Cross-Market Bankruptcy Prediction: An Interpretable Ensemble Learning Framework Using SHAP Analysis

The increasing complexity and interconnectedness of global financial markets demand more robust and transparent methods for predicting corporate failures across diverse environments. This study introduces a novel computational framework for cross-market bankruptcy prediction that integrates ensemble learning and interpretable artificial intelligence. Our approach combines Natural Gradient Boosting (NGBoost), Categorical Boosting (CatBoost), and Light Gradient Boosting (LightGBM), leveraging the strengths of each algorithm, and incorporates Shapley Additive Explanations (SHAP) for enhanced interpretability. By integrating capital structure optimization principles from trade-off theory with Shapley value frameworks, we establish a three-phase computational process adaptable to market-specific characteristics. Empirical validation using comprehensive financial data from US and Chinese markets (2015–2022), encompassing the COVID-19 crisis period, demonstrates the superior predictive performance of our framework. Our ensemble approach achieves 70-90% prediction accuracy across all the markets and periods. This demonstrates robust performance stability even during the COVID-19 crisis. The SHAP analysis provides granular insights into market-specific bankruptcy indicators, revealing the distinct influence of factors like operational efficiency in the US and leverage in China. These findings underscore the framework’s effectiveness in identifying subtle financial distress signals across heterogeneous market structures and during periods of economic volatility, offering valuable insights for risk management and investment strategies.