## wpiea2021167-print-pdf

## Source details

**Canonical URL:** [wpiea2021167-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021167-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021167-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021167-print-pdf.pdf.json)

---

### I. INTRODUCTION — Overview and contributions
- Digital financial services (DFSs), enabled by fintech, have become an important driver of financial inclusion in emerging markets and developing economies (EMDEs).
- Two key contributions of the paper:
  - Cross-country examination of digital financial inclusion on economic growth.
  - Exploration of key drivers of digital financial inclusion using new indices developed by Khera et. al. (2021) across 52 EMDEs across time.

### Empirical strategy and methodology
- Outcome variable:
  - Growth rate of the real per capita gross domestic product (GDP).
- Main explanatory variable:
  - Level of usage of DFSs measured by the digital financial usage index in Khera et. al. (2021).
- Conditioning variables:
  - Broad set of macroeconomic, socio-economic, and financial sector indicators and a measure of traditional financial inclusion.
- Identification approaches over 2011–18 for a sample of 52 countries:
  - (i) Cross-country OLS regression of GDP per capita growth indicators (averaged over 2011–18) on digital financial usage index.
  - (ii) Use digital financial usage in 2011 to predict subsequent average growth over 2011–18 to mitigate reverse causality.
  - (iii) Cross-country instrumental variable estimator (LIML) to extract exogenous components of the digital financial usage index. Instruments: access to mobile money agents and access to the internet.

### Core empirical findings on growth
- Digital financial usage has a significantly positive relationship with GDP per capita growth.
- The exogenous component of digital financial usage (from IV/LIML) has a strong positive relationship with long-run growth.
- Key quantitative estimates (LIML, 2011–18):
  - A one standard deviation increase in digital financial usage (equivalent to 0.1205 increase in the index) is expected to boost the annual real GDP growth rate by 1.61 percentage points.
  - A one standard deviation increase in digital financial usage is expected to boost growth by 0.98 standard deviations on average (Table 2, regression over 2011–18).
  - The increase in digital financial inclusion from the 25th percentile to the 75th percentile is associated with an increase in average economic growth of up to 2.2 percentage points (Figure 5).
- LIML diagnostics and robustness:
  - Instruments (log(mobile money agents per 100,000 adults) and percent of population with access to the internet) are statistically significant predictors of average digital usage in first-stage regressions.
  - F-statistics for excluded instruments exceed Stock and Yogo’s (2005) cutoffs in many specifications (rejecting weak instruments when restricting IV bias to 15 percent of OLS bias).
  - Anderson–Rubin test confirms statistically significant partial correlation between the endogenous variable (digital financial usage) and growth.
  - Hansen-J overidentification tests are passed in reported cases (p-values reported in tables).

### II. NEW INDICES OF FINANCIAL INCLUSION: construction and stylized facts
- Indices (Khera et. al., 2021) for 2014 and 2017 covering 52 EMDEs incorporate digital aspects into financial inclusion measures.
- Three-stage principal component analysis (PCA):
  1. Construct “access” and “usage” sub-indices to capture supply-side and demand-side aspects.
  2. Combine access and usage sub-indices into “traditional” and “digital” financial inclusion indices.
  3. Construct an overall “comprehensive financial inclusion index” reflecting both traditional and digital aspects.
- Selected variable weights (as presented):
  - Number of ATMs per 100,000 adults — Weight in Traditional access: 0.25
  - Number of Branches per 100,000 adults — Weight in Traditional access: 0.25 (implied)
  - Number of registered mobile money agents per 100,000 adults — Weight in Digital access: 0.25
  - Mobile subscription per 100 people — Weight in Digital access: 0.125
  - % of population who has access to internet — Weight in Digital access: 0.125

### Regional patterns and changes (2014 to 2017)
- Traditional financial inclusion is relatively high in Asia and the Pacific, Latin America and the Caribbean, and Emerging Europe.
- Traditional financial inclusion index remained broadly unchanged between 2014 and 2017; eight countries experienced a decline.
- Digital financial inclusion is relatively high in Africa and Asia and the Pacific.
- Most countries saw an increase in digital financial inclusion index between 2014 and 2017; improvements particularly large in African countries (e.g., Ghana, Benin, and Senegal) and relatively muted in some countries in Latin America and the Caribbean, and Middle East and Central Asia.
- Comprehensive financial inclusion improves significantly for countries with high digital but low traditional inclusion (examples: Senegal and Myanmar). China, Turkey and Brazil show high inclusion by both measures.
- Comprehensive financial inclusion index improved for Botswana, El Salvador, Mexico, Nigeria, Romania, Rwanda, South Africa, and Zimbabwe between 2014 and 2017.

