The Retail Loan Default Model Research Based On Survival Analysis Method

The root causes of sub-loan crisis in the United States was high default rate of sub-housing mortgage loan. Su-housing mortgage loan is a kind of retail loans. If banks can predict the probability of default of such loans accurately, the possibility of crisis will be lower. There are lots of retail loans in China's commercial banks,and how to identify and measure effectively their credit risk to prevent the recurrence of such crisis, is facing a major challenge for China's commercial banks now. The purpose of this paper is to provide an effective way for China's commercial banks estimating default probability of retail loans.The estimated result of credit scoring mode is a credit score,Which is a far cry from the estimated value of default probability for the need of the New Basel Capital Accord. The structural model is difficult to estimate the threshold, so it is difficult to estimate default probability of retail loan in use of it. Therefore the default model above should not be easy to measure out of default probability of the retail loans. The biggest advantage of survival analysis method takes into account the censored data, which improves the accuracy of measurement. Recalling the basic theoretical knowledge of survival analysis and in view of the operation rules of retail loans and the characteristics of default factors affecting default Behavior, we put forward non-linear proportional default model with time-dependent variables. This model takes into account the non-linear relationship among variables and time-dependent variables while calculating the retail loan default probability of China's commercial banks, which will make the model to be in line with its objective reality better.This paper adopted personal car loan data from for a city commercial bank empirical analysis. The result shown that the nonlinear time-varying default model can better fit non-linear relationship between the age and probability of default,which was also very strong predictive power. At the same time, this paper simulated the 10-year data on individual housing loans, and added time-dependent variables into the mode, which demonstrated it was feasible to measure default probability of the long-term retail loans.