Uncertain Times and Early Predictions of Bank Failure

Abstract

The recent financial crisis and economic recession has shown that bank failure in the United States, while rare is a concern during uncertain times. Understanding the magnitude of banks at risk early in a crisis is a key challenge faced by policymakers. Early warning models are quite accurate at assessing risk using rolling predictions, yet they rely on recent failures for their accuracy. Of interest here is the ability at the start of a crisis to predict future failures, when the recent past has few events to base our inferences. We use logit and survival models to show that banks’ initial conditions at the start of the most recent crisis, along with model estimates from the S & L crisis, are good indicators of failures during the crisis, and that by accounting for uncertainty in our model’s specification we are able to improve our model’s out-of-sample predictions.