Machine Learning Models for Bankruptcy Prediction in Lebanon
1 viewsAbstract
Predicting corporate insolvency is a critical component of financial risk management, particularly in economies experiencing severe macroeconomic shocks. Following the unprecedented 2019 financial collapse in Lebanon, traditional statistical tools for bankruptcy prediction have proven increasingly inadequate. This study investigates the applicability and necessary adaptation of ensemble machine learning (ML) architectures—specifically gradient-boosted algorithms like XGBoost and Random Forests—for assessing corporate bankruptcy risk within Lebanon's highly informal, dollarized economy. Utilizing a mixed-method approach, the research applies the Altman Emerging Market Z''-Score to a cross-section of Lebanese firms (2018–2023) while reviewing contemporary empirical ML literature. The findings identify a critical "Lebanese Paradox": numerous firms exhibiting highly distressed quantitative Z'' scores successfully avoid insolvency by leveraging mechanisms invisible to traditional models, such as unrecorded hard- currency revenues and diaspora capital injections. Consequently, the study concludes that raw ML models trained on standard financial ratios will yield systematically high false-positive rates in Lebanon. To achieve the 82–95% accuracy observed in international studies, a Lebanon-calibrated ML model must inherently integrate macroeconomic stress variables, parallel exchange rates, and shadow economy proxies. Ultimately, while machine learning offers superior diagnostic capabilities for investors and creditors, its successful implementation requires a fundamental restructuring of input variables to accurately reflect the realities of crisis-stricken, unstructured economies.
Keywords
Altman Z-Score, Bankruptcy Prediction, Corporate Insolvency, Ensemble Models, Financial Risk Management, Lebanese Financial Crisis, Machine Learning, Shadow Economy