Digital Financial Inclusion and Economic Growth: Multi-Dimensional Evidence from Coverage, Depth, and Digitisation

Abstract

Using panel data from 278 Chinese prefecture-level cities during 2011–2019, this study employs two-way fixed effects and instrumental variable (2SLS) models to investigate how the distinct dimensions of digital financial inclusion (DFI)—coverage breadth, usage depth, and digitisation level—affect urban economic growth. The results reveal substantial heterogeneity across these DFI dimensions. The expansion of coverage breadth significantly and robustly promotes city-level economic growth. In contrast, greater usage depth exerts a negative effect, possibly due to regulatory lags in internet credit and insurance that intensify financial risks. The digitisation level shows a positive but statistically insignificant impact, indicating that digital infrastructure has not yet been fully transformed into growth-enhancing productivity. Furthermore, the regional heterogeneity analysis reveals a stark divergence: DFI acts as a crucial growth engine in the financially underserved central and western regions, whereas excessive financialisation has exerted a crowding-out effect in eastern cities. These findings suggest that policy efforts should prioritise broadening DFI coverage while strengthening the regulation of usage-related activities, thereby balancing financial innovation with systemic stability.

1. Introduction

Digital financial inclusion (DFI) serves as a critical mechanism for expanding economic opportunity and reducing income inequality. The continuous progression of digitisation accelerates DFI development, extending its socio-economic benefits. This study synthesises recent literature to examine how DFI affects economic growth, focusing on three primary channels: technological innovation, capital deepening, and total factor productivity (TFP).

DFI represents the convergence of digital technology and financial inclusion. Its capacity to stimulate economic growth depends heavily on expanding financial service access via digital technologies. Empirical evidence consistently demonstrates a positive correlation between DFI and economic growth, especially in financially underserved regions. For instance, Chinoda and Kapingura (2023) show that in Africa, robust institutions amplify the benefits of DFI by ensuring the equitable distribution of financial resources. From a broader perspective, Sha’ban et al. (2021) construct a financial inclusion index for 95 countries, identifying a positive correlation with economic indicators such as GDP per capita and employment from 2004 to 2015. This aligns with findings from R. Huang et al. (2021), who report that improved financial access significantly drives growth, particularly in low-income nations and newer EU member states.

The positive externalities of DFI extend to broader economic stability and sustainability. In G20 countries, financial inclusion is linked to long-term improvements in financial efficiency, sustainability, and economic growth (Khan et al., 2022). Pal and Bandyopadhyay (2022) confirm that financial inclusion fosters sustainable development by enhancing both economic growth and bank efficiency. Ozturk and Ullah (2022) find that DFI promotes growth across 42 Belt and Road Initiative (BRI) countries. Similarly, Ali et al. (2023) argue that DFI facilitates the achievement of the Sustainable Development Goals (SDGs) by reducing poverty and inequality. Moreover, recent scholarship indicates that FinTech adoption significantly bolsters ESG performance in the banking sector (Abu Alim & Mansour, 2026), while digital economy development acts as a catalyst for firm-level sustainability performance (Mansour et al., 2026b).

Technological innovation and digitisation serve as fundamental channels through which DFI affects growth. By integrating a diverse range of participants into the financial system, DFI stimulates innovation and social inclusion. Farzana et al. (2024) show that technological improvements in DFI enable broader access to financial services, increasing economic participation. Fintech innovations in digital lending and credit access expand financial inclusion, particularly in rural and underserved areas (Hasan et al., 2021). In China, where the financial system is dominated by state-owned banks, DFI reduces information asymmetry, alleviates financing constraints for private enterprises, and spurs innovation (Yao & Yang, 2022; Luo, 2022; Li & Pang, 2023). In this context, Mansour et al. (2026a) identify digital maturity as a strategic driver of organisational performance, suggesting that the effectiveness of digital transformation is contingent upon institutional readiness. Capital deepening also plays a critical role, with DFI increasing access to financial capital, thereby enhancing TFP and labour productivity (Sethi & Acharya, 2018; X. Huang & Meng, 2023; Du et al., 2024).

