Evolution of Early Warning Models for Bank Failures: A Bibliometric Analysis

Abstract

This bibliometric analysis examines the progression of early warning models in finance literature for predicting banking failures. Using the biblioshiny software in R and RStudio program, this study examined 40 English-language publications from 1989 to 2023 to identify the knowledge development and dissemination of early warning models. Initially, 1441 publications were downloaded from the SCOPUS database using the search criteria “Early Warning Models” and “Banking Failure” and the non-journal publications and articles written other than the English language equal to 1401 were excluded from the analysis. The data were reviewed and peer-reviewed before the bibliometric analysis and the systematic literature review. The results highlight that the application of machine learning models in predicting banking failures became prominent in 2010 and continued with simulation approaches in the next decade due to their higher level of accuracy in out-sample performances. At the same time, conventional logit, probit, signal extraction, and discriminant analysis were extensively used over time. The results contribute to the existing literature by identifying complex early warning models employed to forecast different scenarios of bank failures. The study suggests that terms of investment and financial variables combining the macro-network and institutional risk indicators outperform the models that merely included financial variables. Hence, relying only on balance sheet performance to predict bank failures may be futile. This study guides financial policymakers, analysts in macro-prudential surveillance units of Central Banks, bank managers, and practitioners in rethinking, reconstructing, and reevaluating existing early warning models by incorporating advanced techniques based on machine learning and simulation language.