Stress-Testing Banks’ Credit Risk Using Mixture Vector Autoregressive Models

Tom Pak-Wing Fong and Chun-Shan Wong

Under a stress-testing framework for risk exposures of banks’ loan portfolios, the probability distribution of default rates conditional on an adverse macroeconomic shock is usually underestimated.²²A comprehensive summary for the framework of stress testing can be found in Sorge (2004). This is because the underlying distributions of default rates and macroeconomic variables are assumed to be unimodal. Such an assumption does not discriminate between normal and abnormal situations (including abnormal rises and falls). The resulting distribution may therefore reflect the vulnerability of a financial system due to normal market shocks, but not “exceptional but plausible” macroeconomic shocks, since the number of normal market observations are many more than that in the stressful market situation.³³Some studies have added dummy variables to econometric models to filter such “crisis” effects, however, the stress-testing exercise becomes a test for having a shock under normal market condition. That seems not to be the exact purpose of a “stress” test for banks’ loan portfolios.

Conventional studies have considered tail events of historical episodes to devise scenarios in order to