However, the efficacy of DFI exhibits significant dimensional heterogeneity. While expanding coverage breadth generally yields positive outcomes, rapid growth in usage depth may lead to financial risks. Tang (2019) notes that fintech lending in China expanded rapidly but attracted high-risk borrowers, suggesting that excessive depth can outpace risk management.

Despite this growing body of evidence, several gaps remain in the existing literature. First, most studies treat DFI as a single composite index, obscuring the potentially divergent—or even opposing—effects of its sub-dimensions. The coverage breadth, usage depth, and digitisation level of DFI may operate through fundamentally different mechanisms, yet their distinct impacts on economic growth remain insufficiently examined. Second, existing empirical work predominantly relies on provincial- or country-level data, which may mask substantial within-country heterogeneity. City-level analysis offers a more granular perspective on how DFI translates into local economic outcomes. Third, while individual transmission channels such as innovation or capital accumulation have been explored in isolation, a unified empirical framework that simultaneously examines capital deepening, TFP, and technological innovation as mediating pathways is lacking. Addressing these gaps is critical, as relying on aggregate indices may yield misleading policy conclusions by masking the potentially adverse effects of specific financial dimensions. Accordingly, this study seeks to answer two central research questions: (1) Through which structural channels—capital deepening, productivity, or innovation—does DFI exert its influence on the real economy? (2) Do the three distinct dimensions of DFI contribute to growth in a uniform manner, or do they exhibit divergent impacts? To address these questions, this study utilises panel data from 278 Chinese prefecture-level cities over the period 2011–2019, decomposing DFI into its three sub-dimensions and jointly testing three transmission mechanisms. This timeframe provides a unique empirical window to identify the endogenous growth effects of DFI prior to 2020, when the COVID-19 pandemic, the P2P sector shutdown, and the regulatory crackdown on major fintech platforms collectively reshaped China’s digital financial landscape. From a theoretical perspective, the multidimensional nature of DFI implies distinct economic outcomes. Within the framework of endogenous growth, the coverage breadth facilitates growth by lowering entry barriers and driving capital deepening. Coverage breadth, by expanding access to formal finance, also improves resource allocation efficiency and enhances TFP (Sethi & Acharya, 2018; X. Huang & Meng, 2023), while simultaneously alleviating financing constraints for previously underserved firms and stimulating technological innovation (Yao & Yang, 2022; Li & Pang, 2023). Conversely, the usage depth reflects financial activity frequency; however, during periods of regulatory lag, it may lead to financial misallocation. Additionally, digitisation improves transaction efficiency but may trigger a ‘crowding-out effect’, potentially diverting resources away from the long-term capital required for corporate technological innovation. Based on this analysis, we propose the following hypotheses:

H1. 

Digital financial inclusion has a significant impact on China’s economic growth, but the effects differ across its three dimensions.

H1a. 

Coverage breadth of DFI positively contributes to growth.

H1b. 

Usage depth exerts an adverse effect due to financial risks and regulatory gaps.

H1c. 

The digitisation level positively influences growth by improving financial efficiency.

H2. 

DFI promotes economic growth primarily by enhancing capital deepening, TFP, and technological innovation.

2. Materials and Methods

2.1. Regression Framework

Following the augmented Solow growth model (Mankiw et al., 1992), we investigate the impact of digital financial inclusion on economic growth using the following two-way fixed effects specification:

$${\mathrm{G}\mathrm{r}\mathrm{o}\mathrm{w}\mathrm{t}\mathrm{h}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_{\mathrm{o}}+{\mathsf{\alpha }}_1{\mathrm{D}\mathrm{F}\mathrm{I}}_{\mathrm{i}\mathrm{t}}+{\sum }_{\mathrm{j}=2}^5{\mathsf{\alpha }}_{\mathrm{j}}{{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}^{\mathrm{j}}}_{\mathrm{i}\mathrm{t}}+{{\mathrm{Y}\mathrm{e}\mathrm{a}\mathrm{r}}_{\mathrm{t}}+{\mathrm{C}\mathrm{i}\mathrm{t}\mathrm{y}}_{\mathrm{i}}+\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

