A systematic literature review of bank failure prediction models

Jev, Japheth and Akanbi, Lukman Adewale and Bilal, Muhammad and Sukumar, Arun and Babatunde, Simeon A. and Owolabi, Hakeem and Kolovos, Vasileios (2026) A systematic literature review of bank failure prediction models. Finance Research Open, 2 (4). p. 100186. ISSN 3050-7006

Preview

Text 1-s2.0-S3050700626000976-main.pdf - Published Version Available under License Creative Commons Attribution. Download (2MB)

Abstract

Following the Global Financial Crisis of 2008–2009, bank-failure prediction and Early Warning Systems (EWS) have received renewed attention from researchers and banking regulators seeking to identify vulnerabilities in financial institutions before failure occurs. However, the literature employs heterogeneous definitions of bank failure, predictor variables, modelling approaches, prediction horizons, validation designs and performance measures, making comparison and interpretation of reported predictive performance challenging.
This systematic literature review identifies, evaluates and synthesises a focused body of post-Global Financial Crisis research on bank-failure prediction models, with particular attention to the methodological characteristics, validation practices and comparative predictive performance of conventional econometric and artificial intelligence/machine learning (AI/ML) approaches. Following a PRISMA-informed study-selection process, a sequentially refined search of the Scopus database identified 23 publications meeting the review criteria, comprising 21 primary empirical studies and two review studies. The narrative synthesis examined model type, operational definition of bank failure, data and predictor variables, prediction horizon, validation design, performance metrics and comparative predictive performance.
The findings reveal substantial methodological heterogeneity across the empirical literature. Bank-level financial and accounting variables remain prominent predictors, while broader macroeconomic and non-financial risk factors are less consistently represented. Several studies reported competitive or higher predictive performance for AI/ML approaches relative to conventional econometric models; however, the magnitude and consistency of these differences varied across datasets, failure definitions, prediction horizons, class distributions, evaluation metrics and validation designs. Greater interpretive weight was therefore placed on within-study comparisons conducted under broadly comparable empirical conditions. Overall, the evidence indicates that AI/ML methods can provide incremental predictive value in particular settings but does not establish their universal superiority over conventional econometric approaches. The review further identifies limitations in the treatment of class imbalance, model validation, performance evaluation and model interpretability. These findings support the need for more transparent and methodologically comparable evaluation frameworks and provide directions for future research on robust and regulator-relevant bank-failure Early Warning Systems.

Item Type:	Article
Identification Number:	10.1016/j.finr.2026.100186
Dates:	Date Event 13 October 2026 Accepted 19 September 2026 Published Online
Uncontrolled Keywords:	Banking, Bank failures, Artificial intelligence, Banking crisis, Bank regulation, Ensemble AI, Machine learning, Early Warning System
Subjects:	CAH17 - business and management > CAH17-01 - business and management > CAH17-01-02 - business studies
Divisions:	Business School > Management, Business and Marketing
Depositing User:	Gemma Tonks
Date Deposited:	06 Nov 2026 12:05
Last Modified:	06 Nov 2026 12:05
URI:	https://www.open-access.bcu.ac.uk/id/eprint/17298

Actions (login required)

View Item