A Neural Network For Financial Distress Prediction

Financial distress prediction is one of major business classification tasks because it greatly affects the financial decision making of investors, credits, and bank officers. Auditors also need information on financial distress prediction for the going-concern judgment. There are many statistical procedures to handle this financial distress prediction problem. The most widely used classification technique is statistical methods including MDA, logit, and probit methods. Since Altman introduced the use of MDA to financial distress prediction, MDA has been widely applied to the business classification, including bankruptcy prediction, credit rating, and bank loan classification, etc. The studies using MDA have encountered, however, some methodological problems. The violation of the underlying normality assumption of independent variables causes the biased results.. The decision tree, logit, and probit methods have been used as alternative statistical methods. However, they also require different kinds of statistical assumptions which limit the usefulness of their application.As one of alternative methods, it is well known that neural network approach is very promising for the financial distress prediction problem. A neural net represents a nonlinear discriminant fuction as a pattern of connections between its processing units. In this paper, we propose a new neural network model to the financial distress prediction problem. Based on the financial data of the specially treated companies during the year from 1999 to 2002, we use 25 finacial ratios as input and the probability as output. Empirical results show that neural network is a promising method of financial distress prediction in terms of predictive accuracy, adaptability, and robustnee. Limitations of using neural nets as a general moding tool are also discussed.