where ${\mathrm{G}\mathrm{r}\mathrm{o}\mathrm{w}\mathrm{t}\mathrm{h}}_{\mathrm{i}\mathrm{t}}$ is the logarithm of real GDP per capita for city i in year t, ${\mathrm{D}\mathrm{F}\mathrm{I}}_{\mathrm{i}\mathrm{t}}$ represents the digital financial inclusion level, ${{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}^{\mathrm{j}}}_{\mathrm{i}\mathrm{t}}$ denotes a vector of control variables (including CapInten, Labour, Edu, and Open), ${\mathrm{Y}\mathrm{e}\mathrm{a}\mathrm{r}}_{\mathrm{t}}$ and ${\mathrm{C}\mathrm{i}\mathrm{t}\mathrm{y}}_{\mathrm{i}}$ capture city and year specific fixed effects, and ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ is the idiosyncratic error term.

To examine the hypothesized transmission channels, we specify the following mediation models:

$${\mathrm{M}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_{\mathrm{o}}+{\mathsf{\beta }}_1{\mathrm{D}\mathrm{F}\mathrm{I}}_{\mathrm{i}\mathrm{t}}+{\sum }_{\mathrm{j}=2}^5{\mathsf{\beta }}_{\mathrm{j}}{{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}^{\mathrm{j}}}_{\mathrm{i}\mathrm{t}}+{{\mathrm{Y}\mathrm{e}\mathrm{a}\mathrm{r}}_{\mathrm{t}}+{\mathrm{C}\mathrm{i}\mathrm{t}\mathrm{y}}_{\mathrm{i}}+\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

where ${\mathrm{M}}_{\mathrm{i}\mathrm{t}}$ represents one of the three mechanism variables: capital deepening (CapDeep), total factor productivity (TFP), or technological innovation (Innov).

2.2. Data and Variables

This study employs an unbalanced panel dataset of 278 Chinese prefecture-level cities from 2011–2019. DFI data are obtained from the Peking University Digital Inclusive Finance Index Research Centre (https://en.idf.pku.edu.cn/ (accessed on 10 January 2024)), while city-level economic and demographic indicators are drawn from the China City Statistical Yearbook, the China Statistical Yearbook, and the CNRDS database.

The dependent variable is economic growth (Growth), measured as the logarithm of real GDP per capita deflated to the 2011 base year. The core independent variable is the digital financial inclusion index (DFI). To capture potential variations across its constituent dimensions, the aggregate index is further decomposed into three sub-indices: coverage breadth (Breadth), which measures the reach of electronic accounts; usage depth (Depth), which reflects the actual frequency of digital financial service use; and digitisation level (Digit), which captures the convenience and cost-efficiency of financial services. Logarithmic transformations are applied to the aggregate index and all sub-dimensions to reduce skewness and address heteroscedasticity.

Regarding the mechanism variables, capital deepening (CapDeep) is measured as the logarithm of the capital-to-labour ratio. Capital stock is estimated via the Perpetual Inventory Method (PIM), ${\mathrm{K}}_{\mathrm{i}\mathrm{t}}={\mathrm{K}}_{\mathrm{i},\mathrm{t}-1}\left(1-\mathsf{\delta }\right){+\mathrm{I}}_{\mathrm{i}\mathrm{t}}$, where the depreciation rate $\mathsf{\delta }$ is set at 9.6% based on the dataset specifications. The initial capital stock is calculated as ${\mathrm{K}}_{\mathrm{i}0}={\mathrm{I}}_{\mathrm{i}0}/\left(\mathrm{g}+\mathsf{\delta }\right)$, where g represents the average investment growth rate during the sample period. Total factor productivity (TFP) is derived using the Solow residual method. Technological innovation (Innov) is measured as the logarithm of annual patent applications.

Control variables include capital intensity (CapInten), labour participation (Labour), human capital (Edu), and economic openness (Open). Due to city-level data availability, Edu is measured using provincial-level average years of schooling. Descriptive statistics for all variables are presented in

Table 1. Variable selection and data sources.

