A neural network model for bankruptcy prediction

Bankruptcy prediction is studied widely since it can impact on bank lending decisions significantly and then effect bank profitability ultimately. It's well known that statistical techniques are very popular for this purpose, such as univariate and multivariate discriminant analysis. In this paper we develop a neural network model for bankruptcy prediction. Novel sets of financial ratios that can lead to significant improvement in prediction accuracy are also introduced. These ratios are extracted from expert knowledge. An experiment is shown to evaluate this neural network model. The training data and test data are obtained from Guangxi Branch of Industry and Commerce Bank of China. Test results show that neural network outperforms statistical techniques in accuracy of prediction and robustness. And the novel set of financial ratios presented in this paper perform much better than traditional financial ratios of Altman's.