum premium, the difference of average returns on the two highest and two lowest prior t - 12 to t - 2 months return portfolios of the size quintile, is significant for all size quintiles (see Panel C of Table 3). Panel D of Table 3 shows that the short-term contrarian premium, the difference of average returns on the two lowest and two highest prior-month-return portfolios, is also significant for all size groups. Since the 25 premiums summarize those anomalies which have not been well explained, we focus on them for the remainder of the study. If foregone diversification explains the persistence of market anomalies, then Diversification should not only be significant, but also subsume most of the Selectivity associated with the size, value, long-term reversal, momentum, and short-term contrarian premiums, i.e., net selectivity should be insignificant. The following subsections investigate this plausible explanation for the anomalies. B. The Size Effect The size effect is one of the most notable patterns in average returns. Banz (1981) and Reinganum (1981) examine the relationship between stock returns and market capitalization to find that smaller stocks have tendency to yield higher risk adjusted returns than larger stocks. Panel A of Table 4 shows the results of the decomposition of the abnormal returns of the size premiums within B/M quintiles. The size Selectivity premium obtained from the CAPM seems to increase monotonically across the B/M quintiles, to the extent that it is reliably positive solely for the quintile of extreme value stocks (33 basis points per month; t = 2.25). This size premium of the extreme value stocks is also significant in the four-factor regression (21 basis points per month; t = 2.65). The existence of this significant abnormal size premium does not guarantee a free lunch, since the costs of foregone diversification may hinder profitable arbitrage. If the portfolios used to obtain this significant size Selectivity were well diversified, the market correlation with the size premium should be close to one and Diversification close to zero. Such is not the case, however, because the market correlation is very small (about 16.59%) and the Diversification component of the abnormal size premium in the extreme value stocks is significant (about 29 basis points per month—just 4 basis points shy of the abnormal return estimated in the CAPM regression).12 Since none of the size premiums delivers a reliably positive Net Selectivity, we conclude that size investing does not exhibit significant incremental profitability beyond the cost of foregone diversification. C. The Value Effect The value effect—the tendency of value stocks to outperform growth stocks—is considered one of the most significant challenges to the CAPM (see Fama and French (2006) and the references therein). Panel B of Table 4 shows the results of the decomposition of the abnormal value premiums within size quintiles. Throughout 1963-2004, value investing triggers a significantly positive CAPM Selectivity in all but the largest stocks quintile.13 The four-factor model explains the value premiums in the three largest size quintiles but fails in the remaining two smaller-stocks groups. However, since the portfolios used to generate the abnormal value premiums are not well diversified, the Diversification component of investment performance is always higher than 37 basis points per month and represents the most important component of the abnormal value premiums across the five 12 Because under the CAPM, the market is the only factor used in the decomposition of investment performance, the t-statistic of Diversification will be constant and equal to that of the market excess return. 13 This is consistent with Loughran (1997, p.249), who states, “in the largest size quintile of all firms (accounting for 73% of the total market value of all publicly traded firms), book-to-market has no significant explanatory power on the cross-section of realized returns during the 1963-1995 period.” See also Fama and French (2006) for a discussion of the robustness of Loughran’s findings. 