Var Name	Symbol	Definition/Source	Mean	SD	Min	Max
Economic Growth (log)	Growth	Real GDP per capita	−1.480	0.610	−3.208	−0.523
Digital Financial Inclusion (log)	DFI	PKU digital finance index	5.003	0.507	3.525	5.652
Coverage Breadth (log)	Breadth	Index of coverage breadth	4.931	0.542	3.061	5.634
Usage Depth (log)	Depth	Index of usage depth	4.992	0.497	3.526	5.664
Digitisation Level (log)	Digit	Index of digitisation level	5.171	0.598	3.084	5.743
Capital Deepening(log)	CapDeep	Capital stock per worker	4.661	0.591	3.199	5.924
Total Factor Productivity (log)	TFP	Solow residual method	−2.887	0.595	−4.409	−2.079
Technological Innovation (log)	Innov	Number of patent applications	7.697	1.551	4.443	11.469
Capital Intensity(log)	CapInten	Ratio of capital stock to GDP	1.242	0.379	0.251	2.098
Labour Participation	Labour	Labour force growth rate	1.758	8.881	−22.697	35.707
Human Capital	Edu	Average years of education	9.577	0.643	7.897	11.124
Economic Openness	Open	Ratio of FDI to GDP	0.016	0.016	0.000	0.073

Note: Economic data for prefecture-level cities are obtained from the China City Statistical Yearbook and the China Statistical Yearbook; digital financial inclusion (DFI) data are provided by the Peking University Digital Inclusive Finance Index Research Centre (https://en.idf.pku.edu.cn/ (accessed on 10 January 2024)).

. All empirical analyses in this study were conducted using Stata 17.0 (StataCorp LLC, College Station, TX, USA).

3. Results

3.1. Benchmark Test

Table 2. Regression results for digital inclusive finance and economic growth.

	(1)	(2)	(3)
Variables	Growth	Growth	Growth
DFI	0.121 ***	0.263 ***	0.258 ***
	(4.997)	(20.014)	(6.666)
CapInten		−0.229 ***	−0.486 ***
		(−9.044)	(−19.399)
Labour		−0.006 ***	−0.004 ***
		(−14.680)	(−13.205)
Edu		0.053 ***	−0.092 ***
		(6.248)	(−5.161)
Open		−0.360	1.014 ***
		(−0.987)	(3.108)
Constant	−2.086 ***	−3.004 ***	−1.300 ***
	(−17.153)	(−32.975)	(−4.946)
Year FE	No	No	Yes
City FE	Yes	Yes	Yes
R-squared	0.010	0.942	0.957
F-statistic	24.972	233.263	116.548
Observations	2502	2502	2502

Note: *** represent the significance levels of 1%. t statistics are in parentheses.

presents the baseline regression results regarding the relationship between DFI and economic growth. Column (1) presents the results without control variables, the coefficient on DFI is 0.121, which is statistically significant at the 1% level. This indicates that an increase in the level of digital financial inclusion can effectively stimulate economic growth. In Column (2), after incorporating city-level control variables (CapInten, Labour, Edu, and Open), the coefficient on DFI increases to 0.263, remaining significant at the 1% level. Furthermore, Column (3)—our preferred specification controlling for both year and city fixed effects—shows a coefficient of 0.258, which is significant at the 1% level. These empirical results confirm that DFI plays a significant and robust role in promoting China’s economic growth, supporting Hypothesis 1.

Furthermore, to rule out potential multicollinearity concerns among the explanatory variables, we conduct a variance inflation factor (VIF) test. The mean VIF is 1.24, and the maximum VIF is 1.49; both values are well below the conventional strict threshold of 5, indicating that our models do not suffer from severe multicollinearity. To validate the choice of fixed effects over random effects, we conduct a Hausman test. The resulting chi-squared statistic of 192.76 (p < 0.001) firmly rejects the null hypothesis of no systematic difference between FE and RE estimates, confirming that city-level unobserved heterogeneity is correlated with the regressors. This result validates our use of the two-way fixed effects estimator as the appropriate strategy for addressing time-invariant omitted variable bias.

