Policy uncertainty and bank stress testing

Abstract

The accuracy of dynamic stress-test capital models remains undocumented. Three methodologies: a CLASS-style approach, Bayesian model averaging, and a Lasso specification are used to forecast the performance of 14 large US banks during the financial crisis. Individual bank models are calibrated using bank historical data while regulatory models are calibrated using representative bank data. Representative bank model forecasts differ dramatically from the forecasts from bank-specific models and from actual outcomes. The Lasso methodology is most accurate, but its superiority may be sample-specific and is only apparent ex post. The results highlight the policy uncertainty inherent in regulatory stress tests.