euro-mediterranean economics and finance review
Transcription
euro-mediterranean economics and finance review
Euro-Mediterranean Economics and Finance Review ISSN 1967-502X Editors Mondher Bellalah and Jean-Luc Prigent Aims and Scope The Euro-Mediterranean Economics and Finance Review is a peer-reviewed research journal of the Mediterranean Association of Finance Insurance and Management (AMFAM). It is intended to develop research in economics, finance and management with a special emphasis on the main issues and problems regarding the Euro-Mediterranean zone. The journal is committed to excellence by publishing high quality research papers in economics and finance with theoretical and empirical contents as well as invited viewpoints (2000 to 4000 words) written by well-known experts. The journal’s editorial policy is to publish original articles that obey the accepted standards and to improve communications between academies practitioners and policymakers at both national and international levels. While recognizing the Euro-Mediterranean origins of the research papers, the journal is also open to research that shows diversity in theoretical and methodological underpinning. Editorial Office Eric Simon (Managing Editor) Sandrine Clais (Editorial Assistant) 22, Boulevard du Fort de Vaux 75017 Paris France Email: [email protected] Phone: +33 1 40 53 99 99 | Fax: +33 1 40 53 98 98 For any information or subscription, please contact Sandrine Clais at [email protected]. Access this journal electronically The current and past issues of the journal can be found at www.emefir.org 2 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Editors Mondher Bellalah, University of Cergy-Pontoise, France Jean-Luc Prigent, University of Cergy-Pontoise, France Advisory Board Harry Markowitz, Nobel Prize Laureate, University of California, San Diego, USA Edward Prescott, Nobel Prize Laureate, Arizona State University, USA Associate Editors Michael Adler Columbia University, USA Rudy Aernoudt Brussels Business School, Belgium Aman Agarwal Indian Institute of Finance, India Gordon Alexander UCLA, USA Mohamed Arouri University of Auvergne, France Mohamed Ayadi HEC Montreal, Canada Giovanni Barone-Adesi University of Lugano, Switzerland Hatem Ben Ameur HEC Montreal, Canada Jean-François Boulier CA Asset Management, France Michael Brennan UCLA, USA Eric Briys Cyberlibris, Belgium Harvey R. Campbell Duke University, USA K.C. Chen California State University, USA Ephraim Clark Middlessex University, UK Georges Constantinides University of Chicago, USA Manuel José Da Rocha Armada University of Minho, Portugal Gabriel Desgranges University of Cergy-Pontoise, France Joao Duque, ISEG Portugal Alain Finet ULB, Belgium Philippe Foulquier EDHEC Business School, France Bertrand Jacquillat IEP, France Frank Janseen Catholic University of Louvain, Belgium Cuong Le Van PSE & University of Paris 1, France Michel Levasseur University of Lille 2, France Patrick Navatte University of Rennes 1, France André de Palma University of Cergy-Pontoise, France Bernard Paranque Euromed Management, France Kuntara Pukthuanthong San Diego University François Quittard-Pinon University of Lyon 1, France Richard Roll UCLA, USA Olivier Scaillet HEC, Genève, Switzerland Stefan Straetmans Maastricht University, Netherlands Héracles Vladimirou University of Cyprus, Cyprus Jose Scheinkman Princeton University, USA Paul Willmott Editor Derivatives, UK Table of Contents Volume 7 Number 4 2013 Page Editorial 5 The performance of hybrid models in the assessment of default risk Sami Zouari 7 Does co-integration and causal relationship exist between the non-stationary variables for Chinese bank’s profitability? Empirical evidence Omar Masood, Marc Bradford, Jean-Jacques Levy 25 Dependence structure between the equity market and the foreign exchange market after the Greek crisis: Evidence from France and Germany Adel Boubaker, Jaghoubbi Salma 41 Financial development and income inequality in the MENA region Zied Saadaoui, Maher Gassab 53 Editorial This special volume handles new emerging problems in Finance and comprises some selected papers from the IFC international Finance conference, organized by ISC Paris - the University of Cergy-Pontoise and REMEREG in Paris, in March 2013. The first paper, written by Sami Zouari, is interested in the performance of hybrid models in the assessment of default risk. The paper uses simultaneously fundamental analysis and contingent claim analysis into a hybrid credit risk model of French companies. It provides an answer to the following question: how the combination of continuous market assessments and the values from financial statements improve the estimation of the probability of default. The second paper by Omar Masood, Marc Bradford and Jean-Jacques Levy provides an answer to the co-integration and causal relationship between the non-stationary variables for bank’s profitability. The empirical analysis used the co-integration and other tests. The third paper by Adel Boubaker and Jaghoubbi Salma studies the dependence structure between the equity market and the foreign exchange market after the Greek crisis. The study investigates the dependence structure and tests the degree of dependence between financial returns. It uses two measures: correlations and copula functions. The findings could have some impact for global investment risk management. The last paper, written by Zied Saadaoui and Maher Gassab handles the financial development and income inequality in the MENA region. Using a sample of 11 Middle East and North Africa countries, the research tests the impact of financial development on income inequality. Thanks for your interest in the Review. Editors in charge: Mondher Bellalah and Jean-Luc Prigent The performance of hybrid models in the assessment of default risk Sami Zouari*,THEMA, University of Cergy-Pontoise Abstract In this paper, we have combined fundamental analysis and contingent claim analysis into a hybrid model of credit risk measurement with French companies listed on the Paris Stock Exchange (Euronext Paris). Our goal is to assess how the combination of continuous assessments provided by the market and the values derived from financial statements improve our ability to forecast the probability of default. During a first phase, the probability of defaults are estimated using both methods separately, and subsequently, the probability of default of the structural model are integrated at each point in time in the non-structural model as an additional explanatory variable. The appeal of the hybrid model allows the probability of default to be continuously updated by integrating market information via the probabilities of default extracted from the structural model. Our results indicate that default probabilities extracted from the structural model contribute significantly to explaining default risk when included in a hybrid model with accounting variables. Keywords: Default risk, structural model, non-structural model, hybrid model, probit model, Probability of default. JEL Classification: G21, G33 1. Introduction Credit risk refers to the risk due to unpredicted changes in the credit quality of a counter party or issuer and its quantification is one of the major frontiers in modern finance. The creditworthiness of a potential borrower affects the lending decision and the credit spread, since it is uncertain whether the firm will be able to perform its obligation. Credit risk measurement depends on the likelihood of default of a firm to meet it’s a required or contractual obligation and on what will be lost if default occurs. When one considers the large number of corporations issuing fixed income securities and the relatively small number of actual defaults might regard default as rare event. However, all corporate issuer have some positive probability of default. Models of credit risk measurement have focused on the estimation of the default probability of firms, since it is the main source of uncertainty in the lending decision. We may distinguish two large classes of credit risk models on the basis of the analysis they adopt. The first class, the set of traditional models assume the fundamental analysis, called the non-structural model. The goal of these models that goes back to Beaver (1966) and Altman (1968) is to distinguish which factors are more significant in * Sami Zouari is at THEMA, University of Cergy-Pontoise, 33 bd du port, 95011, Cergy, France, email: [email protected] 8 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW assessing the credit risk of a firm. The second class, called structural models assume the contingency claim analysis. The philosophy of these models goes back to Black-Scholes (1973) and Merton (1974) and assumes corporate liabilities as contingent claims on the assets of the firm1. In this paper, we investigate the hybrid contingent claim approach with French companies listed on the Paris Stock Exchange (Euronext Paris). Our goal is to assess how the combination of continuous assessments provided by the market and the values derived from financial statements improve our ability to forecast the probability of default. The structural model of Merton has the advantage of being flexible, since the probability of default can continually be updated with changes in the value of corporate assets. Its main drawback is that it may over-or underestimate the probability of default, since asset values are unobservable and must be extrapolated from the share prices. On the other hand, the non-structural model of Altman is more accurate because it uses the accounting data of companies, but it is less flexible. Because the frequency of information is generally annual, the probabilities of default cannot be updated during the fiscal year. The quarterly financial statements can be found, but they are not always audited by an external accounting firm. The Bank of England estimated the hybrid model with data from British companies and found some interesting results. During a first phase, the probability of defaults are estimated using both methods separately, and subsequently, the probability of default of the structural model are integrated at each point in time in the non-structural model as an additional explanatory variable. The appeal of the hybrid model allows the probability of default to be continuously updated by integrating market information via the probabilities of default extracted from the structural model. In this paper, we apply the hybrid model to French companies listed on Paris stock exchange (Euronext Paris). This paper is organized as follows. Section 2 reviews the main models in the literature. Section 3 presents the estimated structural model and describes the data used and finally section 4 presents the estimation of the hybrid model and summarizes the main results. 2. Review of key models for risk assessment of default 2.1 Non-structural models Traditional non-structural models adopt fundamental analysis and try to find which factors are important in explaining the credit risk of a company. They assess the significance of these factors, mapping a reduced set of financial ratios, accounting variables and other information into a quantitative score. The latter, can be interpreted as a probability of default and can be used as classification system2. 1 Another widely used category of credit risk models is the reduced form approach where the dynamics of default are given exogenously by an intensity or compensator process. For a review of these models see Jarrow, and Turnbull (1995), Jarrow, Lando and Turnbull (1997), Duffie and Singleton (1999). 2 For a review of traditional models see: Jones (1987) and Cauette, Altman and Naraynan (1998), Saunders (2002). VOLUME 7, NUMBER 4, 2013 9 In 1966, Beaver has introduced the univariate approach of discriminant analysis in the default risk of firm’s explanation. Altman in 1968 has extended it to a multivariate context and developed the Z-Score model. It weights the independent variables (financial ratios and accounting variables) and generates a single composite discriminant score. In 1977 Altman, Haldeman and Narrayman have developed the ZETA model, which integrated some improvements to the original Z-Score approach. Then the binary dependent variables models, known as the logit and probit model, have been used in bankruptcy prediction3. Ohlson (1980) used logit methodology to derive a default risk model known as O-Score. Probit (Logit) methodology weights the independent variables and allocates scores in a form of failure probability using the normal (logistic) cumulative function. Mester (1997) have recognized the prevalent use of the binary credit risk models: 70 % of banks have used them in their non-listed firm lending procedure. Several banks use this method for privately and publicly traded companies, either by buying a model, such as RiskCalc Moody’s, or by programming their own estimate. One problem they often face is to build an appropriate proper database. Very often, credit files are not computerized or do not contain historical data. The main advantage of non-structural models is their accuracy in estimating probabilities of default. In addition, they are easy to use for financial institutions equipped with solid management systems of database and may produce very accurate default probabilities. Nonetheless, these models are not flexible, because they need information from financial statements. Thus, it is very difficult to update the probabilities of default over a year. Some financial institutions may require reporting on a quarterly basis, but they are rarely audited by accounting firms. 2.2 Structural Models The original Merton model is based on some simplifying assumptions about the structure of the typical firm’s finances. The event of default is determined by the market value of the firm’s assets in combination with the liability structure of the firm. When the value of the assets falls below a certain threshold, the firm is considered to be in default. The main criticism leveled at Merton’s model is that it does not account for the possibility that the firm may default before the debt matures. To improve this basic model, several extensions have been suggested in the literature. Crosbie and Bohn (2002) summaries KMV’s default probability model. KMV’s default probability model is based on a modified version of the Black-ScholesMerton framework in the sense that KMV allows default to occur at any point in time and not necessarily at the maturity of the debt. In this model multiple classes of liabilities are modeled. There are essentially three steps in the determination of the default probability. The first step is to estimate the market value and volatility of the firm’s assets, the second step is calculate the distance-to-default, the number of standard deviations the firm is away from default, and the third step is to transform the distance-to-default into an expected default frequency (EDF) using an empirical default distribution. 3 Jones (1987) in his review of bankruptcy literature, concludes that binary dependent variable models do not led to notable improvements in the predictive power of fundamental analysis when compared to the earlier LDA models. 10 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Brockman and Turtle (2003) propose using barrier options. Thus, rather than stockholders who wait for the debt to mature before exercising a standard European call option, we have a down-and-out option on the assets in which lenders hold a portfolio of risk-free debt and a short put option combined with a long down-and-out call option on the firm’s assets. The last part gives them the right to place the company into bankruptcy when they anticipate that its financial health can only deteriorate. Wong and Choi (2004) demonstrate that estimating the parameters of the Brockman and Turtle (2003) model by maximum likelihood yields results that resemble those from the iterative estimation method used in this literature when the theoretical model is Merton’s. The appeal of the maximum likelihood method is that it allows for statistical inference or, more specifically, calculating descriptive statistics for the estimated parameters, such as the value of the firm. Tudela and Young (2003) present an application of the hybrid model. This application uses barrier options with a down-and-out call option. The authors estimate various models on data from non-financial English firms for the period 1990–2001. They use data on firms that did, and did not, default, for their estimates of probabilities of default in the structural model. First, they verify whether the two firm types represent different predicted probabilities of default. Second, they compare their hybrid model with other non-structural models to verify whether the additional probability of default (PD) variable is significant for explaining probabilities of default. Third, they measure the performance of their model with power curve and accuracy ratio type instruments. 3. Estimation of the probabilities of default with the structural model: application of the tudela and young model (2003) (the bank of england model) 3.1 Model description In this model, the authors use the theory of barrier options4 and more precisely the call option down-and-out, which vanishes when the underlying asset reaches the barrier. In this model we assume that the capital structure consists exclusively of debt and equity (as Merton). The level of debt is denoted B and (T-t) represents the time remaining to maturity of the debt, the value of the firm is At and the value, at time t, of the debt maturing at time T is V (A, T, t). The share value at time t is f (A, t). Therefore the total value of the firm at time t is: At = V (A, T, t) + f (A, t) (1) To derive the probability of default using a barrier option we suppose that the value of the firm’s underlying assets follows the following stochastic process: uA = A dA dt + σA A dz (2) 4 Other equity-based models of credit risk that use the concept of barrier options are Black and Cox (1976), Longstaff and Schwartz (1995) and Briys and de Varenne (1997). VOLUME 7, NUMBER 4, 2013 11 As to the liabilities, assume, on one hand, that the firm’s liabilities L are the sum of short-term liabilities plus one-half of long-term liabilities. On the other hand, we assume that L follows a deterministic process: dL = μL L dt (3) We note the asset-liability ratio by k: k= A L (4) ~ A default occurs when k falls below the default point called k it at any time. To estimate the probability of default we need to model how k changes over time. Differentiate (4) and use (2) and (3) we get: dk = (μA - μL) k dt + σA k dz (5) We define: μA – μL = μk And σA = σk The values of μk and σk are needed to calculate the probabilities of default. Maximum likelihood techniques are used to obtain estimates of those two parameters, but to build the maximum likelihood function, we need first to derive an expression for the density function of k. Given equation (5) we can derive the density function of ln It can be shown that the defective density function is given by {ln following expression: . } by the (6) 12 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Equation (6) represents the probability density of not crossing the barrier and being at the point in at time T. This expression is used to construct the likelihood function that we must maximize in order to obtain estimates of μk and σk. These estimates will be used to calculate the probability of default us shown below. The probability of the firm is not defaulting until date T is given by the probability of ~ kT > k conditionally and N is the cumulative density function of the normal distribution. In the case of a European call option, the probability of default equals N (u1). However, for the barrier option we see that the term w [1- N (u2)] adjusts the probability of default to take into account that the firm can default before the horizon date T. ~ The Bank of England set k = 1.We shall adopt this normalization. On the other hand we assume that the ratio, y= X , where X represents the market capitalization of the firm and L is its L liabilities as a proxy for the ratio k= A since the value of the firm’s assets is unobservable. L VOLUME 7, NUMBER 4, 2013 13 We use Matlab to estimate μk and σk with the maximum likelihood method, then we calculate the probabilities of default. Parameters μk and σk are estimated on the basis of a 24-months window for all firms. (As starting value we take σk = 0,4 and μk =0,3). Finally, Tudela and young find that if they add some account variables in their model, the model performance increases slightly. The final model of the Bank of England is as follows: P D= f [probability of default (1 -2 years), profitability, Debt over assets, Cash over liabilities Sales Growth, log number of employees, GDP] This model will be the subject of our research, the authors have applied this model to calculate the probability of default on data from non-financial English firms. We will try to applying it to a sample of French listed companies but retaining other explanatory variables for the hybrid model. 