Predicting Bank Insolvency with Random Forest Classification

Sammanfattning: The aim of this paper is to evaluate a machine learning technique, Random Forest, to predict default rates for banks in the United States. This study extends the findings of a Random Forest model first introduced by Petropoulos et al. (2017) by extending their model by evaluating a longer sample period and adding macroeconomic variables to analyze how current market conditions impact the prediction of default rates of U.S. Banks from 1994-2016. Petropoulous et al. (2017) evaluated multiple traditional and artificial intelligence models to find that Random Forest produced the best results utilizing quarterly data from the FDIC from 2008-2014. Numerous studies have suggested that the financial condition of banks is purely determined by bank specific variables. Our empirical results confirm that theory as a bank default prediction model utilizing Random Forest classification performed worse with the addition of macroeconomic variables when compared to a model based purely on bank specific variables.