Machine Learning and Credit Risk: Empirical Evidence from small- and mid-sized businesses

In this paper, we compare two different approaches to estimate the credit risk for small- and mid-sized businesses (SMBs), namely a classic parametric approach, by fitting an ordered probit model, and a non-parametric approach, calibrating a machine learning historical random forest (HRF) model. The models are applied to a unique and proprietary dataset comprising granular firm-level quarterly data collected from a European investment bank and an international insurance company on a sample of 464 Italian SMBs over the period 2015–2017. Results show that the HRF approach outperforms the traditional ordered probit model, highlighting how advanced estimation methodologies that use machine learning techniques can be successfully implemented to predict SMB credit risk, i.e. when facing high asymmetries of information. Moreover, by using Shapley values, we are able to assess the relevance of each variable in predicting SMB credit risk.