The Stress test of Household Loan Sector considering Heteroscedasticity, Autocorrelation and Conditional Loss at Given Default(LGD) (in Korean)

After financial crisis in 2008, it was found in several research that the estimated value of the unexpected loss calculated by stress test have a possibility of being underestimated. In this regard, recent studies have focused on developing an appropriate model to address this problem(BIS(2009), Foglia(2008), Haldane(2009)). This paper examines the potential underestimation problem in the process of stress test using CreditPortfolioView, which is suggested by Wilson(1997a), and provides some alternatives to deal with this problem. Heteroscedasticity and autocorrelation are taken into account in the regression model of the probability of default which is the core of the CreditPortfolioView method. This is in order to reflect particularities in crisis period and additional risk factors. Then, credit loss distribution and VaR(Value at Risk) in normal and crisis period are estimated respectively by using the basic model and the revised model. In case of the normal condition, the result shows negligible difference in the outcome of VaR between these two models. In the case of the crisis, however, it shows that the estimated value of VaR in the revised model is approximately 1.75 times larger than the one in the basic model. This result indicates that if stress test does not consider particularities and additional risk factors in crisis, it is likely to underestimate the value of unexpected loss in crisis. Therefore, it implies that it is necessary to consider the particularities and additional risk factors in performing stress test.