Institutional Foundations of Digital Financial Inclusion: Governance, Financial Development, and Infrastructure Legacy in 38 OECD Countries, 2000–2022

1. Introduction

Financial inclusion research has, for the past decade, run in one direction: from inclusion to outcomes. Studies link account ownership and digital payment adoption to poverty reduction, resilience to shocks, and formalization of economic activity, most influentially through the World Bank’s Global Findex program (Demirgüç-Kunt & Klapper, 2012). Far less attention has gone to the reverse question — what produces cross-country differences in financial inclusion in the first place — and what attention exists concentrates on developing and emerging economies, where inclusion gaps are large, and the policy stakes are obvious (Allen et al., 2016; Asongu & Nwachukwu, 2018).

This creates a gap for advanced economies. Among the 38 members of the Organization for Economic Co-operation and Development (OECD), account ownership is already high — the sample mean is 87.3 percent, with a median above 95 percent — which might suggest the inclusion question is settled. It is not: the sample still contains meaningful variation, from a low of 27.4 percent in an early-period observation to a ceiling of 100 percent in several small, high-income economies. The persistence of that variation, in a set of countries with broadly comparable income levels and financial systems, is itself informative. It suggests that whatever remains unexplained by income is being absorbed by something slower-moving — institutional quality, the maturity of the financial sector, or the historical timing of digital infrastructure rollout.

This paper tests that proposition directly. We ask: does institutional quality — proxied by government effectiveness — predict digital financial inclusion in OECD countries, and does it do so as a structural, cross-country condition or as something that moves with year-to-year governance improvements within a country? We further ask whether financial-sector development (captured by the IMF’s Financial Institutions sub-index) and the historical timing of broadband and submarine-cable infrastructure add independent explanatory power once institutional quality and income are controlled for.

The empirical strategy is built around two features of the data that a standard fixed-effects panel model handles poorly. First, the dependent variables — account ownership and digital payment use — are fractions bounded between 0 and 100 and are heavily concentrated near the ceiling for this sample; several OECD countries have been at or near saturation for the entire observation window, which biases linear models estimated near a boundary (Papke & Wooldridge, 1996). Second, two of the theoretically central regressors — the timing of broadband infrastructure rollout and distance to the nearest submarine cable landing station — are time-invariant by construction. A country fixed-effects model, the default choice in cross-country panel work, mechanically absorbs both variables and cannot estimate them at all, and as we show in Section 5, it also erases the institutional-quality relationship along the way. We instead use a within-between (Mundlak/Bell-Jones) decomposition inside a fractional-response quasi-likelihood framework, which lets us separate persistent cross-country institutional effects from within-country dynamics while still retaining the time-invariant infrastructure variables in the model.

The contribution is threefold. Substantively, we show that in a high-inclusion setting, institutional quality operates almost entirely through persistent between-country differences rather than through within-country governance improvements — a finding with direct implications for how much inclusion gain a government can expect from governance reform within an electoral or policy-relevant horizon. Methodologically, we demonstrate, using the same data, how a conventional two-way fixed-effects specification would have missed this relationship entirely, which is a caution relevant well beyond this application. Empirically, we bring together five distinct data sources — World Bank Global Findex, the IMF Financial Development Index, the Worldwide Governance Indicators, ITU broadband statistics, and TeleGeography submarine cable data — into a single OECD panel that, to our knowledge, has not been assembled and analyzed together in this way.

