Deep Learning: Credit Default Prediction from User-Generated Text

22 Pages Posted: 15 Feb 2020

See all articles by Johannes Kriebel

Johannes Kriebel

University of Muenster

Lennart Stitz

University of Muenster

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

Kriebel, Johannes and Stitz, Lennart, Deep Learning: Credit Default Prediction from User-Generated Text (January 13, 2020). Available at SSRN: or

Johannes Kriebel (Contact Author)

University of Muenster ( email )

Universitätsstraße 14-16
Münster, D-48143

Lennart Stitz

University of Muenster ( email )

Schlossplatz 2
Muenster, D-48149

Here is the Coronavirus
related research on SSRN

Paper statistics

Abstract Views
PlumX Metrics