Modeling Credit Losses for Multiple Loan Portfolios

Abstract

We propose a dynamic structural model of credit risk of multiple loan portfolios. In line with Merton, Vasicek and Pykhtin, we assume that a loan defaults if the assets of the debtor fall below his liabilities, and that the subsequent loss given default is determined by the collateral value. For each loan, the assets, liabilities and the collateral value each depends on a common and an individual factor. The common factors are interdependent, possibly dependent on exogenous (macroeconomic) variables. Consequently, the credit risk of each portfolio, quantified by default rate and loss given default, possibly depends not only on the exogenous variables, but also on historical credit losses of the portfolios. By applying our model to two nationwide US loan portfolios with real estate collateral, we demonstrate its considerable predicting power and we show that the hypothetical economic capital needed to withstand the 99.9 quantile loss is lower compared to the regulatory IRB approach; this suggests that using our model could lead to considerable savings for loan providers.