3.2 Data In this section we present the used data and explain how we built it to calculate probabilities of default. This data is used also to estimate the hybrid model in section 4. Our initial database contains 20 companies that did not default and 14 companies that did. The study period for the probabilities of default is from January 2004 to December 2005. The methodology we use to compute the probabilities of default with the structural model requires that our data window extend 24 months prior to the estimation period for the predicted probabilities of default in order to ensure statistical reliability. Market capitalization has a monthly frequency while the values of debt are observed annually, thus the value of debt is considered during the year. 3.2.1 Companies that have defaulted Data on companies that have defaulted are from DIANE. However, 6 companies that defaulted were removed from the database because of lack of data (accounting and / or market) or because too large a shift between the date of publication of the last financial statement and effective date of default. Indeed many of these companies have significant gaps between these two dates. This is explained by the fact that most of the firms do not publish their financial statements during the last year prior to bankruptcy. Another explanation lies in the slow process of putting in default of certain companies. Thus we eliminated firms with a lag of more than 18 months. 3.2.2 Companies that did not default Accounting data on companies that did not default for the year 2005 and the monthly market capitalizations are from Diane. 3.2.3 Various statistics Financial firms were eliminated from the database because they do not generally have the same structure of financial statements as non-financial firms. Thus the final database contains a total of 23 non-financial companies, 8 of them have defaulted .The following table presents the descriptive statistics of firms retained for analysis. 14 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Table 1 Descriptive statistics of all firms retained for analysis (in million Euro) NOT-DEFAULT DEFAULT Statistic Market value Liabilities Market value Liabilities Mean 130,48 44,924 38,292 23,193 Median 101,128 38,714 17,073 9,248 Maximum 386,65 156,147 190,854 120,568 Minimum 27,65 3,955 11,473 7,492 Standard deviation 94,013 36,76 61,698 39,383 Skewness 1,464 1,829 2,259 2,2591 Kurtosis 4,7749 6,7184 6,122 6,1207 Number of observations 360 30 180 16 3.3 Estimation results Estimating probabilities of default by the structural model allowed us to obtain the following results: for companies that have defaulted, the mean of probabilities of default is 33.97% while for companies that did not default is 13.54%. The following figures show the evolution of the probabilities of default predicted for several firms. Figure 1 show the evolution of the probabilities of default for the ones defaulted. Modeling the probabilities of default of these companies seems consistent with the model since it takes the form of probabilities of default predicted higher when approaching the year of default. Indeed, the most of the companies that did default presents a similar evolution of the probabilities of default. However, the results in Figure 2 are somewhat surprising. They represent an extreme case and show an example of the overstatement of the probabilities of default in the structural model. The model appears very sensitive to significant fluctuations in the values of this firm’s stocks and provides the rationale for using the hybrid model, which contains more information for conditioning the estimates of the probabilities of default. Two smoother examples are featured in Figure 3. VOLUME 7, NUMBER 4, 2013 Figure 1 Monthly default probabilities (2 years) of defaulting firms Figure 2 Monthly default probabilities (2 years) of non-defaulting firms 15 16 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Figure 3 Other PDs (2 years) of non-defaulting firms 4. Hybrid model 4.1 Methodology We did not estimate the model with a simple linear regression, since we know that it must reflect non-linear behavior of the explanatory variables for defaults. In addition, it is well documented that simple linear models are inappropriate when the dependent variable is a probability. This model has the advantage of being easy to estimate but he has the disadvantage that it leads to PDs estimated to be out of the interval [0,1]. Thus, we must use other models which keep the probability of default (PD) in the considered interval. This is particularly the probit model. In this type model the dependent variable is a dichotomous variable taking the value 1 if an event occurs and 0 otherwise. In our case, the variable Yi assumes the following values: Yi = 1 if firm I defaults, and Yi = 0 otherwise. VOLUME 7, NUMBER 4, 2013 17 The vector of explanatory variables (financial ratios and accounting variables…) for firm i is denoted Xi, while β is the vector of weights of these variables. The probit model assumes that there is a qualitative response variable (Yi*) defined by the following equation: Yi* = β’ Xi + εi. (7) However, in practice Yi* is an unobservable latent variable. We rather observe the dichotomous variable Yi such that: Yi = 1 if Yi* >0; Yi = 0 otherwise. (8) In this formulation, β’Xi is not E(Yi / Xi) as in the simple linear model, but rather E (Yi* / Xi). From equations (7) and (8), we get. Prob (Yi = 1) = Prob (εi > - β’Xi) = 1 - F (β’Xi)(9) Where F is the cumulative distribution function of εi. The functional form of F in equation (11) depends on the retained assumptions regarding the distribution of the residual errors ( εi ) in equation (7). The probit model is based on the assumption that these errors are independently and identically distributed (i.i.d.) and follow a standard normal distribution N(0,1). The functional form can thus be written: (10) In this case, the observed values Yi are simply the realizations of a binomial process whose probabilities are given by (9) and vary from one observation to the next (with Xi). The likelihood function can be defined as follows: (11) And the parameter estimates β are those that maximize i. 4.2 Variable selection The main objective of this study is to verify whether combining the structural and the non-structural model into a hybrid model yields a better measure of the default risk than those obtained from structural and traditional non-structural models estimated separately. To accomplish this, we aim to explain defaults deficiencies by estimating a probit model in which the explanatory variables are the estimated probabilities of default from the structural model, financial ratios and other accounting data. The dependent variable is binary taking the value of 1 if the default occurs and 0 otherwise. Using the same methodology, we also estimate a model with only ac- 18 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW counting data as explanatory variables (non-structural model) and a third probit model in which the only exogenous variable is the probability of default from the structural model (the model that contains only structural information). Thus, we examine the predictive power of the PD variable to explain corporate bankruptcy by integrating it in the non-structural model as an explanatory variable. If we find that the estimated coefficient of the variable PD (resulting from the structural model) is statistically different from zero, the probabilities of default obtained by the structural model in this case would contain additional information that complements that of accounting data, and we will be able to use its coefficient to update the probabilities of default when the PD from the structural model changes. As to the choice of accounting variables and financial ratios used in the nonstructural and hybrid models, we were faced with difficulties in the selection of variables given the scarcity of accounting and financial data on French listed companies that did default. To make a sound choice, we estimated the probit model on each variable accounting separately. This enabled us to retain the most significant ones. 4.3 Estimation results 4.3.1 Estimation of the probit model with different specifications In this section we analyze the characteristics and performance of three models: the hybrid model, the non-structural model and the model containing only structural information. We summarize the results of these estimations in Table 2. In Model 1, we only use the information from the structural model by considering the mean PD(2 years) from the structural model as an explanatory variable. The coefficient of PD is 0.15 per cent, and has the expected sign. It is a significant factor for predicting probabilities of default, with a p-value of less than 5 per cent and a high corrected pseudo-R2 (52.56 per cent). In Model 2, we estimate the-non-structural model with 2 variables (the turnover and profitability ratio). Examination of Model 2 reveals that the non-structural specification largely outperforms the one using only information from the structural model (Model 1) in terms of its ability to explain corporate bankruptcy. The likelihood ratio is 17.63 for the non-structural model, versus 15.62 for the structural model with only PD as an exogenous variable (the corresponding values of R2 are 59, 34 per cent and 52.56 per cent). In Model 3, we estimate the hybrid model by adding the probabilities of default calculated from the structural model to the explanatory variables of Model 2. An analysis of the results reveals that the probabilities of default from the structural approach have an additional predictive power for corporate defaults than the firms’ financial statements. We observe that the likelihood ratio increased from 17.63 for the non-structural model to 24.6 for the hybrid model (the corresponding values of corrected pseudo-R2 are 59.34 per cent and 82.77 per cent). Furthermore, to acquire to acquire a better understanding of the contribution of model 3 relative to that of Model 2, we repeat this analysis in Models 4 and 5, but this time by changing the variables used in the non-structural model and in Model 19 VOLUME 7, NUMBER 4, 2013 5 we estimate the hybrid model by adding default probabilities calculated from the structural model as an explanatory variables to Model 4. Analysis of the results gives us the same findings confirming the predictive power provided by incorporating the variable PD from the structural model to the non-structural model. Indeed, the likelihood ratio has increased from 18.24 for the non-structural model (model 4) to 21.23 for the hybrid model (the corresponding values of corrected pseudo-R2 are 61.37 per cent and 71.43 per cent). Table 2 Analysis of the maximum-likelihood estimators Parameters Model 1 Model 2 Model 3 Model 4 Model 5 Constant -4,2571 (0,0245) 0,6808 (0,2001) -7,91 (0,3719) 0,7242 (0,4283) -3,8512 (0,2486) PD (2 years) 0,1506 (0,0226) 0,2934 (0,3252) Profitability -0,0405 (0,1079) -0,0229 (0,4589) Turnover -0,0161 (0,0518) -0,0213 (0,1988) 0,1571 (0,1886) -0,0069 (0,4615) -0,0164 (0,3557) Equity/Total assets -0,0394 (0,0659) -0,00709 (0,8033) Debt/Equity 0,0029 (0,8891) 0,0055 (0,8328) Number of observations 23 23 23 23 23 Number of Default 8 8 8 8 8 0,5256 0,5934 0,8277 0,6137 0,7143 Likelihood ratio 15,6209 <0,0001 17,6367 0.0001 24,6009 <0,0001 18,2408 0.0003 21,2300 0.0002 Log likelihood -7,0496 -6,0416 -2,5596 -5,7396 -4,2450 McFadden’s R squared *Into parenthesis are the p-value of estimated parameters 4.3.2 Various tests In Figure 4, we reproduce the mean of the probabilities of default for companies that did default of the five models estimated so far. The mean of the probabilities of default for the model containing only the structural information is 71.41 per cent. This probability is maximized at 89.18 per cent for the hybrid model with two accounting variables (turnover and profitability ratio) (model 3). The same model but without the probabilities of default from the structural 20 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW approach, comes in at 75.83 per cent. This confirms the results from the previous section. Figure 4 Probabilities of default of firms that did default In Figure 5, we reproduce the mean of the probabilities of default for companies that did not default of the five models estimated. The mean of the probabilities of default for the model containing only the structural information is 15.73 per cent. This probability is minimized at 5.56 per cent for the hybrid model with two accounting variables (turnover and profitability ratio) (model 3). The same model but without the probabilities of default from the structural approach, has a mean of 15.3 per cent. This also confirms the results from the previous section. Figure 5 Probabilities of default of firms that did not default To better visualize the performance of the hybrid models, the following figures illustrate a comparison between the predictions of the probabilities of default of two of the five models used (green curves) and the actual situation of firms (red curves). This graphs clearly show that the hybrid model dominates the oth- VOLUME 7, NUMBER 4, 2013 21 ers models. We again observed that the hybrid models are the best estimates of probability curves of defects. Figure 6 Accuracy of score models (Model 4) Figure 7 Accuracy of hybrid models (models 3) 22 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW 5. Conclusion Credit risk measurement is an area of great and renewed interest for both academicians and practitioners. In this paper, our goal was to study a major component of credit risk, the probability of default using the methodology proposed by Tudela and Young (2003) to a sample of French companies whose shares are traded on the Stock Exchange Paris. This model has investigated the ability of hybrid models to calculate the default risk of UK companies by verifying whether combining the structural and the non-structural model into a hybrid model yields a better measure of the default risk than those obtained from structural and traditional non-structural models estimated separately. To accomplish this, we have aimed to explain defaults deficiencies by estimating a probit model in which the explanatory variables are the estimated probabilities of default from the structural model, financial ratios and other accounting data. The dependent variable is binary taking the value of 1 if the default occurs and 0 otherwise. Using the same methodology, we have also estimated a model with only accounting data as explanatory variables (non-structural model) and a third probit model in which the only exogenous variable is the probability of default from the structural model (the model that contains only structural information). Thus, we have examined the predictive power of the probabilities of default from the structural model to explain corporate bankruptcy by integrating it in the non-structural model as an explanatory variable. Our results indicate that the predicted probabilities of default (PDs) contribute significantly to explaining default probabilities when they are included alongside the retained accounting variables. This confirms the results of the study of Tudela and Young (2003) and those of Dionne and al (2005). We note that the main limitation of our work was to fixing the default barrier at any level (equal to 1 in our study). Thus, it would be interesting in future work to make endogenous default barrier and to estimate it by the maximum likelihood method may increase the predictive ability model. References Altman, E. I., (1968). Financial Ratios, Discriminant Analysis and Prediction of Corporate Bankruptcy, The Journal of Finance: p 589–609. Altman E., Haldeman R., Naranan P., (1977). ZETA analysis: a new model to identify bankruptcy prediction risk of corporations, Journal of Banking and Finance, pp. 29-54. Beaver, (1966). Market Prices, Financial Ratios and the Prediction of Failure, Journal of Accounting Research, pp. 179-192. Black, F and Cox, J., (1976). Valuing corporate securities: some effects of bond indenture provisions, Journal of Finance, pp.351–67. VOLUME 7, NUMBER 4, 2013 23 Black, F and Sholes, M., (1973). On the pricing of options and corporate liabilities, Journal of Political Economy, pp.637-659. Brennan, M.J. and E.S. Scwartz (1978). Corporate Income Taxes, Valuation, and the Problem of Optimal Capital Structure, Journal of Business, pp. 103–14. Briys, E. and F. de Varenne (1997). Valuing Risky Fixed Rate Debt: An Extension, Journal of Financial and Quantitative Analysis, pp. 239-48. Brockman, Paul ET H.J. Turtle (2003). A Barrier Option Framework for Corporate Security Valuation, Journal of Financial Economics, pp. 511-529. Crosbie P., and Bohn J., (2003). Modeling Default Risk, Journal of Derivative, pp.9-24. Caoutte J., Altman E., Naraynan P., (1998). Managing Credit Risk: The Next Great Financial Challenge, (John Wiley & Sons). Dionne, Georges. Laajimi, Sadok. Mejri, Sofiane and Petrescu, Madalina, (2005). Estimation of the Default Risk of Publicly Traded Companies, working paper, Bank of Canada, p 63. Duan, Jin-Chuan, Gautier Geneviève and Simonato Jean-Guy (2004). On the Equivalence of the KMV and Maximum Likelihood Methods for Structural Credit Risk Models, mimeo, HEC Montréal, p.19. Jones (1987). Current Techniques in Bankruptcy Prediction, Journal of Accounting Literature, pp. 131-164. Longstaff, F A and Schwartz, E S (1995). A simple approach to valuing risky and floating rate debt , Journal of Finance, pp. 789–819. Merton, R C (1974). The pricing of corporate debt: the risk structure of interest rates, Journal of Finance, pp. 449–470. Mester L (1997). What’s the Point of Credit Scoring?, Federal Reserve Bank of Philadelphia , Business Review, pp.3-16. Ohlson J (1980). Financial Ratios and the Probabilistic Prediction of Bankruptcy, Journal of Accounting Research, pp. 109-131. Saunders, A (2002). Credit Risk Measurement: New Approaches to Value at Risk and Other Paradigms , (John Wiley & Sons). Sobehart, J.R., S.C. Keenan, and R.M. Stein. (2000). Benchmarking Quantitative Default Risk Models: A Validation Methodology, Moody’s Investor Service Rating Methodology. Tudela, Merxe et Garry Young (2003). A Merton-model Approach to Assessing the Default Risk of UK Public Companies, Bank of England, working paper , p19441. Wong, H. and T. Choi, (2004). The Impact of Default Barrier on the Market Value of Firm’s Asset, Working Paper, Chinese University of Hong Kong. Does co-integration and causal relationship exist between the non-stationary variables for chinese bank’s profitability? Empirical evidence Omar Masood*, University of East London Marc Bradford†, ISC Paris Jean-Jacques Levy‡, ISC Paris Abstract This study aims to give the analysis of the determinants of banks’ profitability in the Kingdom of China over the period starting 2003. The paper investigates the co-integration and causal relationship between total assets (TA) and total equity (TE) of Saudi banks. The analysis employs Augmented Dickey Fuller (ADF) test, Johansen’s cointegration test, Granger causality test. Analyzing the cointegration and other tests on Saudi Arabian banking sector over the study period, the relationships between the two variables are examined. The empirical results have found strong evidence that the variables are co-integrated. Keywords: Banking, bank profitability, total assets, total equity, co-integration. JEL Classification: E50, F30, F17, F65, G01, G21 1. Introduction In the last two decades economists have developed a number of tools to examine whether economic variables trend together in ways predicted by theory, most notably cointegration tests. Cointegration methods have been very popular tools in applied economic work since their Introduction about twenty years ago. However, the strict unit-root assumption that these methods typically rely upon is often not easy to justify on economic or theoretical grounds. The multivariate testing procedure of Johansen (1988, 1991) has become a popular method of testing for cointegration of the I(1)/I(0) variety, where I(1) and I(0) stand for integration of orders one and zero, respectively. In the Johansen methodology, series are pre-tested for unit roots; series that appear to have unit roots are put into a vector auto regression from which one can test for the existence of one or more I(0) linear combinations. Utilizing the cointegration