In recent times there has been a rapid development of Deep Learning techniques capable of extracting information from unstructured sources of data such as images, sound and text. In our presentation, we explore the application of the latest Deep Learning techniques to assess the feasibility of automating some of the Micro and SME lending process. We apply the most recent developments from the ﬁeld of Deep Learning including the current state of the art in Natural Language Processing (NLP), the Google BERT model, for the prediction of loan default over 40,000 Micro and SME loans. Our initial results suggest that the text loan assessment are surprisingly predictive; however, when combined with traditional credit scoring variables, no additional performance improvement is gained in terms of AUC of Accuracy. Despite no observed increase in the performance metrics, we ﬁnd that the inclusion of the text results in better-calibrated predicted probability outputs, which in practice leads to more robust and interpretable results for credit lenders.
Cristián Bravo is Associate Professor and Canada Research Chair in Banking and Insurance Analytics at Western University, Canada. Previously he served as Associate Professor of Business Analytics at Department of Decision Analytics and Risk, University of Southampton, Research Fellow at KU Leuven, Belgium; and as Research Director at the Finance Centre, Universidad de Chile. His research focuses on the development and application of data science methodologies in the context of credit risk analytics, in areas such as deep learning and text analytics.