### Drivers of digital financial inclusion — approach and estimators
- Two empirical models:
  - Random effects panel regression (feasible GLS) with mobile agents per capita (mma) as dependent variable to identify supply-side drivers.
  - Fractional logit panel regression for the digital usage index (퐹퐼_F_u) to identify demand-side drivers (handles dependent variable in [0,1], including 0 and 1).
- Baseline access regression (simplified):
  - mma_it = α0 + γ_t + α1(퐹퐼_F_u)_{i,t−3} + α2(퐹퐼_T_a)_{i,t−3} + α3(퐹퐼_T_u)_{i,t−3} + Σ β_k X_{k,i,t−1} + ε_it.
- Baseline usage regression (simplified):
  - (퐹퐼_F_u)_{i,t} = θ0 + φ_t + θ1(mma_{i,t}) + θ2(퐹퐼_F_a)_{i,t−3} + θ3(퐹퐼_T_a)_{i,t−3} + θ4(퐹퐼_T_u)_{i,t−3} + Σ δ_k X_{k,i,t−1} + v_{i,t}.
- Explanatory variables (lagged where applicable):
  - log GDP per capita; financial sector efficiency (overhead costs to assets); bank concentration (assets share of three largest banks); financial stability (log of NPL as % of gross loans); rule of law (0–1); share of urban population; digital infrastructure measures.

### Main findings on drivers (demand and supply)
- Demand-side:
  - Digital payment usage is higher where a culture of using financial services exists: traditional usage index positive and significant.
  - Digital usage is higher where access to traditional financial institutions is constrained: traditional access index negative and significant — interpreted as fintech “filling the gap.”
  - Access to foundational digital infrastructure (mobile phones, mobile data services, broadband internet) and higher financial literacy/familiarity facilitate usage.
- Supply-side:
  - Greater competition among traditional providers (lower bank concentration) is associated with higher numbers of mobile money agents.
  - Bank inefficiencies (higher overhead costs to assets) are associated with higher numbers of mobile agents—implying profitable opportunities for fintech providers when banks are inefficient.
  - Higher quality of institutions has a positive impact on mobile agent supply in several specifications.
- Interaction effects:
  - Traditional access negatively and significantly affects digital usage, while traditional usage positively affects digital usage — implying substitution where physical access is limited but complementarity where usage norms exist.
- Statistical evidence (selected reported coefficients, preserved verbatim):
  - Mobile money agents (t-1) coefficient examples in digital usage regressions: 0.00302*** with standard error (0.00061) in column (1) (Table 4).
  - Traditional usage coefficient examples in digital usage regressions: 4.692*** with standard error (0.870) in column (1) (Table 4).
  - Traditional access coefficient examples in digital usage regressions: -3.658*** with standard error (1.359) in column (1) (Table 4).
  - Log(GDP per capita) (t-1) coefficient example in mobile agents regressions: -121.6*** with standard error (35.95) in column (3) (Table 5).

### Robustness checks and alternative specifications
- Methods and checks performed:
  - Pooled OLS instead of random effects for Equation (3).
  - Additional explanatory variables tested (found insignificant): GSMA mobile money regulatory index; fixed broadband internet monthly subscription rate; percent of gross secondary school enrollment.
  - Inclusion of lags to avoid multicollinearity between financial inclusion sub-indices and other explanatory variables.
  - Controlling for aggregate trends over time made the coefficient on access to digital infrastructure smaller and insignificant in some specifications.
  - Coefficients remain significant when regional dummies are included; on average, countries in Africa have significantly higher usage of DFSs compared to other regions.
  - Weak identification statistics improve when two aggregated instruments are used instead of a single generated instrument.
- Appendix confirmatory results:
  - Appendix Table I.1 and I.2 present multiple OLS specifications with digital usage F.I Index coefficients across columns (selected values preserved verbatim, e.g., 0.228** (0.0939); 0.130** (0.0536); 0.152* (0.0787)).
  - Random effects estimates for digital usage (Appendix II) show Trad. access FI Index (t-3) negative and significant in many columns (e.g., -0.527*** (0.171)), Trad. usage FI Index (t-3) positive and significant (e.g., 0.779*** (0.139)), and Mobile Money Agents (t-1) positive and significant (e.g., 0.000563*** (0.000102)). N reported: 96, 96, 80, 79, 79, 79, 79, 79, 96, 95, 95, 95, 95.