and error correction models on all Chinese’s banks over the study period, various potential internal and external determinants are examined to identify the most important determinants of profitability. Cointe* Omar Masood is at University of East London, Docklands Campus, University Way, E16 2RD, UK † Marc Bradford is at ISC Paris, 22, Boulevard du Fort de Vaux, 75017 Paris, France ‡ Jean-Jacques Levy is at ISC Paris, 22, Boulevard du Fort de Vaux, 75017 Paris, France (Corresponding author), E-mail: [email protected] 26 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW gration methodology has been extensively used as a convenient way of testing for the weak-form of asset market efficiency, which states that no asset price should be forecastable from the prices of other assets The Johansen (1988) method of testing for the existence of co-integrating relationships has become standard in the econometrics literature. Since unit-root tests have very limited power to distinguish between a unit-root and a close alternative, the pure unit-root assumption is typically based on convenience rather than on strong theoretical or empirical facts. This has led many economists and econometricians to believe near-integrated processes. Near-integrated and integrated time series have implications for estimation and inference that are similar in many respects. Cointegration, however, simply requires that cointegrating linear combinations have lower orders of integration than their parent series Granger (1986). Granger and Joyeux (1980) and Hosking (1981), where continuous orders of integration from the real line are considered, the case where there exists an I(d − b) linear combination of two or more I(d) series has become known as fractional cointegration. The cointegration approach is one of the recent methodologies employed to identify the determinants of profitability in banking. It enables the estimation of a relationship among non-stationary variables by revealing the long-run equilibrium relationship among the variables. This paper will help to determine the most important factors of profitability in Saudi Arabian banks, and is supposed to help banks’ stakeholders especially the managers and regulatory authorities to improve the sector soundness by boosting the impact of positive factors and lessening the impact of the negative factors. A good econometric practice to always include tests on the cointegrating vectors to establish whether relevant restrictions are rejected or not. If such restrictions are not tested, a non-zero cointegrating rank might mistakenly be taken as evidence in favour of cointegration between variables. This is particularly relevant when there are strong prior opinions regarding which variables “have to” be in the cointegrating relationship. Unit root tests are performed on unvaried time series in order to test the order or integration. If individual time series are found to be integrated of same order after the unit root tests, then these variables may be cointegrated. Cointegration deals with relationships among the group of variables where each has a unit root. Application of cointegration test in the estimation of money demand were analyzed by Johansen and Juselius (1990) and Dickey, Thansen and Thornton (1991). The purpose of this paper is to investigate the effect of deviations from the unitroot assumption on the determination of the cointegrating rank of the system using Johansen’s (1988, 1991) maximum Eigen value and trace tests. The paper will contribute towards the existing literature by interrogating the determinants of profitability of Chinese bank’s using a co-integration approach. First we test for the stationary roots using augmented Dickey-Fuller test, then the Johansen’s unit root test and granger causality test are applied to these variables. The paper is divided into five sections. Section 2 will describe about the previous existing literatures, Section 3 describes an overview of Chinese banking system, Section 4 will give a complete description about the methodologies of the vari- VOLUME 7, NUMBER 4, 2013 27 ous tests performed in this paper, and Section 5 contains the empirical results, finally section 6 concludes with a short summary. 2. Literature review Despite an extensive literature on savings behavior, there are not many studies, which focused primarily on the factors that determine the level of deposits made by various categories of depositors at the commercial banks. These studies, however, concentrated mainly on private and household savings and not on the business and government sectors. Lambert and Hoselitz (1963) were among the first researchers to compile the works of others on savings behavior. They extended the works of researchers who studied the savings behavior of households in Sri Lanka, Hong Kong, Malaysia, India, Philippines. Snyder (1974) and Browning and Lusardi (1996) also presented a similar study which reviewed micro theories and econometric models. Masood, Akhtan & Chaudhary (2009) studied the co-integration and causal relationship between Return on Equity and Return on Assets of Saudi Arabia, they found that there are stable long run relationships between the two variables. They also argued that unidirectional causality from ROE to ROA implies that sustainable development strategies with higher levels of ROE may be feasible and fast economic growth of Saudi Arabia may be achievable. Loayza et al. (2000b) listed papers and publications of the saving research project of a particular country and gave general reference in this area. Thereafter, lots of work has been done on this area Cárdenas and Escobar (1998), Rosenzweig (2001), Kiiza and Pedreson (2001), Athukorala and Kunal Sen (2003), Dadzie et al. (2003), Ozcan, et al. (2003), Athukorala and Tsai (2003), Qin (2003) and Hondroyiannis (2004) have studied the savings behavior of a particular country. A large empirical literature was developed on the cross country comparison, which was contributed by Doshi (1994), Masson et al. (1998), Loayza et al. (2000a), Agrawal (2001), Anoruo (2001), Sarantis and Stewart (2001), Cohn and Kolluri (2003), Ruza and Montero (2003). The first works on cointegration methods were first studied by, Wallace and Warner (1993), Malley and Moutos (1996) that cointegration-based tests of foreign exchange market efficiency, Cardoso (1998), Bremnes et al. (2001), Jonsson (2001), Khamis and Leone (2001) and Bagchi et al. (2004). Studies arguing the stationary of these variables include Song and Wu (1997, 1998), Taylor and Sarno (1998), Wu and Chen (2001) and Basher and Westerlund (2006). Sephton and Larsen (1991) showed that inference based on Johansen cointegration tests of foreign exchange market efficiency suffers from structural instability. Cheung and Lai (1993) argued that significant finite-sample bias in the performance of the Johansen test statistics when asymptotic critical values are used for inference in finite samples. As unit-root tests have very limited power to distinguish between a unit-root and a close alternative, the pure unit-root assumption is typically based on convenience rather than on strong theoretical or empirical facts Stock (1991), Cavanagh et al., (1995) and Elliott (1998) argued that near-integrated processes, which explicitly allow for a small deviation from the pure unit-root assumption, to be 28 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW a more appropriate way to describe many economic time series. Phillips (1988) concluded that spurious regressions are a problem when variables are nearintegrated as well as integrated and presented an analytical discussion Elliott (1998) shows that large size distortions can occur when performing inference on the cointegration vector in a system where the individual variables follow nearunit-root processes rather than pure unit-root processes. The banks profitability is generally classified into two broad categories i.e. internal and external. The internals factors are in the control and framework of the bank for instance number or employees, investments etc whereas the external factors are out of control and framework of the bank for instance, market share, competition, inflation etc. Lots of literature has already been developed interrogates the profitability of banks of the particular country in question. Hester and Zoellner (1966) argued that the balance sheet structure has a significant impact on profitability. Smirlock (1985) found a significant positive relationship between demand deposits and profits. Lambert and Hoselitz (1963) were among the first researchers to compile the works of others on savings behavior. Heggested (1977) interrogated the profitability of commercial banks and reports that time and savings deposits have negative impact on profitability. Steiner and Huveneers (1994) found similar association while studying overhead expenditure. Bourke (1989), and Molyneux and Thorton (1992) found that capital and staff expenses are positively related to bank’s profitability. Mullineaux (1978) found a positive impact for bank’s size on profitability. Studies of Pelzman (1968), Vernon (1971), Emery (1971), Mullineaux (1978) and Smirlock (1985) concluded that regulation have a significant impact on banks’ profitability. Emery (1971) examined the effect of competition on banks’ profitability and find insignificant association between the two variables. Smirlock (1985) further examined the effect of concentration on profitability and the findings of these studies were mixed and inconclusive. Demirgüç-Kunt and Huizinga (1998) concluded that that the well-capitalized banks have higher net interest margins and are more profitable. Keynes (1936), despite arguing the quantitative importance of the interest rate effect, believes that in the long run substantial changes in the rate of interest could modify social habits considerably, including the subjective propensity to save. The importance of the rate of interest on consumption, many researchers using various methodologies tried to establish the strength of relationship between these two elements. Wright (1967), Taylor (1971), Darby (1972), Heien (1972), Juster and Watchel (1972), Blinder (1975), and Juster and Taylor (1975) in their studies found an inverse relationship between interest rate and consumption. Modigliani (1977) based on his works and after seeing evidence on the effect of interest rate on consumption concludes that the rate of interest effects on demand, including the consumption component, are pervasive and substantial. Alrashdan (2002) found that the return on asset (ROA) is positively related to liquidity and total assets while ROA is negatively related to financial leverage and cost of interest. Naceur (2003) examined the determinants of Tunisian banks’ profitability over the period 1980-2000, and found that the capital ratio, loans VOLUME 7, NUMBER 4, 2013 29 and stock market development have positive impact on profitability while the bank’s size has a negative impact. Hassan and Bashir (2003) stressed on the fact that on the importance of customer and short-term funding, non-interest earning assets, and overhead in promoting profits. They also argued that profitability measures respond positively to increases in capital ratio and negatively to loan ratios. Haron and Azmi (2004) also investigated the determinants of Islamic Banks across various countries using time series techniques of cointegration and errorcorrection mechanism (ECM). The study concludes that liquidity, deposit, asset structure, total expenditures, consumer price index and money supply to have significant impact on profitability while capital structure, market share and bank size to have no impact. Alkassim (2005) examined the determinants of profitability in the banking sector of the GCC countries and found that asset have a negative impact on profitability of conventional banks but have a positive impact on profitability of Islamic banks. They also observed that positive impact on profitability for conventional but have a negative impact for Islamic banking. Liu and Hung (2006) examined the relationship between service quality and long-term profitability of Taiwan’s banks and found a positive link between branch number and long-term profitability and also proved that average salaries are detrimental to banks’ profit. 3. An overview of chinese banking system More than 85% of financial resources in China are allocated through the banking system hence the economic reforms of China are dominated by Chinese commercial banks. The measurement of profitability and competitive conditions of banking system depicts the economic growth of China. Since 1978, the Chinese economy has experienced an impressive annual growth rate of about 10 percent. China’s financial assets have grown at the annual rate of about 18 percent, or more than twice the growth rate of GDP. Since the People’s Republic of China was founded in 1949, the People’s Bank of China (PBOC) has been the only bank in Mainland China until 1978. During this period, the PBOC played a dual role in China’s financial system: as a central bank and as a commercial bank. Upon nationalization of private banks in the 1949, bank management in China essentially followed the approach of centrallyplanned economy. During the period of 1953-1978, four of the Chinese banks had operated intermittently as separate units as four Chinese banks had operated intermittently as separate units or as a single entity. During the period 1978-1984, People’s Bank of China (PBC), Agriculture Bank of China (ABC), Bank of China (BOC), and Construction Bank of China (CBC). During the period 1978-1984, the PBC retained the central banking activities, while four other banks – ABC, BOC, CBC, and ICBC, were carved out from the PBC to provide specialized services. The modern Chinese banking system comprises or four state-owned commercial banks, as well as several joint-stock commercial banks, city commercial banks, rural credit cooperatives, finance companies, and trust and investment companies, foreign banks have been allowed to be an integral part of the banking system. 30 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW As the economy started shifting toward market-orientation, profitability and liquidity – if any – of state-owned enterprises suffered due to their inability to adapt to the challenges imposed. State-owned banks have been hampered by non-paying or delinquent loans to other state-owned enterprises. Both the stateowned and private banks had to extend credit to the inefficient, monopolist, state-owned enterprises or public projects. In 1994, three policy banks - China Development Bank (CDB), Agricultural Development Bank of China (ADBC), and the Export-Import Bank of China (China Eximbank) - were established. They undertook most part of policy loans business from the four national specialty banks. Meanwhile, the four national specialty banks became state-owned commercial banks. In 2003, China Banking Regulatory Commission (CBRC) was established to take over most supervisory function of PBOC, becoming a main regulator of China’s banking industry. By the end of 2007, there were 3 policy banks, 5 large state-owned commercial banks, 12 jointstock commercial banks, 124 city commercial banks, and 29 locally incorporated foreign bank subsidiaries as well as other banking institutions. Chinese authorities have adopted more flexible approach in seeking help from foreign banks in rescuing weak banks. Foreign banks, when they are well-capitalized and have an access to external markets, are less likely to turn off the credit when monetary authorities pursue tight monetary policies. Foreign banks’ presence also provided needed competition to domestic banks unburdened of non-performing loans. Competitive pressure in turn provided incentives to local banks to improve management practices. According to Commercial Banking Law of the People’s Republic of China which began to be in effect on July 1st of 1995, one of the prerequisites to establish commercial banks is “Having directors and senior management personnel with professional knowledge for holding the post and work experiences”. In June 2002, the People’s Bank of China promulgated Guidance on Independent Directors and External Supervisors of Joint-Stock Commercial Banks, which aims to establish and enhance the arrangement of independent directors. For the Chinese banking, most crucial issues is to convert the four state-owned commercial banks, with 70 percent of the nation’s financial assets and loans, into share-holding companies. During system-wide crisis, state-owned banks often become safe havens because the public perceives that their funds will be fully guaranteed by the state. Chinese banks have suffered from bad loans and high operating costs for a long time. A sound banking industry is essential for development of efficient financial markets, in turn, efficient markets are crucial for ensuring effectiveness of the market-oriented economy. Excessive bad debts reflecting inefficient bank management are thus antithetical to the goal of market orientation for the Chinese economy. They were underperforming their counterparts in other countries. In recent years, Chinese government has carried out a series of reforms aiming at making banks more market driven, more profitable, and well managed. One important reform among these is to establish a board of directors system in existing banks to improve corporate governance. In this context, how effective the role of board of directors have played in the profitability of Chinese banks is up to close examination. VOLUME 7, NUMBER 4, 2013 31 4. Methodology The estimation of the long run relationship between the variables, time series properties of the individual variables are examined by conducting Augmented Dickey Fuller (ADF) stationary tests, then the short run dynamic and long run co-integration relationship are investigated by using the multivariate Johansen’s co-integration test and Granger Causality test. 4.1 Unit root tests The Augmented Dickey-Fuller (ADF) unit root test method put forward by American scholars Dickey and Fuller is widely used in the academia to examine the stationarity of the time series and determine the integration order of non-stationary time series. Unit root tests are first conducted to establish the stationary properties of the time series data sets. Stationary entails long run mean reversion and determining a series stationary property avoids spurious regression relations. It occurs when series having unit roots are regressed into one another. The presence of non-stationary variables might lead to spurious regressions and nonobjective policy implications. Augmented Dickey Fuller (ADF) tests are used for this purpose in conjunction with the critical values, which allows for calculation of critical values for any number of regressors and sample size. The ADF model used is describes as follows: (1) Here Y: variable used for unit root test, α is the constant, T represents the trend, ω = p-1 and ε is the white noise series. The null hypothesis is HO: ω =0. If the ADF value of the lnY is bigger then the McKinnon value at 5% significant level, the null hypothesis is accepted, which means lnY has unit root and is non-stationary. If it is less then the McKinnon value then the H0 is rejected and lnY is stationary. As for the non-stationary series, we should test the stationary of its 1st difference. If the 1st difference is stationary, the series has unit root and it is first order integration I (1). 4.2 Johansen’s co-integration test According to the co-integration theory, there may be co-integration relationship between the variables involved if they are 1 order integration series, i.e. their 1st difference is stationary. There are two methods to examine this cointegration relationship, one is EG two-step procedure, put forward by Engle and Granger in 1987, the other is Johansen cointegration test (Johansen(1988) and Juselius1990) based on Vector Auto Regression (VAR). 32 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW For co-integration test, we will conduct the Johansen’s multivariate co-integration tests. The Johansen’s multivariate co-integration test involved testing the relationships between the variables following vector auto-regression (VAR) model: (2) Yt represents n*1 vector of I (1) variables. Γ and Π are n*n matrix of coefficients to be tested. B denoted n*h matrix and Xt denoted h*1 vector of I(0) variables. Π denoted the rank of the matrix and interrogates the long-run relationships in the variable and is equal to the number of independent co-integrating vectors. If rank of Π is 0, the variables in are not co-integrated. Johansen developed two test statistics: the trace test and the maximum eigen value test. λtrace statistic tests the null hypothesis that r= 0 (no co-integration) against a general alternative hypothesis of r>0 (co-integration). The Kmax statistic tests the null hypothesis that the number of co-integrating vectors is r against the specific alternative of r+1 co-integrating vectors. The test statistics obtained from λtrace and Kmax tests are compared against the asymptotic critical values of the two test statistics by Johansen and Juselius. 