Troubled Banks : Why Don't They All Fail?

In this paper we examine troubled banks - those that receive a poor safety-and-soundness rating when examined - in order to predict future bank state. Besides failure, we see three alternative outcomes for these banks: recovery, acquisition, or continuation as a problem. The determinants of bank failure have been much researched, as has failure prediction. Most of this research uses a binary approach, dividing banks into two groups (those that fail and those that do not) or predicting one of two states (failure or nonfailure). Because our sample contains only troubled banks, we can go beyond a two-state approach. First we use univariate trend analysis to determine whether financial variables differ within this group of banks depending on the banks' future states. This analysis suggests that meaningful relationships exist between these future states and prior-period financial conditions. We then use financial ratios as explanatory variables in a unified model of bank states, with the goal of improving predictions of future bank condition. We gauge the model's effectiveness by testing the out-of-sample forecasting accuracy. Our results show that our model compares favorably with the standard binary failure-prediction model, yet has the added feature of predicting recovery, merge r, or continuation as a problem bank