### Interpretation, limitations, and causality caveat
- Usage of DFSs is positively associated with economic growth in cross-country OLS specifications and is a statistically significant predictor for subsequent growth in some initial-period specifications (notably 2014–2018).
- Causality is not definitively established: reverse causation from GDP per capita growth to digital financial inclusion is possible; digital financial services may develop in anticipation of future economic activity (financial inclusion may be a leading indicator rather than a fundamental cause).
- Data limitations noted (e.g., assumptions about relative measures of digital financial inclusion across countries between 2011 and 2014). Emphasis is on usage (not merely access) because usage affects savings and consumption smoothing channels relevant for growth.

### Policy implications and risks (high-level priorities)
- Tailor policies to country context:
  - Countries with high traditional usage but low access: encourage technological innovation in payments to fill bank infrastructure gaps; high potential for fintech to spur inclusion.
  - Countries with high traditional access but low usage: prioritize improving financial literacy and trust in the financial system; address social barriers including high informality.
  - Countries with high voluntary and involuntary exclusion: tackle social and economic barriers and encourage competition and fintech entry.
- Demand-inducing measures needed to translate supply into inclusion:
  - Financial and digital literacy programs.
  - Consumer protection.
  - Ensuring cybersecurity to promote trust in financial services.
- Regulatory and supervisory priorities:
  - Strengthen consumer protection and crisis management tools where DFSs expand without such safeguards.
  - Adapt regulatory frameworks to fintech-related risks, including cyber security risks, misuse of payment channels for money laundering and terrorist financing, and liquidity risks.
  - Recognize that fintech firms often rely on traditional banks to store customer funds, posing potential insolvency and custodial risks.
- Address inequality in access to technology (mobile phone, the internet, electricity, digital ID) to prevent a digital divide within and across countries.
- Filling regulatory gaps is increasingly important as technology applications expand from payments to digital credit and other financial services.

*Content derived from wpiea2021167-print-pdf (selected sections, tables, and appendices) as provided.*

### References _______________________________________________________________ 24

### I.   INTRODUCTION

### Overview
- Digital financial services (DFSs), enabled by fintech, have become an important driver of financial inclusion in emerging markets and developing economies (EMDEs).
- Country-based case studies and regional studies cited: Jack and Suri, 2011; 2014; Tarazi and Breloff, 2010; Sy et. al., 2019; Berkmen et. al., 2019; Loukoianova et al, 2019; Lukonga, 2018; Blancher et al., 2019.
- Adapting existing measurement methodology, Sahay et. al. (2020) and Khera et. al. (2021) quantify progress in digital financial inclusion and find advancement in most of the EMDEs between 2014 and 2017, even where traditional financial inclusion retreated.
- This paper’s two key contributions:
  - Cross-country examination of digital financial inclusion on economic growth.
  - Exploration of key drivers of digital financial inclusion using new indices developed by Khera et. al. (2021) across 52 EMDEs across time.

### Empirical strategy and methodology
- Outcome variable: growth rate of the real per capita gross domestic product (GDP).
- Main explanatory variable: level of usage of DFSs measured by the digital financial usage index in Khera et. al. (2021).
- Conditioning variables include a broad set of macroeconomic, socio-economic, and financial sector indicators and a measure of traditional financial inclusion.
- Three approaches to address endogeneity and causality over the period 2011-18 for a sample of 52 countries:
  - (i) Cross-country OLS regression of GDP per capita growth indicators (averaged over 2011-18) on digital financial usage index.
  - (ii) Relate digital financial usage in 2011 to subsequent average growth over the period 2011-18 to avoid reverse causality.
  - (iii) Cross-country instrument variable estimator to extract exogenous components of the digital financial usage index. Instruments used: access to mobile money agents and access to the internet.

### Core empirical findings on growth
- Results show a significantly positive relationship between digital financial usage and GDP per capita growth.
- The positive link between digital financial usage and growth is not solely due to growth driving digital financial inclusion; the exogenous component of digital financial usage has a strong positive relationship with long-run growth.

### Drivers of digital financial inclusion: approach
- Most prior studies focused on traditional financial inclusion; few cross-country analyses exist for digital financial inclusion.
- Two empirical models employed:
  - Random effects panel regression with mobile agents per capita as the dependent variable to identify supply-side drivers.
  - Fractional logit panel regression with the digital usage index as the dependent variable to identify demand-side drivers.
- Explanatory variables include access to digital infrastructure, efficiency and competition in the financial sector, and other macroeconomic and socio-economic indicators.