314 size quintiles. Consistent with the hypothesis that the value premium mainly reflects the effects of foregone diversification, none of the value premiums delivers a significantly positive Net Selectivity at the 5% level for both the CAPM and the four-factor model, suggesting that the marginal benefit of arbitraging it does not significantly exceed the cost of foregone diversification. Hence, we conclude that the value effect is not an anomaly. Table 4 Breakdown of Investment Performance in the United States, 1963:7–2004:12 Quintile CAPM Selectivity Four-factor model Diversification Net Selectivity Selectivity Diversification Net Selectivity Panel A. Size premium within B/M quintiles Low -0.29 (-1.66) 0.29 (2.35) -0.58 (-2.71) -0.45 (-5.50) 0.97 (6.00) -1.42 (-7.84) 2 0.16 (0.94) 0.32 (2.35) -0.16 (-0.74) 0.03 (0.42) 0.97 (5.95) -0.94 (-5.33) 3 0.22 (1.47) 0.29 (2.35) -0.07 (-0.37) 0.08 (1.33) 0.83 (5.88) -0.74 (-4.84) 4 0.25 (1.80) 0.27 (2.35) -0.02 (-0.11) 0.13 (2.10) 0.78 (6.02) -0.65 (-4.54) High 0.33 (2.25) 0.29 (2.35) 0.04 (0.20) 0.21 (2.65) 0.84 (6.16) -0.63 (-4.00) Panel B. Value premium within size quintiles Small 0.76 (5.99) 0.50 (2.35) 0.26 (1.06) 0.39 (5.14) 0.77 (3.60) -0.38 (-1.66) 2 0.63 (5.23) 0.45 (2.35) 0.17 (0.76) 0.15 (2.43) 0.57 (3.00) -0.42 (-2.10) 3 0.51 (4.01) 0.46 (2.35) 0.05 (0.21) 0.05 (0.85) 0.62 (3.08) -0.57 (-2.70) 4 0.46 (3.74) 0.40 (2.35) 0.05 (0.26) 0.02 (0.28) 0.54 (3.07) -0.52 (-2.76) Big 0.21 (1.71) 0.37 (2.35) -0.17 (-0.84) -0.23 (-3.40) 0.49 (3.05) -0.72 (-4.14) The table decomposes the abnormal return (Selectivity) of the size, value, long-term reversal, momentum, and short-term contrarian premiums into Diversification and Net Selectivity components (see Table 2). Selectivity is the intercept of the regression of the premium on either the market factor (CAPM) or the market, SMB, HML, and WML factors (four-factor model). Diversification is the difference of average returns between a diversified benchmark with the same total risk as the portfolio and a diversified benchmark with the same systematic risk as the portfolio. Net Selectivity is the difference between Selectivity and Diversification. The t-statistic is the average component divided by its standard error; t-statistics reported in boldface are significantly positive at the 5% level using a two-sided test. The premiums used are described in Table 3. The data for the premiums and factors are from Professor Ken French’s data library. The sample period is July 1963 to December 2004 (498 observations). All returns are in percent per month. (continued on next page) 315 Table 4 (continued) Breakdown of Investment Performance in the United States, 1963:7–2004:12 Quintile CAPM Selectivity Diversification Four-factor model Net Selectivity Selectivity Diversification Net Selectivity Panel C. Long-term reversal premium within size quintiles Small 0.40 (4.35) 0.21 (2.35) 0.19 (1.46) 0.19 (2.27) 0.41 (5.87) -0.22 (-2.05) 2 0.22 (2.32) 0.21 (2.35) 0.01 (0.08) 0.00 (0.04) 0.43 (5.74) -0.42 (-3.65) 3 0.27 (2.52) 0.27 (2.35) 0.00 (0.00) 0.06 (0.57) 0.53 (5.35) -0.47 (-3.38) 4 0.24 (2.19) 0.31 (2.35) -0.06 (-0.36) 0.05 (0.48) 0.61 (5.08) -0.56 (-3.59) Big 0.34 (2.48) 0.35 (2.35) -0.01 (-0.04) 0.09 (0.68) 0.71 (5.28) -0.63 (-3.42) Panel D. Momentum premium within size quintiles Small 1.06 (7.67) 0.35 (2.35) 0.71 (3.47) 0.52 (7.27) 0.43 (3.21) 0.09 (0.60) 2 0.97 (6.83) 0.36 (2.35) 0.60 (2.88) 0.34 (5.45) 0.37 (2.79) -0.03 (-0.20) 3 0.75 (4.69) 0.42 (2.35) 0.34 (1.40) 0.03 (0.37) 0.41 (2.71) -0.38 (-2.31) 4 0.67 (3.93) 0.45 (2.35) 0.23 (0.88) -0.08 (-1.15) 0.46 (2.83) -0.54 (-3.05) Big 0.40 (2.12) 0.47 (2.35) -0.07 (-0.26) -0.44 (-5.31) 0.47 (2.75) -0.91 (-4.77) Panel E. Short-term contrarian premium within size quintiles Small 0.69 (6.09) 0.20 (2.35) 0.49 (3.48) 0.85 (8.10) 0.88 (7.08) -0.03 (-0.21) 2 0.58 (4.92) 0.21 (2.35) 0.37 (2.52) 0.71 (6.26) 0.88 (7.06) -0.17 (-1.03) 3 0.59 (5.11) 0.20 (2.35) 0.39 (2.74) 0.77 (7.12) 0.92 (7.00) -0.15 (-0.86) 4 0.58 (4.51) 0.23 (2.35) 0.35 (2.16) 0.75 (6.02) 0.99 (7.07) -0.25 (-1.31) Big 0.23 (1.63) 0.28 (2.35) -0.05 (-0.27) 0.38 (2.78) 1.07 (7.17) -0.69 (-3.39) 316 D. The Long-Term Reversal Effect In their initial paper, De Bondt and Thaler (1985, p. 804) conclude that “thirty-six months after portfolio formation, the losing stocks have earned about 25% more than the winners, even though the latter are significantly more risky.” The significance of the long-term reversal effect is also evident from the abnormal returns obtained in Panel C of Table 4. Indeed, the CAPM Selectivity premiums for the long-term reversal effect range between 22 and 40 basis points per month within the size quintiles, and are always statistically significant at the 5% level. The four-factor model does a good job explaining most of the abnormal reversal premiums but fails in one special case: for the long-term reversal effect of the smallest stocks (t = 2.27). If the losing and winning portfolios are not well diversified, then this abnormal long-term reversal premium in small stocks could simply reflect the cost of bearing diversifiable risk. The breakdown of performance confirms this explanation for the long-term reversal effect for the small stocks because the significant abnormal return vanishes when the effects of both systematic and idiosyncratic risk are taken into account (Net Selectivity is always insignificant at the 5% level for both the CAPM and the four-factor model.) E. The Momentum Effect The momentum effect is described as one of the most remarkable anomalies. In their recent review of the theory and evidence on asset pricing, Fama and French (2004, p. 40) state, The three-factor model is hardly a panacea. Its most serious problem is the momentum effect of Jegadeesh and Titman (1993). Stocks that do well relative to the market over the last three to twelve months tend to continue to do well for the next few months, and stocks that do poorly continue to do poorly. This momentum effect is distinct from the value effect captured by book-to-market equity and other price ratios. Moreover, the momentum effect is left unexplained by the three-factor model, as well as by the CAPM. The inability of the asset-pricing models to explain momentum makes it even more interesting to investigate the potential role of foregone diversification (see Jegadeesh and Titman (2001) for recent evidence on this effect). Panel D of Table 4 reports decomposition of the abnormal momentum premiums within size quintiles. Momentum investing always yields a significant Selectivity in the CAPM regression. The momentum Selectivity premium is largest for the quintile of the smallest stocks, with about 106 basis points per month (t = 7.67). The four-factor model does eliminate some of the CAPM Selectivity premiums but fails to explain the abnormal momentum premiums in the two smallest size quintiles. However, these significant abnormal returns cannot be taken as anomalies, since the portfolios are not well diversified; the Diversification premium is always higher than 35 basis points per month across the five momentum portfolios. When the cost of foregone diversification is included with the use of Net Selectivity, only the two smallest size quintiles deliver positive abnormal returns for the CAPM. However, none of the momentum premiums delivers a significantly positive abnormal performance for the four-factor model. Therefore, we conclude that no arbitrageur will find it worthwhile to bet on the difference of returns between winning and losing stocks over the medium term, if the cost of holding idiosyncratic risk is considered. F. The Short-Term Contrarian Effect Jegadeesh (1990, p. 882) shows a short-term contrarian premium in average returns: the difference between the risk-adjusted excess returns on the extreme decile portfolios formed on the basis of one-month lagged returns is 1.99% per month for 1934-1987. Albeit less spectacular, the short-term contrarian premiums obtained from Fama-French’s quintile portfolios over 1963-2004 range between 29 and 77 basis points per month and are highly statistically significant (see Table 3). Interestingly, Panel E of Table 4 shows that the short-term contrarian effect is not entirely explained by the asset-pricing models, as the Selectivity of the contrarian premium remains statistically significant for all but the largest size quintile. At first glance, this result is puzzling. If the short-term contrarian effect is profitable after the systematic risk associated with this strategy is compensated, then why don’t market arbitrageurs exploit this apparent free lunch and quickly eliminate the mispricing? 