Predicting financial distress of banks using random subspace ensembles of support vector machines

Abstract

Models for financial distress predictions of banks are increasingly important tools used as early warning signals for the whole banking systems. In this study, a model based on random subspace method is proposed to predict investment/non-investment rating grades of U.S. banks. We show that support vector machines can be effectively used as base learners in the meta-learning model. We argue that both financial and non-financial (sentiment) information are important categories of determinants in financial distress prediction. We show that this is true for both banks and other companies.