### Key findings on drivers
- Digital financial inclusion is higher where:
  - The use of traditional financial services is high, but access to these services is limited — i.e., fintech supply (access) is “filling the gap” left by traditional financial institutions.
  - Individuals have higher financial awareness and trust in the financial system — i.e., fintech demand (usage) is higher with greater financial literacy and trust.
- Statistically significant drivers include:
  - Access to digital infrastructure (measured by accessibility to internet services and mobile phone).
  - Inefficiency of traditional financial service providers.
- Implication: fintech can expand inclusion in countries with gaps in traditional financial access, but supply-side expansion alone is insufficient without demand-inducing policies.

### Policy implications
- Supply-focused policies where traditional access gaps exist:
  - Promote an enabling environment for innovation and competition in the financial sector.
- Demand-inducing policies needed to translate supply into inclusion:
  - Financial and digital literacy programs.
  - Consumer protection.
  - Ensuring cybersecurity to promote trust in financial services.
- Regulatory responses should address emerging risks:
  - Digital divide across the population.
  - Data biases.
  - Ensuring competition among fintech companies (Sahay et al., 2020).

---

### II.   NEW INDICES OF FINANCIAL INCLUSION: STYLIZED FACTS

### Index construction methodology (Khera et. al. (2021))
- Enhanced measurement for 2014 and 2017 covering 52 EMDEs by incorporating digital aspects into financial inclusion measures.
- Indices capture indicators on access and usage provided by financial institutions and by DFSs including mobile money operators, fintech companies, and other new entrants.
- Three-stage principal component analysis (PCA):
  1. Construct “access” and “usage” sub-indices to capture supply-side and demand-side aspects.
  2. Combine access and usage sub-indices into “traditional” and “digital” financial inclusion indices to capture financial inclusion through financial institutions and enabled by technology separately.
  3. Construct an overall “comprehensive financial inclusion index” reflecting both traditional and digital aspects.

### Selected variables and weights (as presented)
- Overall Financial Inclusion Index components include both Traditional Financial Inclusion Index and Digital Financial Inclusion Index.
- Access variables (examples listed):
  - Number of ATMs per 100,000 adults — Data Source: IMF FAS — Weight in Traditional access: 0.25
  - Number of Branches per 100,000 adults — Data Source: IMF FAS — Weight in Traditional access: 0.25 (implied by structure)
  - Number of registered mobile money agents per 100,000 adults — Data Source: GSMA / Staff est. — Weight in Digital access: 0.25
  - Mobile subscription per 100 people — Data Source: ITU — Weight in Digital access: 0.125
  - % of population who has access to internet — Data Source: ITU — Weight in Digital access: 0.125

---

*Source: Excerpt from the provided content unit.*

### 0.25 Usage

### 0.25 Usage

### Regional patterns of financial inclusion
- Traditional financial inclusion is relatively high in Asia and the Pacific, Latin America and the Caribbean, and Emerging Europe.
- Traditional financial inclusion index remained broadly unchanged between 2014 and 2017; eight countries experienced a decline.
- Digital financial inclusion is relatively high in Africa and Asia and the Pacific.
- Most countries saw an increase in digital financial inclusion index between 2014 and 2017, driven by both access and usage; improvements were particularly large in African countries (e.g., Ghana, Benin, and Senegal) and relatively muted in some countries in Latin America and the Caribbean, and Middle East and Central Asia.
- Comprehensive financial inclusion improves significantly for countries with high digital but low traditional inclusion; examples: Senegal and Myanmar show similar traditional inclusion but large gaps in the comprehensive measure. China, Turkey and Brazil show high inclusion by both measures.
- Comprehensive financial inclusion index improved for Botswana, El Salvador, Mexico, Nigeria, Romania, Rwanda, South Africa, and Zimbabwe between 2014 and 2017.

### Impact of digital financial inclusion on macroeconomic growth — methodology
- Sample: cross-section of 52 developing and emerging economies.
- Dependent variable: 푦̇ (per capita GDP growth averaged over 2014 to 2018; also estimated for 2011 to 2018).
- Main regressors: traditional usage index (퐹퐼_T_u) and digital financial inclusion usage index (퐹퐼_F_u).
- Control variables (X): log of GDP per capita; government consumption (% of GDP); FDI (% of GDP); log of private credit (% of GDP); population growth rate; regional dummy variables (Asia, Middle East and Central Asia, Latin America, Emerging Europe, Sub‑Saharan Africa).
- Estimation strategy: instrumental variables using Limited Information Maximum Likelihood (LIML) to treat digital financial usage as endogenous.
- Instruments: initial level of log of mobile money agents per 100,000 adults (mma) and percent of population with access to the internet (internet).
- First-stage equation: (퐹퐼_F_u̅) = 훼0 + 훼1(mma) + 훼2(internet) + 훼3 X + μ.
- Robustness: tests of instrument strength (Cragg-Donald Wald statistics compared to Stock-Yogo critical values), Anderson–Rubin test, Hansen-J (overidentification) tests reported.

