Model Construction And Empirical Study On Credit Risk Evaluation In Chinese Companies

As the acceleration of global economic integration, credit risk in the form of diversity, complex and the intensification of globalization, credit risk measurement and management has increasingly become the focus of attention for financial sector. A long time, expert system method, rating method, and some other traditional means have been used to measure credit risk. But most of these methods are based on the subjective experience and adopt qualitative analysis method to manage credit risk, such as third-party guarantees, the establishment of bad debt reserve funds, maintaining long-term friendly customer relations and so on. These credit risk management methods often cause unnecessary losses by human factors (such as misjudged). The lack of enterprises credit will bring about directly loss to the economic subjects. For example, if loan companies can not pay the debt on time, it will directly affect the lending institutions normal economic activity and cause loss of investors. At the same time, deficiency of enterprises credit will reduce mutual trust relationship between enterprises and contract party, and increase difficulty of raising funds.Firstly, the paper defined the definition and classification of credit risk, and then reviewed theoretical method of credit risk assessment at home and abroad. At the same time, some important and classical model were elaborated. Secondly, combing with practical of the listed companies and the of credit evaluation in China, this article regards 70 the listed companies that was treated specially(ST)from 2009 because of'unusual financial condition'and 70 normal companies in the proportion of 1:1 as the modeling samples in A-share Shanghai and Shenzhen stock market. Take statistical analysis software for SAS9.1 and Matlab7.0 as the auxiliary tool to set up logistic model and neural network model based on the financial information. At last,108 ST (including *ST) company and 130 normal company which have complete financial information were selected to test the constructed model. The empirical research shows that logistic model has good prediction ability and the BP neural network model is more ideal prediction effect before the listed company being special treatment.This study also obtained:(1) Some financial index of listed companies do not follow a normal distribution, so as we test "problem sample" and "normal sample" differences of financial indicators, parameters T test and non-parametric test should be used to weed out index that have no significant difference.(2) Five financial indicators are absorbed into Logistic regression model that is used by stepwise regression method, which are asset-liability ratio, rate of return on assets, gross profit margin, ratio of expenses to sales and operating income growth rate. There are three index reflecting profitability, one index reflecting debt paying ability and the last one reflecting development capacity. It can be seen profitability indicators exert great effect in identifying good financial situation and bad financial situation of enterprises account. So if a business had deteriorated profitability index, which indicate the great default possibility.(3) After the financial indicators screened, the Logistic regression model and the BP neural network model are created. Positive test results of both models are ideal, and the BP neural network model has a better accuracy. The two predicted models can be compared and added to reduce the false caused by the model itself.