Credit Evaluation Approach And Application Based On The Perspective Of Binary Quantile Regression

Credit evaluation plays an important role in market economy and has received highly attention from both practice and research field, which leads to the rapid development of technology of credit measurement and management. However, the traditional method of credit assessment is established under the framework of mean regression, which is failed to reveal the heterogeneity of economic behavior. As an alternative way, quantile regression is able to reveal the heterogeneous effects of the covariates on the different quantiles of the response distribution. Considering the complexity of Credit data and the heterogeneity of credit behavior, this thesis mainly is deviated to the following two researches through quantile regression approach.In the first place, we establish a credit assessment method of listed company in China based on general binary quantile regression. Binary quantile regression, which extends the binary mean regression to the quantile framework, can reveal the heterogeneous affects of independent variables on the response variable across different quantiles. Therefore, it can describe and predict the behavior of binary choice more accurately than binary mean regression. We compare the performance of binary quantile regression model with binary mean regression model in both simulation studies and real data analysis. The results indicate that the binary quantile regression model is not only effective in terms of classification ability and robustness, but also is able to reveal the heterogeneous affects of influence factors on the credit across different quantiles. In the second place, we apply Lasso binary quantile regression to the credit evaluation of listed company in China. It can identify the key factors from vast influence factors through the Lasso variable selection function. Further, it can provide more comprehensive and detailed evaluation information through quantile regression, which is helpful to discover the heterogeneous affects of the key factors on credit status. We compare Lasso binary quantile regression model with Logit model, Lasso-Logit model and support vector machine through simulation studies and real data analysis. The numerical findings show that the former not only has outstanding capability of variable selection but also can get the best evaluation performance.The contribution of this thesis is three-fold. First, our method is able to provide the accurate evaluation of enterprises’ credit through the outstanding evaluation performance. Second, our results is helpful for policy makers to improve the level of credit through targeted company governance policies according to the heterogeneous affects of the factors on credit. Third, our method can identify the key factors influencing the credit through the Lasso variable selection, which helps to reduce the cost of collecting and managing irrelevant factors...