This study aims to predict bank failures in Turkey between years 1990 and 2013 by utilizing continuous time parametric survival models and controlling for the unobserved heterogeneity at the bank level, i.e. shared frailty in survival models. In order to examine the determinants of a bank failure, a range of financial ratios of 79 banks along with some macroeconomic variables are obtained. Two parametric distributions, the Exponential and the Gompertz, are employed for the models with and without shared frailty. Based on test statistics, unobserved heterogeneity exists at the bank level in almost all models. According to the estimation results, the most important financial ratios belong to the groups of capital ratios, profitability, asset quality and branch ratios. Macroeconomic variables and crisis dummy variable are also significant in all models. Prediction accuracies of the exponential models improve when the models are estimated with shared frailty while for the Gompertz distribution model, shared frailty improves the predictions only when the macroeconomic variables are included. However, shared frailty (and without shared frailty) models excluding macroeconomic variables have better prediction results than the same models with macroeconomic variables.
Key words: Bank Failure, Survival Analysis, Unobserved Heterogeneity. JEL Classification: C34, E44, G21. Article Language: EnglishTurkish
|
