Predicting Formal Financial Account Ownership Using Machine Learning: A Development Perspective

Abstract

Financial inclusion has rapidly increased globally, yet large gaps persist between developed and developing countries in the ownership of accounts at formal financial institutions. Although several cross-country studies exist, few have specifically studied financial inclusion in a development context. Even so, demand-side studies have been constrained to demographic and socioeconomic variables and have included measures of key enablers to financial inclusion using country-level indicators. This limits our understanding of important enablers and barriers to financial inclusion. Our study overcomes these constraints by exploiting rich surveys from Financial Inclusion Insights. Using machine learning, we take a data-driven approach to identifying variables that vitally contribute to the predictive performance of our Random Forest and LASSO classifiers. We then compare our findings to hypotheses about perceived enablers and barriers to financial inclusion. Our analysis reveals that contrary to prior studies, various demographic and socioeconomic characteristics such as literacy, living in an urban area, gender, and poverty status are not consistently predictive of financial inclusion. Instead, we find that the most important predictors are distance to financial points of service, trust in financial service providers, and consistent sources of income, such as regular government-to-person and peer-to-peer transfers. Despite important variation in predictors across the markets examined, each of these indicators is the first or second most important predictor of financial inclusion in every country in our sample. This suggests data captured in individual-level questionnaires is meaningful for understanding priority enablers and barriers to financial account access. These results also provide policymakers and industry practitioners with evidence of actionable ways to increase financial inclusion. This research is joint work with Seth Garz.