317 A potential answers to this question is that the short-term contrarian premiums are not well diversified, so that their abnormal returns contain diversification premiums. In the four-factor model, the Diversification premium is always higher than 88 basis points per month, suggesting that the short-term contrarian premiums should be subjected to a much higher standard if their level of diversification is taken into account. As a result, none of the short-term contrarian premiums achieves a positive abnormal return if it is compared to a naively selected portfolio of the four factors with the same level of total risk. Conclusion The empirical literature has documented numerous market anomalies that are unexplained by the existing theoretical models. However, the key question is whether investors can exploit them in practice. In this paper, we investigate one of the numerous costs that investors have to cover in order to make the trading on the anomalies worthwhile: the cost of foregone diversification. In particular, we derive a multifactor breakdown of investment performance and examine the role of foregone diversification in the reported market anomalies. Because the portfolios used to generate the anomalies are less than perfectly diversified, idiosyncratic risk does matter. We decompose the abnormal returns of the size, value, long-term reversal, momentum, and short-term contrarian premiums in the U.S. and five international capital markets (Canada, France, Germany, Japan, and the U.K.) to find not only that diversification is the most significant component of abnormal performance, but also that the abnormal returns vanish when the cost of foregone diversification is taken into account. The evidence supports the premise that market anomalies largely reflect the costs to arbitrage mispriced securities. References Ali, A., L. S. Hwang, and M. A. Trombley, 2003, Arbitrage risk and the book-to-market anomaly, Journal of Financial Economics 69, 355-373. Banz, R. W., 1981, The relationship between return and market value of common stocks, Journal of Financial Economics 9, 3-18. Barber, B., and T. Odean, 2000, Trading is hazardous to your wealth: The common stock investment performance of individual investors, Journal of Finance 55, 773-806. Benett, J. A., and R. W. Sias, 2007, Why do Anomalies Survive? The Idiosyncratic Risk of Arbitrage Portfolios, Working paper, Washington State University. Bloomfield, T., R. Leftwich, and J. B. Long, 1977, Portfolio strategies and performance, Journal of Financial Economics 5, 201-218. Blume, M. E., and I. Friend, 1975, The asset structure of individual portfolios and some implications for utility functions, Journal of Finance 30, 585-603. Campbell, J. Y., 2000, Asset pricing at the millennium, Journal of Finance 55, 1515-1567. Campbell, J. Y., M. Lettau, B. G. Malkiel, and Y. Xu, 2001, Have individual stocks become more volatile? An empirical exploration of idiosyncratic risk, Journal of Finance 56, 1-43. Carhart, M. M., 1997, On persistence in mutual fund performance, Journal of Finance 52, 57-82. De Bondt, W. F. M., and R. H. Thaler, 1985, Does the stock market overreact?, Journal of Finance 40, 793-805. Douglas, G. W., 1969, Risk in the equity markets: An empirical appraisal of market efficiency, Yale Economic Essays 9, 3-45. Fama, E. F., 1972, Components of investment performance, Journal of Finance 27, 551-567. Fama, E. F., and K. R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 3-56. Fama, E. F., and K. R. French, 1996, Multifactor explanations of asset pricing anomalies, Journal of Finance 51, 55-84. Fama, E. F., and K. R. French, 2004, The capital asset pricing model: Theory and evidence, Journal of Economic Perspectives 18, 25-46. Fama, E. F., and K. R. French, 2006, The value premium and the CAPM, Journal of Finance 61, 21632185. Fu, F., 2005, Idiosyncratic risk and the cross-section of expected stock returns, Working paper, William E. Simon Graduate School of Business Administration, University of Rochester. 