Using PCA-based neural network committee model for early warning of bank failure

As the Basel-II Accord is deemed to be an international standard to require essential capital ratios for all commercial banks, early warning of bank failure becomes critical more than ever. In this study, we propose the use of combining multiple neural network models based on transformed input variables to predict bank failure in the early stage. Experimental results show that: 1) PCA-based feature transformation technique effectively promotes an early warning capability of neural network models by reducing type-I error rate; and 2) the committee of multiple neural networks can significantly improve the predictability of a single neural network model when PCA-based transformed features are employed, especially in the long-term forecasting by showing comparable predictability of raw features models in short-term period.