Thuy Tu Pham

Type of the article: Research Article

Abstract
Digital financial inclusion (DFI) has become a critical driver of sustainable growth and financial resilience in the digital era, yet its implications for bank stability remain ambiguous, particularly across heterogeneous institutional contexts. This study examines whether and under what conditions DFI fosters bank stability, using data from 65 emerging and advanced economies during 2010–2022. Employing Double Machine Learning (DML) and Causal Forests to address endogeneity and treatment heterogeneity, together with Panel Threshold Regression (PTR) to capture nonlinear dynamics, the paper provides a causal and structural assessment of the DFI–stability nexus. Results reveal that, on average, DFI exerts no statistically significant impact on bank stability across the full sample. However, substantial heterogeneity emerges in financially developed and institutionally strong economies, DFI significantly enhances stability (CATE = +0.0165, p < 0.001), while in underdeveloped systems it weakens it (CATE = –0.0082, p < 0.001). The PTR model identifies a critical DFI threshold (–1.3611), below which DFI undermines stability and above which its effect becomes neutral, confirming nonlinear regime behavior. These findings highlight that DFI alone cannot guarantee stability; its benefits materialize only within robust institutional and financial ecosystems. Methodologically, the integration of causal machine learning and threshold modeling offers a novel framework for examining digital finance policies and contributes to a deeper understanding of conditional digital effectiveness in modern banking systems.

Type of the article: Research Article

Abstract
The study analyzes the role of the Funding Gap (FGAP) as a dynamic structural liquidity indicator that influences bank financial stability in emerging markets, particularly amid heightened post-COVID-19 financial volatility. It aims to forecast banking stability by integrating advanced econometric and machine-learning techniques using a balanced panel dataset of 63 commercial banks from six ASEAN countries over the period 2010–2023. The methodological framework combines Ridge regression for variable selection, Particle Swarm Optimization (PSO) for hyperparameter tuning, and SHapley Additive exPlanations (SHAP) for interpretability within a Gradient Boosting model. The PSO-optimized specification achieves an R2 of 92.2%, substantially outperforming traditional fixed-effects and random-effects regressions. Empirical results indicate that persistent negative FGAP values significantly reduce Z-scores, confirming that structural liquidity imbalances constitute a key transmission channel from funding stress to systemic fragility. The analysis further reveals the moderating role of macroeconomic shocks, particularly inflation and the COVID-19 pandemic, in amplifying liquidity-induced instability. The proposed framework functions as an operational early warning system that enhances forecasting accuracy, model interpretability, and regulatory transparency, while repositioning FGAP as a forward-looking liquidity metric and offering both theoretical and practical contributions to financial risk management and supervisory practices in emerging economies.