Deep Learning: Credit Default Prediction from User-Generated Text
22 Pages Posted: 15 Feb 2020
Date Written: January 13, 2020
The digital transformation produces vast sources of unstructured data that are storable by and accessible to traditional banks and fintechs. Prior literature indicates that this unstructured information is valuable for decisions of accepting and pricing credit contracts. While processing this kind of information has been very difficult in the past, deep learning offers tools to process parts of these unstructured sources automatically and use it to predict credit defaults. We employ deep learning techniques to extract credit relevant information based on loan descriptions from Lending Club. Our results confirm that even short pieces of user-generated text can improve credit default predictions significantly. The additional information extracted by deep learning is robust towards controlling for credit scores, structured application information, and common theoretically suggested text characteristics.
Keywords: Deep learning, natural language processing, credit risk, fintechs
JEL Classification: G21, C14, C45
Suggested Citation: Suggested Citation