2. Theoretical Framework, Literature Review, and Hypotheses

2.1. Theoretical Framework: Institutions, Transaction Costs, and the Diffusion of Digital Finance

The paper’s organizing framework is New Institutional Economics (NIE), which treats institutions as the humanly devised rules of the game — formal rules, informal constraints, and their enforcement characteristics — that structure economic exchange by reducing the uncertainty and transaction costs of transacting with strangers (North, 1990). Digital financial services are a textbook case of a transaction-cost-intensive exchange: a digital payment or a bank account is, at bottom, a bet that a counterparty (a bank, a mobile-money operator, a merchant) will honor an obligation that is not settled instantaneously or in person, and that if it is not honored, some enforceable mechanism exists to make the aggrieved party whole. Where contract enforcement is unreliable, financial regulation is unpredictable, or corruption distorts the licensing and supervision of payment providers, the transaction costs of adopting digital finance rise relative to cash or informal arrangements, and adoption should lag even where income and technology are not binding constraints. Williamson (2000) formalizes this logic as a nested hierarchy of institutional levels, from informal norms that change over centuries down to the resource-allocation decisions economic actors make period to period; critically for the empirical strategy below, Williamson places the formal “rules of the game” — the level government effectiveness and its constituent Worldwide Governance Indicators are designed to proxy — at a decadal-to-generational time scale, well above the annual frequency of firm- and household-level adoption decisions. This yields a first, non-obvious theoretical prediction that motivates our within-between econometric strategy rather than following from it after the fact: if institutions genuinely operate at NIE’s slow-moving, decadal time scale, their explanatory power for a rapidly diffusing technology like digital payments should surface as a persistent, cross-country ranking effect — a between-country phenomenon — rather than as a force that tracks year-to-year fluctuations in a country’s own measured governance quality.

A second and more specific institutional channel is the legal-origin and investor-protection literature associated with La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1998), who show that the strength and enforcement of formal legal rules protecting financial claims is a more precise predictor of financial-sector development than diffuse governance capacity. Applied to digital finance, this suggests the relevant institutional channel is narrower than “good government” in general: it is specifically the enforceability of contracts and the predictability of financial-sector regulation that lowers the cost digital payment providers face in extending services and that lowers the risk consumers face in trusting those services with their money. This distinction between general state capacity and the specific legal-financial institutions La Porta et al. identify is what motivates decomposing government effectiveness into its constituent governance dimensions in Section 6.2, rather than treating the aggregate index as a black box.

A third theoretical channel operates through the economics of network infrastructure rather than governance. Rogers’s (2003) diffusion-of-innovations framework describes technology adoption as a path-dependent process shaped by the timing of an innovation’s introduction into a social system and by the cumulative infrastructure that early adoption leaves behind; broadband and submarine-cable infrastructure exhibit the classic features of network goods with high fixed costs and low marginal costs of extension once built; the earlier a country crosses the threshold of meaningful broadband penetration, the longer it has had a payments-relevant digital substrate in place, and the closer a country sits to a submarine cable landing station, the lower the marginal cost of extending that substrate to the last mile. This channel is theoretically independent of institutional quality — a country can have excellent governance and still face a geographic or historical infrastructure disadvantage — which is why both are entered as separate, non-nested regressors rather than treated as alternative proxies for the same underlying construct.

Together these three channels generate the hypotheses tested below: an institutional-quality channel operating primarily between countries (H1), a financial-sector-development channel that is conceptually distinct from retail inclusion and need not move in the same direction (H2), and an infrastructure-legacy channel operating through the historical timing and geography of network build-out (H3).

2.2. Financial Inclusion as an Outcome of Institutional Quality

The dominant strand of the financial inclusion literature treats inclusion as an explanatory variable for growth, poverty, and resilience outcomes (Demirgüç-Kunt & Klapper, 2012; Allen et al., 2016). A smaller but growing strand inverts the question and treats institutional quality as the explanatory variable. Vo (2024), using a panel smooth-transition regression across 110 countries from 2004 to 2020, finds an asymmetric effect of institutional quality on financial inclusion that is positive and significant in high- and middle-income countries but largely absent in low-income countries, consistent with a threshold logic in which institutional reform pays off in inclusion terms only once a baseline of financial infrastructure and income already exists — precisely the OECD setting examined here. Ouechtati (2023) similarly documents that institutional quality is a robust correlate of financial inclusion once income inequality is controlled for, while more recent evidence specific to digital channels — Ben Mbarek (2025), using panel data with instrumental variables across 32 emerging economies from 2017 to 2023, and Meniago (2025), using system GMM across the SADC bloc from 2010 to 2023 — both find that institutional quality does not merely correlate with digital financial inclusion but actively moderates the strength of the relationship between fintech adoption and realized inclusion outcomes, reinforcing the view that institutions function less as a direct driver and more as a precondition that determines how much of a given digitalization push translates into actual uptake.

