Modeling with scarce data: Credit and

Commentaires

Transcription

Modeling with scarce data: Credit and
Modeling with small data:
Credit and Operational Risk
13th congrès des actuaires, Peter Middelkamp, Small Data
Example 1: Operational Risk
13th congrès des actuaires | Peter Middelkamp | Small Data
2
Operational Risk1 : (typically) Available Data
• Internal sources:
0.99
– Own loss history collected over
the past 5-10 years
0.98
0.97
– Expert judgement of loss
potentials
– Industry loss data from loss
data exchanges (eg: ORX)
– Public information
(eg: corporate filings, rating
agencies, news, Algo First)
0.96
F(x)
• External sources:
Empirical CDF
1
0.95
0.94
0.93
0.92
model
internal loss data
external data
0.91
0.9
0
500
1000
x
1500
2000
13th congrès des actuaires | Peter Middelkamp | Small Data
3
severity distribution per scenario and legal entity
• truncated Pareto with one important input: A = Anticipated Loss Potential
for a given confidence level e.g. f = 99.5%.
13th congrès des actuaires | Peter Middelkamp | Small Data
4
aggregation with dependence
• Dependence within operational risk:
– Dependence between legal entities
– Dependence between event categories
• External dependencies:
– Dependence to financial markets (more prevailing in banks)
– Dependence to pandemic events
13th congrès des actuaires | Peter Middelkamp | Small Data
5
Conclusion Operational risk
• All parameters are mainly based on expert judgement that might be inspired
by historic events in the firm and that happened to peers.
• A simple dependence method is a must (e.g. mixture of full dependence,
independence)
• Number of parameters should be kept at a minimum to have a stable, easy
to understand "well behaved" model.
• Continue to collect data and use it as benchmark for the model.
• Model alternatives (rejected):
– LDA (not enough data)
– Standard Formula (based on industry averages)
– causal models with explicit control modelling (complicated, potentially unstable)
– combining loss data sources and expert judgement using Bayesian inference
(potentially unstable)
13th congrès des actuaires | Peter Middelkamp | Small Data
6
Example 2: Credit Risk
13th congrès des actuaires | Peter Middelkamp | Small Data
7
Compared to operational risk, there is plenty of data
AAA and BB history 1920-2013
• Default history
exposes large spikes.
10
8
BB
6
AAA
4
2
0
1920
1924
1928
1932
1936
1940
1944
1948
1952
1956
1960
1964
1968
1972
1976
1980
1984
1988
1992
1996
2000
2004
2008
2012
• Rating information
alone is not enough
to determine default
risk
12
Default rate [%]
• Some rating
categories never
experienced a
default within one
year
14
13th congrès des actuaires | Peter Middelkamp | Small Data
8
Multi year default probabilities can be used to estimate
one year default probabilities
Cumulative default probability
AAA rating
Fit
Data
10.000%
1.000%
0.100%
0.010%
0.001%
1
3
5
7
9 11 13 15 17 19
Years
• Note that a minimum default probability of 0.03% is applied to corporates,
0.01% to sovereigns.
13th congrès des actuaires | Peter Middelkamp | Small Data
9
Modelled default probabilities should be able to reproduce
and go beyond historic realisations
• Credit portfolio risk
models capture historic
default history
(back-test)
DP
13th congrès des actuaires | Peter Middelkamp | Small Data
10
Credit portfolio risk models should also capture historic
dependencies to financial markets (back-test)
• Default rates are higher
when equity markets
are down/ credit
spreads widen (not that
there is sometimes the
default rate lags behind
financial markets)
13th congrès des actuaires | Peter Middelkamp | Small Data
11
13th congrès des actuaires | Peter Middelkamp | Small Data
12
Legal notice
©2014 Swiss Re. All rights reserved. You are not permitted to create any modifications
or derivative works of this presentation or to use it for commercial or other public purposes
without the prior written permission of Swiss Re.
The information and opinions contained in the presentation are provided as at the date of
the presentation and are subject to change without notice. Although the information used
was taken from reliable sources, Swiss Re does not accept any responsibility for the accuracy
or comprehensiveness of the details given. All liability for the accuracy and completeness
thereof or for any damage or loss resulting from the use of the information contained in this
presentation is expressly excluded. Under no circumstances shall Swiss Re or its Group
companies be liable for any financial or consequential loss relating to this presentation.
13th congrès des actuaires | Peter Middelkamp | Small Data
13