4.3 Granger Causality test The pair wise Granger causality tests are used to examine whether the past value of a series Xt , will help to predict the value of another series at present Yt taking into account the past value of the previous value of Yt. The two series are first tested for stationary using the ADF unit root test, followed by the Johansen co integration test before performing the Granger causality test. If the time series of a variable is stationary or I(0) from the ADF test, or if the time series are found to be I(1) and co integrated. The Granger causality test is as follows: (3) (4) VOLUME 7, NUMBER 4, 2013 33 Where Xt is the log of the first variable at time t, and Yt is the log of the second variable at time t. µx,t and µy,t are the white noise error terms at time t. αx,i is the parameter of the past value of X, which tells us how much past value of X explains the current value of X and βx,i the parameter of the past value of Y, which tells us how much past value of Y explains the current value of X. Similar meanings apply to αy,i and βy,i . 4.4 Data The data used in this paper was collected from 16 most significant banks of China, which includes banks like Agriculture Bank of China, Agricultural Development bank of China, China Development Bank, China Merchant bank, Bank of Communication, Indrustrial and Commercial bank of China, China Everbright bank, China Construction Bank, Bank of China, Hua Xia Bank, Export-Import bank of China, Shen Zhen Ping An, Shen Zhen Development Bank, Xia Men International bank, Min Sheng, Shangai Pudong Development bank. The dataset was developed by collecting the information from these banks. The pooled data was made by combining the datasets from all the banks, and the regression analysis was performed on this pooled data to obtain the results which are mentioned in the next section. The mean square and double accounting techniques were also used on the dataset, wherever required. 5. Empirical analysis 5.1 Unit root test We test for the presence of unit roots and identify the order of integration for each variable using the Augmented Dickey–Fuller (ADF). The null hypothesis is considered as non-stationary. The test on the variable total assets gave the following result. The computed ADF test-statistic (1.331162) is greater than the critical values (-8.033476, -4.541245, -3.380555 at 1%, 5% and 10% significant level, respectively), thus we can conclude that the variable total assets has a unit root i.e. it is a non-stationary series. Table 1 ADF test statistics Null Hypothesis: total assets has a unit root Exogenous: Constant, Linear Trend. Lag Length: 1 (Fixed) Augmented Dickey Fuller Test Statistic Test critical values 1% level 5% level 10% level t-statistic Prob* 1.331162 0.9773 -8.033476 -4.541245 -3.380555 * MacKinnon (1996) or Augmented Dickey-Fuller Test Equation Dependent Variable: D(total assets). Method: Least Squares 34 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Augmented Dickey fuller test equation Dependent variable - D(total assets) Method – Least Squares Variable Total assets Coefficient 0.523410 R-squared Adjusted R-squared S.E. of regression Sum squared resid. Log likelihood F-statistic Prob (F-statistic) Standard error 0.393198 0.639249 0.278497 52281516 2.73E + 15 -55.92536 1.771993 0.410164 T-statistic Prob. 1.331162 0.4102 Mean dependent variable S.D. dependent variable Akaike info Criterion Schwarz Criterion Hannan Quinn Criterion Dublin Watson stat 1.29E + 08 615510168 38.61691 38. 01598 37.40897 2.997813 In order to eliminate the heteroskedasticity of total assets and total equity as, we take their natural logarithm and define them as LnTA and LnTE. Similarly, ADF tests were conducted on total equity and the logged variables of total assets and total equity differentiated by their order of integration are reported in Table 2. Table 2 Results of ADF unit root test Variable ADF-statistic Critical value (5%) Total assets Total equity LnTA LnTE 1.331162 -0.831217 0.646288 0.599704 AIC SC Result -4.541245 38.61691 38. 01598 non- stationary -4.542245 32.61347 32.01255 non- stationary 3.850555 -2.032422-2.633347 stationary 3.380555 9.1059548.504869 stationary The lag is added to make the residual be white noise, AIC is Akaike Info.Criterion and SC is the Schwarz Criterion. As shown in Table 3, for the variables of total assets and total equity, the results shows that it is evident that we found the presence of a unit root at conventional levels of statistical significance for the variables of total assets and total equity. To see whether they are integrated of order one I(1) at the 1% level, we performed augmented Dickey–Fuller tests on their first difference. The results of the unit root test show that the first differences of both series are stationary which are found to reject the null hypothesis of unit root. Therefore we can conclude that all series involved in the estimation procedure are regarded as I(1), and it is suitable to make co integration test. VOLUME 7, NUMBER 4, 2013 35 5.2 Johansen’s Cointegration test As proved by previous test the variables under analysis are integrated of order 1 (namely I(1)), hence now the co-integration test is performed. The proper way to test for the relationship between total assets and total equity is certainly to test for a co-integrating equation. In testing co-integration relationships, we use the Johansen and Juselius method of testing. For selecting optimal lag length for the co-integration test, we adopt the Schwartz Information Criterion (SIC) and Schwarz criterion (SC) Criterion. The co-integration tests results performed on the variables gave the following result. Table 3 Cointegration rank test Series: total_assets-total_equity Lags interval (in first differences): 1 to 1 Unrestricted Cointegration Rank Test (Trace) Hypothesized No. of CE(s) None** At most 1** Eigen values 0.874563 0.598298 Trace statistic 44.57638 8.746002 Trace test indicates 2 cointegrating eqn(s) at the 0.05 level ** denotes rejection of the hypothesis at the 0.05 level * MacKinnon-Haug-Michelis (1999) p-values 0.05 critical value 12.64738 4.983491 Prob* 0.0010 0.0194 Unrestricted Cointegration Rank Test (Maximum Eigenvalue) Hypothesized No. of CE(s) None** At most 1** Eigen values 0.874563 0.598298 Max Eigenstatistic 38.98723 8.746002 Trace test indicates 2 cointegrating eqn(s) at the 0.05 level ** denotes rejection of the hypothesis at the 0.05 level * MacKinnon-Haug-Michelis (1999) p-values 0.05 critical value 11.56473 4.983491 Prob* 0.0010 0.0194 Therefore, by applying Johansen test on total assets and total equity series we found the presence of two cointegration vectors. Therefore, by applying Johansen decision rule, we conclude that there are two co-integration vectors for the model. Hence our findings imply that there are stable long run relationships between the two variables i.e. total assets and total equity. The results for the Johansen’s test are concluded in Table 4. Table 4 Results of Johansen’s cointegration test Eigen-value t-statistic 0.874563 0.598298 44.57638 8.746002 Critical value (0.05) 12.64738 4.983491 Trace test indicates 2 co-integrating eqn(s) at the 0.05 level. Prob. 0.0010 0.0194 Null-hypothesis r =0 r ≤1 36 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW 5.3 Granger Causality Test Granger causality test demands that the economic variables should be stationary series. So we need to examine the stationarity of the 1st difference. Hence we test variables LnTA and LnTE so as to observe the causality between total assets and total equity. As the sample of observation for this test is small, we take the lag to be 1. The results of Granger Causality test are shown in table 5. Table 5 Pairwise granger Causality test Null Hypothesis F-statistic 3.91826 0.1632 LN_TE does not granger cause LN_TA 6.02545 0.0984 LN_TA does not granger cause LN_TE Prob Hence by applying the granger causality test to the variables can interpret that total assets is a granger cause to total equity and total equity is also a granger cause to total assets . In other words total assets can affect total equity input, similarly total equity can also affect the total assets in the Chinese Banking sector. Therefore, there exist a bi-direction cause-effect relationship between total assets and total equity. The results of the Granger Causality are concluded in table 6. Table 6 Results of Granger Causality Test Lag 1 Ho LnTA F-value P-value 0.1632 Accept Ho 1 LnTE 6.02545 0.0984 Accept Ho 5.4 Graphical Comparison 3.91826 Result VOLUME 7, NUMBER 4, 2013 37 To further illustrate the relationship between total assets and total equity in the Chinese Banking sector, we also conducted a graphical comparison of the two variables over a four year period. The above figure depicts that both the variables show similar kind of trend till 2006. After 2006, the total assets observed a positive growth while total equity experienced a small decerement. 6. Conclusion In testing the co-integration and causal relationship between total assets and total equity, the time series model of ADF unit-root test, Johansen co-integration test, Granger causality test and graphical comparison model are employed. The empirical results have found strong evidence that the variables are co-integrated and feedback. By applying Johansen decision rule, we found that there are two co-integration vectors for the model. Hence our findings imply that there are stable long run relationships between the two variables i.e. total assets and total equity. Furthermore after the granger causality test to the variables we found that there exist a bi-directional cause-effect relationship between total assets and total equity in the Chinese banking sector. By applying the granger causality test to the variables we found that total assets are a granger cause to total equity and total equity is also a granger cause to total assets. The evidences of long-run bi-directional causality from total assets to total equity implies that sustainable development may be feasible and fast economic growth of China may be achievable. Furthermore by graphical comparison we found that both the variables were observed having similar kinds of trends over the period of last four years. References Anoruo, E., 2001. Saving-Investment Connection: Evidence from the Asean Countries’, American Economist, spring, Vol. 45, N°1, pp. 46-53. Athukorala, Prema-C, and Long Pang Tsai, 2003. Determinants of Household Saving in Taiwan: Growth, Demography and Public Policy, Journal of Development Studies, Vol. 39, Issue. 5, pp. 69-88. Agrawal, Paradep, 2001. The Relation between Savings and Growth: Cointegration and Causality Evidence from Asia, Applied Economics, Vol. 33, pp. 499-513. Athanasoglou, Panayiotis, S. Brissimis and M. Delis., 2008. Bank-specific, industry-specific and macroeconomic determinants of bank profitability, Journal of International Financial Markets, Institutions & Money, Vol.18,Iss.2;p.121-136. Bashir, Abdel-Hameed, 2003. Determinants of Profitability in Islamic Banks: Some Evidence from the Middle East, Islamic Economic Studies, Vol. 11, N°1, 3160. Blonigen, B. A., 1997. Firm-Specific Assets and the Link Between Exchange Rate and Foreign Direct Investment, American Economic Review, Vo1. 87, N°3, pp. 447465. 38 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Bourke, Philip, 1989. Concentration and Other Determinants of Bank Profitability in Europe, North America and Australia, Journal of Banking and Finance, 13, 65-67. Boldla, B., R. Verma, 2006. Determinants of profitability of banks in India: A multivariate analysis, Journal of Services Research, Vol.6, Issue .2; pp.75-89. Browning, Martin and Annamaria Lusardi, 1996. Household Saving: Micro Theories and Micro Facts, Journal of Economic Literature, Vol XXXIV, December, pp. 1797-1855. Cardenas, Mauricio and Andreas Escobar, 1998. Saving Determinants in Colombia: 1925-1994, Journal of Development Economics, Vol. 57, Issue. 1. pp.5-44. Cesaratto, Sergio, 1999. Savings and Economic Growth in Neoclassical Theory, Cambridge Journal of Economics, November, Vol. 23, N°6, pp. 771-793. Cheung, Yin-Wong, and K.S.Lai, 1993. Finite-sample Sizes of Johansen’s Likelihood Ratio Tests for Cointegration, Oxford Bulletin of Economics and Statistics, Vol. 55, pp.313-328. Doshi, Kokila, 1994. Determinants of Saving Rate: An International Comparison, Contemporary Economic Policy, January, Vol. 12, Issue. 1, pp. 37-45. Demirgüç-Kunt, Asli and Huizinga, Harry, 1998. Determinants of commercial bank interest margins and profitability: some international evidence, (Electronic Version), The World Bank economic Review, Oxford University Journal Vol. 13, May 2, 1999: PP. 379-408. Darby, M.R., 1972. The Allocation of Transitory Income among Consumers’ Assets, American Economic Review, December, pp. 928-41. D.A. Dickey, W.A. Fuller, 1979. Distribution of the Estimates for Autoregressive Time Series with a Unit Root, Journal of the American Statistical Association 74 (1979) 427-431. D.A. Dickey, W.A. Fuller, 1981. Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root, Econometrica, 49 1057-1072. El-Bdour, Radi and Cengiz Erol, 1989. Attitudes, Behaviour and Patronage Factors of Bank Customers Towards Islamic Banks, International Journal of Bank Marketing, Vol. 7, N°6, pp. 31-37. Engle, R.F. And C.W.J. Granger, 1987. Co-integration an Error Correction Representation, Estimation, and Testing, Econometrica, Vol. 55, pp. 251-276. Erol, C., and El-Bdour, R., 1989. Attitudes, Behaviour and Patronage Factors of Bank Customers towards Islamic Banks, International Journal of Bank Marketing, Vol 7, N°6, pp. 31-39. Friedman, Milton, 1957. A Theory of the Consumption Function. General Series 63, National Bureau of Economic Research, Cambridge, Mass. Processed. Fraser, Donald R. and Peter S. Rose, 1972. Bank Entry and Bank Performance. Journal of Finance, Vol. 21, No 1 (March), 65-78. VOLUME 7, NUMBER 4, 2013 39 Granger, C.W.J., 1980. Testing for Causality: A Personal Viewpoint, Journal of Economic Dynamics and Control, 2, 329-352. Granger, C.W.J., 1969. Investigating Causal Relation by Econometric Models and Cross-Spectral Methods, Econometrica, 37, 424-438. Haron, Sudin, 1996a. The Effects of Management Policy on the Performance of Islamic Banks, Asia Pacific Journal of Management, Vol. 13, N°2, 63-76. Haron, Sudin and Azmi, Wan Nursofiza Wan, 2004. Profitability Determinants of Islamic Banks: A Co integration Approach, (Electronic Version), Islamic Banking Conference, Union Arab Bank, Beirut, Lebanon, 5-7 December, 2004. Heggested, Arnold A., 1977. Market Structure, Risk, and Profitability in Commercial Banking. Journal of Finance, 32 (September), 1207-16. Heien, D.M., 1972. Demographic Effects and the Multiperiod Consumption Function, Journal of Political Economy, Vol 80, no 1, pp. 125-88. Hondroyiannis, George, 2004. Estimating Private Savings Behaviour in Greece. Journal of Economic Studies, Vol. 31, Iss. 5, pp.457-476. H. Pesaran, Y. Shin, 1998. Generalized Impulse Response Analysis in Linear Multivariate Analysis, Economics Letters 58 17-29. Johansen, S. And K. Juselius, 1990. Maximum Likelihood Estimation and Inference on Cointegration with Application to the Demand for Money, Oxford Bulletin of Economics and Statistics, Vol. 52, pp.169-210. Johansen, S., 1988. Statistical Analysis of Cointegration Vectors, Journal of Economic Dynamics and Control, Vol. 12, pp. 231-254. Juster, F.T. And Taylor, L.D, 1975. Towards a Theory of Saving Behaviour, American Economic Review, Papers and Proceedings, May, pp. 203-9. Kwast, Mayron L., and John T. Rose, 1982. Pricing, Operating Efficiency, and Profitability among Large Commercial Banks. Journal of Banking and Finance, Vol. 6 N°2 (June), 233-254. Kahf, Monzer and Khursid Ahmad, 1980. A Contribution to the Theory of Consumer Behaviour in Islamic Society, Studies in Islamic Economics, Leicester: The Islamic Foundation. Kiiza Barnabas and Glen Pederson, 2001. Rural Household Savings Mobilization: Empirical Evidence from Uganda, Journal of African Economics, Vol. 10, N°3, pp. 390-409. Loayza, Norman and Rashmi Shankar, 2000. Private Savings in India, World Bank Economic Review, Vol. 14, N°3, pp. 571-594. Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Serven, 2000a. What Drives Private Saving Across the World, Review of Economics and Statistics, May, Vol. 14, N°3, pp. 393-414. Liu, Yong, and Jung-Hua Hung, 2006. Services and the long-term profitability in Taiwan’s banks, Global Finance Journal, Vol. 17, Iss.2; p. 177-191. 40 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Li- Zhou, WU Yu-ming, LI Jian-xia, 2007. Co-integration and Causality between R&D Expenditure and Economic Growth in China: 1953 – 2004. International Conference on Public Administration Molyneux, Philip and John Thornton, 1992. Determinants of European Bank Profitability: A Note, Journal of Banking and Finance, 16, 1173-1178. Metwally, M.M., 1997. Differences between the Financial Characteristics of Interest-Free Banks and Conventional Banks, European Business Review, Vol 97, N°2, pp. 92-98. Masood O., Akhtan B. and Chaudhary S., 2009. An empirical study on the banks profitability in the KSA: a co integration approach, African journal of Business Management Vol 3(8) August, pg 374-382 Mullineaux, Donald J., 1978. Economies of Scale and Organizational Efficiency in Banking: A Profit-Function Approach, Journal of Finance, 33, 259-280. Modigliani, F. and R. Brumberg, 1954. Utility Analysis and the Consumption Function: An interpolation of the Cross-Section Data, in Post-Keynesian Economics, (Ed.) K. Kurihara, Rutgers U. Press, New Brunswick, NJ, pp. 388- 436. Ozcan, Kivilcim M., Asli Gunay, and Seda Ertac, 2003. Determinants of Private Savings behaviour in Turkey, Applied Economics, Vol. 35, Iss. 12, pp. 1405- 1416. Ruza, Cristina and Jose, M. Montero, 2003. Empirical Analysis of Savings Behaviour in European Countries: New Insights, International Advance in Economic Research, Vol. 9, N°4, November, pp. 279-287. Snyder, Donald W., 1974. Econometric Studies of Household Saving Behaviour in Developing Countries: A Survey’, Journal of Development Studies, Vol. 10, 2, January, pp. 139-154. Short, Brock K., 1979. The Relation Between Commercial Bank Profit Rates and Banking Concentration in Canada, Western Europe and Japan, Journal of Banking and Finance, 3, 209- 219. Smirlock, M., 1985. Evidence on the (Non) Relationship Between Concentration and Profitability in Banking.” Journal of Money, Credit and Banking, Vol. 17, N°1 (February), 69-83. Steinherr, A., and Ch. Huveneers, 1994. On the Performance of Differently Regulated Financial Institutions: Some Empirical Evidence. Journal of Banking and Finance, Vol. 18, 271-306. Uster, F.T. and Wachtel, P., 1972. A Note on Inflation and the Saving Rate. Brooking Papers on Economic Activity, N°3, pp. 765-78. Wright, C., 1967. Some Evidence on the Interest Elasticity of Consumption. American Economic Review, September, pp. 850-4. Yu Qiao, 2005. PPP, real exchange rate and international competitiveness – theoretical approach about calculating Chinese weighted real exchange rate index, Financial Research, N°1, pp. 57-62. Dependence structure between the equity market and the foreign exchange market after the greek crisis: evidence from france and germany Adel Boubaker*, University of Tunis el Manar, Tunis Jaghoubbi Salma†, University of Tunis el Manar, Tunis Abstract The main objective of this study is to investigate the dependence structure and to test the degree of the dependence between financial returns using two measures of dependence: correlations and copula functions. Five candidates, the Gaussian, the Student’s t, the Frank, the Clayton and the Gumbel copulas, are compared. Our empirical results imply that both of the degree of the dependence and the dependence structure between each pair of stock-FX returns vary when the financial Greek crisis occurs. Our findings have important implications for global investment risk management by taking into account joint tail risk. Keywords: Greek financial crisis, Stock return, Foreign exchange rate, Rank correlation, copula approach. JEL Classification: G01, G12, F31, C1 1. Introduction Understanding the dependence structure across international financial markets remains a crucial issue for risk management and portfolio management. Many researchers have focused on the co-movement