Bank Failure Prediction: A Deep Learning Approach

Abstract

As with any business, the bankruptcy of a bank manifests itself in the cessation of payments. At this point, the bank must be closed and put into liquidation. Such a situation would cause considerable prejudice to the customers and to to the economy as a whole. Hence, over the years there has been many works in the literature in regards to bankruptcy prediction. However research exploring deep learning for this problem are scares. This paper presents an empirical study of banks failures prediction, by proposing the use of deep learning in comparison with traditional methods. Using a sample of 1100 of FDIC-insured US commercial banks, over a 14-year period from 2004 to 2018, we extracted and constructed 40 performance ratios known to have an influence on banks performance and forecasting bankruptcy. We then investigated the efficiency of prediction techniques already used in the literature and the performance of a Deep Autoencoder in relation to these methods. Experimental results prove that our proposed model, based on a deep neural network, outperforms the typical statistical and machine learning methods, in terms of the Matthews Correlation Coefficient and F1Score.