Credit Rating Models Considering Sample Selection Biases

In general, credit-scoring models suffer from a sample-selection bias. This paper uses the bivariate probit approach to estimate an unbiased models scoring model. The data set with large commercial loans data provided by a commercial bank of China to estimate the model contains some financial and firm information on both rejected and approved applicants. In the bivariate probit model, we find a significant cross equation between loans rejected and loans granted. The results show that the variables with a positive (negative) effect on the probability of granting a loan have the same effect on default risk, implying that the bank lends loans in a way that is consistent with default minimization, not consistent with profit maximization.