Latent Similarity-Enhanced Credit Risk Prediction1

Given the sheer size of the consumer credit market and the huge number of consumer credit users, credit risk prediction, or how to predict delinquent (or default) probabilities of consumer credits to aid financial institutions in granting and managing them, has become a critical problem in the consumer credit industry. While it is desirable to employ both users’ intrinsic characteristics and similarities among them for effective credit risk prediction, existing studies rely solely on similarities derived from their observed characteristics and fail to account for unobserved similarities among them. To address this challenge, we propose a latent similarity-enhanced credit risk prediction model, which operationalizes the similarity between a pair of users as a combination of the observed and latent similarities between them. We then design a new method that estimates the model parameters, learns latent similarities among users, and integrates both observed and latent similarities among users with their intrinsic characteristics for credit risk prediction. We further extend our method to the multiclass and numerical credit risk prediction problems. Extensive empirical evaluations with real world data demonstrate the superior predictive power of our method over benchmark methods for a broad spectrum of credit risk prediction problems. We also show substantial economic value generated from the superiority of our method through a case study.