Causal and explainable machine learning insights into ROA’s impact on bank stability amid liquidity and pandemic shocks

Abstract

Amid growing systemic vulnerabilities in emerging financial markets, this study examines the causal impact of bank profitability, proxied by Return on Assets (ROA), on institutional stability, measured by the Z-score, under liquidity stress and pandemic-induced shocks. Using data from 63 commercial banks across six ASEAN countries (2010–2023), we develop a hybrid framework integrating causal inference, machine learning, and model interpretability. Regularized regressions (Lasso, Ridge, Elastic Net) enhance feature selection, while Double Machine Learning (DML) estimates both average and conditional treatment effects. Optimized via Particle Swarm Optimization (PSO), CatBoost and XGBoost models achieve R² above 0.93. SHAP-based analysis reveals nonlinear interactions between ROA, liquidity gaps (FGAP), and COVID-19, indicating that profitability’s stabilizing role weakens under high liquidity stress. By bridging explainable AI and causal reasoning, this study offers a novel, interpretable framework for early-warning systems and prudential supervision in volatile and capacityconstrained financial environments.