Stress testing and the representative bank model

Abstract

Since the Supervisory Capital Assessment Program, US regulators have used representative bank stress test models to benchmark individual bank forecasts made using models calibrated to match a bank’s own historical performance. Stress scenario forecasts from a representative bank model are compared with bank-specific models when all models are estimated using an identical methodology and only differ by the historical bank-specific calibration data utilised. Under an identical stress scenario, forecasts from the representative model differ dramatically from bank-specific model forecasts and actual bank outcomes. The results highlight inconsistencies that can arise when a regulator uses a representative bank model benchmark.