among worldwide financial markets and have investigated the relationship of world exchange rates during a worldwide financial crisis. Besides, some literature studies the co-movements across markets of different asset types, such as the stock market and foreign exchange rates. As to the relation between equity prices and exchange rates, economic theory insists two different approaches, namely, the ‘flow-oriented’ approach and the ‘stock-oriented’ approach. First, the flow-oriented approaches insist that FX rate changes affect international competitiveness and trade balance. Hence local currency depreciation works to strengthen their competitiveness of domestic companies as their exports will be cheaper in international trade. As a result, the flow-oriented approaches claim appositive linkage between equity prices and the FX rate. Under the stock-oriented approaches, the portfolio balance models are often considered. (Frankel 1983, Branson and Henderson 1985, for example) These approaches consider an internationally diversified portfolio, and these models suggest that the FX rate dynamics function to balance the demand * Adel Boubaker is at the University of Tunis el Manar, Tunis, B.P 248 El Manar II 2092 Tunisia † Jaghoubbi Salma is at the University of Tunis el Manar, Tunis, B.P 248 El Manar II 2092 Tunisia, (Corresponding author), E-mail: [email protected] 42 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW and supply of domestic and foreign financial assets. Thus in these approaches, an increase in domestic equity prices will lead an appreciation of the domestic currency since investors’ demand for domestic currency increases in order to purchase domestic equities. Therefore, these approaches suggest a negative relationship between FX rates and equity prices. The empirical literature provides conflicting findings regarding the dynamic linkage between Foreign Exchange market and stock prices. In this context, early studies including Jorion (1990, 1991) suggest that foreign exchange rate changes offer no predictive power for stock returns volatility, although others such as Dumas and Solnik (1995) and Roll (1992) claim the existence of a strong linkage between Foreign Exchange rate changes and stock market volatility. Moreover, Yang and Doong (2004) investigate the asymmetries in the volatility transmission mechanism between stock prices and exchange rates for the G7 countries over the period 1979–1999. Using a multivariate EGARCH model, they showed that exchange rate changes have a direct impact on future changes of stock prices. Phylaktis and Ravazzolo (2005) study the long-run and short-run dynamics between stock prices and exchange rates using cointegration approach and multivariate Granger causality tests for some pacific basin countries. Their results indicate that stock prices and FX markets are positively linked. Recently, Aloui (2007) explores the nature of mean, volatility and causality transmission mechanisms between stock and FX markets for the United States and some major European markets for the period pre- and post-euro, using a multivariate EGARCH model. Results found show that movements of stock prices affect exchange rate dynamics for the two periods pre- and post-euro. However, stock markets are less influenced by exchange rates movements for the two periods. In a more recent paper, Zhao (2010) analyzes the dynamic relationship between the real effective exchange rate and the Chinese stock price, using a VAR with a multivariate GARCH model. The results show that there is no stable long run equilibrium relationship between the two financial markets. Furthermore, the paper reveals that bidirectional causality exists between volatility on the two markets. Using a copula based approach; Ning (2010) investigates the dependence structure between the equity market and the foreign exchange market for the period pre- and post-euro for the G5 countries (US, UK, Germany, Japan, and France). Ning (2010) finds significant and positive tail dependence between the foreign exchange market movements and the stock market in each country for the two sub-periods. Using Latin America countries, Diamandis and Drakos (2011) examine dynamic linkages between exchange rates and stock prices. Their empirical results show that there is a significant long run relationship between the local stock market and the foreign exchange market but that the stability of the relationship is affected by financial and currency crises such as the Mexican currency crisis of 1994 and the subprime crisis. In this paper, the nature of the relationship and the dependence structure between exchange rates and stock prices is investigated for two developed European countries namely France and Germany, and also for Greece, the native country of the crisis, during the period 2007-2011. These two states, France and Germany, represent almost one quarter of the territory of the European Union and third of the population and almost half of Gross Domestic Product of 27 countries. Both countries provide 36% of the EU budget, and in terms VOLUME 7, NUMBER 4, 2013 43 of decision making, France and Germany have 31% of votes in the European Council. Besides, these two States represent the founding members of the European Economic and Monetary Union which is created in 1999. On January 1, 2001, Greece became the twelfth member of the euro area. This Eurozone now consists of seventeen countries with heterogeneous economic structures and sizes. The common monetary policy cannot be adapted to all. Therefore, real differences occur. In addition, the slowdown after 2001, as the financial crisis of 20072008, has not affected European countries in the same way. The Greek financial crisis has also highlighted the fragility of the functioning of the euro area. This crisis has revealed inconsistencies and macroeconomic policies that characterize the area, so an exit scenario for the euro area of one or several countries is no longer regarded as fanciful. Failure Greek may trigger a panic both on other highly indebted European countries like Italy, Ireland and Portugal, and also for the developed countries of the area, triggering further deterioration of the agencies. This context has revived current events to the traditional questions about the nature and the degree of the dependence across markets of different asset types, such as the stock market and foreign exchange rates. The purpose of this paper is to investigate the dependence structure and to test the degree of the dependence between financial returns, after the occurrence of the current financial Greek crisis. For that, we use two measures of dependence: correlations and copula functions. More precisely, our main research questions are: 1) what is the dependence structure between European stock market and exchange rates? 2) Is the dependence symmetric or asymmetric? 3) Are the comovements of stock-currency markets affected by the happen of the Greek crisis? This present study contributes to the related literature in that we provide a framework for understanding the co-movements and describing the dependence structure between the financial markets, in the context of the recent financial Greek crisis. This is important since the knowledge of the nature of the dependence structure is very important in describing the reality of the behavior of financial markets. Indeed, it indicates the potential of simultaneous extreme events in both the stock and foreign exchange market. This property of dependence structure is important to international investment decisions in foreign stock markets. The remainder of the paper is organized as follows. In Section 2, we introduce the methodology used in the empirical application. In Section 3, we present the data and discuss our empirical results. Finally, section 4 summarizes our results and concludes. 2. Methodology Our methodology is based, primarily, on the calculation of linear and rank correlation coefficients between the European market returns. We get series of correlation coefficients between these markets and we study their dynamics changes. Secondly, such as measurements based on linear correlation may lead to 44 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW misspecification of the dependence structure with its nonlinear portion, copula approach is employed to provide the robust measures of dependences based on the entire joint distributions of variables and also to estimate dependence focuses on the entire structure rather than correlation. 2.1 Correlations: Correlations are the most familiar measures of dependence in finance. If properly specified, correlations tell us about average diversification opportunities over the entire distribution. 2.1.1 Traditionnel correlation : The Pearson correlation coefficient is the covariance divided by the product of the standard deviations: (2.1) Although most studies have focused on measuring the dependence between financial markets have used the Pearson correlation, this coefficient is only reliable when the random variables are jointly Gaussian. Therefore, we consider two other measures of dependence: the Kendall’s tau and the Spearman’s Rho, which are measures of concordance, generalize the linear correlation, taking into account the joint distribution (and not just marginal) and are dependent on copulas. 2.1.1 The rank correlation: The rate of Kendall and Spearman’s rho are two measures of concordance well known in statistics. They provide a measure of the correlation between the ranks of the observations, unlike the linear correlation coefficient which assesses the correlation between the values of observations. They also offer the advantage to be simply expressed in terms of the copula associated with the couple of random variables. 2.1.1.1 The Kendall correlation coefficient: Let (X, Y) a couple of random vectors and (X’, Y’) (a copy of (X, Y) that is to say a pair of vectors in all respects identical to (X, Y) the Kendall’s tau is then: (2.2) The Kendall’s tau is simply the difference between the probability of concordance and of discordance. 2.1.1.1 The Spearman correlation coefficient: Let X and Y are two random variables of marginal distributions Fx and Fg. VOLUME 7, NUMBER 4, 2013 The correlation coefficient Spearman rank coefficient between Fx (X) and Fg (Y): 45 is the Pearson correlation (2.3) The correlation coefficients of Spearman and Kendall have the property of being based on copula functions. Thus, if X and Y are two random variables with continuous marginal distributions and the copula C associated, their expressions in terms of copula is: (2.4) Like the linear correlation coefficient, the rank correlation coefficients reflect the extent to which high (respectively low) of a variable are associated with high (respectively low) for the other variable. However, the rank correlation has the advantage of being preserved under strictly increasing transformations. 2.2 A copula model for asymmetry dependence: Copulas are multivariate distribution functions with standard uniform marginal distributions. Am-dimensional copula is represented as follows: C (u) = C (u, …, um ) (2.5) Where u, …, um are standard uniform marginal distributions.In such a context, copulas can be used to link margins into a multivariate distribution function. The copula function extends the concept of multivariate distribution for random variables which are defined over [0,1]. This is possible due to the Sklar (1959) theorem which states that copulas may be constructed in conjunction with univariate distribution functions to build multivariate distribution functions. Sklar’s Theorem: Let FXY be a joint distribution function with margins and. Then there exists a copula C such that for all x, y in R, (2.6) If FX and Fy are continuous, then C is unique; otherwise, C is uniquely determined on Ran FX×Ran Fy and C is invariant under strictly increasing transformations of the random variables. Here we study five copulas with different dependence structure: the Gaussian copula, the Student-t copula, the Frank copula, the Clayton and the Gumbel copula. From them, the Gaussian copula is the most popular in finance and used as the benchmark. 46 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW • The Gaussian copula: The multivariate Gaussian copula applied to a joint distribution function with correlation matrix R, is defined by: (2.7) WhereCR is the distribution function of joint variables, these variables are normal, standardized and have a correlation matrix R. Since the majority of models in finance use this dependence structure, managers must adapt their model by modifying the structure of dependence if they consider the extreme risks. The use of copulas consistent with the extreme value theory is a modeling technique that allows analysis of rare events without being rigid methods based on the extension to several dimensions of a univariate distribution. This copula has no tail dependence and does not correlate extreme values. Modeling the dependence structure with a Gaussian copula is consistent with the extent of this dependence by the linear correlation coefficient. • The Student- t copula: The Student-t copula is defined by: (2.8) Where Tv,m,∑ is the multivariate student distribution function with a degree of freedom v and variance-covariance matrix ∑. • Archimedean copula: The Gaussian and Student copulas are called elliptic. They apply to the pattern of symmetric distributions. However, the Clayton, Gumbel and Frank copulas are called Archimedean copulas. They have the great advantage of being able to describe a variety of dependence structures including the asymmetric dependencies, where the coefficients of the lower tail and upper tail are different. We present as follow the characteristics of the best known models. The variables u and v are cumulative distribution functions. The parametermeasures the degree of dependence between risks. (2.9) (2.10) (2.11) VOLUME 7, NUMBER 4, 2013 47 3. Data and results: 3.1 Descriptive statistics: We use daily market data from three stock market indices, which are France (CAC40), Germany (DAX) and Greek (CYSE) and the foreign exchange rates for a sample period of February 1, 2007 to December 21, 2011. We choose this period to investigate the impact of the 2009 Greek crisis on the relationship between these European returns with the exchange rates. The total number of observations is 1253 for the full sample. We briefly overview summary statistics, then discuss the correlation and copula estimates. Table 1 Descriptive statistics of daily stock prices and foreign exchange rates Stock and FX returns Mean S.D Skewness Kurtosis Jarque-Bera CAC40 -0.021238 0.779427 0.139538 7.969773 1289.906 [0.000] DAX -0.005244 0.746523 0.118501 8.201018 1414.069 [0.000] CYSE 0.018518 0.974732 -0.021023 7.124780 885.2031 [0.000] EURO/USD 9.05E-05 0.316340 -0.192636 6.300257 577.9667 [0.000] The returns are in national currencies. The sample contains daily market returns from February 1, 2007 until December 21, 2011. The values in parenthesis are the probability values. The descriptive statistics for daily returns shown in table 1 suggest that all returns show excess kurtosis implying the properties of asymmetry, leptokurtosis, and tail dependence; hence, the normality assumption has been severely challenged. Furthermore, Jarque-Bera tests on log returns data indicate that the normality hypothesis cannot be accepted for these stocks. 3.2 Empirical results: 3.2.1. Correlation estimates of dependence: Table 2 presents linear correlations, the Kendall’s tau and the Spearman’s rho rank correlations between the stock and the exchange rate return pairs, before the financial Greek crisis. We observe that the pair wise correlations are positive for both France and Germany, indicating that the increase (decrease) of the local stock market is associated with the appreciation (depreciation) of the exchange rate EURO/USD. However, it is not the same case for Athens which has a negative relationship with the FX market. The Kendal’s Taus for our pairs of stock market returns and stock exchange rate are positive expect for Athens; showing the probability of concordance is significantly higher than the probability of discordance. The Spearman’s Rhos for the pairs in each country are also positive, with the exception for Athens. From these results, we can conclude that there are strong rank correlations. The German pair has the strongest dependence, followed by the French pair. 48 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Table 2 Correlation measures (2007-2009) Pairs Pearson correlation Kendall’s Tau Spearman’s Rho Pre-crisis French pair 0.217* 0.120* 0.179* German pair 0.241* 0.127* 0.191* Greek pair -0.038 -0.023 -0.029 This table gives different correlation measures for each stock-EUR/USD exchange rate daily return pair over the period February 1, 2007 to October 15, 2009. * denote significance level at the 1 %. Total observations are 691. In table 3, we present these linear correlations and rank correlations measures for each stock-exchange rate return pair after the current financial Greek crisis. The linear correlation, Pearson coefficients, for our pairs of returns are all positive, showing that, for these European markets, the increase (decrease) of the local stock market is associated with the appreciating (depreciating) of the exchange rate EURO/USD. The Kendall’s Taus for our pairs are all positive indicating that the probability of concordance is higher than the probability of discordance. The Spearman’s Rhos indicate strong rank correlations. The values of taus and Rhos are consistent with each other and the linear correlation. The French market has the strongest dependence with the EURO/USD exchange rate. Further, the correlation increase and became strong in the post-crisis period. Thus, the stockexchange rate returns become more dependent when financial extreme events (Greek crisis) occurs. Table 3 Correlation measures (2009-2011) Pairs Pearson correlation Kendall’s Tau Spearman’s Rho Pre-crisis French pair 0.399* 0.261* 0.389* German pair 0.366* 0.241* 0.355* Greek pair 0.356* 0.230* 0.345* This table gives different correlation measures for each stock-euro/usd exchange rate daily return pair over the period October 16, 2009 to December 21, 2011. Total observations are 562. *Indicates statistical significance at the 1% level. In order to see the dependence structure between European stock markets and the exchange rate market, we present below the empirical copula. VOLUME 7, NUMBER 4, 2013 49 Figure 1 The empirical copula for stock returns and exchange rate returns over the period February 1, 2007 to December 21, 2011. From the figure above, we can conclude that there exists symmetric upper and lower tail dependence. Thus, the dependence between the stock returns and exchange rate return is symmetric, implying both markets boom and crash together. As follows, we will run the test of adequacy on copula Gaussian (Normal), Student, Clayton, Frank and Gumbel to select the best copula which is more appropriate to the dependence structure shown in the empirical copula. Copula results: To better assess the degree as well as the dependence structure, we will examine the relationship between each pair of stock-FX returns separately, for the two sub period. Table 4.A bellow, reports parameters estimates of bivariate copulas for each pair, before the occurrence of the financial Greek crisis. We note that the parametermeasure the degree of dependence between returns and DoF is the degree of freedom in the Student-t copula. Table 4.A Estimation of copula parameters for the pre-crisis period: Parameters Pairs Copula models ρ France Student-t 0.187 German Student-t 0.199 Athens Student-t 0.210 DoF Information criteria q SIC AIC HQIC -32.08 -41.14 -37.65 6 -34.16 -43.22 -39.72 4 -47.92 -56.98 -53.48 5 50 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW As we see in the table above, the dependence between all pairs of financial returns used are described by the student-t copula. The dependence parameters (the correlation coefficient ρ) are all positive in the pre-crisis period. The correlation coefficient ρ from the Student-t copula is close to the usual correlation coefficient. The DoF of the Student-t copulas are from 4 to 6, indicating the presence of extreme co-movements and tail dependence. The tail dependence parameter q for pre-crisis period is zero. Thus, we can conclude that any pair has asymmetric tail dependence. The stock market returns have elliptical symmetric dependence structure (the case of the Student-t copulas) with the foreign exchange rate. In order to appreciate both, the dependence structure and the degree of this dependence, after the Greek crisis; we estimate the copula parameters in the post-crisis period. Table 4.B bellow, reports parameters estimates of bivariate copulas for each pair after the occurrence of the financial Greek crisis. Table 4.B Estimation of copula parameters for the post-crisis period: Parameters Information criteria Pairs Copula models ρ DoF q SIC AIC HQIC France Gumbel 1.36 -85.73 -94.37 -91.01 German Student-t 0.369 7 -73.82 -82.46 -79.10 Athens Gaussian 0.359 -69.99 -74.32 -72.63 For all pairs, the dependence parameters; the correlation coefficient ρ in both Gaussian and Student-t copulas, the degree of freedom DoF in the Student-t copula and the asymmetric dependence parameter q in the Gumbel copula are positive. The German return has the highest correlation coefficient with ρ = 0.369. The DoF of the Student-t copulas is 7 for German, indicating the presence of extreme co-movements and tail dependence. The tail dependence parameter q for post crisis period is 1.36 for the French pair. Moreover, the dependence structure between each stock index returns and exchange rate returns is largely changed from a symmetric structure with symmetric tail dependence to an asymmetric structure with non-zero and asymmetric upper tail dependence. From our results, we find The Gumbel copula which is limited to the description of a positive dependence structure. Thus, it allows only positive dependence structures or upper tail dependence, for which the parameter belongs to the interval [1,+∞). Consequently, the degree of the dependence varies when the financial Greek crisis occurs. Indeed, as we see in tables 4(A; B) above, it increased after the crisis. VOLUME 7, NUMBER 4, 2013 51 Despites France and Germany are two developed countries belonging to the Economic and Monetary Union, their dependence after the Greek crisis was different. The dependence of the French pair becomes asymmetric with tail dependence. The significance of the tail dependence implies that the stock market and exchange rate tend to experience concurrent extreme shocks. However, the Germany pair remains symmetric which approves that German stock market and currency market boom and crash together. Thus, this finding improves our understanding of the market dependence. 