Constructing Early Warning Indicators for Banks Using Machine Learning Models

This study focuses on using supervised machine learning models to provide banks with early warnings of liquidity stress using market-based indicators. The study uses publicly available data from 2007 to 2021 to train the machine learning model and transform market stress into a classification problem. The St. Louis Fed Financial Stress Index was used to define the level of stress in the market, and an ensemble model with a random under-sampling boosting algorithm (RUSBoost) was used to improve predictions from imbalanced data. The study shows that the developed model can predict 83% of “red” risk days and can contribute to bank risk management.