Predicting bank failure using a simple dynamic hazard model

We compare the out-of-sample forecasting accuracy of the time-varying hazard model developed by Shumway (2001) and the one-period probit model used by Cole and Gunther (1998). Using data on U.S. bank failures from 1985" 1 992, we find that, from an econometric perspective, the hazard model is more accurate than the probit model in predicting bank failures, but this improvement in accuracy results from incorporating more recent information in the hazard, but not the probit, model. When we limit both models to the same information set, we find that the one-period probit model is slightly more accurate than the time-varying hazard model. We also find that a parsimonious specification of the one-period probit model fit to data from the 1980s performs surprisingly well in forecasting bank failures during 2009" 2010.