Financial state analysis and bankruptcy risk forecasting for banks

Abstract

The problem of banks financial state analysis and bankruptcy risk forecasting is considered. In this study, financial indices of 170 Ukrainian banks were chosen, the training sample included 120 banks, and the test sample included 50 banks. Financial indices were taken one and two years before the 2008-2009 crisis of the bank system in Ukraine. Taking into account the uncertainty of the input data, the following fuzzy methods for solving this problem are suggested: fuzzy neural networks (FNN) ANFIS, TSK and fuzzy GMDH. The experimental investigations of the suggested methods were performed and their efficiency was estimated for the bank system of Ukraine. The comparative analysis of the suggested fuzzy methods with conventional classical methods was performed. The results of experiments showed that FNN TSK gave a better forecast than ANFIS. Also, the increase in the number of rules in FNN does not improve the forecasting accuracy. While comparing different fuzzy methods, it was found that FNN TSK gives a more accurate forecast at the short-term forecast (one year), while fuzzy GMDH gives a better forecast at the middle and long-term intervals (two and more years). In a whole, the fuzzy methods give a better forecast than classical methods in the problem of Ukrainian banks bankruptcy risk forecasting. The most essential financial indices for bankruptcy risk forecasting were determined.