Analyses on digital financial services inclusion and adoption with cross-country macro panel and micro-level data

Abstract

This dissertation presents two sides of digital financial services inclusion in macro and microanalysis, dividing them into two sub-studies. The first study focused on identifying the determinants and growth of digital financial services inclusion through cross-country analysis using panel data. Also, there is a division of countries by income into three groups to understand the influence of variables and determine the main trends of these factors. The study identified five determinants from previous studies and six new additional factors, so 11 hypotheses were developed. Pooled OLS, LSDV, linear regression and sigmoid growth curve function are used as research models to test hypotheses. The study collected data two-years (2014 & 2017) for 123 countries. As a result, the study found that GDP per capita, number of ATM machines, basic skills, mobile coverage, mobile broadband, mobile device prices, political stability, control of corruption, online security and fixed broadband were all demonstrated at significant levels, and only a factor like an affordable mobile internet tariff did not show much importance in all research models. Overall, the adjusted r-squared values were greater than 0.8 in the pooled OLS and LSDV regressions, while the sigmoid growth curve function had 0.86 due to the training of each country's data in the neural network with TensorFlow by 20000 epochs. Moreover, the s-shaped growth function showed that most high-income countries are already in the final stages of DFS inclusion, while upper-middle and lower-middle income countries are in the early stages or only halfway through DFS inclusion. The second study focuses on micro-level analysis, in particular, of the personal behavior of digital financial service users during a pandemic. The study used the UTAUT model, by extending with two constructs: fear of COVID-19 as a moderator and social isolation as a direct factor in determining behavioral intentions. A survey method is used, in which primary data is collected more than 400 participants aged 18 and over. This study uses structural equation modeling (SEM) to test constructs and hypotheses, analyze relationships, and determine the overall fit of the model. The study used factor analysis (EFA, CFA) and covariance-based SEM. As a result, the study found that all UTAUT constructs and social isolation positively influence behavioral intent and actual use of DFS during a pandemic. Testing the moderating role of COVID fear for other constructs plays a positive role, with the exception of the relationship between expected performance and behavior intensity and the relationship between behavioral intent and actual use.