Government effectiveness is one of six Worldwide Governance Indicators constructed by Kaufmann, Kraay, and Mastruzzi using an unobserved-components model that aggregates several hundred underlying governance signals into a single latent estimate with an explicit margin of error (Kaufmann, Kraay, & Mastruzzi, 2010). It is a natural proxy for the state capacity that plausibly underlies digital financial inclusion: an effective bureaucracy standardizes identification systems, enforces contract and consumer-protection law for digital payment providers, and coordinates the public-private rollout of payment rails — all preconditions for account ownership and digital payment use to spread. Because government effectiveness in the WGI framework changes slowly and is highly persistent within a country, we hypothesize that its explanatory power will show up primarily as a between-country effect.

H1. 

Government effectiveness is positively associated with digital financial inclusion across OECD countries, and this association is driven predominantly by persistent between-country differences rather than within-country changes over time.

2.3. Financial-Sector Development

Financial-sector development is conceptually distinct from financial inclusion — the former describes the depth, access, and efficiency of financial institutions and markets, while the latter describes the share of the population actually using them. The IMF’s Financial Development Index decomposes this into institutions and markets sub-indices, each further split into access, depth, and efficiency dimensions, precisely to avoid conflating deep financial systems with narrow measures like private credit to GDP (Svirydzenka, 2016). A well-developed financial institutions sector should, in principle, make account ownership easier and cheaper to obtain, predicting a positive relationship with inclusion.

H2. 

Financial institutions development (FI_index) is positively associated with digital financial inclusion, controlling for income and institutional quality.

2.4. Infrastructure Legacy: Broadband Rollout and Submarine Cable Proximity

A separate literature treats internet and broadband infrastructure as a supply-side driver of financial digitalization. Instrumental-variable studies have used lagged broadband coverage, fixed-telephone penetration, and mobile/fixed broadband penetration as instruments for fintech and digital-payment adoption, generally finding a robust positive causal effect of connectivity on adoption. Niu, Jin, Wang, and Zhou (2022), using a difference-in-differences design around China’s nationwide rural broadband rollout, find that broadband construction raises the coverage dimension of digital financial inclusion substantially more than the usage dimension, with the usage effect concentrated in areas that already have higher human capital and existing bank-branch penetration — evidence that infrastructure and institutional/financial preconditions interact rather than substitute for one another. Calzada and Pablo (2026), using an instrumental-variable strategy based on lagged broadband coverage in a four-year household panel, similarly find that fixed broadband access raises the probability of adopting online banking, though the gain is concentrated among younger, wealthier, and more educated users, suggesting infrastructure expansion alone does not close all dimensions of the inclusion gap. Because roughly 99 percent of international internet traffic still travels through submarine cables (ITU, 2024), and because private investment in cable landing stations concentrates in high-traffic, commercially attractive markets, geographic proximity to a cable landing station functions as a quasi-exogenous constraint on connectivity quality that is independent of a country’s contemporaneous policy choices (World Bank, 2024). We treat both the historical timing of broadband rollout and cable landing distance as capturing a lasting infrastructure legacy rather than a contemporaneous policy variable, consistent with their time-invariant construction in the data.

H3. 

Countries with an earlier broadband rollout and shorter distance to a submarine cable landing station exhibit higher digital financial inclusion, net of income, institutions, and financial development.

