## 1. Two Dimensions of Financial Service Digitalization

## Source details

**Canonical URL:** [1. Two Dimensions of Financial Service Digitalization](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021046-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021046-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021046-print-pdf.pdf.json)

---

### Overview and Key Findings
- Digital transformation is reshaping the financial ecosystem and the future of banks; new entrants such as fintechs and bigtechs and digitally advanced customers are major drivers.
- The COVID-19 pandemic has accelerated the adoption of digital technologies in financial services.
- Digitalization could increase banks’ profitability but may favor larger banks, potentially leading to a more concentrated banking system.
- Financial inclusion can improve through digital services, but less digitally advanced customers may face reduced access; some bank employees may lose jobs due to automation.
- Cross-country comparison shows a global digital divide: banks’ digital services are more widely used in high-income economies than in middle- and low-income countries.
- Estimation results (cross-sectional fractional model) identify supportive and impeding factors:
  - Supportive: advanced digital infrastructure; good legal and business environment; presence of fintechs and bigtechs (positive correlation).
  - Impeding: weak bank balance sheets (low profitability, high NPLs).
  - Not important: age structure of the population does not appear to be an important factor.
  - Mixed: maturity of the banking industry is negatively correlated with bank digitalization in high-income economies but positively correlated in low-income countries.

### Definition and Two Dimensions of Digital Transformation
- Digital transformation: the use of new and fast changing digital technology to transform business activities, competencies, and business models.
- Two dimensions of financial service digitalization:
  - Financial Services impacted: Payments, clearing, settlement; Credit, deposits, and capital-raising; Wealth Management; Investment banking; Communication.
  - Technologies utilized: Cloud; AI/machine learning/advanced data analytics; Big data; Distributed ledger (DLT); Application programming interfaces (APIs); Robot advisor; Mobile technology.

### History of Digital Transformation and Banks’ Performance
- Selected adoption milestones:
  - By 1984, 8 percent of U.S. households owned a personal computer.
  - By 2000, 51 percent of U.S. households owned a computer.
  - In late 2005, internet users reached one billion.
  - By the end of 2010, 3 billion people worldwide were using cellphones.
  - By 2015, tablet computers and smartphones had exceeded personal computers in internet usage.
- Banks historically led adoption of some technologies (example: Citibank installed the first ATM in 1977), but leadership weakened since the Global Financial Crisis (GFC).
- New entrants (fintechs and bigtechs) expanded rapidly since 2015; global capital invested in fintech and global deals increased markedly (figure referenced in source).

### Implications for Competitiveness, Costs, and Structure
- Cost and profitability effects:
  - Initial investments can be large (internal and external hardware, software, services, hiring IT-skilled staff).
  - Recurrent costs expected to be lower in medium to long term as legacy systems are replaced.
  - Online distribution channels reduce investment in branches, branch staff, and back-office departments.
  - Citi (2019) estimate: digitalization could cut banks’ operational cost by 30 percent to 50 percent; revenues could decline by 10 percent–30 percent due to enhanced competition and transparency.
  - Accenture (2019) survey: digital advanced banks saw an overall increase of return on equity (ROE) of 0.9 percent between 2011 and 2017; less digitally advanced banks saw a ROE decline of 1.1 percent over the same period.
- Industry structure:
  - Large initial investment needs and potential increasing returns to scale imply digitalization could lead to a more concentrated banking industry with larger banks gaining market share; smaller or local banks could exit.
- Financial inclusion and consumer effects:
  - Mobile wallets in Africa have granted millions without a banking account access to financial services.
  - Digital financial services significantly enhance client well-being directly and by enabling a broader ecosystem.
- Labor market and access risks:
  - Ernst and Young (2019): of 121 job roles in financial services, 40 out of the 121 job roles are highly impacted by Robotic Process Automation, Advanced Analytics, and AI.
  - Branch closings and digital transitions may shut out less tech-inclined customers or those in remote areas.

### Drivers and Barriers—Empirical Evidence and Model Insights
- Empirical literature: micro- and firm-level studies suggest ICT capital accumulation is strongly correlated with shifts in cost and profit frontiers and efficiency gains (examples cited).
- Cross-country fractional logit regressions find:
  - Positive associations: advanced digital infrastructure; good rule of law and credit market regulation; debit card ownership; mobile phone use for utility payments; presence of mobile money accounts (constructive role for fintechs and bigtechs).
  - Negative associations: higher NPL ratios; weak bank profitability indicators.
  - Age structure (millennials and post-millennials) shows mixed or non-significant relationships across income groups.
  - Banking industry maturity: negative correlation with bank digitalization in high-income economies; positive correlation in low-income countries.

### Data, Methodology, and Key Variables
- Primary cross-country usage proxy (GFd, 2017): "used a mobile phone or internet to access a financial institution account" (% of age 15+ with a financial institution account); sample of 139 countries (2017).
- Other data sources: Enhanced Digital Access Index (EDAI) by Alper and Miktus (2019); Cellular subscriptions and broadband (OECD/WB); R&D, education, STEM graduates (World Bank); Rule of Law (World Bank governance); Financial Stability Indicator (IMF FSI); Financial Access Survey (FAS, IMF).
- Estimation method: fractional logit regressions (Papke and Wooldridge, 1996); stepwise variable selection with lagged explanatory variables; GDP per capita included as control.
- Caveats: cross-sectional data, limited observations, potential endogeneity, and lack of instruments constrain causal interpretation.

