An Applied Study Of Financial Distress Warning Of Listed Companies Base On Logistic Model And Bayesian Network

In recent years, Chinese securities market has rapidly developed. The number of listed companies in China has reached 1976, and the value of total stock market has reached 23.874 trillion yuan by the September 2010. The healthy and stable development of the securities market impact the macroeconomic stability. List companies are the fundamental elements of the securities market, and the quality of its management will directly affect the healthy development of the securities market. Moreover, equity investors, creditors and banks are subject to the impact of the financial situation of listed companies. Due to the fierce competition, a lot of companies were in financial trouble because of the poor management. Some of the companies had to face the debt restructuring and the assets reorganization, which made the equity investors, creditors and banks lose their money. Therefore, it's very important to build an early warning system of financial distress for various stakeholders of listed companies.This paper studies if the non-financial indicators and the DEA indicators could improve the accuracy of predicting result of financial early warning models and compares the predicting efficiency of logistic regression model with the Bayesian network. The full paper includes four chapters, and the main contents of each chapter are as follows:The first chapter discusses the background and the significance of the research, then classifies and summarizes the research of domestic and foreign scholars. The second chapter defines the concept of the financial distress first, then clarifies the theoretical basis of using the ST companies as the financial distress. At last, I illustrated the theory of financial distress predicting models which are used in this paper: logistic regression, Bayesian network and DEA.In the third chapter I selected the samples and variables. First of all, I selected the listed manufacturing companies as the research objects, and divided them into two groups. One of them is for the model training and the other is for the model testing. The basic variables were selected from both financial indicators and non-financial indicators. Financial indicators are reflected in the companies'solvency, profitability, operation capability and the growth capacity. Non-financial indicators are reflected in the companies'governance structure, market returns, equity indicators. First,all variables were tested by independent-sample T test. Then, variables were selected whose sig. is lower than the critical value of the level 0.05. Finally,I did correlation test of the selected variables and excluded the variables which were highly related to others. Then, I selected six financial variables and five non-financial variables into the models.In the fourth chapter, I used the selected variables to do the empirical analysis.First,I use the six financial indicators to establish the logistic regression model, and I called it logitⅠmodel. The adding five non-financial indicators into logitⅠmodel, then I called the new model logitⅡmodel. Next I need to compare the goodness of fit and the predicting accuracy of the two models. In the second subsection, I used the Tree Augmented Na?ve Bayesian network (TAN) to build the models. First, I selected the financial indicators to build the Bayesian network and classify the samples. Then add non-financial indicators into the above mentioned model and calculate the accuracy of the new model. At last, determine whether the addition of the non-financial indicators could improve the predicting accuracy of Bayesian network. In this paper I attempted to use DEA method into the financial crisis early warning system. Study if the DEA efficiency variable could improve the predicting efficiency of the models.The empirical results indicate that non-financial variables could be used as an effective complement to the financial variables, and improve the prediction accuracy of the financial early warning model. But if added the DEA efficiency variable into the logistic model and TAN model, the accuracy of prediction could not be effectively improved. The logistic regression model and TAN model had advantages and disadvantages. If I only selected the financial variables, the logistic model was more efficient than TAN predicting model. If non-financial variables were added into the model, the prediction accuracy of the TAN model was slightly higher than the prediction accuracy of the logistic model. The two models had good stability, and TAN model was more stable than the logistic model.