Research Of Credit Risk Prediction Models Based On Quantile Regression

Credit risk is one of the traditional class risks in commercial bank. In recent years, the financialglobalization is ceaseless aggravate, credit derivative product is emerge in large numbers, bank’sbusiness scope has expanded, therefore, the credit risk faced by banks are also increasing. Atpresent, the credit risk management of bank management activities gradually becomes the mostimportant content. An effective method of Credit risk management is the use of a credit riskforecast model for credit risk measurement. There was widespread concern about credit riskforecasting model such as Z score model based on accounting and financial data and KMV modelbased on stock price. Although these two models play a crucial role in predicting corporateborrowers default probability, but because they rely on different data from different sources, leadingto both applied separately to the different scope. Quantile regression based on the dependentvariables of different conditional quantile regression to independent variable, it describes theresearch object in different quantile corresponding to different distributions, and the parameterestimation is also different, this will show the same kind of influence factors on different levels in aresearch object have different impact. Especially for analysis with a thick tail distributioncharacteristic of financial data, quantile regression method can provide more accurate and detailedinformation.This paper applys quantile regression model in credit risk model research, than constructs adefault probability prediction model which based on quantile regression model with Z score and DDvalue, aims to do a regression analysis by quantile Regression Method on the model constructedbefore, further to determine the Z score and DD value in different quantile of default probability,then verify the effect of Z score model and KMV model on different credit quality of listedcompany’s default risk. This paper selects40listed companies’ data to empirically analysis theconstructed model, results showed that at high levels of default probability the DD value has strongcorrelation, the Z score has weak correlation; at low levels of default probability the Z score hasstrong correlation, the DD value has weak correlation. It shows that Z score model in predict creditquality is better when the enterprise with high accuracy, whereas KMV model in prediction of lowcredit quality of enterprise with high accuracy. Therefore in the prediction of counterparty default risk, we can consider the company’s credit quality, choosing suitable model, to obtain more accurateprediction of the risk of default.