2.5. Which Dimension of Governance Matters?

Government effectiveness is one of six correlated but conceptually distinct Worldwide Governance Indicators, alongside voice and accountability, political stability, regulatory quality, rule of law, and control of corruption (Kaufmann, Kraay, & Mastruzzi, 2010). The broader law-and-finance literature suggests these dimensions are not interchangeable for financial outcomes specifically: La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1998) show that legal protection of investors and creditors, and the quality of its enforcement, is a stronger and more direct predictor of financial-sector development than general governance capacity, with rule-of-law-type measures carrying most of the explanatory power in their cross-country evidence. Contemporary comparative work on fintech regulation specifically reinforces this: Vijayagopal, Jain, and Viswanathan (2024) compare the regulatory response to fintech across the US, UK, and India and find that the predictability and coherence of the regulatory regime — not the volume of regulation, and not government capacity in general — is what determines whether fintech innovation translates into expanded access rather than regulatory arbitrage or consumer harm. This motivates treating the aggregate government effectiveness result in Section 5 as a starting point rather than an endpoint, and testing whether the institutional relationship documented for digital financial inclusion is similarly concentrated in the rule-of-law and regulatory dimensions of governance, or is instead a more diffuse function of state capacity in general. We return to this decomposition as a robustness check in Section 6.

2.6. The Measurement Problem: Inclusion as a Bounded, Near-Ceiling Variable

A methodological point follows directly from the OECD setting: account ownership and digital payment use are not continuous, unbounded outcomes. They are percentages bounded in [0,100], and in a high-income sample many observations sit close to the upper bound. Papke and Wooldridge (1996) show that ordinary least squares estimated on fractional dependent variables performs poorly precisely in this situation, and propose a quasi-maximum-likelihood approach — fractional probit or logit — that respects the bounded support without the ad hoc transformations required by log-odds approaches. We adopt this framework for both dependent variables.

3. Data and Variables

The panel covers all 38 OECD member countries from 2000 to 2022 (874 country-years). Account ownership (has_account) and digital payment use (digital_payment_any) are drawn from the World Bank Global Findex Database, which surveys nationally representative samples of adults aged 15 and above in five waves (2011, 2014, 2017, 2021, 2024) and is linearly interpolated across intervening years, consistent with standard practice for between-wave Findex panels (Global Findex Database 2025; Demirgüç-Kunt & Klapper, 2012). Government effectiveness is the corresponding Worldwide Governance Indicator (Kaufmann, Kraay, & Mastruzzi, 2010). Financial institutions development (FI_index) is one of the two second-level sub-indices of the IMF Financial Development Index (Svirydzenka, 2016). GDP per capita (logged) and trade openness are from the World Bank World Development Indicators. The infrastructure head-start variable is constructed as the number of years between a country’s broadband rollout threshold year (the year fixed broadband subscriptions first exceeded five per 100 people, from ITU data via WDI) and the end of the sample (2022); it is time-invariant by construction. Submarine cable distance is the straight-line distance in kilometres from each country’s main economic centre to the nearest cable landing station, drawn from the TeleGeography Submarine Cable Map, and is zero for coastal countries with direct landings and positive for landlocked or geographically isolated countries.

Table 1 reports descriptive statistics. Account ownership and digital payment use are both left-skewed, with means near 85–87 percent, medians well above the mean, and several countries observed at the 100 percent ceiling — the pattern that motivates the fractional-response specification in Section 4. Government effectiveness ranges from −0.77 to 2.32 on the WGI’s approximately −2.5-to-+2.5 scale, reflecting genuine institutional heterogeneity even within the OECD (Greece and Mexico anchor the low end of the distribution; the Nordic countries and Switzerland anchor the high end). FI_index has 76 missing country-years (8.7 percent of the panel), concentrated in 2022, the last year the IMF has updated the series.

