Forecasting credit card portfolio losses in the Great Recession: a study in model risk

José J. Canals-Cerdá and Sougata Kerr

Abstract

ABSTRACT

The Great US Recession of 2007-9 offers a unique opportunity to analyze the performance of credit risk models under conditions of economic stress. We evaluate three potential sources of model risk: model specification, sample selection and stress scenario selection applied to credit card portfolios, a major portion of a bank's balance sheet. Our analysis indicates that model specifications that incorporate interactions between macroeconomic variables and core account characteristics generate more accurate loss projections across risk segments than specifications that do not. The sensitivity of modeled losses to macroeconomic factors depends on the severity of the downturn in the model development sample. Pre-2007 models estimated over a period that included the 2001-2 recession fail to project levels of credit loss consistent with those experienced during the Great Recession. These models estimated over a time period that did not include a significant economic downturn severely underpredict credit loss for some segments and the levels of forecast error are significantly affected by model specification assumptions. Prime borrower segments of the portfolio are more severely affected by downturn economic conditions and model specification assumptions relative to the subprime or near-prime segments. The selection of the stress scenario can have a dramatic impact on projected loss.