Your Posts Expose You : Theory-Driven Approach to Credit Risk Prediction for Microloans Based on Social Media Content

We investigated the value of credit borrowers’ social media posts, such as microblog posts, in the prediction of their credit risk. We used a mixed methodology to enhance credit assessment models with theory-driven features. On the basis of general strain theory, we proposed a set of psychological features to explain borrowers’ default risk. We conducted a field survey to empirically verify the reliability and validity of these features in a theoretical model. Next, we extracted borrowers’ psychological features from their social media posts and used them to enhance traditional features in credit assessment models. The results indicate that the theory-driven psychological features can considerably enhance the prediction performance of credit assessment models. The contribution of these features is higher than that of traditional features. We compared the theory-driven psychological features with data-driven features extracted from social media posts and observed that the former not only exhibited interpretability but also had a similar ability to predict credit risk to data-driven features. This paper describes a comprehensive theoretical framework for explaining credit default behavior and a methodology of theory-driven feature engineering based on unstructured social media data so that interpretable machine learning models can be trained for the online microloan industry