Failure Prediction in Commercial Banks with a Hybrid Prediction Model

In this paper, a new prediction model which was combined by trait recognition and SVM is examined through the use of accounting data measured 1 year prior to the commercial bank failure. To identity the features of problem banks and predict them as early as possible seems to be crucial for the stability of financial systems. Failure prediction models help to identify causes of failure, and thus lead to make a better understanding of bank operation. The contribution of six variable categories: profit, liquidity, loan risk, interest rate risk, capital and size are examined. The empirical data are collected from the Federal Deposit Insurance Corporation. Nine different comparing models, logistic regression method, C-SVM, Bayesian Network and so on were employed to justify the advantages of the hybrid method. In addition, the new method outperformed nine prediction models in overall accuracy. It was also shown that ROA, liquidity assets and short-term gaps are sound predictors for bank failure prediction.