3.2. Transmission Mechanism Testing

Having established the beneficial effect of DFI on economic growth, we further analyse the transmission channels.

Table 3. Regression results of mechanism tests.

	(1)	(2)	(3)	(4)	(5)
Variables	CapDeep	TFP	Innov	Growth	Growth
DFI	0.343 ***	0.129 ***	0.641 ***	−0.021	0.253 ***
	(11.568)	(3.600)	(7.813)	(−0.498)	(6.443)
CapDeep				0.271 ***
				(10.495)
Innov					0.008
					(0.808)
CapInten	0.545 ***	−0.602 ***	−0.174 ***		−0.485 ***
	(27.192)	(−24.786)	(−3.283)		(−19.294)
Labour	−0.000	−0.005 ***	−0.002 ***	−0.004 ***	−0.004 ***
	(−0.280)	(−13.468)	(−2.875)	(−9.951)	(−13.130)
Edu	−0.146 ***	−0.035 **	−0.251 ***	−0.059 ***	−0.090 ***
	(−10.330)	(−2.059)	(−6.674)	(−2.971)	(−4.996)
Open	2.046 ***	0.271	1.716 **	−0.951 ***	1.000 ***
	(8.346)	(0.912)	(2.485)	(−2.779)	(3.061)
Constant	3.633 ***	−2.448 ***	7.092 ***	−2.064 ***	−1.357 ***
	(17.693)	(−9.832)	(12.746)	(−6.982)	(−4.985)
Year FE	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes
R-squared	0.974	0.963	0.970	0.956	0.957
F-statistic	291.191	154.614	28.918	54.152	97.217
Observations	2376	2376	2502	2376	2502

Notes: t statistics in parentheses. *** p < 0.01, ** p < 0.05. In Column (4), CapInten is excluded to avoid multicollinearity with CapDeep, as both variables are capital-based measures.

presents the regression results of DFI on the mechanism variables: capital deepening (CapDeep), total factor productivity (TFP), and technological innovation (Innov). Columns (1) to (3) report that the regression coefficients on DFI are 0.343, 0.129, and 0.641, respectively, all significant at the 1% level.

To rigorously verify the transmission chain while mitigating endogeneity concerns, we apply a mixed-step identification strategy tailored to the specific characteristics of each mechanism variable:

For capital deepening, we conduct a standard three-step mediation test. As shown in Column (4) of Table 3, when CapDeep is incorporated into the baseline growth model, it exhibits a significantly positive direct effect (0.271, p < 0.01). Meanwhile, the coefficient of DFI drops sharply from 0.258 to −0.021 and becomes statistically insignificant. This confirms that capital deepening plays a complete mediating role.

For TFP, a standard three-step mediation test is methodologically precluded. Since TFP is constructed as the Solow residual from GDP, the independence assumption underlying Baron and Kenny’s (1986) mediation framework is violated by construction. Introducing TFP alongside DFI in a joint regression on GDP would generate mechanical endogeneity rather than identify a causal channel. Instead, following the two-step identification approach, we establish the DFI→TFP link (Column 2). Given the extensively documented role of TFP as a fundamental driver of long-run growth (Romer, 1990), this provides a complete and rigorous transmission chain.

Conversely, for technological innovation, the three-step test reveals a different dynamic. As shown in Column (5), when both DFI and Innov are controlled for, the coefficient of Innov is 0.008 and statistically insignificant, whilst the DFI coefficient remains highly significant (0.253, p < 0.01). This indicates that the mere quantity of patent applications does not serve as an effective short-term transmission channel. This empirical finding aligns with the existing literature on China’s institutional context, which notes that the surge in patent quantities is often driven by government subsidies and administrative targets and does not necessarily translate into immediate productivity gains (Hu & Jefferson, 2009; Boeing, 2016).

These results indicate that DFI significantly enhances capital accumulation and improves production efficiency. While it fosters innovation activities, the immediate growth effects are primarily driven by capital deepening and TFP.

In Column (4), the coefficient of Open flips to a negative sign. This is a common statistical artefact caused by multicollinearity. In China, foreign direct investment is a major driver of fixed asset investment and is thus highly correlated with capital deepening. When both are included, the partial effect of Open captures residual variation, leading to the observed sign change.