### Selected Key Statistics and Summary Measures (from Appendices)
- Dependent variable: Electronic access to financial accounts — Ob.Mean 0.30, Std. Dev. 0.20. Source: GFd database (WB).
- DAI: internet usage correlations and summary metrics cited in bivariate analysis: correlation 0.7 (internet usage with bank digitalization highest reported).
- Country examples: Top ten countries by share of banking customers who use mobile or internet to access their accounts: Norway, Denmark, Finland, Sweden, the Netherlands, New Zealand, the United States, Estonia, South Korea, and Canada — share ranges from 70 percent to 85 percent (as reported).
- Global connectivity context:
  - "barely 50 percent of the world’s population has access to the internet today" (verbatim from source).
  - Mobile cellular subscriptions in 2018: 7.86 billion; China: 1.65 billion; India: 1.18 billion (as reported).

### Selected Multivariate Regression Results (preserved coefficients and standard errors)
- Full sample (Table 3, Column 1; Observations concatenated in table as 9848412829392618):
  - DAI: internet usage (0–100): coefficient 0.0201*** (standard error 0.00618).
  - DAI: infrastructure (0–100): coefficient -0.00534 (standard error 0.00325).
  - Millennials and post millennials (%): coefficient 0.732 (standard error 1.016).
  - Rule of Law (index): coefficient 0.338** (standard error 0.135).
  - Credit market regulations (rating): coefficient 0.102* (standard error 0.0563).
  - NPL ratios (%): coefficient -0.0117* (standard error 0.00607).
  - ROA (%): coefficient -0.0549 (standard error 0.0364).
  - Credit card ownership (%): coefficient 0.002170 (standard error 0.00435).
  - Debit card ownership (%): coefficient 0.0139*** (standard error 0.00429).
  - Mobile phone to pay utility (%): coefficient 0.0265*** (standard error 0.00390).
  - ln(GDP/capita): coefficient -0.222** (standard error 0.0996).
- Mobile money account (Column 2, Observations 84):
  - Mobile money account (%): coefficient 0.00998* (standard error 0.00562).
- Income-specific highlights (Table 3):
  - High-income (Column 3): Millennials coefficient 2.501*** (standard error 0.895); DAI: internet usage 0.0118* (standard error 0.00693); NPL ratios -0.0136** (standard error 0.00590); ROA 0.171*** (standard error 0.0591); credit card ownership -0.0134*** (standard error 0.00519).
  - Low-income (Column 4): DAI: internet usage -0.0342 (standard error 0.0208); labor: mandated cost worker (rating) 0.0792*** (standard error 0.0257); NPL ratios -0.0861*** (standard error 0.0233); mobile phone to pay utility 0.0412*** (standard error 0.00809); ln(GDP/capita) 0.611 (standard error 0.422).
  - Middle-income (Column 5): DAI: internet usage 0.0334*** (standard error 0.00898); millennials -4.905*** (standard error 1.591); debit card ownership 0.0335*** (standard error 0.00792).
- Regional examples:
  - Europe (Column 6): DAI: internet usage 0.0178* (standard error 0.00930); NPL ratios -0.0432*** (standard error 0.0109); mobile phone to pay utility 0.0361*** (standard error 0.0106); ln(GDP/capita) 0.321* (standard error 0.187).
  - Euro area (Column 8): DAI: internet usage 0.0296*** (standard error 0.00792); mobile phone to pay utility 0.0503*** (standard error 0.00725); ln(GDP/capita) 0.615*** (standard error 0.200).

### R&D-specific Findings (Table 4 selected coefficients)
- Full sample (Observations 74):
  - R&D expenditure (% of GDP) coefficient 0.228*** (standard error 0.0629).
- High-income:
  - R&D expenditure coefficient 0.252*** (standard error 0.0850).
- Low- and middle-income (reported with caution due to small samples):
  - Low-income R&D coefficient -0.381** (standard error 0.161).
  - Middle-income R&D coefficient -2.582*** (standard error 0.000742).

### Bivariate Correlations and Stepwise Results (selected magnitudes)
- Strong positive bivariate correlations with bank digitalization:
  - DAI: internet usage — correlation 0.7.
  - DAI: knowledge — correlation 0.6.
  - R&D expenditure — correlation 0.7.
  - Credit card ownership — correlation 0.7.
  - Debit card ownership — correlation 0.8.
  - Mobile money accounts — correlation 0.7.
  - Rule of law — correlation 0.8.
- Negative bivariate correlations observed:
  - Millennials and post-millennials — correlation -0.5.
  - NPL ratios — negative correlation (magnitude smaller in bivariate analysis).
- Selected stepwise regression coefficients (Table A3–A6 excerpts):
  - DAI: infrastructure — 0.0108*** (0.00384).
  - DAI: internet usage — 0.0158** (0.00652).
  - R&D expenditure — 0.707*** (0.0979).
  - Rule of Law — 0.628*** (0.165).
  - ROA — −0.522** (0.211).
  - NPL — −0.0488** (0.0215).
  - Credit card ownership — 0.0341*** (0.00303).
  - Debit card ownership — 0.0262*** (0.00196).
  - Mobile money account — 0.0313*** (0.00330).
  - Mobile phone for utility payment — 0.0338*** (0.00672).
- Statistical significance notation: *** p<0.01, ** p<0.05, * p<0.1.