318 Goetzmann, W., and A. Kumar, 2003, Diversification decisions of individual investors and asset prices, ICF working paper, Yale School of Management. Goyal, A., and P. Santa-Clara, 2003, Idiosyncratic risk matters!, Journal of Finance 58, 975-1007. Jegadeesh, N., 1990, Evidence of predictable behavior of security returns, Journal of Finance 45, 881898. Jegadeesh, N., and S. Titman, 1993, Returns to buying winners and selling losers: Implications for stock market efficiency, Journal of Finance 48, 65-91. Jegadeesh, N., and S. Titman, 2001, Profitability of momentum strategies: An evaluation of alternative explanations, Journal of Finance 56, 699-720. Jensen, M. C., 1968, The Performance of mutual funds in the period 1945-1964, Journal of Finance 23, 389-416. Jiang, X., and B. Lee 2004, On the dynamic relation between returns and idiosyncratic volatility, Working paper, University of Northern Iowa. Kelly, M., 1995, All their eggs in one basket: Portfolio diversification of US households, Journal of Economic Behavior and Organization 27, 87-96. Lehmann, B., 1990, Residual risk revisited, Journal of Econometrics 45, 71-97. Lintner, J., 1965, The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets, Review of Economics and Statistics 47, 13-37. Loughran, T., 1997, Book-to-market across firm size, exchange, and seasonality: Is there an effect?, Journal of Financial and Quantitative Analysis 32, 249-268. Malkiel, B. G. and Y. Xu, 2002, Idiosyncratic risk and security returns, Working paper, University of Texas at Dallas. Merton, R. C., 1973, An intertemporal asset pricing model, Econometrica 41, 867-888. Merton, R., 1987, A simple model of capital market equilibrium with incomplete information, Journal of Finance 42, 483-510. Modigliani, F., and L. Modigliani, 1997, Risk-adjusted performance, Journal of Portfolio Management 23, 45-54. Mossin, J., 1966, Equilibrium in a Capital Asset Market, Econometrica 34, 768-783. Polkovnichenko, V., 2005, Household portfolio diversification: A case for rank-dependent preferences, Review of Financial Studies 18, 1467-1501. Pontiff, J., 2006, Costly arbitrage and the myth of idiosyncratic risk, Journal of Accounting and Economics 42, 35-52. Reinganum, M. R., 1981, 'Misspecification of capital asset pricing: Empirical anomalies based on earnings yields and market values, Journal of Financial Economics 9, 19-46. Rosenberg, B., K. Reid, and R. Lanstein, 1985, Persuasive evidence of market inefficiency, Journal of Portfolio Management 11, 9-16. Ross, S. A., 1976, The arbitrage theory of capital asset pricing, Journal of Economic Theory 13, 341360. Schwert, G. W., 2003, Anomalies and market efficiency, in G.M. Constantinides, M. Harris, and R. Stulz (eds.), Handbook of the economics of finance, Elsevier Science. Sharpe, W. F., 1964, Capital asset prices: A theory of market equilibrium under conditions of risk, Journal of Finance 19, 425-442. Shleifer, A., and R. Vishny, 1997, The limits of arbitrage, Journal of Finance 52, 35-55. Solnik, B., 1974, An equilibrium model of the international capital market, Journal of Economic Theory 8, 500-524. Spiegel, M. I., and X. Wang, 2005, Cross-sectional variation in stock returns: Liquidity and idiosyncratic risk, ICF working paper, Yale School of Management. Statman, M., 2000, Socially responsible mutual funds, Financial Analysts Journal 56, 30-39. Statman, M., 2004, The diversification puzzle, Financial Analysts Journal 60, 44-53. Tinic, S, and R. West, 1986, Risk, return and equilibrium: A revisit, Journal of Political Economy 94, 126-147. 319