An efficient and functional model for predicting bank distress: In and out of sample evidence

We examine the failures of 132 U.S. banks over the 2002-2009 period using discriminant analysis and successfully distinguish between banks that failed and those that didn't 92% of the time using in-sample quarterly data. Our two most important variables are related to bank capital and loan quality, as one might expect; although bank profitability is also important. The resulting model is then used out-of-sample to examine the failure of 191 banks during 2010-11, with predictive accuracy in the 90-95% range. Our results demonstrate that our model can also easily be applied to a large number of firms (even those that don't fail) and does an excellent job of distinguishing healthy from distressed banks. Combining this effectiveness with its ease of implementation makes it very functional. Such a model should be of obvious interest to regulators, analysts, and all those with a direct interest in assessing bank financial health.