### Policy Recommendations (as presented)
- Invest in digital infrastructure: widely available, affordable, high-quality internet.
- Strengthen the rule of law and the broader business environment.
- Maintain healthy bank balance sheets and accelerate NPL resolution.
- Foster an enabling environment for non-bank fintech and bigtech entrants to encourage competitive digital development.
- Monitor labor market impacts and promote worker retraining where automation displaces roles.
- Address digital inclusion to avoid shutting out less tech-inclined customers or those in remote areas.

### Conclusions and Open Questions
- Digital transformation is important for banks’ competitiveness, financial stability, and inclusion.
- High-income economies: banks tend to dominate digital service provision; lower-income economies: non-banks often lead digital finance.
- Adequate digital infrastructure, better legal and business environments, and healthy bank balance sheets support bank digital advancement.
- Development of non-bank digital financial services may encourage banks to adopt newer digital technologies.
- COVID-19 has accelerated digital migration; countries lagging in digital development risk widening disadvantages for citizens.
- Open questions for future research include drivers of fintech and bigtech development, cyber security implications of banks’ digital transformation, and the net effect of digitalization on financial inclusiveness.

*Source — wpiea2021046-print-pdf, "1. Two Dimensions of Financial Service Digitalization" (chapter text supplied).*

### 1.  Two Dimensions of Financial Service Digitalization.............................................................7

### 1. Two Dimensions of Financial Service Digitalization

### Overview and Key Findings
- Digital transformation is reshaping the financial ecosystem and the future of banks; new entrants such as fintechs and bigtechs and digitally advanced customers are major drivers.
- The COVID-19 pandemic has accelerated the adoption of digital technologies in financial services.
- Digitalization could increase banks’ profitability but may favor larger banks, potentially leading to a more concentrated banking system.
- Financial inclusion can improve through digital services, but less digitally advanced customers may face reduced access; some bank employees may lose jobs due to automation.
- Cross-country comparison shows a global digital divide: banks’ digital services are more widely used in high-income economies than in middle- and low-income countries.
- Estimation results (cross-sectional fractional model) identify several supportive and impeding factors for banks’ digital advancement:
  - Supportive: advanced digital infrastructure; good legal and business environment; presence of fintechs and bigtechs (positive correlation).
  - Impeding: weak bank balance sheets (low profitability, high NPLs).
  - Not important: age structure of the population does not appear to be an important factor.
  - Mixed: maturity of the banking industry is negatively correlated with bank digitalization in high-income economies but positively correlated in low-income countries.

### Definition and Two Dimensions of Digital Transformation
- Digital transformation: the use of new and fast changing digital technology to transform business activities, competencies, and business models.
- Two dimensions of financial service digitalization (Table 1):
  - Financial Services impacted:
    - Payments, clearing, settlement
    - Credit, deposits, and capital-raising
    - Wealth Management
    - Investment banking
    - Communication
  - Technologies utilized:
    - Cloud
    - AI/machine learning/advanced data analytics
    - Big data
    - Distributed ledger (DLT)
    - Application programming interfaces (APIs)
    - Robot advisor
    - Mobile technology

### History of Digital Transformation and Banks’ Performance
- Timeline highlights and adoption statistics preserved from the source:
  - By 1984, 8 percent of U.S. households owned a personal computer.
  - By 2000, 51 percent of U.S. households owned a computer.
  - In late 2005, internet users reached one billion.
  - By the end of 2010, 3 billion people worldwide were using cellphones.
  - By 2015, tablet computers and smartphones had exceeded personal computers in internet usage.
- Banks historically led adoption of some technologies (e.g., Citibank installed the first ATM in 1977), but their leadership weakened since the Global Financial Crisis (GFC) as balance-sheet repair and regulation became priorities.
- New entrants (fintechs and bigtechs) expanded rapidly since 2015, backed by swift adoption of newer technologies; global capital invested in fintech and global deals increased markedly (Figure 2 referenced).

### Implications for Competitiveness, Costs, and Structure
- Cost and profitability effects:
  - Initial investments can be large (internal and external hardware, software, services, hiring IT-skilled staff).
  - Recurrent costs expected to be lower in medium to long term as legacy systems are replaced.
  - Online distribution channels reduce investment in branches, branch staff, and back-office departments.
  - Citi (2019) estimate preserved from source: digitalization could cut banks’ operational cost by 30 percent to 50 percent; revenues could decline by 10 percent–30 percent due to enhanced competition and transparency.
  - Accenture (2019) survey result preserved: digital advanced banks saw an overall increase of return on equity (ROE) of 0.9 percent between 2011 and 2017; less digitally advanced banks saw a ROE decline of 1.1 percent over the same period.
- Industry structure:
  - Large initial investment needs and potential increasing returns to scale imply digitalization could lead to a more concentrated banking industry with larger banks gaining market share; smaller or local banks could exit.
- Financial inclusion and consumer effects:
  - Mobile wallets in Africa have granted millions without a banking account access to financial services.
  - Digital financial services significantly enhance client well-being directly and by enabling a broader ecosystem (Karl an and others (2016) referenced).
- Labor market and access risks:
  - Ernst and Young (2019) preserved finding: of 121 job roles in financial services, 40 out of the 121 job roles are highly impacted by Robotic Process Automation, Advanced Analytics, and AI (potential for convergence or displacement).
  - Branch closings and digital transitions may shut out less tech-inclined customers or those in remote areas (Financial Times (FT, 2019) cited example of U.K. branch closings).