### Impact of digital financial inclusion on macroeconomic growth — main findings
- LIML estimates imply a statistically stronger and larger impact of digital financial usage on GDP per capita growth compared with OLS.
- Interpretation of key estimate (2011–18 LIML): a one standard deviation increase in digital financial usage (equivalent to 0.1205 increase in the index) is expected to boost the annual real GDP growth rate by 1.61 percentage points.
- Alternative expression: a one standard deviation increase in digital financial usage is expected to boost growth by 0.98 standard deviations on average (Table 2, regression over 2011–18).
- The increase in digital financial inclusion from the 25th percentile to the 75th percentile is associated with an increase in average economic growth of up to 2.2 percentage points (Figure 5).
- Higher initial level of traditional inclusion is found to be associated with slower subsequent growth in this sample and period; possible interpretation: decreasing returns to traditional inclusion in recent years and that benefits from traditional inclusion may have been reaped prior to the sample period.
- LIML diagnostics and robustness:
  - First-stage: log(mobile money agents) and internet (2014) are statistically significant predictors of average digital usage; F-statistics for excluded instruments exceed Stock and Yogo’s (2005) cutoffs for rejecting weak instruments when restricting IV bias to 15 percent of OLS bias in many specifications.
  - Andersen–Rubin test confirms statistically significant partial correlation between the endogenous variable (digital financial usage) and growth.
  - Hansen-J overidentification tests are passed in all reported cases (p-values reported in tables).

### Drivers of digital financial inclusion — methodology
- Dependent variables: mobile money agents per 100,000 adults (mma) as proxy for supply of DFSs; digital usage index (퐹퐼_F_u) for usage.
- Estimators:
  - Random effects regression (feasible generalized least squares) for access/supply (mobile money agents), with time fixed effects γ_t. Financial inclusion regressors lagged by three periods where applicable.
  - Fractional logit regression (Papke and Wooldridge, 1996) for digital payments usage (to handle dependent variable in [0,1] including 0 and 1).
- Baseline access regression (equation 3): mma_it = α0 + γ_t + α1(퐹퐼_F_u)_{i,t−3} + α2(퐹퐼_T_a)_{i,t−3} + α3(퐹퐼_T_u)_{i,t−3} + Σ β_k X_{k,i,t−1} + ε_it.
- Baseline usage regression (equation 4): (퐹퐼_F_u)_{i,t} = θ0 + φ_t + θ1(mma_{i,t}) + θ2(퐹퐼_F_a)_{i,t−3} + θ3(퐹퐼_T_a)_{i,t−3} + θ4(퐹퐼_T_u)_{i,t−3} + Σ δ_k X_{k,i,t−1} + v_{i,t}.
- Explanatory variables considered (lagged one period): i) log GDP per capita; ii) financial sector efficiency: overhead costs to assets; iii) bank concentration (assets share of three largest banks); iv) financial stability: log of NPL as % of gross loans; v) rule of law (0–1); vi) share of urban population.
- Sample: 52 low income and developing economies; time periods t = 2014 and 2017 (financial inclusion indices lagged by three periods due to availability).

### Drivers of digital financial inclusion — main findings
- Usage of digital payment services is higher where:
  - There already exists a culture of using financial services (traditional usage index positive and significant).
  - Access to traditional financial institutions is constrained (traditional access index negative and significant), suggesting digital services fill gaps left by limited traditional supply.
- Supply-side drivers:
  - Greater competition among traditional providers (lower bank concentration) is associated with higher number of mobile money agents.
  - Inefficiencies in bank operations (higher overhead costs) are associated with higher numbers of mobile agents—implying banks’ inefficiencies create profitable opportunities for fintech providers.
  - Higher quality of institutions has a positive impact on mobile agent supply.
- Digital infrastructure and supply matter:
  - Supply of DFSs (mobile money agents per population) and better access to digital infrastructure (digital access index) have positive and statistically significant impacts on digital usage in most specifications.
- Factors with limited or no significance:
  - Financial sector efficiency and quality of institution do not have statistically significant impacts on digital usage in most specifications.
- Interaction of traditional and digital inclusion:
  - Traditional access index negatively and significantly affects digital usage, while traditional usage positively affects digital usage—interpreted as digital services substituting where traditional access is limited but being adopted more readily where people already use financial services.
- Robustness checks:
  - Using single instruments instead of both; detrending GDP with HP filter; alternative control variables (literacy, overall financial development index); results remained similar.
  - Weak identification statistics improve when two aggregated instruments are used instead of a single generated instrument.

