Spurious Default Probability Projections in Credit Risk Stress Testing Models

Abstract

Credit risk stress testing has become an important risk management device which is used both by banks internally and by regulators. Stress testing is complex because it essentially means projecting a bank's full balance sheet conditional on a macroeconomic scenario over multiple years. Part of the complexity stems from using a wide range of model parameters for, e.g., rating transition, write-off rules, prepayment, or origination of new loans. A typical parameterization of a credit risk stress test model specifies parameters linked to an average economic, the through-the-cycle, state. These parameters are transformed to a stressed state by utilizing a macroeconomic model. It will be shown that the model parameterization implies a unique through-the-cycle portfolio which is unrelated to a bank's current portfolio. Independent of the stress imposed to the model, the current portfolio will have a tendency to propagate towards the through-the-cycle portfolio. This could create unwanted spurious effects on projected portfolio default rates especially when a stress test model's parameterization is inconsistent with a bank's current portfolio.