Predicting Bank Failures: Comparing and Extending Existing Models

Abstract

We replicate three bank failure models (Martin (1977), Cole and White (2012), and DeYoung and Torna (2013)) and introduce a new predictive model along with several evaluation methods to compare their out-of-sample predictive accuracy. We find that the models are highly accurate individually, and despite their similarities, are significantly different. Our new model is compared to the existing models using evaluation methods which are new to the extant field of bank failure research. Our model is significantly different from and more accurate than the existing models. Our paper makes an important contribution to bank failure research by demonstrating and validating evaluation methods which allow researchers and regulators to determine if a new bank failure model is different from and better than what has already been published.

Keywords: bank, bank failure, CAMELS, FDIC, McNemar’s Test, early warning system

JEL Classification: C52, G01, G17, G21, G28