### Drivers and Barriers—Empirical Evidence and Model Insights
- Empirical literature is limited, but micro- and firm-level studies suggest ICT capital accumulation is strongly correlated with shifts in cost and profit frontiers and efficiency gains (e.g., Casolaro and others (2007) for Italian banks).
- SWIFT adoption evidence (Scott and others (2017)): adoption had a large impact on profitability in the long term for banks studied.
- Cross-country fractional logit regressions (details of estimation in later sections) find:
  - Positive associations: advanced digital infrastructure; good legal and business environment; presence of fintechs and bigtechs.
  - Negative associations: weak bank profitability; high NPLs.
  - Age structure of population not significant.
  - Banking industry maturity: negative correlation with bank digitalization in high-income economies; positive correlation in low-income countries—possibly reflecting entrenched older technologies in advanced economies.

### Data and Methodology (introductory points)
- Cross-country databases differentiating bank digital advancement from non-bank digital advancement are lacking; the paper uses proxy databases.
- The study uses Global Findex database (GFd) produced by the World Bank every three years since 2011 as a primary data source for cross-country comparison (more details and caveats discussed in the paper).
- The paper employs a cross-sectional fractional model to study factors associated with the global digital divide in bank digital services.

_Italic: Source — wpiea2021046-print-pdf, "1. Two Dimensions of Financial Service Digitalization" (chapter text supplied)._

### Appendix I).  Among a few hundreds GFd indicators, only one qualifies as a proxy for the

### Appendix I

### Bank digitalization proxy and cross-country usage
- Proxy used: "used a mobile phone or internet to access a financial institution account" (% of age 15+ with a financial institution account), data available for 2017 and a sample of 139 countries.
- Top ten countries by share of banking customers who use mobile or internet to access their accounts: Norway, Denmark, Finland, Sweden, the Netherlands, New Zealand, the United States, Estonia, South Korea, and Canada — where the share ranges from 70 percent to 85 percent.
- Digital payment usage for utility bills (2017) is reported as "% total payment on utility bills" (World Bank Global Findex Database); regional patterns: Scandinavia high, many European countries lag, China and some African countries among the most advanced.

### Bank versus non-bank digital financial services
- Two GFd indicators compared: “digital (mobile or internet) access to financial account” (accounts with financial institutions) and “digital (mobile or internet) access to account” (accounts including non-bank accounts such as mobile wallets).
- Interpretation: If non-banks provide a significant share of digital financial services, “access to all accounts” > “access to financial institution account.” Sub‑Saharan African countries are prominent examples where usage is mainly through non-bank accounts (countries above the 45 degree line in Figure 6).
- China does not appear as a highly non-bank dominated case in the 2017 data; possible reasons: 2017 data may not capture later mobile payment growth, or the indicator captures only basic mobile/internet usage.

### Potential explanatory factors for the global digital divide (thematic categories)
- Digital ecosystem: (i) digital infrastructure; (ii) technological know-how; (iii) digital technology adoption rate; (iv) demographic factors (age, education, gender, income).
  - Noted statistics: world internet access “barely 50 percent of the world’s population has access to the internet today.”
  - Mobile cellular context: in 2018, mobile cellular subscriptions amounted to 7.86 billion; subscriptions in China ranked number one at 1.65 billion, followed by India at 1.18 billion.
- Broader business environment: competition, enforceable property rights, flexible labor market, business regulation and policy timing.
- Financial sector development: maturity of banking system, prevalence of credit/debit card usage, competitive pressure from non-banks (fintechs, bigtechs).
- Bank characteristics: size, profit margins, capital positions, business models; large initial investment in new technologies may favor larger banks.

### Data sources used
- Enhanced Digital Access Index (EDAI) by Alper and Miktus (2019), including five sub-indices: infrastructure, knowledge, affordability, quality, actual internet usage.
- Cellular subscriptions and broadband subscriptions (OECD Innovation and Technology database).
- Education, share of STEM graduates, R&D expenditures (World Bank).
- Share of young people (millennials and post-millennials) (United Nations population database).
- Rule of Law (World Bank governance database); legal system indicators (Economic Freedom of the World, Fraser Institute); property rights (Heritage Foundation, Fraser Institute).
- Bank conditions and financial sector development: Financial Stability Indicator (FSI) of the IMF and GFd database (World Bank).
- Main dependent variable: Electronic Access to Financial Accounts (0–1), measured as mobile or internet access to financial institution accounts (2017).

### Bivariate correlation patterns (summary)
- Strong positive correlations with bank digitalization:
  - DAI: internet usage — correlation 0.7 (highest reported).
  - DAI: knowledge — correlation 0.6.
  - R&D expenditure — correlation 0.7.
  - Credit card ownership — correlation 0.7.
  - Debit card ownership — correlation 0.8.
  - Mobile money accounts — correlation 0.7.
  - Rule of law — correlation 0.8.
- Negative correlations observed:
  - Millenials and post-millenials — correlation -0.5.
  - NPL ratios — negative correlation (magnitude smaller in bivariate analysis).
- Bank condition indicators show mixed/weak correlations:
  - Total capital ratio — correlation 0.2.
  - Tier 1 capital ratio — correlation 0.2.
  - NPL ratios — correlation -0.2.
  - ROA — correlation -0.2.
  - ROE — correlation -0.1.
- Note: correlation magnitudes were grouped by quartiles in Table 2.

