The purpose of this paper is to split loan customers to different credit ratings to ensure the results that show that customers with lower credit ratings have higher loss rates, and the number of customers that satisfies the bell-shaped distribution. Hence, the number of credit ratings, the distribution of the rated obligors among ratings can achieve a meaningful differentiation of risk, which can avoid the loan pricing confusion.
The authors introduce a multi-objective programming to establish the credit rating model. Objective function 1 minimizes the absolute difference between the obligor number proportion and perfect client proportion, following a standard normal distribution. Objective function 2 minimizes the total difference of the deviation between two adjacent credit ratings’ loss rates. This study combines the two objective functions to ensure the obligor number distribution and the monotonicity of the loss rate, and applies genetic algorithm to solve the model.
This study’s analysis is based on data from 6,155 enterprises, provided by a Chinese bank and Prosper P2P loan data. The empirical results reveal that the proposed approach can ensure the balance between both criteria and avoid undue concentration of obligors in particular grades.
The proposed credit model could help building a reasonable credit rating system, which is the prerequisite of loan pricing; thus, inaccurate credit rating can cause incorrect loss rate estimates and loan pricing.
