Empirical Research Of Credit Risk Evaluation Of Listed Companies

With the rapid growth of credit in today's global economy, problems related to credit risk have attracted much attention. Credit risk has become one of the important risks which financial institutions have to face. Market economy naturally is credit economy. The listed companies are a major participant of market economy, so the credit of listed companies should be paid more attention to. It is practical significance to security market supervision, to investor's interests' protection and to credit organization's risk control. As one of means in order to avoid credit risk, credit evaluation is the outcome for the development of market economy. The importance and the improved function of the market economy have been proved by the research and the practice in developed countries.First of all, the paper reviews the literatures about the credit risk measurement of other countries and China in this field. Second of all, this paper introduces the related concept of credit risk of the listed companies, and proposed the reasons for credit risk and the problems of china's current assessment. Third of all, the paper introduces the developing of the credit evaluated models briefly. Basing on this, the paper choose the models which are more suitable to measure the credit risk of our country's listed companies, and explaining the theories of these models detailed. The last part is the empirical research, which is the core of the paper. The contents as follows: choose 61 listed companies which are special treated and 61 which are not special treated as the samples; establish the indexes system and the original indicators are condensed by the method of factor analysis. Six factors which are significantly affect the credit risk of the listed company are chosen; appliy the BP-Neural Network model, Support Vector Machine model, SVM-Logistic mixed binary discriminate rule and the Adaboost classification model for empirical research. Finally, the empirical results of the four models are compared with each other.The conclusion shows that: SVM uses structural risk minimization principle to substitute the experience risk. It can effectively avoid over-fitting phenomenon, so it has a higher classification accuracy rate comparing with the BP Neural Network; SVM-Logistic mixed binary discriminate rule integrated the merits of SVM and Logistic regression model, so it can improve the accuracy rate of SVM. The Adaboost model has overall superiority, so it has the higher accuracy rate than another three models.