Study On Listed Companies’ Financial Distress Prediction Based On Bayesian Network Combined Model

Chinese security exchange market has gone through 25 years since Shanghai and Shenzhen Stock Exchange have been established. Its prosperity and development is huge, especially the stock market of our country. Although the stock market still can’t avoid the impact from policies of the government, it is improving increasingly, moving closer to the mature stock markets of developed countries, and makes great contribution to the development of the domestic real economy. However, even if listed companies will not be developing sustained and healthily forever, a small portion will face the risk of financial distress. This study is to predict the financial distress of listed companies, which are of great significance for the business owners, investors, creditors and the whole economy.95 companies listed in Shanghai and Shenzhen Stock Exchange, specially treated because of abnormal financial status between 2008 and 2013, are chosen as the financial distress company samples, and 190 healthy listed companies are chosen as paired samples. Meanwhile, 22 financial indicators of listed companies are chosen as the initial independent variables, and then these indicators are refined by the Stepwise method. The result shows that the Stepwise method is capable of reducing the dimension of model variables, and variables after refining cover the sufficient information which reflects the company’s future financial status.This study proposes a combined forecasting model which predicts the companies’ financial status of year T with financial data of year(T-2). Firstly, a classification layer is added to optimize the output based on the traditional Adaptive-network-based Fuzzy Inference System(ANFIS). Then the data of the training sample is used to train the improved ANFIS and Mahalanobis distance discriminant method, which respectively predict consequence of the checking sample. Finally, the consequence and prediction accuracy rate of these two models are input into the Bayesian network, getting the prediction consequence of the combined models.The empirical analysis results show that Mahalanobis distance discriminant method can discriminant those financial distress companies and healthy finance companies effectively, which proves that some financial indicators of companies before being involved in financial distress is obvious different from healthy finance companies and it exists different clustering phenomena in space; the prediction accuracy rate of the financial distress companies and the healthy finance companies are counted separately, and the prediction performance of the proposed combined model is better than that of the Neural Network, Support Vector Machine, Logit model and pure Bayesian Network.