Determinants of Trust in Mexican Banks: Machine-Learning Analysis of the 2021–2024 Financial Inclusion Surveys

Abstract

Public confidence in banks' deposit protection and compliance with obligations is essential to foster financial inclusion. Accordingly, this research studies individual-level predictors, regional/national infrastructure and safety indicators, demographic and socioeconomic covariates, focusing on the 2021 and 2024 surveys of Mexico's National Financial Inclusion Survey (ENIF). The research question is: "What factors predict Bank Trust (TIB) among Mexican adults?" An end-to-end machine learning process was employed using Random Forest, XGBoost, LightGBM, and CatBoost set frameworks with stacking, and a survey-weighted logistic regression specification, to address statistical and predictive relevance. The variables which augment confidence in a bank are related to a higher financial literacy rate, better risk diversification strategies, greater per capita incomes per household, as well as proximity to banks. Age and internet access are also important variables for such assessments. The CatBoost model has a robust AUC value for the accuracy recovery of 0.734. The proposed study contributes to two different major objectives: (i) the application of a combination of econometric models with modern machine learning algorithms.