4. Methodology

4.1. Fractional Response Specification

Because has_account and digital_payment_any are proportions bounded in [0,1] once rescaled, we model both using the Papke-Wooldridge fractional probit quasi-maximum-likelihood estimator (Papke & Wooldridge, 1996):

E(y_it | x_it) = Φ(x_it′β)

where y_it is the fraction of the population with an account (or making/receiving a digital payment) in country i, year t; Φ(·) is the standard normal cumulative distribution function; and x_it is the regressor vector described below. The model is estimated by quasi-maximum likelihood using the Bernoulli log-likelihood, with standard errors clustered at the country level to allow for arbitrary within-country serial correlation.

4.2. Within-Between (Mundlak / Bell-Jones) Decomposition

A standard country fixed-effects specification would absorb the two time-invariant infrastructure variables entirely and would conflate within-country and between-country institutional effects into a single coefficient. We instead follow the within-between random-effects (REWB) approach of Bell and Jones (2015), which decomposes each time-varying regressor x_it into a between-country component (the country’s own time-mean, x̄_i) and a within-country component (the deviation from that mean, x_it − x̄_i), and enters both simultaneously:

y_it = α + β_W(x_it − x̄_i) + β_B x̄_i + γ z_i + λ_t + ε_it

Here β_W is the within-country effect — how a country’s own governance, financial development, income, and trade openness changes predict changes in its own inclusion rate — and β_B is the between-country effect — how a country’s average level of these variables predicts its average inclusion rate relative to other countries. z_i is the vector of time-invariant regressors (infrastructure head start and submarine cable distance), which survive in this specification precisely because there is no country fixed effect to absorb them. λ_t is a full set of year fixed effects, included to net out the common global trend in digital payment adoption — a trend that is substantial in this sample, rising sharply after 2012 as smartphone-based payment systems diffused across the OECD.

The time-varying regressors entered in this decomposition are government effectiveness, FI_index, GDP per capita (log), and trade openness. We initially included years-since-broadband-rollout as an additional time-varying regressor, but because the panel is close to balanced (23 years for most of the 38 countries) and broadband rollout year is fixed per country, its within-country deviation is nearly collinear with the year fixed effects by construction; we therefore retain only the time-invariant infrastructure head-start term and the year fixed effects, and drop the redundant within-country broadband-timing term from the reported specification.

4.3. Estimation and Inference

All models are estimated via statsmodels’ generalized linear model framework with a binomial family and probit link, on the complete-case sample (N = 798 country-years, 38 countries) after listwise deletion for FI_index missingness. Standard errors are clustered at the country level. As a diagnostic contrast, Section 5 also reports a naive two-way (country and year) fixed-effects OLS specification on the same sample, to illustrate what a conventional panel specification would have concluded. The within-between decomposition is not the only estimator capable of retaining time-invariant regressors alongside unobserved heterogeneity; the Hausman-Taylor (1981) instrumental-variables estimator offers a closely related alternative that partitions regressors into exogenous and potentially endogenous time-varying and time-invariant groups and instruments the latter internally. We adopt the within-between approach as the primary specification for its more transparent decomposition of within- versus between-country effects, and report a Tobit alternative to the fractional probit functional form, together with a disaggregation of government effectiveness into its constituent Worldwide Governance Indicators, as robustness checks in Section 6.

5. Results

Table 2 reports the within-between fractional probit estimates for both dependent variables. The central result concerns government effectiveness: its between-country coefficient is positive and statistically significant in both specifications (0.628, p = 0.026, for account ownership; 0.686, p = 0.006, for digital payment use), while its within-country coefficient is small, statistically insignificant, and even wrong-signed for account ownership (−0.186, p = 0.140). This pattern supports H1 only partially and with an important qualification: institutional quality does predict digital financial inclusion in OECD countries, but the effect is structural — it distinguishes countries with persistently higher-capacity governments from those with persistently lower-capacity governments — rather than dynamic. A country that improves its measured government effectiveness over the sample period does not, on this evidence, see a corresponding rise in inclusion. This is consistent with a ceiling interpretation: OECD inclusion rates are already high enough on average that within-country governance improvements have limited room to move the outcome, while the cross-country ranking established by longer-run institutional capacity remains informative.

