Research On Credit Risk Stress-testing Of Commercial Banks In China Based On A CPV Model With Partial Least Squares Optimization

Nowadays credit risk has become the most significant risk that commercial banks need to manage. To identify and prevent credit risk, banks have to construct an effective risk mea-surement system. Traditional credit risk measurement methods are typically qualitative ones, like credit rating. Quantitative credit risk measurement models hadn’t obtained rapid develop-ment until the 1990s. This paper firstly completed a comparative analysis of several modern models for the measurement of credit risk:KMV model, Credit Risk+ model, Credit MetricsTM model and Credit Portfolio View(CPV) model. Despite some technical differences, the main function of the four models are the same, which use historical data observed from financial market to predict the future risk level of a portfolio. Unfortunately, however, the past is past; it’s impossible to foresee some sort of extreme events in the future such as great financial crisis. Once that situation happened, the banks and even the whole financial system would suffer a lot. So the banks and their regulators have to take the possible loss under extreme situations into account. The stress-testing method was developed as a result. The paper uses CPV model, which involves a variety of macroeconomic factors for stress testing. When the relationship be-tween these factors and the probability of default built, a seemingly unrelated regression (SUR) method is used, which would give out a wrong estimation of parameters when there is a linear relationship between variables. This paper proposes to use partial least squares(PLS) method instead of SUR method to overcome this problem, and then build the stress-testing model with the help of Monte Carlo method, to predict the extent of changes of credit risk level when macroeconomic is hurt under various situations. The paper also tries to fulfill some empirical analysis, but finds that the data of default probability does not exist in mainland China and some researchers suggested to replace the probability of default by the non-performing loan ratio. This paper finally dismisses that notion after further research and creates some simulated data to accomplish empirical analysis and stress testing in details. In the end, we brought for-ward some suggestion for improving the accuracy of the model in the future.