### Multivariate analysis: fractional logit regressions (method and general approach)
- Estimation method: fractional logit regressions to account for fractional dependent variable bounded in [0,1]; stepwise regressions were used to select variables (significance threshold 5 percent).
- Lagged explanatory variables used to mitigate reverse causality concerns; GDP per capita included as a control in all regressions.
- Caution: limited observations and cross-sectional data constrain causal interpretation; potential endogeneity and lack of instruments noted.

### Key multivariate findings (selected coefficient estimates from Table 3)
- Full sample (Column 1, All; Observations: 98 84 41 28 29 39 26 18 reported as concatenated Observations9848412829392618 in the table):
  - DAI: internet usage (0–100): coefficient 0.0201*** (standard error 0.00618).
  - DAI: infrastructure (0–100): coefficient -0.00534 (standard error 0.00325).
  - Millennials and post millennials (%): coefficient 0.732 (standard error 1.016) — not significant.
  - Rule of Law (index): coefficient 0.338** (standard error 0.135).
  - Credit market regulations (rating): coefficient 0.102* (standard error 0.0563).
  - NPL ratios (%): coefficient -0.0117* (standard error 0.00607).
  - ROA (%): coefficient -0.0549 (standard error 0.0364).
  - Credit card ownership (%): coefficient 0.002170 (standard error 0.00435).
  - Debit card ownership (%): coefficient 0.0139*** (standard error 0.00429).
  - Mobile phone to pay utility (%): coefficient 0.0265*** (standard error 0.00390).
  - ln(GDP/capita): coefficient -0.222** (standard error 0.0996).
- Non-bank digital services (Mobile money account %) (Column 2, All with mobile money available; Observations 84):
  - Mobile money account (%): coefficient 0.00998* (standard error 0.00562).
- Income-specific patterns (Table 3 columns 3–5):
  - High-income (Column 3): Millennials and post-millennials coefficient 2.501*** (standard error 0.895); DAI: internet usage 0.0118* (standard error 0.00693); NPL ratios: -0.0136** (standard error 0.00590); ROA: 0.171*** (standard error 0.0591); credit card ownership: -0.0134*** (standard error 0.00519).
  - Low-income (Column 4): DAI: internet usage coefficient -0.0342 (standard error 0.0208); labor: mandated cost worker (rating) coefficient 0.0792*** (standard error 0.0257); NPL ratios: -0.0861*** (standard error 0.0233); mobile phone to pay utility: 0.0412*** (standard error 0.00809); ln(GDP/capita): 0.611 (standard error 0.422).
  - Middle-income (Column 5): DAI: internet usage coefficient 0.0334*** (standard error 0.00898); millennials coefficient -4.905*** (standard error 1.591); NPL ratios: -0.00448 (standard error 0.0112); debit card ownership 0.0335*** (standard error 0.00792).
- Regional regressions (Europe, EU, Euro area — Columns 6–8) show heterogeneous coefficients; examples:
  - Europe (Column 6): DAI: internet usage 0.0178* (standard error 0.00930); NPL ratios -0.0432*** (standard error 0.0109); mobile phone to pay utility 0.0361*** (standard error 0.0106); ln(GDP/capita) 0.321* (standard error 0.187).
  - Euro area (Column 8): DAI: internet usage 0.0296*** (standard error 0.00792); mobile phone to pay utility 0.0503*** (standard error 0.00725); ln(GDP/capita) 0.615*** (standard error 0.200).

### R&D-specific regressions (selected findings from Table 4)
- R&D expenditure (% of GDP) included (Table 4):
  - Full sample (Column 1, All; Observations 74): R&D expenditure coefficient 0.228*** (standard error 0.0629).
  - High-income (Column 3): R&D expenditure coefficient 0.252*** (standard error 0.0850).
  - Low- and middle-income (Columns 4–5): R&D expenditure coefficients reported as -0.381** (standard error 0.161) for low-income and -2.582*** (standard error 0.000742) for middle-income — interpreted with caution due to possible omitted variables and small sample sizes.

### Interpretation and policy implications (as presented)
- Key drivers positively associated with bank digitalization:
  - Ease of internet use and quality of internet (DAI: internet usage).
  - Better rule of law and credit market regulation.
  - Financial sector development measures (debit card ownership; mobile phone use for utility payments).
  - Development of non-bank digital services (mobile money accounts) is associated with higher bank digitalization, suggesting a constructive role for fintechs and bigtechs.
- Negative or impeding factors:
  - Higher NPL ratios (weak bank balance sheets) impede digital advancement.
  - Entrenched use of older technologies (e.g., credit cards) in high-income economies may reduce pressure to innovate.
- Policy recommendations emphasized:
  - Invest in digital infrastructure (widely available, affordable, high-quality internet).
  - Strengthen the rule of law and business environment.
  - Maintain healthy bank balance sheets and speed NPL resolution.
  - Foster an enabling environment for non-bank fintech and bigtech entrants to encourage competitive digital development.

