Answers are grounded in this paper and cite their sources.
This study develops an early‐warning framework for South African bank failure using 570 bank‐year observations for 19 banks from 1991 to 2020 and five verified failures. Conventional and Firth logistic regression, time‐varying Cox models and Bayesian model averaging (BMA) are examined across 1‐, 2‐ and 3‐year forecast horizons. Expanding‐window walk‐forward validation distinguishes full‐sample fit from prospective predictive performance. Bank‐specific fundamentals provide the most persistent warning signals, particularly at the 2‐ and 3‐year horizons, with profitability, liquidity and management quality emerging as the most prominent indicators of vulnerability. Profitability is negatively associated with failure, while liquidity and management quality receive stronger support at longer horizons. Macroeconomic information contributes selectively, with exchange‐rate conditions particularly relevant at the short horizon. The combined BMA‐logit model performs strongest at 1 year, while bank‐only BMA‐logit performs strongest at 2 and 3 years. The findings support a parsimonious, model‐uncertainty‐aware framework for bank‐failure monitoring in a rare‐event emerging‐market setting.
No takes yet. Share an insight, caveat, or question.
Sonono et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: