A sample selection model of credit rating: Evidence for a commercial bank of China

This paper describes a bivariate probit selection model of large loan credit rating that accounts for the cross equation correlations of the errors. The data set provided by a commercial bank is used to estimate the model. We find a significant cross equation between loan settlement and loan default. Based on these results, the bivariate selection model provides more efficient estimates than does a single equation model because the bivariate selection model accounts for the sample selection bias associated with loan settlements. In addition, the sample selection model predicts much lower default rate for the data set as a whole than the single equation model does. The results strongly, show that the bivariate probit selection model is acceptable when constructs the credit rating model for practice used.