### Conclusion (summary of overarching messages)
- Digital transformation is important for banks’ competitiveness, financial stability, and inclusion.
- High-income economies: banks tend to dominate digital service provision; lower-income economies: non-banks often lead digital finance.
- Adequate digital infrastructure, better legal and business environments, and healthy bank balance sheets support bank digital advancement.
- The development of non-bank digital financial services may encourage banks to adopt newer digital technologies.
- COVID-19 has accelerated digital migration; countries lagging in digital development risk widening disadvantages for citizens.
- Open questions for future research include: drivers of fintech and bigtech development, cyber security implications of banks’ digital transformation, and the net effect of digitalization on financial inclusiveness.

*Source: wpiea2021046-print-pdf Appendix I (IMF staff analysis, based on World Bank Global Findex Database and additional datasets as described in the appendix).*

### REFERENCES

### REFERENCES

### Theoretical and empirical findings on technology and firm performance (Appendix I)
- The Shumpeterian theory (Shumpeter, 1943) frames innovation as generating “creative destruction,” enabling firms to introduce new products, services and organizational processes and gain market share at the expense of non-innovating competitors.
- Early literature documented a “productivity paradox”: Solow (1987) — “you can see the computer age everywhere but in the producitivty statistics” — and Brynjolfsson (1993) summarized studies finding little relationship between IT investment and observable productivity.
- Firm-level evidence from the early 1990s:
  - Brynjolfsson and Hitt (1995, 1996) and Lichtenberg (1995) used data from over 300 large firms for 1988–92 and found that ICT capital generates up to 10 times more output than other forms of capital.
- Macro- and cross-firm considerations:
  - Technologically advanced countries may leverage new competitive positions to accumulate “monopolistic rents” and increase profitability (Cainelli et al., 2006).
  - Digital platforms enable new business models and rapid scaling of digital ventures (Huang et al., 2017).
  - The nature of new technologies often favors large firms, potentially triggering “winner takes all” dynamics that benefit a minority of leading frontier firms (OECD 2015; Brynjolfsson et al., 2008).
- Other salient points:
  - Digital ventures can grow at a massive rate and scale and founders can create temporary monopolies or oligopolies with less external capital (Huang et al., 2017; Kurz 2017).
  - The literature links technological change to changes in entrepreneurial culture and industry structure consistent with Schumpeterian tradition.

### Data on financial service digitalization (Appendix II)
- Primary micro-level usage dataset:
  - Global Findex database (GFd), World Bank:
    - Published every three years since 2011.
    - Data are collected through nationally representative surveys of more than 150,000 adults in over 140 economies.
    - The 2017 edition (the latest edition described) added new data on financial technology usage, including use of mobile phones and internet to conduct financial transactions.
    - The GFd dataset contains 14 broad cateogoies of indicators that represent a limited scope of digital finance.
    - Five of these fourteen GFd indicators could potentially represent digital services offered by banks (see Banking service indicators below).
    - Limitation: indicators narrowly focus on internet and mobile phone usage to access financial services; no information on back-end technologies, cloud usage, AI, and other technologies; indicators do not differentiate between bank and non-bank digital financial services.
    - Example: mobile payment channels vary by region—mobile wallets provided by technology firms (e.g., Mpesa in Africa, Tencent and Alibaba in China) versus mobile applications linked to bank accounts (e.g., Apple Pay in the U.S. and Europe).
- Supply-side dataset:
  - Financial Access Survey (FAS), IMF:
    - Launched in 2009.
    - Covers 189 countries spanning more than 10 years and contains 121 time-series on financial access and usage.
    - Based on administrative data collected by central banks and other financial regulators.
    - Evolution and coverage issues:
      - In 2014, country-level data on mobile money were introduced and coverage of innovations in traditional banking services (including branchless banking and debit and credit cards in circulation) was expanded.
      - These expanded indicators are only available for at most 56 countries, with a concentration on low-income economies.
      - The 2019 FAS introduced new data series on mobile and internet banking for deposit-taking microfinance institutions.
      - The number of countries reporting mobile money data increased from 66 to 71, but substantial gaps remain.

### Global Findex indicator summary (selected facts and availability)
- GFd features and temporal coverage:
  - 14 broad categories of indicators in the GFd.
  - Made or received digital payments in the past year (% age 15+) — available in 2011, 2014, 2017.
  - Mobile money account (% age 15+) — available in 2011, 2014, 2017.
  - Most other listed GFd indicators' year available: 2017 (as specified in Table A1).
- Examples of GFd indicator definitions and exact years available (verbatim phrasing preserved):
  - "Used the internet to pay bills in the past year (% age 15+)" — 2017.
  - "Used the internet to buy something online in the past year (% age 15+)" — 2017.
  - "Sent or received domestic remittances: through a mobile phone (% age 15+)" — 2017.
  - "Used a mobile phone or the internet to access a financial institution account in the past year (% age 15+)" — 2017.
  - "Made or received digital payments in the past year (% age 15+)" — 2011, 2014, 2017.
  - "Mobile money account (% age 15+)" — 2011, 2014, 2017.

### Banking service digitalization indicators (GFd subset noted in the text)
- Five GFd indicators identified as potentially representing digital services offered by banks (verbatim list as described):
  1. used internet to pay bills in the current year;
  2. paid online for internet purchases;
  3. used a mobile phone or internet to access a financial institutions account in the past year, as a share of the adult population;
  4. used a mobile phone or internet to access a financial institutions account in the past year, as a share of adults with financial institution accounts;
  5. used a mobile phone or the internet to check account balance in the past year.

