A Cox proportional hazards-cash flow model for bank failure prediction

This research uses the Cox proportional hazard model and bank cash flow information to determine whether adding cash flow information will improve current bank failure prediction methods. In two published papers it has been shown that the Cox model performs as well as other models with respect to error rates and predictive accuracy. The Cox model, however, provides more information than models using the logit, probit or multiple discriminant analysis techniques. Proportional hazards models estimate both the probability that a bank will fail and the probable time to failure.To date, no attempt has been made to include cash flow information in bank failure prediction models. In the corporate failure prediction literature it has been shown that including cash flow variables improves the performance of the models. This paper tests whether a similar result occurs for bank failure prediction models.Models were estimated for several different time horizons and their temporal stability tested. Ideally, a predictive model will be stable over time so that it may be applied to new data without re-estimating the coefficients.Data was drawn from the FDIC Call Report tapes. Failed banks were taken from the FDIC Annual Reports: a nonfailed matching sample was drawn. The time frame of the study is 1985-1990.The results indicate that adding the cash flow information does not result in greater predictive accuracy. The models using cash flow information were not significantly superior to those using only regular accounting information.U of I OnlyETDs are only available to UIUC Users without author permissio