FI_index performs against expectation. Its between-country coefficient is negative in both specifications and only marginally significant for digital payment use (−0.886, p = 0.102) and account ownership (−1.095, p = 0.123). This does not support H2 as stated. We interpret this cautiously: FI_index measures the depth and efficiency of formal financial institutions (e.g., bank branch density, private credit, insurance penetration), which is conceptually closer to traditional banking depth than to the ease of opening a basic transaction account or using a mobile payment app. It is plausible that some of the OECD countries with the highest FI_index scores (large, bank-dominated financial centres) do not lead on account ownership or digital payment use, which could already be near-universal via simpler and more recently built systems in other member states. This is a genuine empirical finding rather than a modeling artifact — the sign is stable across both dependent variables — and it argues against assuming that financial-sector depth and financial inclusion move together even within a relatively homogeneous group of advanced economies.

The infrastructure legacy variables support H3, with digital payment use responding more strongly than account ownership. Infrastructure head start is positive and significant for digital payment use (0.066, p = 0.044) and positive but not significant for account ownership (0.065, p = 0.108); submarine cable distance is negative in both specifications and marginally significant for digital payment use (−0.0004, p = 0.096). The pattern is intuitive: account ownership is closer to a policy-mandated minimum (e.g., basic bank accounts, government transfer accounts) that spreads even in countries with a later digital start, while active digital payment use depends more directly on the maturity of the payment infrastructure a country has had time to build out.

Table 3 reports the diagnostic contrast: a naive two-way fixed-effects OLS specification, with the same time-varying regressors, estimated on the identical sample. The two time-invariant infrastructure variables cannot be estimated at all — they are perfectly absorbed by the country fixed effects — and government effectiveness loses all statistical significance (coefficient −0.970, p = 0.695). The within R² of 0.965 looks impressive but is mechanical: with 38 country dummies and 22 year dummies absorbing the bulk of cross-sectional and temporal variation, there is very little residual variation left for any substantive regressor to explain. This comparison makes the methodological point directly: a researcher who ran the conventional fixed-effects specification on this exact dataset would have concluded that institutions do not matter for digital financial inclusion in the OECD, and would have been unable to test the infrastructure-legacy hypothesis at all. The relationship documented in Table 2 is not an artifact of the estimator choice — it is only visible because the estimator choice was made deliberately to preserve between-country variation.

Variance inflation factors for the core regressors range from 1.06 to 7.68, with the highest values on the between-country components of government effectiveness (6.69) and GDP per capita (7.68) — expected, since richer OECD countries also tend to have higher-capacity governments, and both correlate with an early broadband rollout. None exceed the conventional threshold of 10, but the moderate collinearity among the between-country institutional and income variables means the individual between-country coefficients should be read as suggestive of a joint institutional-and-income cluster of causes rather than as fully separable, independent effects.

6. Robustness Checks

6.1. Functional Form: Tobit as an Alternative to the Fractional Probit

The fractional probit specification in Section 5 is one defensible response to the ceiling problem in the dependent variables, but it is not the only one. As a robustness check we re-estimate both outcomes as a Type I (upper-censored) Tobit model on the original percentage scale, censored at 100, using maximum likelihood with cluster-robust standard errors. Thirty-eight of the 798 has_account observations are censored at the ceiling; digital_payment_any has none. Table 4 reports the results, using the identical within-between decomposition and regressor set as the main specification.