*Source: IMF staff calculations, as reported in the PDF chapter "0.25 Usage" (wpiea2021167-print-pdf).*

### Appendix II confirm our baseline findings. In addition, we conducted the following

### Appendix II confirm our baseline findings. In addition, we conducted the following

### Robustness checks and alternative specifications
- Pooled OLS estimation was used instead of random effects for Equation (3).
- Additional explanatory variables tested (found insignificant for DFS adoption):
  - GSMA mobile money regulatory index (score scale 0 to 100; comprises 27 indicators aggregated into six dimensions).
  - Fixed broadband internet monthly subscription rate (internet costs).
  - Percent of gross secondary school enrollment (level of literacy).
- Inclusion of lags to avoid multicollinearity between financial inclusion sub-indices and other explanatory variables.
- Controlling for aggregate trends over time made the coefficient on the access to digital infrastructure variable smaller and insignificant.
- Coefficients remain significant when regional dummies are included; on average, countries in Africa have significantly higher usage of DFSs in comparison to other regions.
- Results did not change substantially under alternative specifications and aligned with baseline estimates.

### Key empirical findings (usage and access)
- Digital financial inclusion in payments is found to boost annual economic growth by up to 2.2 percentage points.
- Cross-sectional instrumental variable analysis indicates the exogenous component of digital financial inclusion is positively associated with economic growth; the large positive link is unlikely due to omitted variables, simultaneity, or reverse causation.
- Drivers of digital financial inclusion:
  - Demand-side: higher where demand for financial services exists, but gaps in supply of traditional financial services.
  - Usage facilitators: access to foundational infrastructure (mobile phones, mobile data services, broadband internet), and financial literacy/familiarity.
  - Supply-side: higher competition, inefficiencies in financial institutions, and rule of law are important.
- Mobile money agents (t-1) show positive and statistically significant coefficients in digital usage regressions (see Table 4; coefficients e.g., 0.00302*** with standard error (0.00061) in column (1)).
- Traditional usage is positively associated with digital usage (see Table 4; coefficients e.g., 4.692*** with standard error (0.870) in column (1)).
- Traditional access is negatively associated with digital usage (see Table 4; coefficients e.g., -3.658*** with standard error (1.359) in column (1)).
- Log(GDP per capita) (t-1) has negative and statistically significant associations with mobile money agents in several specifications (see Table 5; e.g., -121.6*** with standard error (35.95) in column (3)).

### Model, sample, and estimation approaches (Appendix I)
- Baseline cross-country OLS specification (Equation (1)):
  - ẏ_i = β0 + β1(FI_Tu)_i + β2(FI_Fu)_i + β3 X_i + ε(i)
  - ẏ is per capita GDP growth averaged over 2014 to 2018.
  - FI_T and FI_F are traditional and digital financial inclusion usage indices respectively.
  - X is a vector of control variables affecting growth.
- Sample: 52 developing and emerging economies.
- Growth periods analyzed: average growth over 2014-18 and alternative estimation over 2011-18.
- Control variables included:
  - log of GDP per capita.
  - Government consumption as percent of GDP.
  - Foreign Direct Investment (FDI) as percent of GDP.
  - log of private credit as percent of GDP.
  - Population growth rate.
  - Regional dummy variables: Asia, Middle East and Central Asia, Latin America, Emerging Europe, Sub-Saharan Africa.
- Two empirical approaches employed:
  1) Cross-country OLS regression with financial inclusion measures averaged over 2014-18 and growth averaged over 2014-18.
  2) Instrumental variable analysis to identify exogenous component of digital financial inclusion.

### Policy implications and risks
- Policy priorities should be country-specific, considering individual country circumstances:
  - Countries with high traditional usage but low access: encourage technological innovation in payments to fill bank infrastructure gaps; high potential for fintech to spur inclusion.
  - Countries with high traditional access but low usage: prioritize improving financial literacy and trust in the financial system; address social barriers including high informality.
  - Countries with high voluntary and involuntary exclusion: tackle social and economic barriers and encourage competition and fintech entry.
- Address inequality in access to technology (mobile phone, the internet, electricity, digital ID) to prevent a digital divide within and across countries.
- Regulatory and supervisory priorities:
  - Strengthen consumer protection and crisis management tools where DFSs expand without such safeguards.
  - Adapt regulatory frameworks to fintech-related risks, including cyber security risks, misuse of payment channels for money laundering and terrorist financing, and liquidity risks.
  - Recognize that fintech firms often rely on traditional banks to store customer funds, posing potential insolvency and custodial risks.
- Filling regulatory gaps is increasingly important as technology applications expand from payments to digital credit and other financial services.

