Research On The Forecast Of Corporate Credit Default Probability For Commercial Banks

As one of the most important risks which commercial banks are faced with in our country, credit risk is composed mainly of the risk of commercial banks giving loans to enterprises. Probability of Default (PD) is the possibility that the borrower cannot fully repay the principal and interest of each loan on schedule in accordance with the stipulations of loan contracts or perform the relevant legal obligations. It influences the quantification of bank credit risk, the implementation of the new Basel Capital Accord and the accounting of regulatory capital. Meanwhile, PD is also the basis to calculate the Value at Rist (VaR) and then estimate the overall VaR of the banks’ assets. This research focuses on the study of prediction model, method and technique of the PD, which are the core issues of risk quantitative management. This research topic has great theoretical significance and practical value.Based on the literature review of researches and theories regarding to quantification of bank credit risk and the PD forecasting, as well as the practice problem of "monitoring the credit capital flow", this research has proposed a new method which could predict the VaR according to customers’ behavior of credit capital transaction directly. The major works and innovatory results are as follow.(1) The mechanism of clients’ capital abnormal flows is explained systematically in this paper. Firstly, the business modes of some domestic commercial banks for monitoring and managing capital flows have been analyzed to discover the critical control points. Secondly, on the basis of the concept of clients’ capital abnormal flows which defined in this paper, the motivation of clients’ behavior of capital abnormal flows have been analyzed from the perspectives of both risk aversion and profit seeking, in order to explain the causes of the behavior. Finally, the characteristics of clients’ capital abnormal flows have been concluded by automatic monitoring and analyzing the improper trading event data from the clients’ capital transaction database, which could provide the comprehensive, objective and real time information for monitoring the clients’ credit capital transaction behavior.(2) A new method to perform the real-time prediction of PD is proposed based on the data of clients’ credit capital transaction. Firstly, the factors which could be used for the prediction of PD is selected by statistical approaches including the significance analysis of single factor based on SPSS as well as the principal component analysis from the dimensions of rationality, validity and combination necessity. The selection results show that, there are six factors about the clients’credit capital transaction which have significant correlation with the customer default. The indicators defined by these factors are effective for the real-time prediction of PD. Compared with the other PD prediction methods, the modeling approach proposed in this paper can utilize the real-time data, which have the advantages of high reliability and sound objectivity. So the prediction of PD based on this approach could be more accurate and timely.(3) According to the condition of the real clients’ credit capital transaction data, a PD prediction method based the XDT-SVM is conducted in this paper. As the prediction method mentioned above needs the massive amounts of real-time data, the traditional SVM modeling tool are not suitable for the method. Based on the in-depth study of SVM, a modified DT-SVM algorithm, which named as XDT-SVM algorithm, is applicable to the prediction method in this paper. The XDT-SVM algorithm uses the method of decision-tree for rough classification of samples, which could shorten the time cost of modeling effectively and develop the parallel processing capability.(4) With the application of the XDT-SVM algorithm, the real-time PD prediction method could be used to handle the massive data. The actual company loan business data and customer funds transaction data in 2011 from a commercial bank are chosen, and XDT - SVM algorithm modeling is applied to forecast the probability of loans default. The research results show that the default probability forecasting model constructed in this article, based on the customer funds trading behavior, has higher prediction accuracy.