4. Conclusion This paper examines the dynamics relationship between foreign exchange rates, Athens and the two developed stock markets namely German and French, after the occurrence of the Greek crisis, using daily data from February 2007 to December 2011. We employ five multivariate copulas; the multivariate normal, multivariate Student-t, multivariate Gumbel, multivariate Clayton and the multivariate Frank to directly model the underlying dependence structure. We find that, during the pre-crisis period, the stock-foreign exchange market returns have elliptical symmetric dependence structure. However, the degree of the dependence becomes stronger when the financial Greek crisis occurs. Indeed, the dependence structure for the French pair becomes asymmetric with upper tail dependence after the crisis. For both the German and Athens stock-FX, the dependence remains symmetric however the degree of freedom DoF become higher. Our findings may have important implications in the risk management. First, symmetric dependence structure with zero tail dependence can specify different levels of correlation between the marginals; however, it must possess radial symmetry which doesn’t allow to extreme values correlation. Thus, in this case, the dependence has the linear correlation coefficient as measure of dependence. Second, asymmetric dependence structure can have upper tail dependence, lower tail dependence, or both; as such, they can better describe the reality of the behavior of financial markets. Additionally, it indicates the potential of simultaneous extreme events in both the stock and foreign exchange market. This property of dependence structure is important to international investors who invest in foreign stock markets. References Aloui C., 2007. Price and volatility spillovers between exchange rates and stock indexes for the pre- and post-euro period. Quantitative Finance 7, 1–17. Branson, W.H., Henderson, D.W., 1985. The specification and influence of assets markets. In: Jones, R.W., Kenen, P.B. (Eds.), Handbook of International Economics, 2. Amsterdam, Elsevier. Diamandis, P.F., Drakos, A.A., 2011. Financial liberalization, exchange rates and stock prices: exogenous shocks in four Latin America countries. Journal of Policy Modeling 33, 381–394. 52 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Dumas, B., Solnik, B., 1995. The world price of foreign exchange risk. Journal of Finance 50, 445–477. Frankel, J.A., (1983). Monetary and portfolio-balance models of exchange rate determination. Bhandari, J., Putnam, S. (Eds.), Economic Interdependence and Flexible Exchange Rates. MIT Press, Cambridge, MA. Jorion, P., 1990. The exchange rate exposure of the U.S. multinationals. Journal of Business 63, 331–345. Jorion, P., 1991. The pricing of exchange rate risk in the stock market. Journal of Financial and Quantitative Analysis 26, 363–376. Ning, C., 2010. Dependence structure between the equity market and the foreign market - a copula approach. Journal of International Money and Finance 29, 743–759. Phylaktis, K., Ravazzolo, F., 2005. Stock prices and exchange rate dynamics. Journal of International Money and Finance 24, 1031–1053. Roll, R., 1992. Industrial structure and the comparative behavior on international stock market indices. Journal of Finance 47, 3–41. Yang, S.Y., Doong, S.C., 2004. Price and volatility spillovers between stock prices and exchange rates: empirical evidence from the G-7 countries. International Journal of Business and Economics 3, 139–153. Zaho, H., 2010. Dynamic relationship between exchange rate and stock price: Evidence from China. Research in International Business and Finance 24, 105112. Financial development and income inequality in the mena region Zied Saadaoui*, University of Sfax Maher Gassab†, University of Manouba Abstract We study a sample of eleven Middle East and North Africa (MENA) countries, observed between 1963 and 2002, to test the impact of financial development on income inequality. Estimation results using system GMM, show that the development of deposit and savings instruments that the financial systems of MENA countries participate in reducing income inequality, by stimulating the demand for real money balances. The results also show that increasing the share of credits granted to government is likely to complicate the access of poor households to credit. Finally, financial instability decreases, rather than increases, income inequality between rich and poor households. Keywords: MENA countries, Income inequality, Financial development, Financial instability, System GMM. JEL Classification: O15, O16 1. Introduction Since the 1960’s, the majority of the Middle East and North Africa countries5 have adopted new policies to develop their financial sector. In the first stage of development, governments were highly involved in these policies until mid 1980, with the liberalization of their financial systems, after the Washington consensus. But, while several studies focused on the causality between financial development and economic growth, especially since the release of the World Bank’s report in 1989, few works were dedicated to the relationship between financial development and income inequality. Moreover, in explaining income inequality, the literature often considered financial development as exogenous, assuming that financial systems remain unchanged with time (Demirgüç-Kunt and Levine, 2009). Academic researches focused mainly on the impact of fertility, education and redistributive policies on inequality. However, financial systems are in continuous mutation, affecting the decisions of economic agents and their incomes, by altering human and physical capital, public politics, etc. Researches in this area are too recent and their number increased since the World Bank published its report in 2007 on the reduction of poverty through “financial inclusion”. * Zied Saadaoui is at University of Sfax, Tunisia. Address : Route de l’Aéroport Km 4 Sfax 3018 Tunisie (casier # 17). Corresponding author : [email protected] / [email protected] † Maher Gassab is at University of Manouba, Laboratoire de Recherche en Microéconomie Appliquée (LARMA) 5 Henceforth called MENA countries. 54 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW The recent international crisis revived the debate over economic and financial policies that developing countries should apply in order to deal with income inequality and to reduce vulnerability of poor people to cyclical downturns. Although MENA countries were relatively protected from financial crises, thanks to their weak international financial integration, growing globalization of financial services will certainly exerts positive impacts on their economies but with growing risk of financial crises that may aggravate poverty rate. In this context, it is interesting to focus on the effects that financial development may exert on income inequality in the MENA countries. The understanding of such effects is crucial to make financial systems of these countries work more efficiently in favour of the United Nation’s millennium development goals. MENA countries don’t suffer from high poverty or inequality rates, because many of them profit from important oil rents and are classified among upper middle income countries. But economic indicators are not always good in this region of the world, due to the fluctuation of oil prices. According to the United Nation Development Program’s report published in 2009 and dedicated to the Arab countries, this region grew very slowly since 1980, with an annual growth rate less than 0,5% according to the World Bank’s data. The same report indicates that 40% of the population in this region was below the poverty threshold and that inequality was increasing over time. The purpose of this paper is to explore the determinants of income inequality in the MENA region, by focusing mainly on the impact exerted by financial development and instability. The paper is organized as follow: Section 2 offers a description on how financial development and instability can influence income inequality. Section 3 presents the principal empirical works dealing with this field of research, while the fourth section focus on the case of the MENA countries by studying a sample of 11 countries observed from 1963 to 2002. Finally, section 5 concludes this paper. 2. Financial development, instability and income inequality 2.1. Financial development and the conduit effect Financial development tends to reduce income inequality through two principal effects: a direct effect, by ameliorating access to financial services and an indirect effect, by stimulating economic growth. Through the direct effect financial development can influence the situation of the poor and the distribution of income. To better explain these arguments, let’s consider the opposite case, i.e. the underdevelopment of financial system which is a major problem for many developing countries. In this case, poor households will be excluded from financial services, simply because they can’t pay the cost of these services (Beck et al., (2006, 2007); World Bank, (2008)). Market imperfections, like moral hazard and adverse selection, are likely to complicate access to credit for poor people who don’t possess required collaterals, in contrast to the rich people who possess sufficient assets to pledge (Greenwood and Jovanovic, (1990) ; Claessen and Perotti, (2007); Beck et al., (2009)). This situation is likely to deepen income inequality. VOLUME 7, NUMBER 4, 2013 55 Besides, during the two last decades, several studies tried to empirically demonstrate the existence of a positive relationship between financial development and economic growth (King and Levine, (1993); Arestis and Caner, (2004); Levine, (2004)). In parallel, other empirical approaches tried to test the impact of economic growth on income inequality and on poverty, like Dollar and Kray (2002). These authors tested the relationship between per capita income and the income share of the poorest quintile, using a sample of 80 countries observed during four decades, they find a positive and significant relationship between the two measures. According to these studies it is possible to hypothesise that financial development can be a robust determinant of income inequality, through his effect on economic growth, as demonstrated empirically by Beck et al. (2007), Clark et al. (2003) and also Guillaumont Jeanneney and Kpodar (2008). Beck et al. (2007) make the assumption that financial development, by impacting economic growth, may have two opposite effects on the poor. On the one hand, by reducing credit market imperfections, by enhancing transparency or lowering transaction costs, and by allocating resources more efficiently, financial development facilitate access to credit for poor households who cannot pledge personal collateral. This permits also to lower inequalities in the society given that poor borrowers wishing to start their own business will not be deprived from credits. But in the other hand, financial development can profit mainly to rich people, since they are more likely to own collaterable assets in the first stage of economic development, particularly when credit constraints are not reduced (Greenwood and Jovanovic, (1990)). This situation may worsen if the rich minority have the capacity to lobby for economic policies in order to keep safe its wealth, which can be detrimental to the distribution of income and to economic growth (Li et al., (1998)). Moreover, financial liberalization may help the poor to have better access to financial services, by encouraging competition between banks, by creating new sources of capital and by cutting borrowing costs. But capital account opening may also profit to rich minority given their political relationships that enable them to capture foreign capital allocations (Li et al., (1998); Arestis and Caner, (2004)). However, there may be a problem of endegeneity between finance and income inequality. Indeed, financial development can be influenced by the level of poverty. For example, the reduction of poverty may stimulate the demand of credit and enhance the development process of financial services. Then, the reduction of inequality may constraint policy makers to make financial markets more transparent and more efficient, by preventing political relationships and favouritism in financial transactions (Beck et al., (2009)). We now turn to another hypothesis that may explain how financial services can be a solution to persistence of income inequality. According to McKinnon (1973), in developing countries where credits are often not accessible to poor people, investment is often confined to self-financing. In this case, deposit and saving instruments become an attractive source of self-financing and the demand for real money balances become positively related to real interest rates on savings and deposits. This makes capital and money complementary rather than substitutable. Consequently, financial development can stimulate demand for money and quasi-money balances which play the role of a supplementary financial conduit for capital accumulation and for economic growth (Kpodar, 2006 ; Guillaumont Jeanneney and Kpodar, 2008). This conduit effect will turn to be inefficient 56 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW in case of financial repression, i.e. interest rate restrictions, higher legal reserves requirements, political interference, etc., which restrain investment opportunities, especially for poor households. 2.2. Financial instability and banking inefficiency Despite its negative effect on poverty and inequalities, financial development may be the resultant of economic instability with adverse effects on household’s income. MENA countries have not experienced extreme episodes of financial and economic recession these last years. But, if we suppose that there is a strong linkage between finance and income inequality, it will become interesting to test if credit and liquidity turbulence, even minimal ones, hurting the financial and the payment system, can increase income inequalities. In fact, during recession periods, several problems can hurt the payment system and lead to an accumulation of non-performing loans or even to a deposit freezing. In these circumstances, it is the households with limited budgets that will be the most affected because they are unable to diversify their assets. Furthermore, financial instability can occur through the accumulation over time of non-performing loans, which is the Achilles’ heel of supervisors in developing countries. Credit losses are a sign of inefficiency in banking which, in the presence of market imperfections, can aggravate the problem of adverse selection and credit crunch6 (Krueger and Tornell, (1999) ; Berger and DeYoung, (1997) ; Clair and Tucker, (1993)). This is likely to complicate the access to financing for the poor and to prevent them to exploit business opportunities, especially during periods of recession. Furthermore, as demonstrated by Hauner (2008), government involvement in banking activities and the directing of credits towards non-productive political projects, can explain the high level of loans losses suffered by developing countries’ banks, including those of the MENA region (Caprio and Honohan, 1999 ; Caprio and Martinez-Peria, 2000 ; International Monetary Fund, 2010). In other words, banking sector’s health explains its dynamism, and the more this sector is dynamic the more the economy benefits from financial openness and financial development. Thus, it would be interesting to estimate the impact of financial instability and banking inefficiency on income inequality. On the empirical side, there are few studies dealing with financial development effects on poverty and income inequality. In order to construct credible hypotheses, these studies relied mostly on earlier works examining the relationship between financial development and economic growth. In the following section we try to give a succinct survey on these studies. 3. Review of empirical studies beck et al. (2007) investigate the impact of financial development on income distribution, measured by the Gini coefficient and by absolute poverty rate. According to the authors, financial system development must be accompanied by a 6 The credit market is not in a situation where the distribution of credits obeys to the law of supply and demand (the equilibrium interest rate) but rather in a situation where the supply of credits is less than the demand (credit rationing). VOLUME 7, NUMBER 4, 2013 57 more efficient capital allocation and by a reduction of market imperfections. The study focuses on a sample of 72 countries observed between 1960 and 2005. To take into account the endogeneity problem, the authors proceed to and estimation of a dynamic panel using the generalized method of moments (GMM). The results show a significant negative relationship between the indicator of financial development and growth of the Gini coefficient. The results also show the existence of a significant negative relationship between financial development and the income share of the poorest quintile7. By studying the case of India during the period going from 1951 to 2004, Ang (2008) ask whether the financial development and financial liberalization had a significant impact on the poor’s situation in this country. Financial development is measured by the ratio of credit to the private sector, by the ratio of quasimoney (M3 - M1) to GDP and by the ratio of commercial bank assets to commercial bank plus central bank assets. While financial liberalization is measured by three indicators: interest rate restriction, requirements of cash reserves and directed credit programs. The estimation of an error correction model shows that there is a long-term positive relationship between financial development and the Gini coefficient. In addition, per capita GDP growth reduces inequality while financial openness produces the opposite effect. Ang (2008) incorporates also a measure of bank density (number of branches per population) to measure financial development, not in terms of depth, but in terms of wideness. He finds a significant negative relationship between this indicator and Gini coefficient growth. Li et al. (1998) analyze the determinants of inequality for a sample of 49 developed and developing countries from 1947 to 1994. They regress the Gini coefficient (averaged over 5 years) on a set of financial indicators: financial system liquidity, M2/GDP and a set of control variables. Using the OLS and the instrumental variables to estimate the model, the authors conclude that financial development exerts a negative impact on inequality and this impact is the largest among those exerted by the selected variables. Following the theoretical work of Kuznets (1955) and Greenwood and Jovanovic (1990), Clark et al. (2003) test the hypothesis of non-linearity of the relationship between financial development and inequality. The sample includes 44 developed and developing countries observed during the period from 1960 to 1995. The authors regress the natural logarithm of the Gini coefficient on the ratio of credit to private sector to GDP and on the value added in industry and services sectors over GDP to measure the degree of modernization of the economy. An interaction term between this variable and the indicator of financial development is also introduced into the model to test the hypothesis of Kuznets (1955), namely that financial development of the economies based on modern productive sectors, i.e. excluding agriculture, reinforces the unequal distribution of income. Other control variables were also introduced in the model specification estimated by the ordinary least squares (OLS) and GMM techniques. The results show that there is a negative linear relationship between financial de7 The authors decompose the effect of financial development on income growth in two distinct effects: a growth effect and a distribution effect. They found that 40% of financial development effect is a distribution effect, while 60% of this effect is a direct effect on income growth. 