The Early Warning Model Of Listed Companies Financial Distress Based On MDA And BP Neural Network

The financial distress is not only a common phenomenon of the market economy, but also a global problem. Especially after China's accession to the WTO, The enterprises of China have competed with the enterprises from other countries in the world, which have greatly increased the possibility of financial difficulties. Practice shows that most companies in financial distress are a gradual process, which is the exception to the gradual deterioration of the financial position, eventually resulting in the company bankrupt. Therefore, the business model for financial distress early warning is possible and necessary. Predicting financial accurately distress has important practical significance to prevent risks for the operators, to protect the interests for investors and creditors, to strengthen supervision on the economy for the government departments.This paper uses Fisher discriminant multivariate analysis model and BP neural network model for early warning of financial distress of listed companies and applies the standards of misclassification cost and resubstitution risk to compare the empirical results of two models, the specific content of this arrangement is as follows:Chapter 1 is the preface of this paper, and introduces the research backdrop, import and frame of this paper.Chapter 2 first discusses the domestic and foreign scholars on the definition of financial distress, but domestic and foreign scholars on the company's financial difficulties and there is no uniform definition, this article will be special treatment (ST) as a listed company in financial distress sign. , This chapter discusses in detail the plight of foreign and domestic aspects of the financial results of some studies and research methods. Among them, the research methods of financial distress univariate analysis, multivariate discriminant analysis, Logit models and neural networks, survival analysis, rough set analysis, and so on.Chapter 3 describes the first empirical study of this article is from the Shanghai and Shenzhen listed companies were selected in 2006-2010 special treatment of 100listed companies, and in order to eliminate the industry and the adverse effects of asset size, industry and selected ST listed company with assets at or near the 100 listed companies as paired samples, all samples will be randomly divided into model group and test group, which group is used to build a business model of financial distress early warning models, test group with model to determine the ability of corporate financial distress prediction. Then, from the profitability, solvency, operational capacity, growth capacity and the company's stock value point of view, selected preliminary financial difficulties to build early-warning model of listed companies, 21 financial indicators. At last, we do a simple exposition to the basic principles of Fisher discriminant method and BP neural network.Chapter 4 is the empirical study of this paper. Firstly, we conduct the normality test and mean difference test of the financial variables, the variables which are non-significant are removed. Then we conduct the correlation test of the remaining financial variables, the variables which have strong correlation are removed. Finally we screened out the financial variables which are suitable for the establishment of model, respectively construct Fisher multiple discriminant model of the previous 1 to 3 year. Then we use all of the 21 financial indicators to construct BP neural network model, we found that whether Fisher multiple discrimination model or BP neural network model, the prediction accuracy rate will be lower from the previous 1 to 3 year, and the prediction accuracy of BP neural network model is higher than the Fisher multiple discriminant model in the previous 1 to 3 year. Finally, we apply the standards of misclassification cost and rusubistitution risk to compare the empirical results of the two models. The results showed that: both the previous 1 to 3 year, under these two criteria, BP neural network model's performance is better than the Fisher multiple discriminant model.The conclusion is summary of the empirical analysis. It point out the shortage of this research and make suggestions for future research.