Table 1.
Sample demographics.
Fig 1.
Degree of financial digitalization: Financial online activities (% of internet users).
Fig 1 illustrates the degree to which consumers use various financial services: receiving emails, paying bills, making payments and checking the account balances.
Fig 2.
Dimensions of the financial digitalization.
Fig 2 plots the dimensions of bank customers’ digitalization: adoption of digital banking, diversification of use, and adoption of bank and non-bank payment instruments.
Fig 3.
Consumers classification by dimensions (number of surveyed individuals).
Fig 3 reports the number of surveyed participants for each dimension considered. The total number of surveyed participants is 3,005.
Table 2.
Models’ performance.
Table 3.
Random forest hyperparameters and cross-validation of the algorithm.
Fig 4.
Variable importance for the random forest model on online banking adoption.
Fig 5.
Variable importance for the random forest model on diversification of online banking uses.
Fig 6.
Variable importance for the random forest model on diversification of mobile banking uses.
Fig 7.
Variable importance for the random forest model on debit card adoption.
Fig 8.
Variable importance for the random forest model on credit card adoption.
Fig 9.
Variable importance for the random forest model on adoption of non-bank payment methods.
Figs 4 to 9 report the plots showing the relative statistical importance of each feature in the classification. The left-hand side graph shows the Top 20 features by Mean Decrease in Accuracy. The centered graph shows the Top 20 features by Mean Decrease in Gini. The righ-hand side graph shows the Top 20 features by Total Score [109].
Fig 10.
Fig 10 plots the predicted accuracy of the random forest algorithm across the different subsamples: based on gender (male vs female), age (young vs old) and habitat (rural vs urban areas).
Fig 11.
Tree: Adoption of digital banking.
Fig 12.
Tree: Diversity of digital use—online banking.
Fig 13.
Tree: Diversity of digital use–mobile banking.
Fig 14.
Tree: Debit card use.
Fig 15.
Tree: Credit card use.
Fig 16.
Tree: Use of non-bank payment instruments.
Figs 11 to 16 plot the decision trees of bank customer digitalization by estimating a conditional inference tree using those features having the largest predictive power according to the random forest algorithm.
Fig 17.
Average treatment effects using causal forests.
Fig 17 shows the average treatment effect estimations (ATEs) computed using the causal forest algorithm. The ATEs are shown for each dimension of bamk customers’ digitalization and for those variables with the largest predictive power according to the random forest.
Fig 18.
Subsample analysis of supply-side explanations.
Fig 18 reports the relative importance—measured by mean decrease in accuracy—of those variables with the largest predictive power for the adoption of online banking by banks’ characteristics: size (large banks’ customers—Santander, BBVA, and CaixaBank—vs other banks’ customers) and branch closure (customers whose main bank closed at least one branch in their province vs customers whose main bank has not closed any branch in their province). The bottom panel shows the predicted accuracy.
Fig 19.
Sample matrix (Heatmap) by dimensions and socio-demographics features.
Fig 19 shows the main characteristics of the survey participants by degree of digitalization, degree of financial digitalization, perceptions on mobile and online banking and the use of non-bank services and social networks. Results are presented by gender, age, and employment situation. Each cell represents the percentage of people over the total number of people belong to this category.