Credit risk prediction based on loan profit: Evidence from Chinese SMEs

Abstract

Credit risk prediction should maximize a bank’s loan profit. This paper performs modified profit-based logistic regression (MPLR) by constructing an objective function with the maximum profit as the objective. The optimal weights of two kinds of samples are obtained by constructing an objective function based on the sum of the weighted profit acquired in default and nondefault cases. To obtain greater loan profit, each customer's optimal discrimination threshold is determined by comparing the expected profit that the customer is predicted to produce in the default and nondefault scenarios. The research results show that the predicted and real profits obtained by our model are significantly higher than those obtained by 16 other classification models. The utilized weights can improve the accuracy and profit of the MPLR model, but the discrimination threshold is more important than the weights. The sample balancing process may not necessarily improve the classification accuracy and profit because it can reduce the Type-II error while increasing the induced Type-I error.