58 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW velopment and inequality. In contrast, the model estimation rejects the hypothesis of a nonlinear relationship between financial development and inequality. Estimations results show, however, that there is a positive and significant effect exerted by the interaction term on inequality, confirming the existence of a nonlinear relationship derived from the hypothesis of Kuznets (1955) Arestis and Caner (2009) focus on whether capital account openness had a significant impact on poverty. They study a panel of 64 developing countries, observed annually from 1985 to 2005. The authors regress on poverty8 on an index of capital account openness (KAOPEN), developed by Chinn and Ito (2002)9. Estimation of a dynamic panel model, using the system GMM technique, shows that capital account liberalization paradoxically deteriorates the situation of the poor. Indeed, capital account openness is negatively related to the income share of the poor and positively related to poverty gap. Guillaumont Jeanneney and Kpodar (2008) identify two channels through which financial development can affect poverty: the conduit effect and financial instability (cf. supra). They regress poverty indicators on two indicators of financial development: M3/GDP, measuring the supply of savings and deposit instruments in the economy and the ratio of credit to the private sector. They introduce also a variable of financial instability measured by the volatility of the two forementioned indicators of financial development. The sample includes 75 developing countries observed between 1966 and 1999. Estimation of the Model by OLS and system GMM techniques showed a significant negative relationship between M3/GDP and the different measures of poverty. Financial instability, measured by M3/GDP, tends to aggravate the situation of the poor and to increase poverty. In contrast, credit to the private sector is not significantly related to poverty indicators. In the next section, the main hypotheses of this study will be exposed. These hypotheses were inspired from the different theoretical and empirical studies presented throughout this paper. We also present the selected variables, the estimation technique and the selected sample. 4. Empirical study 4.1 Hypotheses, variables and model 4.1.1 Hypotheses There are four hypotheses to test: • Financial development and better access of households to financial services reduces income inequality through the stimulation of economic growth and the creation of new sources of income for the poor, like entrepreneurship. 8 Three measures of poverty have been selected: the index of poverty, measured by the number of households living below the $ 2.15 per day threshold (PPP $ of 1993), the poverty gap index, which measures the intensity of poverty, i.e. distance from the poverty line, and finally, the income share of the poorest quintile. 9 This index is the first principal component of four binary variables selected from the IMF Annual Report on foreign exchange restrictions (Annual Report on Exchange Arrangements and Exchange Restrictions). VOLUME 7, NUMBER 4, 2013 59 • In cases where access to credit is difficult for poor households, financial development induces banks to offer more profitable financial opportunities for savings, which under the hypothesis of the conduit effect, can stimulate accumulation of monetary deposits and constitute an alternative solution for low-income households to invest and get rich. • Financial development may be accompanied by episodes of instability which may worsen the situation of poor households and aggravate income inequality. • Increasing the proportion of bank loans granted to the government, leads to the accumulation of non-performing loans and to more inefficient allocation, which may complicates the access of poor households to credit, especially in times of crisis. The second and the third hypothesis, namely the effect of financial system liquidity and financial instability on income inequality have been inspired by the study by Guillaumont Jeanneney and Kpodar (2008). The main contribution of this study lies in the fourth hypothesis, which we consider as more appropriate to the context of MENA countries, whose banking and financial systems have not experienced remarkable episodes of instability. Therefore, it is useful to assess the fragility of these systems through their relationship with their government which could impede their efficiency and depth. 4.1.2 Income inequality indicator The indicator of income inequality used in this study is the natural logarithm of the Gini coefficient (GINI), observed annually from 1963 to 2002. This coefficient, derived from the Lorenz curve, measures inequality of income distribution within a country. The larger its value, the higher inequality is. To have a sufficient number of observations, we used the Estimated Household Income Inequality (EHII) database developed under the University of Texas Inequality Project. This database measures inequality using the average income by group of population, eg. by industry, sector or region. These semi-aggregated data may contain sufficient information on the evolution of inequality and income distribution. The advantage of this method, based on the Theil index, is that it overcomes two important obstacles, in contrast to the database developed by Deninguer and Squires (1996): data availability and imprecision of measures of income. Wage inequality, derived from the Theil index, is converted in order to adequately approximate inequality of household’s incomes, resulting in a dense and consistent data set of income inequality measures in Gini format (Galbraith , 2008). 4.1.3 Financial development indicators The measures of financial development should reflect the consolidation of the financial system over time and the reduction of market imperfections, which make finance more beneficial for the poor. Financial development variables used in this study are extracted from the International Financial Statistics database (IFS CD-ROM, (2007)). Following previous studies, we use the indicator of credit to the private sector on GDP (CSP) to estimate the improvement of households and private sector access to financial services offered by banking 60 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW and non-banking financial institutions in the MENA region. This indicator is a more accurate measure of market imperfections and of development of payment instruments. Indeed, CSP does not take into account credits granted by the central bank and development banks, credits to the public administration and interbank loans. This indicator is considered more efficient than the ratio of commercial bank assets to commercial bank plus central bank assets, proposed by King and Levine (1993), for two reasons: first, because the mission of the central bank can change from one country to another and can concentrate on directing bank loans toward political development goals and toward privileged sectors and firms; Then, because commercial banks are not the unique financial actor able to grant loans, other non-bank financial intermediaries may also offer this service (Beck et al. 2007). In addition, the indicator of credit to private sector is often found to be positively and significantly related to economic growth (Clark et al., 2003). We also introduce a second indicator of financial development (M2), whose interest is to measure the development of savings instruments and demand for real money balances. This indicator is measured by the ratio of M2 (cash items, demand and savings deposits, etc.) on GDP. Using the M3 indicator would be more accurate because it takes into account also money market instruments (mutual funds, certificates of deposit, commercial papers, etc.), But the database used in this study (International Financial Statistics) does not provide these data for the MENA region. However, M2 indicator remains quite relevant in estimating the financial conduit effect and its impact on income inequality. 4.1.4 Financial instability indicators To measure financial instability, we follow Guillaumont Jeanneney and Kpodar (2008) which calculate the volatility of each financial development indicator (CSP and M2) by assuming that these variables do not follow a stochastic trend. Thus, we regress a variable of financial development, denoted x, on its lagged value and a time trend (t). If we denote Vx the measure of financial instability, then: (1) Where εt is the error term of the following equation estimated by OLS: (2) Thus, the absolute value of the residuals measure the volatility of the variable x, assuming that this variable follows a time trend (t), contrary to other measures of volatility such as the standard deviation. Financial instability indicators, VCSP (for CSP) and VM2 (for M2), are obtained following two steps: first, we estimate equation (2) for each selected country, in order to obtain the residual series for each country. Then, at each point of income inequality data we associate a five-year average of the residuals absolute value (the year of observation plus the four precedent years ), i.e. we suppose that n = 5 in equation (1). VOLUME 7, NUMBER 4, 2013 61 4.1.5 Banking inefficiency As noted by Honohan (2004a), macroeconomic indicators, like CSP and M2, may not provide accurate estimates, as these indicators do not integrate all the elements that would promote financial development in a country, like the legal and judicial system efficiency, information transparency, regulatory mechanisms, etc. In China, for example, the indicators CSP and M2 can not reflect, with a high confidence level, the depth of the financial system, given the strong government involvement in the banking system (Honohan, 2004b)10. Consequently, we propose to add a measure of loans granted to the government relative to total loans granted by the banking system (GOV)11. We make the assumption that in economies where the government capture a significant share of bank loans, the financial system would suffer underdevelopment and liquidity problems. This result was empirically demonstrated by Hauner (2008) on a sample of 142 developing countries’ banks. Results show that the increase of credit to the government has a negative and significant impact on liquidity of the banking system, i.e. the depth of the financial system and its productive efficiency. This problem is due to a misallocation of these credits, often directed to finance unproductive projects (Boyreau-Debrey and Wei, 2005). The dominance of credit to government may also involve higher political interference, often associated with a lower efficiency of the banking sector, and thus to an underdevelopment of the financial system that could hamper economic growth and increase inequality (King and Levine, 1993; Claessens and Perotti, 2007). 4.1.6 Control variables The model also takes into account a set of control variables obtained from the World Development Indicators database (WDI CD-ROM, 2007). This vector of variables includes the natural logarithm of per capita GDP in constant 2000 U.S. dollars (GDP), inflation measured by the natural logarithm of the consumer prices index (base 100 in 2000), denoted CPI; trade openness (TRADE), measured by the sum of exports and imports of goods and services divided by GDP; and government consumption (GOVEXP), measured by the ratio of government consumption expenditure as a percentage of GDP. The evolution of per capita GDP is supposed to reflect the growth of national income. It is expected that the higher the GDP per capita, the higher is the income of poor households and the lower is income inequality. It is also assumed that the poor and middle class population is more vulnerable to inflation than the class of wealthy households who benefit from easier access to finance that allow them to smooth their consumption through time. So, it is expected that inflation is positively related to income inequality. We also expect that trade integration is associated with an increase in inequality (Barro, 2000). However, the relationship between government consumption and the Gini coefficient is less obvious, since it can be positive or negative. It depends on the government redistribution policies (the tax system and social transfer). If State interventions are pro-poor, in this case this variable will be negatively related to income inequality. However, if rich minority exerts its lobbying power to profit from redistributive policies in order to 10 In 2004 the four biggest state banks in China owned more than 50% of banking assets (Jia, 2009). 11 Loans granted by commercial banks and other credit institutions, such as savings banks, development banks, saving and loans associations, offshore banks, etc. 62 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW get richer, then the relationship between government consumption and the Gini coefficient may be positive. 4.2. Econometric methodology The model specifications adopted in this study are as follows: (3) (4) The subscripts i and t refer, respectively, to the countries and the year of observation. The error terms ε and μ are assumed to follow a random walk. This model admits that the lagged values of the dependent variable affect current values. We use the dynamic panel Generalized Method of Moments estimator (system GMM) developed by Arellano and Bover (1995) and Blundell and Bond (1998), used also by Guillaumont Jeanneney Kpodar (2008) and Arestis and Caner (2009). Compared to the OLS, system GMM is more efficient to control for the endogeneity between financial development and inequality, and between the dependent variable and the other explanatory variables (Beck et al. 2007, 2009). Panel techniques are advantageous, since they consider both the individual dimension and the temporal dimension of data. Moreover, the dependent variable appears to the right of equations (3) and (4), which may result in a correlation between specific individual effects and explanatory variables. To get more efficient estimates, it is possible to estimate the model using first differences to remove the countries’ unobserved specific effects and to instrument explanatory variables using their lagged values. This method is called the first-differenced GMM proposed by Arellano and Bond (1991). System GMM technique consists of a combination between a set of first-differenced equations, whose instruments are lagged levels with a second set of equations in levels using lagged first differences as instruments. According to Blundell and Bond (1998), this methodology provides more efficient estimators than first-differenced GMM because, even if the variables are very persistent, the instruments used in the level equation adequately predict the endogenous variables. In addition, Monte Carlo simulations give evidence that the system GMM method is more efficient than the firstdifference GMM, when using a small sample size, which is the case in this study. Finally, to test the validity of the model, we use the Sargan test of over-identifying restrictions to check the validity of instruments and the Arellano and Bond serial correlation test to verify if errors exhibit second order serial correlation. 4.3. Selected sample The study focuses on a sample of eleven MENA countries, observed annually between 1963 and 2002. Sample selection was mainly constrained by data availability. Table 1 shows the number of times the dependent variable Gini was observed for each selected country. VOLUME 7, NUMBER 4, 2013 63 Table 1 Selected sample Country Period Observations Morocco 1967-2000 26 Algeria 1967-1997 28 Tunisia 1963-2001 28 Libya 1964-1980 17 Egypt 1964-2002 36 Syria 1963-1998 36 Iran 1963-2000 38 Jordan 1963-2002 37 Kuwait 1967-2001 35 Oman 1993-2002 10 Qatar 1986-2002 11 Total observations 302 4.4. Empirical results 4.4.1. Descriptive statistics Table 2 shows that for the overall sample, GINI has averaged 44.7. But this variable is quite volatile with a standard deviation of 6.04. The average annual per capita GDP for the whole MENA region amounted to almost U.S.$ 1980, evolving from a minimum of U.S.$ 482.63 and a maximum of U.S.$ 44018.7, with a standard deviation approximately equal to 1.06.The CPI index is measured by setting the year 2000 as base year, which may explain its moderate value (expressed in logarithm) and the low values observed especially at the beginning of the observation period. In fact, inflation changed considerably over the past three decades in MENA countries, which explains its high volatility throughout the observation period, with a standard deviation of 2.13. The variable TRADE amounts to 67.7% on average for the entire sample, with a maximum value of 154.6%. GOVEXP variable, measuring the share of government spending in GDP, is characterized by relatively low volatility (standard deviation equal to 0.089) and amounting approximately to 21%. Financial development indicators are relatively stable, characterized by a standard deviation of 0.24 and 0.23 for CSP and M2 respectively. Credits to the private sector (CSP) are on average 35% of GDP for the whole sample, while the liquidity of the banking system (M2) reached nearly 55% of GDP. The variable GOV, measuring the share of lending to the government, averaged 16.5% and is characterized by a relatively stable variation, with a standard deviation of 0.15. 