Evaluating the Non-linear Impact of Digital Literacy on Financial Inclusion: Evidence from Emerging Economies via Explainable Deep Learning
摘要
The link between digital literacy and financial inclusion is often oversimplified in existing studies that rely on linear frameworks. Such approaches frequently overlook the structural differences and non-linear patterns across various developing regions. In this study, we develop FERENet (Feature Ranking and Explainable Network), a hybrid model combining deep neural networks with random forests to analyze a dataset spanning 56 economies in Africa, Asia, and Latin America. Our results indicate that FERENet significantly outperforms traditional General Method of Moments (GMM) estimations, explaining 94% of the variance. By employing SHAP-based interpretability, we identify specific regional digital barriers that challenge the effectiveness of universal policy designs. The findings suggest that financial inclusion efforts should move away from generic literacy programs toward infrastructure-specific interventions that align with local economic conditions. This study provides a more nuanced understanding of how digital skills translate into financial access in constrained environments.