*Italic line: Content derived from wpiea2021167-print-pdf Appendix II and related sections as provided.*

### Appendix Table I.1 indicates that there is a significant positive relationship between

### wpiea2021167-print-pdf - Appendix Table I.1 indicates that there is a significant positive relationship between

### Main findings on digital financial inclusion and growth
- Appendix Table I.1 indicates that there is a significant positive relationship between usage of DFSs and economic growth, and also holds true for the estimation over 2011-18.
- Appendix Table I.2 indicates that usage of DFSs is a statistically significant predictor of subsequent rates of economic growth in the regression over 2014-18, but that this relationship is not significant in the case of estimation over 2011-18.
- The text cautions that these results do not settle causality: reverse causation from GDP per capita growth to digital financial inclusion is possible, and digital financial services may develop in anticipation of future economic activity (financial inclusion may be a leading indicator rather than a fundamental cause).

### Regression approaches and robustness
- Approach 1 (Appendix Table I.1): Cross-country OLS regressions relating averaged measures of financial usage (traditional and digital) to Real GDP per capita growth (2011–2018 and 2014–2018). Other regressors in X and 퐹퐼
푇 are averaged over the same time period.
- Approach 2 (Appendix Table I.2): Cross-country OLS regressions using financial inclusion measures in the initial period to predict subsequent average growth (2011–2018 and 2014–2018), with other regressors referring to initial conditions, to mitigate reverse causality concerns.
- Robustness checks (Appendix II) use random effects estimation for the Digital usage financial inclusion index (t) with lagged explanatory variables (e.g., Trad. access FI Index (t-3), Trad. usage FI Index (t-3), Mobile Money Agents (t-1), Log (GDP per capita) (t-1), and others). Note: Robust standard errors are reported and significance levels use ***, **, * for 1, 5 and 10% levels.