64 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Table 2 Descriptive statistics – Overall sample Variable Observations Mean SD Min. Max. GINI 302 44.69 6.04 27.21 58.39 GDP per capita 309 1979.36 1.06 498.63 44018.70 M2 (%) 333 0.55 0.24 0.14 1.92 CSP (%) 325 0.35 0.21 0.04 1.04 CPI (log) 301 2.63 2.13 -3.61 4.65 TRADE (%) 334 0.67 0.27 0.13 1.54 GOV (%) 323 0.16 0.15 0 0.89 GOVEXP (%) 334 0.21 0.09 0.07 0.76 Study period extends from 1963 to 2002. The observed sample includes eleven MENA countries. SD = Standard deviation. GINI = Gini Coefficient. Annual per capita GDP is expressed in constant dollars of 2000. M2 = (Money + quasi-money)/GDP. CSP = Credits to the private sector/GDP. CPI = Consumer Price Index (base year : 2000). TRADE = (Importations + exportations)/GDP. GOV = Credits to gouvernement/Total banking loans. GOVEXP = Governement consumption/GDP. The comparison of statistical properties of the selected variables between Arabian Peninsula countries (Kuwait, Qatar and Oman) and the other selected MENA countries is useful given the structural differences between the economies of these two groups of countries. According to the tables 3 and 4, we note, first, the significant difference in the average GINI between the two groups of countries, which explains in part the high volatility of this variable for the overall sample. Table 3 Descriptive statistics – Arabian Peninsula countries Variable Observations Mean SD Min. Max. GINI 56 52.18 2.61 48.42 58.39 GDP per capita 52 16261.99 0.48 7541.12 44018.70 M2 (%) 61 0.58 0.28 0.18 1.92 CSP (%) 61 0.42 0.23 0.09 1.04 CPI (log) 57 4.36 0.25 3.67 4.62 TRADE (%) 62 0.91 0.12 0.65 1.42 GOV (%) 54 0.28 0.27 0 0.89 GOVEXP (%) 62 0.25 0.11 0.07 0.76 Study period extends from 1963 to 2002. The observed sample includes Kuwait, Qatar and Oman. SD = Standard deviation. GINI = Gini Coefficient. Annual per capita GDP is expressed in constant dollars of 2000. M2 = (Money + quasi-money)/GDP. CSP = Credits to the private sector/GDP. CPI = Consumer Price Index (base year : 2000). TRADE = (Importations + exportations)/GDP. GOV = Credits to gouvernement/Total banking loans. GOVEXP = Governement consumption/GDP. VOLUME 7, NUMBER 4, 2013 65 Table 4 Descriptive statistics – MENA countries (excluding Arabian Peninsula) Variable Observations Mean St D Min. Max. GINI 246 42.99 5.25 27.21 55.34 GDP per capita 257 1292.58 0.49 489.62 6075.13 M2 (%) 272 0.54 0.23 0.14 1.37 CSP (%) 264 0.33 0.20 0.03 0.77 CPI (log) 244 2.23 2.18 -3.61 4.65 TRADE (%) 272 0.62 0.27 0.13 1.54 GOV (%) 269 0.14 0.09 0 0.58 GOVEXP (%) 272 0.19 0.07 0.10 0.50 Study period extends from 1963 to 2002. The observed sample includes Morocco, Algeria, Tunisia, Libya, Egypt, Jordan, Syria and Iran. SD = Standard deviation. GINI = GINI Coefficient. Annual per capita GDP is expressed in constant dollars of 2000. M2 = (Money + quasi-money)/GDP. CSP = Credits to the private sector/GDP. CPI = Consumer Price Index (base year : 2000). TRADE = (Importations + exportations)/GDP. GOV = Credits to gouvernement/Total banking loans. GOVEXP = Governement consumption/GDP. For the Arabian Peninsula countries the Gini coefficient reached an average of 52.18, much higher than that observed for the rest of the MENA countries (43 approximately). For this group of countries, the volatility of GINI is still high with a standard deviation of 5.25. The large difference in the level of GINI recorded between the two groups of countries can be explained by the calculation method of this indicator by Galbraith’s team (UTIP), based on wages distribution and not on income, unlike the methodology used by the World Bank (Deininger and Squire, 1996). On the other hand, per capita GDP varies considerably between the two groups of countries, with an average of U.S. $ 16262 in the Arabian Peninsula countries and U.S. $ 1292.5 in the remaining MENA countries. This result is not surprising given that the first group of countries are oil producers and exporters and are much less populated than the first group. It appears also that Inflation, in the Arabian Peninsula countries, is almost two times higher than in the other selected countries. Arabian Peninsula countries seem to be more integrated economically, with an average TRADE ratio equal to 91% against 62% for the other MENA countries. Public spending recorded a higher average in the Arabian Peninsula countries, with a ratio GOVEXP evolving around an average of 25%, whereas this ratio is only 19% in the rest of the MENA countries. In terms of financial development, the statistics do not show a significant difference in the level of M2, which is respectively 58% and 54% for the Arabian Peninsula countries and the remaining MENA countries. In this group of countries, however, the average level of credit to the private sector, amounting to 32% of GDP, is less than that recorded in the Arabian Peninsula countries, which amount to 42%. The share of loans to finance public administration in this latter group of countries stands at 28% of GDP and is on average two times higher than in the other selected countries (14%). 66 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW 4.4.2. Estimation results Table 5 shows the system GMM estimation obtained by using the “xtabond2” command in STATA 9.2. P-values for over-identification and serial correlation tests are quite high, which does not reject the null hypotheses of no correlation between the instrumental variables and error terms (Sargan test) and of no second order correlation between errors (Arellano and Bond test). Columns (1) and (2) present the estimation results of equations (3) and (4) respectively. Column (1) shows that per capita GDP is positively and significantly related to GINI. Although contrary to our expectations, this result can be interpreted in two ways. First, it is possible that economic growth is more beneficial to the wealthy class of the population, than to the poor. Then, according to Clark and al. (2003), who found a similar result using the initial per capita GDP, this positive relationship can be explained through the theoretical arguments of Kuznets (1955), namely that it is possible to find a non-linear (concave) relationship between income inequality and GDP. More specifically, to a certain level of economic development, agriculture, which is a less productive but more egalitarian productive sector, will be gradually dominated by modern, but less egalitarian, sectors, like services and industry sector. So that at the early stage of economic development, which is the case of MENA countries, inequality is likely to increase. Inflation has a positive but non-significant effect on GINI. This result was already found by Clark and al. (2003) and Guillaumont-Jeanneney and Kpodar (2008). Moreover, trade liberalization in MENA countries is positively and significantly related to the Gini coefficient, which is in accordance with our expectations. An increase in TRADE by 1 percentage point (pp) causes an increase in GINI by 0.08465 pp at the 0.01 level. This result therefore confirms the allegations of Barro (2000), namely that economic integration of countries induce a worsening of income inequality. The parameter related to the variable GOVEXP is negative and significant at 95 percent confidence level. An increase in government consumption expenditure by 1 pp reduces inequality by 0.18671 pp. This confirms our initial hypothesis that the redistribution policies adopted by the MENA countries governments are more oriented toward limited-income households. Let’s turn now to the main hypotheses of this study. Column (1) shows a significant relationship between credits to the private sector and the Gini coefficient. Contrary to the results found by Beck and al. (2007) and Clark and al. (2003), the measure of financial development through access to credit does not appear to influence the situation of poor households and income inequality in MENA countries. This result is against our expectations of a negative relationship between CSP and GINI, that can be induced either directly through the easing of credit constraints and the elimination of market imperfections, or through stimulating economic growth. Furthermore, a surprising result appears in column (1). In fact, VCSP is negatively related to the dependent variable GINI, suggesting that the instability of credit availability reduce rather than aggravate income inequality. In their study, Kpodar Jeanneney and Guillaumont (2008) found that credit volatility does not enter significantly. However, the negative relationship between financial instability and inequality can be explained by the differences between countries in terms of redistribution policy. VOLUME 7, NUMBER 4, 2013 67 Table 5 Estimation results Dependent variable: GINI GINIt-1 GDP per capita CPI TRADE GOVEXP GOV CSP (1) (2) 0.65882*** 0.68827*** (0.000) (0.000) 0.01511** 0.00620 (0.046) (0.440) -0.00140 -0.00189 (0.694) (0.564) 0.08465*** 0.12176*** (0.004) (0.001) -0.18671** -0.19683** (0.031) (0.030) 0.10980** 0.14413*** (0.027) (0.009) -0.03212 (0.327) M2 -0.07874** (0.033) VCSP -0.03730** (0.036) VM2 0.11243 (0.560) Constant 1.17426*** 1.11242*** (0.000) (0.000) 228 227 214.678 216.529 p-value (Sargan) 0.307 0.27 Arellano-Bond AR(2) 0.77 0.67 p-value (AR(2)) 0.44 0.50 Observations Sargan statistic * significant at 10%; ** significant at 5%; *** significant at 1% This table shows the estimated parameters α and β of equation (3) and (4). P-values Statistics in brackets. Study period extends from 1963 to 2002. Observations were made on a sample of 11 countries in the MENA region. GINI = GINI Coefficient. Annual per capita GDP is expressed in constant dollars of 2000. M2 = (Money + quasi-money)/GDP. CSP = Credits to the private sector/GDP. CPI = Consumer Price Index (base year : 2000). TRADE = (Importations + exportations)/GDP. GOV = Credits to gouvernement/Total banking loans. GOVEXP = Government consumption/GDP. VCSP = volatility of CSP. VM2= volatility of M2. 68 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW In times of economic and financial downturn, governments can adopt strategies to protect the poor, while rich households bear a heavier income losses (Christiaensen et al., 2003), inducing a decrease of income inequality. Finally, we note from column (1) that the share of bank loans for the government and public entities is positively and significantly related to GINI. An increase of 1 pp of the variable GOV is associated with an increase in GINI by 0.1098 pp. This result is consistent with our expectations because it shows that in economies where the government maintain a strong relationship with the banking system, the financial sector may suffer more from underdevelopment, which may reduce his beneficial impact on economic growth and income distribution. Column (2) of Table 4 shows the estimation results of equation (4). We note, first, that per capita GDP loses its significance, but it is still positively related to GINI. Then, as showed previously, inflation is still non-significant. Moreover, we find a positive and significant relationship between TRADE and the Gini coefficient, at 99 percent confidence level. Whereas, the variable GOVEXP is negatively and significantly related to GINI, which confirms previous results, namely that government spending in the MENA countries appear to be directed more towards poor households in order to reduce income inequality. Column (2) also shows results on the financial conduit effect that may be enhanced through the development of deposits and savings instruments. It appears that M2 is negatively and significantly related to GINI. An increase by 1 pp in M2 is induce a reduction in the Gini coefficient by 0.07874 pp. This confirms the conduit effect hypothesis, according to which greater supply of profitable savings opportunities accelerates the demand for real money balances and allows low-income households and entrepreneurs to accumulate capital to make indivisible investments. This is likely to induce convergence of incomes between rich and poor classes, especially in developing countries such as the MENA countries, where poor households’ access to credit remains difficult. The results show, however, that the second financial instability indicator (VM2) measuring the volatility of M2, does not enter significantly. Finally, we note from column (2) the existence of a negative and significant correlation between GOV and GINI, at 99 percent confidence level. The value and the significance of the coefficient related to the variable GOV are higher than those recorded previously. This confirms that increasing the share of loans to government, which may be also a sign of growing political interference, impede the financial system to reach an advanced stage of development that can allow it to reduce income inequality, through a negative effect on allocative efficiency and productivity of the banking system. 5. Conclusion Through this work we tried to emphasize the importance of financial development in dealing with the problem of income inequality. Indeed, it goes without saying that considering the financial system as a frozen system is probably an important limitation suffered from several works dealing with the issue of income inequality. Following previous studies that have attempted to examine the impact of financial development on poverty and income inequality, we tested VOLUME 7, NUMBER 4, 2013 69 such an impact on a sample of MENA countries. During recent decades, this region has recorded profound economic changes and significant macroeconomic fluctuations that may exacerbate poverty and inequalities. Our empirical results show, first, that under the conduit effect hypothesis (McKinnon, 1973; Kpodar, 2006), financial development, through the stimulation of demand for real money balances, significantly reduces income inequality in the MENA region. Then, the increase in the share of banking loans granted to government, which may reflects a growing political interference, restricts the ability of MENA countries” financial systems to reduce income inequality. Finally, there is a negative relationship between financial instability and income inequality. This paradoxical result can be explained by the fact that MENA countries’ redistributive policies are more directed to protect poor people in times of recession. References Ang, J.B., 2008. Finance and inequality: The case of India. Discussion paper, MONASH University, August. Arellano, M., Bond S., 1991. Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. Review of Economic Studies 58, 277–297. Arellano, M., Bover, O., 1995. Another Look at the Instrumental-Variable Estimation of Error-Components Models. Journal of Econometrics 68, 29–52. Arestis, P., and Caner, A.. 2009. Capital account liberalisation and poverty: how close is the link? Cambridge Journal of Economics 34, 295-323. Arestis, P., and Caner, A., 2009. Financial liberalization and the geography of poverty. Cambridge Journal of Regions, Economy and Society 2, 229-244. Barro, R.J., 2000. Inequality and Growth in a Panel of Countries, Journal of Economic Growth 5, 5-32. Beck, T. Demirgüç-Kunt, A., and. Martinez-Peria, M.S., 2007. Reaching Out: Access to and Use of Banking Services across Countries. Journal of Financial Economics 85, 234–66. Beck, T. Demirgüç-Kunt, A., Levine. R., 2007. Finance, Inequality, and the Poor.. Journal of Economic Growth 12, 27–49. Berger, A.N., De Young R., 1997. Problem loans and cost efficiency in commercial banks. Journal of Banking and Finance 21, 849-870. Blundell, R., Bond, S., 1998. Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of Econometrics 87, 115–43. Boyreau-Debrey G., Shang-Jin W., 2005. The pitfalls of state-dominated financial system: The case of China. NBER working paper, 11214, March. Chinn, M.D., Ito, H., 2002. Capital Account Liberalization, Institutions and Financial Development: Cross Country Evidence. NBER Working Paper, 8967, June. 70 THE EURO-MEDITERRANEAN ECONOMICS AND FINANCE REVIEW Christiaensen, L., Demery, L., Paternostra, S., 2003. Macro and Micro Perspectives of Growth and Poverty in Africa. The World Bank Economic Review 17, 317–47. Claessens, S., Perotti E., 2007. Finance and inequality: Channels and evidence. Journal of Comparative Economics 35, 748-773. Clair, R.T., Tucker, Paula., 1993. Six causes of the credit crunch. Reserve Federal bank of Dallas Economic Review, Third quarter. Clarke, G., Xu L.C., Zou H-F., 2003. Finance and income inequality: Tests of alternative Theories. The World Bank Policy Research Working Paper, 2984, March. Deininger, K., Squire L., 1996. A new data set measuring income inequality. The World Bank Economic Review 10, 565-591. Demetriades, P.O., Luintel K.B., 1996. Financial Development, Economic Growth and Banker Sector Controls: Evidence from India. The Economic Journal 106, 359-374. Demetriades, P.O., Luintel K.B., 1997. The Direct Costs of Financial Repression: Evidence from India. Review of Economics and Statistics 79, 311-320. Demirgüç-Kunt A., Levine R., (2009). Finance and Inequality : Theory and Evidence. The World Bank Policy Research Working Paper, 4967, June. Dollar, D., Kraay, A., 2002. “Growth Is Good for the Poor.” Journal of Economic Growth 7, 195–225. Greenwood, J., Jovanovic B., 1990. Financial development, growth, and the distribution of income. Journal of Political Economy 98, 1076–1107. Guillaumont, J.S., Kpodar, Kangni. 2008. Financial Development and Poverty Reduction : Can there be a benefit without a cost ? International Monetary Fund Working Paper, WP/08/62, March. Hauner, D., 2008. Credit to government and banking sector performance. Journal of Banking and Finance 32, 1499-1507. Honohan, P., 2004a. Financial Sector Policy and the Poor: selected findings and issues. The World Bank Working Paper 43, September. Honohan, P., 2004b. Financial Development, Growth and Poverty: How Close Are the Links? The World Bank Policy Research Working Paper, 3203, February. Jia, C., 2009. The effect of ownership on the prudential behavior of banks – The case of China. Journal of Banking Regulation 33, 77-87. Kaminsky, G.L., Schmukler, S.L., 2003. Short-run pain, long-run gain: the effects of financial liberalization. NBER Working Papers, 9787, June. King, R.G., Levine, R., 1993. Finance and Growth: Schumpeter Might Be Right? Quarterly Journal of Economics 108, 1076–107. Krueger, Anne., Tornell A., 1999. The role of bank restructuring in recovering from crises: Mexico 1995-1998. NBER Working Paper, 7042, March. VOLUME 7, NUMBER 4, 2013 71 Kuznets, S., 1955. Economic Growth and Income Inequality. American Economic Review 45, 1-28. Li, H-Y., Squires L., Zou H-F., 1998. explaining international and intertemporal variation in income inequality. The Economic Journal 108. 26-43. McKinnon, R.I., 1973. Money and Capital in Economic Development. Washington D.C: The Brookings Institution. United Nation Development Programme., 2009. Rapport arabe sur le développement humain. Bureau Régional pour les États Arabes, New York. World Bank., 2007. Finance for all? Policies and Pitfalls in Expanding Access. Policy Research Report, Washington D.C.: World Bank. Submit an Article All submissions to the Euro-Mediterranean Economics and Finance Review must be in English with the understanding that they have not been published and are not under consideration for publication elsewhere. Only the manuscripts that have passed the initial screening review by the editors are sent to a double blind referee process. Authors will be notified of the editorial decision within three months from the submission date. Manuscripts must be prepared following the Guide for Authors and should be sent electronically to [email protected] and [email protected] Guide for Authors Submission Submit your paper electronically. The preferred format for submission is a single PDF (generated from a Word source file) or Word document. For accepted papers, the Word source file of your manuscript is required for further processing. There is no submission fee. Manuscript preparation Please make sure all the characters in the body text, tables, figure legends, footnotes and References are in a single typeface and 12pt Times New Roman. Title: Concise and informative. Abstract: An Abstract of no more than 150 words is required. It should succinctly and accurately describe the contributions and major findings of the paper. Author names and affiliations: Clearly indicate the author names, their affiliations as well as the full contact of the corresponding author (phone, fax, email, and postal address). Keywords and classification codes: Immediately after the Abstract, provide a maximum of 6 keywords and up to 5 standard JEL codes (http://www.aeaweb.org/journal/elclasjn.html). Acknowledgements: Please make any acknowledgements in a separate section at the end of the article before the Références and do not, therefore, include them on the title page, as a footnote to the title or otherwise. List here those individuals who provided help during the research (e.g., providing language help, writing assistance or proof reading the article, etc.). Sections/subsections: they must be numbered consecutively such as 1., 1.1, 1.2, 2., 2.1, 2.2, etc. Math formulas: Formulas should be centered and numbered consecutively throughout the manuscript as (1), (2), etc. They must be edited with equation editor in Word. Footnotes: Footnotes should be numbered consecutively throughout the manuscript using superscript Arabic numbers and should be used sparingly. Figure captions: Please make sure that each illustration has a brief title and a description of the illustration (if any). Tables: Tables should be numbered consecutively in accordance with their appearance in the text. Place footnotes to tables below the table body. Références: Please make sure that only papers that are being cited in the body text are to be included in the Références section and that Références are in journal style. There is no limitation on the number of Références, but it is preferable that it does not exceed 35 in general. Citations in the body text should be as follows: Fama (1991) or (e.g., Fama, 1991). References should appear as follows: • Books: Jorion, P., 2007. Financial Risk Manager Handbook. 4th Edition, John Wiley & Sons. • Journals: Fama, E.F., French, K.R., 2004. The capital asset pricing model: theory and evidence. Journal of Economic Perspectives 18, 25-46. • Forthcoming papers and papers in press: Bekaert, G., Harvey, C.R., Lundblad, C., 2010. Financial openness and productivity. World Development, forthcoming. 02.31.35.76.95 www.Inquarto-conseil.fr Achevé d’imprimer par Corlet Numérique – 14110 Condé-sur-Noireau N° d’imprimeur : XXXXXX - Dépôt légal : 2013 Imprimé en France