DOI: https://doi.org/10.36719/2706-6185/53/76-80
Rafail Bayramli
Akdeniz University
PhD student
https://orcid.org/0000-0002-6186-8714
rafael.bayramli.98@gmail.com
Bankruptcy Prediction Models
Abstract
This paper provides a comprehensive review of the evolution of bankruptcy prediction and financial distress models, tracing the transition from traditional accounting-based static models to dynamic, market-driven, and artificial intelligence-enhanced approaches. Key milestones include the discrete-time hazard model, which corrects biases in classical logit models by using multi-year panel data, and structural models rooted in the Black–Scholes–Merton framework, which consistently outperform accounting-based benchmarks by exploiting forward-looking market information. The review highlights the growing importance of firm structural characteristics (size, diversification) and, particularly in emerging markets such as China, the need for regional systemic risk early-warning systems. Recent literature demonstrates that deep learning and big-data techniques achieve superior accuracy and earlier detection by capturing complex non-linear patterns across accounting, market, macroeconomic, and alternative data sources. The synthesis concludes that the most effective contemporary bankruptcy prediction frameworks integrate the theoretical foundation of structural models, the dynamic richness of hazard specifications, market-based signals, and the representational power of deep neural networks, paving the way for real-time, interpretable, and contagion-aware systems in the future.
Keywords: bankruptcy prediction, distance to default, Merton model, BSM-Prob, KMV model, hazard model, Shumway model, deep learning, financial risk early-warning, China financial regulation