### Data limitations and interpretive cautions (verbatim and preserved implications)
- GFd limitations (verbatim implications preserved):
  - Indicators "narrowly focus on internet and mobile phone usage to access financial services."
  - "No information is available on other aspects of financial industry digital advancement, such as the development of back-end technologies, cloud usage, AI, and other technologies."
  - Indicators "do not differentiate between digital financial services offered by banks and those offered by non-bank financial institutions."
- FAS limitations:
  - Expanded indicators are "only available for at most 56 countries, with a concentration on low-income economies."
  - Despite increases, "there remain substantial gaps" in mobile money reporting (countries reporting mobile money data increased from 66 to 71).

*wpiea2021046-print-pdf - REFERENCES*

### 1. Used the internet to pay bills in the

### 1. Used the internet to pay bills in the past year (% age 15+)

### Usage indicators (definitions and 2017 reference)
- "The percentage of respondents who report using the internet to pay bills in the past 12 months." 2017.
- "Among respondents reporting using the internet to buy something online in the past 12 months, the percentage who report paying online for their internet purchase." 2017.
- "The percentage of respondents who report using a mobile phone or the internet to make a payment, to make a purchase, or to send or receive money through their financial institution account in the past 12 months." 2017.
- "Among respondents with a financial institution account, the percentage who report using a mobile phone or the internet to access their financial institution account in the past 12 months." 2017.
- "The percentage of respondents who report using a mobile phone or the internet to check their balance for a financial institution account in the past 12 months." 2017.

### Country coverage (Appendix III)
- East Asia & Pacific, Europe & Central Asia (cont'd), North America, South Asia, Sub‐Saharan Africa, Latin America & Caribbean, Middle East & North Africa are represented.
- Example country listings shown include: Australia, Cambodia, China, Hong Kong SAR, China, Indonesia, Japan, Korea, Rep., Lao PDR, Malaysia, Mongolia, Myanmar, New Zealand, Philippines, Singapore, Thailand, Taiwan, Province of China, Vietnam; Albania, Armenia, Austria, Azerbaijan, Belarus, Belgium, Bosnia and Herzegovina, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Georgia, Germany, Greece, Hungary, Ireland, Italy, Kazakhstan, Kosovo, Kyrgyz Republic, Latvia, Lithuania, Luxembourg, Macedonia, FYR, Moldova, Montenegro, Netherlands, Norway, Poland, Portugal, Romania, Russian Federation, Serbia, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, Tajikistan, Turkey, Turkmenistan, Ukraine, United Kingdom, Uzbekistan; Canada, United States; Afghanistan, Bangladesh, India, Nepal, Pakistan, Sri Lanka; Benin, Botswana, Burkina Faso, Cameroon, Central African Republic, Congo, Dem. Rep., Congo, Rep., Côte d'Ivoire, Ethiopia, Gabon, Ghana, Guinea, Kenya, Lesotho, Liberia, Malawi, Mali, Mauritania, Mauritius, Mozambique, Namibia, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, Tanzania, Togo, Uganda, Zambia, Zimbabwe; Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Haiti, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Trinidad and Tobago, Uruguay, Venezuela, RB; Algeria, Bahrain, Egypt, Arab Rep., Iran, Islamic Rep., Iraq, Israel, Jordan, Kuwait, Lebanon, Libya, Malta, Morocco, Saudi Arabia, Tunisia, United Arab Emirates.

### Summary statistics and data sources (Appendix IV)
- Dependent variable: Electronic access to financial accounts — Ob.Mean 0.30, Std. Dev. 0.20. Source: GFd database (WB).
- Digital ecosystem (2014–2016 average) metrics (Ob.Mean, Std. Dev., Min, aSource):
  - DAI: infrastructure — 19, 26, 22, 40. Alper and Miktus (2019).
  - DAI: quality — 19, 22, 31, 80. Alper and Miktus (2019).
  - DAI: affordability — 19, 22, 08, 0. Alper and Miktus (2019).
  - DAI: knowledge — 19, 26, 12, 10. Alper and Miktus (2019).
  - DAI: internet usage — 19, 24, 22, 80. Alper and Miktus (2019).
  - EDAI (0–100) — 19, 27, 51, 40. Alper and Miktus (2019).
  - DESI: digital public services — 26, 7, 24. European Commission.
- Connectivity and innovation:
  - Cellular subscriptions (per 100 inhabitants) — 199, 106, 40, 10. WB.
  - Broadband subscriptions (per 100 inhabitants) — 198, 131, 40. WB.
  - R&D expenditure (% of GDP) — 109, 110. WB.
  - Education (% of labor with advanced education) — 115, 787, 58. WB.
  - Stem graduates (% of graduates) — 117, 21, 82. WB.
  - Millennials and post‑millennials (%) — 185, 60, 01434. United Nations.
- Broader Business Environment (2016) example metrics and sources:
  - Legal: Rule of law (index) — 194, 0.0, 1.0, ‐2.3. WB Governance Database.
  - Property rights: property rights (rating) — 178, 42.3, 24.9, 5.0. Heritage Foundation.
  - Various EFW (Fraser Institute) and other indices listed with observed means and deviations as shown in source.
- Banks' own condition (2014–2016 average):
  - Total capital ratio (%) — 136, 18.5, 5.6, 8.6.
  - Tier 1 capital ratio (%) — 136, 16.2, 5.6, 8.2.
  - NPL ratios (%) — 136, 7.5, 7.7, 0.1.
  - ROA (%) — 137, 1.5, 1.5, ‐7.4.
  - ROE (%) — 137, 15.3, 29.2, ‐73.2.
- Payment and account indicators:
  - Credit card ownership (% of age 15+) — 138, 19.3, 21.0, 0.0.
  - Debit card ownership (% of age 15+) — 138, 44.6, 31.4, 1.7.
  - Bank concentration (%, 2016) — 155, 65.9, 19.6, 18.4.
  - Mobile phone to pay utility bills (% of total pay) — 143, 0.1, 0.1, 0.0.
  - Mobile money accounts (% of age 15+) — 77, 0.1, 0.1, 0.0.
  - Income level: Ln (GDP/capita) — 189, 8.6, 1.45, 7. World Economic Outlook (IMF).
- Data sources cited in table: Alper and Miktus (2019), European Commission, WB, Heritage Foundation, EFW (Fraser Institute), World Economic Outlook (IMF), GFd (World Bank), IMF Financial stability indicator (FSI).

### Regression methods (Appendix V)
- Model: Fractional logit model (Papke and Wooldridge, 1996).
- Structure: dependent variable in (0–1), X is set of explanatory variables, link-function specified in source.

### Variable selection procedure (Appendix VI)
- Step 1: Divide variables in each of the four group variables into sub‑groups.
- Step 2: Run fractional logit regressions for each subgroup, and select variables that are significant at 5 percent level.
- Step 3: Run fractional logit regressions for each group with variables selected from step 2.
- Step 4: Select variables that are significant at the 10 percent level from Step 3.
- "Table A3‑6 present regression results from the abovementioned steps, and variables selected from step 4 are highlighted in yellow."

### Selected regression results (stepwise regressions; coefficients with robust standard errors in parentheses)
- Table A3. Digital Ecosystem (Dependent: Electronic Access to Financial Accounts (0–1))
  - DAI: infrastructure — 0.0108*** (0.00384) ; −0.00853** (0.00406)
  - DAI: quality — 0.0103*** (0.00362) ; 0.00302 (0.00354)
  - DAI: affordability — 0.0308** (0.0143) ; −0.00185 (0.0140)
  - DAI: internet usages — 0.0158** (0.00652) ; 0.0205*** (0.00703)
  - Broadband subscriptions — 0.0289** (0.0114) ; 0.0279** (0.0140)
  - DAI: knowledge — 0.0569*** (0.0118) ; 0.00736 (0.0106)
  - Millennials and post‑millennials — −3.882*** (0.546) ; 2.621** (1.115)
  - R&D expenditure — 0.707*** (0.0979) ; 0.268*** (0.0998)
  - Constants reported across model specifications, e.g., −2.476*** (0.294), −1.645*** (0.274), −3.918** (1.918), 1.342*** (0.323), −1.392*** (0.109), −4.028*** (1.063).
  - Observations reported: 136, 132, 72, 135, 96, 91 (as listed).

- Table A4. Broader Business Environment (Dependent: Electronic Access to Financial Accounts (0–1))
  - Legal: Rule of Law — 0.628*** (0.165) ; 0.890*** (0.273)
  - Property rights — 0.0140*** (0.00523) ; −0.00233 (0.00652) ; −0.00748 (0.0391)
  - Legal system property rights — 0.261*** (0.0942) ; 0.0460 (0.129)
  - Labor: hiring regulation — 0.0955** (0.0409) ; 0.0639** (0.0295)
  - Labor: mandated cost worker — 0.120*** (0.0317) ; 0.0549** (0.0237)
  - Business: admin requirements — −0.240*** (0.0741) ; 0.0155 (0.0780)
  - Business: extra payments — 0.403*** (0.0725) ; −0.00362 (0.0997)
  - Business: regulations — 0.216* (0.128) ; −0.409** (0.205)
  - Credit market regulations — 0.167** (0.0795)
  - Constants and observations reported (e.g., Observations 113, 131, 127, 129, 121).

- Table A5. Bank Conditions (select coefficients)
  - ROA — −0.522** (0.211) ; −0.409* (0.220)
  - ROE — 0.0512** (0.0223) ; 0.0240 (0.0253)
  - NPL — −0.0488** (0.0215)
  - Constants and observations (e.g., Observations 108, 108, 107).

- Table A6. Financial Sector Development (Dependent: Electronic Access to Financial Accounts (0–1))
  - Credit card ownership — 0.0341*** (0.00303) ; 0.0161*** (0.00609)
  - Debit card ownership — 0.0262*** (0.00196) ; 0.0129*** (0.00376)
  - Bank concentration index — 0.005730 (0.00528) ; 0.00188 (0.00275)
  - Mobile money account — 0.0313*** (0.00330) ; 0.0216*** (0.00505)
  - Mobile phone for utility payment — 0.0338*** (0.00672) ; 0.0130*** (0.00456)
  - Constants and observations reported (e.g., Observations 135, 135, 130, 74, 139, 68).

### Statistical significance notation used in tables
- *** p<0.01, ** p<0.05, * p<0.1.

*Source: wpiea2021046-print-pdf - 1. Used the internet to pay bills in the past year (% age 15+) (PDF chapter/section).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021046-print-pdf.pdf_