Evaluating the performance of Static versus Dynamic models of credit default: evidence from long-term Small Business Administration-guarenteed loans

Dennis Glennon, Peter Nigro

Abstract

ABSTRACT

The financial crisis exposed the limitations of credit risk models to risk managers, financial regulators, investors and rating agencies. We compare the performance of conventional static-scoring techniques employed in practice with dynamic survival-time models to predict dollar losses on a portfolio of smallbusiness loans.We find that the dynamic models consistently generate more accurate dollar-loss forecasts over multiple time periods and performance horizons. Our results support the hypothesis that seasoning is a key factor in the development of accurate loss forecasts for longer-term amortizing loans (eg, small-business and mortgage loans). Furthermore, our results suggest that banks consider developing capital adequacy, loan-loss provisioning and securitized loan valuation models with a dynamic sample and model design.