The core institutional finding survives the change in functional form for digital payment use: the between-country government effectiveness coefficient remains positive and significant (12.414, p = 0.011), closely tracking its fractional-probit counterpart in both sign and significance, and infrastructure head start remains positive and highly significant (2.597, p = 0.006). For account ownership, the Tobit between-country government effectiveness coefficient is positive but no longer significant at conventional levels (7.284, p = 0.214), a modest weakening relative to the fractional probit result (p = 0.026); infrastructure head start, by contrast, becomes significant in the Tobit specification where it was not in the fractional probit (2.817, p = 0.011). The FI_index within-country coefficient also flips from statistically indistinguishable from zero to marginally positive and significant for account ownership (13.109, p = 0.060). None of these differences reverse the qualitative story — government effectiveness operates through the between-country channel, and infrastructure legacy predicts inclusion independently — but the sensitivity of individual significance levels to functional form is worth flagging rather than obscuring: the has_account result is the less robust of the two dependent variables across specifications, consistent with our reading in Section 5 that account ownership is closer to a policy floor with less remaining variance for any regressor to explain.

6.2. Which Governance Dimension Carries the Effect?

Section 5 uses government effectiveness as the institutional-quality proxy throughout, but Section 2.5 noted that the Worldwide Governance Indicators comprise six correlated but distinct dimensions. Because a full panel reconstruction of all five remaining dimensions (control of corruption, regulatory quality, rule of law, voice and accountability, political stability) at annual frequency back to 2000 was not feasible within this project, we instead construct a country-level cross-section: each WGI dimension is averaged over its most recently available five years (2020-2024) as a proxy for each country’s typical institutional position, and regressed separately against each country’s full-panel (2000-2022) average account ownership and digital payment use, controlling for average income, trade openness, and financial institutions development. This trades panel resolution for coverage of the full governance taxonomy, and should be read as a cross-sectional companion to the panel evidence in Section 5 rather than a replacement for it; because the five WGI dimensions are themselves highly collinear (pairwise correlations of 0.6-0.97, variance inflation factors up to 27 when entered jointly), each is entered in a separate bivariate specification rather than jointly, which is standard practice in this literature.

Table 5 shows that the institutional-inclusion relationship is not spread evenly across governance dimensions. Rule of law is the strongest and most consistent predictor of both outcomes (p < 0.001 in both regressions, and the highest R² of any dimension), followed by regulatory quality and control of corruption. Voice and accountability and political stability are weaker, remaining statistically significant for digital payment use but only marginally so, or borderline, for account ownership. This pattern is consistent with the law-and-finance literature’s emphasis on contract enforcement and regulatory predictability as the specific institutional channels most relevant to financial-sector outcomes (La Porta, Lopez-de-Silanes, Shleifer, & Vishny, 1998), rather than governance capacity or political conditions in general. Substantively, it suggests that the government effectiveness result in Section 5 is not picking up a diffuse state-capacity effect so much as the specific institutional infrastructure — enforceable contracts, predictable regulation of payment providers, low corruption in licensing and supervision — that digital financial services depend on.

7. Discussion

Three findings stand out. First, the institutional-quality effect on digital financial inclusion in the OECD is real but structural rather than dynamic — it operates through the persistent ranking of countries by governance capacity, not through year-to-year governance improvement. This has a direct policy implication: a government seeking to raise financial inclusion within an electoral cycle should not expect governance reform alone to move the needle quickly, because the mechanism this paper identifies operates on a much longer time horizon than most policy evaluation windows. Digital-infrastructure investment, by contrast, shows a more immediate, within-reach lever, particularly for digital payment use.

Second, the divergence between account ownership and digital payment use as outcomes is informative in its own right. Account ownership in the OECD is close to a policy floor — driven by mandated access to basic banking and government transfer accounts — and is correspondingly less sensitive to institutional and infrastructure variation. Digital payment use, which requires active adoption rather than passive account-holding, is more sensitive to both government effectiveness (between-country) and infrastructure legacy. Future work on financial inclusion policy in advanced economies should treat these as distinct outcomes with different determinants, rather than collapsing them into a single “inclusion” measure.

