Stress Testing the Impact of Group Dependence on Credit Portfolio Risk

Steven Vanduffel, Boštjan Aver, Andrew Chernih, Luc Henrard and Carmen Ribas

For the past decade the financial industry has put models in place to assess the default risk of their various credit portfolios and Koyluoglu and Hickman (1998) have classified these as follows

1. the “Merton-based” approach; see JP Morgan’s CreditMetrics (Gupton et al 1997) or MKMV’s PortfolioManager (Zeng and Zhang 2001);

2. the “Econometric” approach; see McKinsey & Company’s Credit-PortfolioView (Wilson 1997a,b); and

3. the “Actuarial” approach; see CreditRisk⁺ (Credit Suisse Financial Products 1997).

While the mathematical properties of these different approaches are now well understood it remains difficult to prove the accuracy of any of these models, especially when measuring upper tails of the portfolio loss distribution. This is essential because the portfolio will be most severely hit when several credit exposures default together. Unfortunately, since a default is a rare event it is difficult to predict default probabilities and even more so for joint default probabilities. At best financial institutions have a good view on the single and pairwise default probabilities, or equivalently, the default correlations but not on the likelihood that three or more