Title: Early warning system for preventing bank distress in Brazil

Authors: Flavio Barboza; Jorge Henrique de Frias Barbosa; Herbert Kimura; Gustavo Carvalho Santos; Paulo Cortez

Addresses: School of Business and Management, Federal University of Uberlândia, Santa Mônica Campus, 38408-100, Uberlândia – MG, Brazil ' Central Bank of Brazil, 70074-900, Brasília – DF, Brazil ' School of Accounting, Business and Economics, University of Brasília, Darcy Ribeiro Campus, 70910-900, Brasília – DF, Brazil ' School of Eletric Engineering, Federal University of Uberlândia, Santa Mônica Campus, 38408 – 100, Uberlândia – MG, Brazil ' Department of Information Systems, ALGORITMI Centre, University of Minho, 4804-533 Guimarães, Portugal

Abstract: The global financial crisis in 2007/2008 showed how important is to be prudent with events related to the banking sector, illustrating emphatically the contagion in the financial system caused by distress in one or more banks. This issue goes beyond competitiveness and the interrelationship among its members, requiring at least signs or warnings of potential problems in such institutions. Thus, the present study presents some early warning system models for bank crises and bank distress, which are empirically tested for Brazilian banks. In addition to the traditional logit, we analyse two machine learning techniques are: random forest (RF) and support vector machine (SVM). The database of Brazilian banks covers 179 events considered as unsound bank. Our findings suggest that RF and SVM underperform the logit model. Moreover, RF models presented greater predictive capacity with the time windows of 32 and 34 months, proving adequate to the regulators' needs.

Keywords: early warning system; EWS; banking crisis; distress prediction; machine learning techniques; Brazil.

DOI: 10.1504/IJBSR.2023.130632

International Journal of Business and Systems Research, 2023 Vol.17 No.3, pp.326 - 346

Received: 09 Oct 2020
Accepted: 19 Mar 2021
Published online: 02 May 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article