Credit growth, institutional transparency, and bank risk under policy uncertainty: A double machine learning analysis

This study examines the relationship between credit growth and bank risk, measured by non-performing loans (NPL), under the conditional influence of institutional quality and global economic policy uncertainty. Anchored in the Institutional-Resilient Allocation Channel (IRAC) framework, it applies Double Machine Learning with moderated mediation analysis to capture nonlinear, context-dependent effects across 93 countries from 2008 to 2023. The results suggest that credit growth generally reduces NPL but reveal counter-intuitive dynamics: the effect is weaker in strong-institution settings and stronger under weak governance, consistent with a Transparency Channel whereby robust institutions enforce stricter NPL recognition. Conversely, during high uncertainty, credit growth exhibits greater stabilizing power, reflecting a Flight-to-Quality Channel as banks allocate lending to safer borrowers. A novel Causal Machine Learning Index quantifies resilience across institutional-policy regimes. These findings refine conventional theories, demonstrating that credit-risk linkages are inherently nonlinear, adaptive, and shaped by transparency and behavioral prudence under uncertainty.