Third, the counterintuitive FI_index result deserves further scrutiny beyond what this paper can resolve. One possibility is that the IMF Financial Development Index, designed for global comparability across countries at very different stages of financial development, loses discriminatory power at the top of the distribution where nearly all OECD countries sit — an analogous ceiling problem to the one motivating the fractional-response choice for the dependent variables. A useful extension would substitute more granular OECD-specific measures of financial-sector structure (e.g., bank concentration, non-bank payment provider penetration) to test whether the negative sign survives.

Methodologically, the contrast between Table 2 and Table 3 is the paper’s clearest general lesson. Cross-country panel researchers routinely default to two-way fixed effects as the “safe” specification against omitted-variable bias. That default is not free: it discards exactly the between-country variation that, in this setting, carries the entire institutional-quality signal, and it mechanically forecloses any question involving a time-invariant regressor. Researchers working with panels that include slow-moving institutional variables or genuinely time-invariant geographic and infrastructure variables should treat the within-between decomposition, or the closely related Hausman-Taylor estimator, as the default choice rather than a robustness check.

8. Limitations and Future Research

Several limitations qualify these conclusions. Findex measures are observed at five discrete waves (2011, 2014, 2017, 2021, 2024) and linearly interpolated for intervening years; this is standard practice but mechanically smooths short-run dynamics and may understate the true within-country variance available to identify β_W, biasing the within-country coefficient toward zero. The IMF Financial Development Index has not been updated beyond 2021 at the time of data collection, which forces the exclusion of 2022 from part of the analysis and limits the ability to capture very recent fintech-driven shifts in financial-sector structure. The WGI sub-dimension decomposition in Section 6.2 is cross-sectional and built from a recent (2020-2024) window rather than the full 2000-2022 panel, because reconstructing five additional governance series at annual frequency for the full period was outside the scope of this project; a full panel version of that decomposition, ideally combined with a within-between split for each dimension separately, is a natural extension once the underlying series can be assembled back to 2000. More generally, while the within-between decomposition addresses the mechanical fixed-effects problem, it does not by itself establish causal identification of the between-country institutional effect, which could still reflect omitted, slow-moving country characteristics correlated with both governance quality and financial-sector history; the submarine-cable and broadband-timing variables were originally compiled as candidate instruments in a related project and remain available for a future instrumental-variable extension of the institutional-quality relationship specifically, potentially combined with a Hausman-Taylor specification as a further cross-check on the within-between results.

9. Conclusions and Policy Implications

This paper reversed the usual causal framing in the digital financial inclusion literature, asking what predicts inclusion in a group of already high-income, already well-governed countries rather than what inclusion predicts. The answer is that institutional quality still matters, but as a structural, between-country condition rather than a lever that responds to short-run governance reform, while historical digital-infrastructure investment — how early a country built out broadband, and how physically close it sits to the global submarine cable network — offers a more immediately actionable channel, particularly for active digital payment use rather than passive account ownership. The robustness checks in Section 6 sharpen this further: the institutional effect survives a change in functional form from fractional probit to Tobit, and when disaggregated across the six Worldwide Governance Indicators, is carried predominantly by rule of law and regulatory quality rather than political stability or voice and accountability — pointing specifically at contract enforcement and predictable regulation of payment providers as the operative channel, not governance capacity or democratic conditions in general. For OECD and OECD-accession governments, this suggests that near-term financial-inclusion policy is better targeted at completing digital-payment infrastructure and provider-level adoption incentives than at governance reform framed as an inclusion strategy, while long-run strengthening of rule of law and payment-sector regulatory quality specifically remains valuable for the structural, cross-country position it appears to secure over time. Methodologically, the paper is a demonstration that the standard two-way fixed-effects panel specification, when applied uncritically to a setting with slow-moving institutional regressors and genuinely time-invariant geographic variables, can silently erase the very relationship a researcher set out to test.