Deep Learning : Credit Default Prediction from User-Generated Text

Digital technologies produce vast amounts of unstructured data that can be stored and accessed by traditional banks and fintechs. Prior literature on the topic indicates that certain aspects of this unstructured data could be valuable for decisions regarding the acceptance and pricing of credit contracts. Both, practitioners and academics are interested in whether this information is, in fact, valuable and how to exploit it for credit risk predictions. We employ deep learning techniques to extract credit-relevant information from user-generated text on Lending Club. Our results confirm that even short pieces of user-generated text can improve credit default predictions significantly. We benchmark four deep neural network architectures and more traditional approaches to retrieving credit-relevant information from text. An average embedding network outperforms all other deep learning architectures. The deep learning models outperform the more traditional approaches in almost all cases. Among the traditional approaches, spelling mistakes are particularly informative