Forecasting bank failures and stress testing: A machine learning approach

摘要: Abstract This paper presents a forecasting model of bank failures based on machine-learning. The proposed methodology defines linear decision boundary that separates the solvent banks from those failed. setup generates novel alternative stress-testing tool. Our sample 1443 U.S. includes all 481 failed during period 2007–2013. set explanatory variables is selected using two-step feature selection procedure. were then fed to support vector machines model, through training–testing learning process. exhibits 99.22% overall accuracy and outperforms well-established Ohlson’s score.