Study Of Commercial Bank Credit Risk Stress Testing Based On Partial Least Squares Method

In previous stress testing models of credit risk in commercial bank, the selection of factors islimited，we could only select a few factors. The defects associating with those models often led to lack ofconsideration of macroeconomic variable. In fact, credit risk in commercial bank would be affected bymultiple macroeconomic variable. This paper uses Partial Least Squares model(PLS) to substituteSeemingly Unrelated Regression model(SUR), and improves the conduction mode in pressure testingmodels for credit risk.Quarter NPL ratios of joint-stock commercial banks in China from2005to2010were chosen as thebaring stress indicators. Ten macroeconomic factors were selected as the pressure indicators, such as thegrowth rate of gross domestic product, the growth rate of fixed investment, total import and export tradegrowth and so on. By processing these factors using statistical analysis method, a conduction model forcommercial bank credit risk stress testing was established based on the relation between NPL andmacroeconomic factors. This paper use SUR method and PLS method to estimate the conduction modelrespectively and comparing the results. Observe that when there is a correlation between the variables,using SUR method will appears coefficient test can t be passed, symbols abnormal. But the PLS rule agood solution to this problem.After building the conduction mode, this paper use Monte Carlo simulation method and constructthe scenario model in credit risk stress testing model of commercial bank. The result shows that in thescenario of declining RGDP, simulation scenarios method is better than the historical scenario methodand hypothetical scenario method.Our commercial banks should apply credit risk stress testing base on PLS model. Besides,commercial bank need to enhance the monitor of the related macroeconomic variable, and build a timelyand effective feedback report mechanism.