Crisis Identification and Prediction using Machine Learning: The Case of U.S. Regional Banks

Abstract

Financial crises have impacts on the economy and society, so their identification and prediction are of interest to investors and policymakers alike. Historically, many financial crises originated in the banking sector. Using a return-based and data-driven crisis definition, we combine clustering, ridge regression, and sequential feature selection to identify and predict U.S. regional banking crises. In addition to a statistical evaluation of the forecasting ability of our approach, we analyze its impact on a trading strategy for a broad range of risk attitudes. Its predictions can be used to avoid market exposure during negative market phases, which improves the risk-adjusted returns of the trading strategy compared to the buy-and-hold benchmark.

Keywords: Finance, Machine Learning, Forecasting

JEL Classification: G01