Comparison of boosting and random forest models in forecasting bank failures : Revisiting the 2008 Financial Crisis from a Supervisory Perspective

Sen, Safa (2024) Comparison of boosting and random forest models in forecasting bank failures : Revisiting the 2008 Financial Crisis from a Supervisory Perspective. ECONOMY AND FINANCE: ENGLISH-LANGUAGE EDITION OF GAZDASÁG ÉS PÉNZÜGY, 11 (3). pp. 258-281. ISSN 2415-9379

Preview

Text 258-281SafaSenE.pdf - Published Version Download (1MB) | Preview

Abstract

This research paper delivers an exhaustive analysis of predictive models for bank failures, a subject of paramount importance for economic stability. Using a dataset from the Federal Deposit Insurance Corporation (FDIC), the study examines 950 banking institutions, including 60 that succumbed to the 2008 financial crisis. The paper employs binary classification analysis using 26 CAMEL ratios and compares boosting algorithms with the Random Forest model family. In classifying non-failed banks, Random Forest variations notably outperform boosting algorithms, achieving a 97% accuracy rate in correctly classified instances, with the Regularized Random Forest model showing exceptional precision with a rate of 0.988. In the context of predicting failed banks, the Random Forest models, particularly the regularized variant, demonstrate a strong capability for accurately identifying true failures. These findings corroborate the efficacy of Random Forest models in predicting bank failures precisely and reliably, highlighting their critical role in reducing false positives and negatives, which is essential for robust forecasting in the banking sector.

Actions (login required)

Edit Item