### Key reported coefficients and statistics (selected, preserved verbatim)
- Appendix Table I.1 (OLS estimation: Real GDP per capita growth rate (2011 – 2018) and (2014 – 2018)):
  - Digital usage F.I Index coefficients (selected columns):
    - 0.0855 (0.109)
    - 0.0670 (0.107)
    - 0.228** (0.0939)
    - 0.187* (0.102)
    - 0.188* (0.104)
    - 0.216* (0.113)
    - 0.0462 (0.0658)
    - 0.0523 (0.0628)
    - 0.130** (0.0536)
    - 0.115* (0.0580)
    - 0.111* (0.0583)
    - 0.164** (0.0668)
  - Log( GDP per capita) coefficients (selected):
    - -0.0418 (0.0254)
    - -0.0518* (0.0258)
    - -0.0755*** (0.0231)
    - -0.0726*** (0.0239)
    - -0.0733*** (0.0242)
    - -0.0524* (0.0266)
    - -0.0194 (0.0192)
    - -0.0314 (0.0189)
    - -0.0531*** (0.0171)
    - -0.0542*** (0.0178)
    - -0.0478** (0.0194)
    - -0.0349 (0.0212)
  - Gov. consumption (% of GDP) coefficients (selected):
    - -0.0187*** (0.00456)
    - -0.0181*** (0.00537)
    - -0.0185*** (0.00419)
    - -0.0179*** (0.00479)
    - -0.0178*** (0.00485)
    - -0.0107* (0.00579)
    - -0.0112*** (0.00270)
    - -0.0120*** (0.00283)
    - -0.0115*** (0.00227)
    - -0.0114*** (0.00250)
    - -0.0111*** (0.00252)
    - -0.00661** (0.00318)
  - Population growth coefficients (selected):
    - -8.758*** (2.302)
    - -8.431*** (2.632)
    - -8.366*** (2.663)
    - -4.801 (3.022)
    - -6.693*** (1.618)
    - -6.212*** (1.778)
    - -6.299*** (1.788)
    - -3.977** (1.872)
  - Sample sizes and fit (selected):
    - N 47, 43, 46, 43, 43, 43, 49, 45, 48, 45, 45, 45
    - adj. R2 0.297, 0.307, 0.508, 0.448, 0.436, 0.476, 0.249, 0.378, 0.538, 0.520, 0.516, 0.572
- Appendix Table I.2 (OLS estimation: Real GDP per capita growth rate (2011 – 2018) and (2014 – 2018)):
  - Digital usage F.I Index coefficients (selected):
    - 0.0693 (0.137)
    - 0.0776 (0.130)
    - 0.180 (0.120)
    - 0.175 (0.120)
    - 0.180 (0.123)
    - 0.196 (0.126)
    - 0.0233 (0.0928)
    - 0.0330 (0.0883)
    - 0.152* (0.0787)
    - 0.135* (0.0785)
    - 0.134 (0.0797)
    - 0.151* (0.0790)
  - Log( GDP per capita) coefficients (selected):
    - -0.0427** (0.0248)
    - -0.0545* (0.0278)
    - -0.0731*** (0.0250)
    - -0.0648** (0.0250)
    - -0.0645** (0.0253)
    - -0.0501* (0.0251)
    - -0.0316 (0.0196)
    - -0.0423** (0.0206)
    - -0.0633*** (0.0181)
    - -0.0611*** (0.0180)
    - -0.0614*** (0.0183)
    - -0.0463** (0.0184)
  - Gov. consumption (% of GDP) coefficients (selected):
    - -0.0185*** (0.00491)
    - -0.0200*** (0.00449)
    - -0.0182*** (0.00441)
    - -0.0181*** (0.00449)
    - -0.0182*** (0.00525)
    - -0.0124** (0.00296)
    - -0.0110*** (0.00300)
    - -0.0118*** (0.00256)
    - -0.0106*** (0.00256)
    - -0.0105*** (0.00261)
    - -0.0104*** (0.00261)
    - -0.00637** (0.00306)
  - Population growth coefficients (selected):
    - -7.887*** (2.099)
    - -7.497*** (2.267)
    - -7.516*** (2.298)
    - -5.726** (2.459)
    - -7.289*** (1.521)
    - -6.601*** (1.565)
    - -6.589*** (1.585)
    - -3.996** (1.873)
  - Sample sizes and fit (selected):
    - N 48, 45, 46, 45, 45, 45, 51, 49, 50, 49, 49, 49
    - adj. R2 0.261, 0.352, 0.455, 0.486, 0.473, 0.543, 0.219, 0.308, 0.487, 0.506, 0.494, 0.566
- Appendix II (Random effects estimation: Digital usage financial inclusion index (t), selected coefficients and significance):
  - Trad. access FI Index (t-3): -0.527*** (0.171), -0.373** (0.171), -0.618*** (0.210), -0.607*** (0.212), -0.615*** (0.210), -0.565** (0.221), -0.547*** (0.207), -0.517** (0.216), -0.237 (0.171), -0.220 (0.174), -0.232 (0.176), -0.249 (0.170), -0.248 (0.174)
  - Trad. usage FI Index (t-3): 0.779*** (0.139), 0.667*** (0.131), 0.738*** (0.137), 0.749*** (0.138), 0.851*** (0.149), 0.853*** (0.153), 0.739*** (0.152), 0.740*** (0.157), 0.773*** (0.132), 0.775*** (0.133), 0.799*** (0.138), 0.679*** (0.141), 0.683*** (0.147)
  - Mobile Money Agents (t-1): 0.000563*** (0.000102), 0.000379*** (0.000112), 0.000355*** (0.000115), 0.000372*** (0.000113), 0.000344*** (0.000115), 0.000272** (0.000117), 0.000260** (0.000118), 0.000464*** (0.000102), 0.000450*** (0.000104), 0.000463*** (0.000106), 0.000361*** (0.000109), 0.000365*** (0.000112)
  - Log (GDP per capita) (t-1): -0.125*** (0.0377), -0.126*** (0.0382), -0.121*** (0.0390), -0.0575 (0.0448), -0.0633 (0.0536)
  - Log (NPL) (t-1): 0.0757** (0.0300), 0.0734** (0.0308), 0.0827*** (0.0309), 0.0779** (0.0321), 0.0572* (0.0318), 0.0556* (0.0327)
  - N reported: 96, 96, 80, 79, 79, 79, 79, 79, 96, 95, 95, 95, 95

### Interpretive summary (as presented in the source)
- Usage of digital financial services (DFSs) is positively associated with economic growth in cross-country OLS specifications (Approach 1) and is a statistically significant predictor for subsequent growth in some initial-period specifications (Approach 2, notably for 2014–2018).
- However, causality is not established: results may reflect financial inclusion as a leading indicator of anticipated economic activity rather than a fundamental cause of growth.
- The authors note data limitations (e.g., assumptions about relative measures of digital financial inclusion across countries between 2011 and 2014) and emphasize that the usage (not merely access) of financial services is included because usage is what would impact growth through savings and consumption smoothing.

*Appendix Tables I.1, I.2, and II (selected coefficients and notes) as presented in the source document.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021167-print-pdf.pdf_