Severe weather and peer-to-peer farmers’ loan default predictions: Evidence from machine learning analysis

Abstract

In theory, climate change affects farmers’ loan default risk because severe weather conditions caused by climate change negatively affect farmlands’ productivity, farmers’ income, and their ability to pay off their loans. In this study, using farmers’ loan data extracted from the Lending Club and U.S. severe weather data, we show that three machine learning algorithms—Artificial Neural Networks (ANNs), Gradient Boosting Trees, and Random Forest—are successful at loan default predictions with accuracies of 70%, 74% and 81%, respectively. Results from the Shapley Additive Explanations (SHAP) also offer evidence of the economic relevance of severe weather and other explanatory variables.