The Study Of Credit Risk Measuring Models And Its Application

The Basel Committee on Banking Supervision published Basel II in June 2004. The object is to encourage its members use more systematic risk (especially Credit Risk) measurement and evaluation systems, such as Moody's KMV. Basel II pays high attention to the function of internal ratings-based (IRB) in risk management and capital supervision. It encourages capable bank establish the IRB model and corresponding computer system. Now, many international advanced banks have begun to use IRB for risk measurement and management. The average growth rate of comprehensive competitive strength in these banks all have significantly increased.In view of the rapid development of IRB, first, this paper draws a sample of stock and financial information of the listed companies in Shanghai Stock Exchange from 2006 to 2007. Construct the KMV model and revise it. Then we use probability statistics analyze the discrimination of KMV combined with samples actual default state in 2007. The result shows that in China the KMV model based on the stock and financial information in year t-1 has fine discrimination to forecast the default state of the listed companies in year t.In order to further study the predictive power of KMV model, this paper constructs the multiple linear regression and Logit regression models respectively with the same sample. Among them, there are 5 explaining variables appear in the multiple linear regression model. They are price-earning ratio, equity turnover, degree of financial leverage, surplus cash cover ratio and debt-equity ratio. Its predictive power is inferior to KMV. In the Logit regression model, the explaining variables are return on net assets, working capital ratio and per share investment net cash flow. Its predictive power is better than KMV.Because the optimal classification point in probability statistics will influence the result of predictive power of models. So in this paper, we use ROC function to further study the model's discrimination. The study shows that in China, under the basic assumption, for the predictive power, the best model is Logit. Second is KMV. And the multiple linear regression model is the worst.