3.3. Analysis of Digital Finance Sub-Dimensions

DFI is a multi-dimensional concept. To reveal the divergent effects proposed in Hypothesis 1, we decompose the aggregate index into three sub-dimensions: coverage breadth (Breadth), usage depth (Depth) and digitisation level (Digit). To rigorously address concerns regarding potential conceptual overlap (multicollinearity), we conducted a variance inflation factor (VIF) test for the sub-dimension models. The results show a mean VIF of 1.22 and a maximum VIF of 1.42, both of which are well below the strict threshold of 5. This confirms that there is no severe multicollinearity between the sub-dimensions and the control variables.

Table 4. Regression results of sub-index and economic growth.

	(1)	(2)	(3)
Variables	Growth	Growth	Growth
Breadth	0.174 ***
	(10.352)
Depth		−0.060 **
		(−2.166)
Digit			0.008
			(0.569)
CapInten	−0.484 ***	−0.449 ***	−0.456 ***
	(−19.782)	(−17.815)	(−18.243)
Labour	−0.004 ***	−0.005 ***	−0.004 ***
	(−13.414)	(−13.188)	(−13.137)
Edu	−0.090 ***	−0.099 ***	−0.100 ***
	(−5.105)	(−5.535)	(−5.572)
Open	0.931 ***	1.349 ***	1.322 ***
	(2.903)	(4.139)	(4.054)
Constant	−0.889 ***	0.309	−0.013
	(−4.653)	(1.424)	(−0.071)
Year FE	Yes	Yes	Yes
City FE	Yes	Yes	Yes
R-squared	0.958	0.956	0.956
F-statistic	132.087	106.701	105.619
Observations	2502	2502	2502

Notes: t statistics in parentheses. *** p < 0.01, ** p < 0.05.

presents the results. The coefficient on coverage breadth (Column 1) is positive (0.174) and significant at the 1% level, suggesting that expanding the reach of financial services is the primary driver of growth. However, the usage depth (Column 2) is significantly negative (−0.060) at the 5% level. This supports Hypothesis 1b, suggesting that an excessive deepening of financial usage, such as internet credit and insurance, might adversely affect growth. This is likely due to regulatory lags and the accumulation of financial risks in the absence of adequate supervision. The digitisation level (Column 3) is positive but not statistically significant, indicating that the mere digitisation of infrastructure has a marginal direct impact on growth compared to coverage expansion.

Table 5. Regression results of sub-index and mechanism variables.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Variables	CapDeep	CapDeep	CapDeep	TFP	TFP	TFP	Innov	Innov	Innov
Breadth	0.170 ***			0.120 ***			0.338 ***
	(13.383)			(7.796)			(9.404)
Depth		−0.023			−0.087 ***			0.028
		(−1.046)			(−3.322)			(0.478)
Digit			−0.020 *			0.003			−0.116 ***
			(−1.758)			(0.187)			(−3.803)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
R-squared	0.975	0.973	0.973	0.964	0.963	0.963	0.970	0.969	0.969
F-statistic	305.610	248.928	249.564	167.619	154.091	151.099	34.599	16.309	19.260
Observations	2376	2376	2376	2376	2376	2376	2502	2502	2502

Notes: t statistics in parentheses. *** p < 0.01, * p < 0.1. Control variables are included in all specifications but omitted for brevity.

further breaks down these effects on the mechanism variables. Notably, while Breadth consistently promotes capital deepening, TFP, and innovation (Columns 1, 4, 7), the Depth shows negative signs for TFP (Column 5) and capital deepening (Column 2). Interestingly, the digitisation level significantly hinders technological innovation (Column 9, −0.116 ***), which may suggest a crowding-out effect or that the current digital infrastructure has not yet effectively translated into innovation outputs.

3.4. Endogeneity Test Results

To address potential endogeneity arising from bidirectional causality and omitted variable bias, this study employs an Instrumental Variable (IV) approach to ensure robust causal identification.

The primary instrument is a Shift-Share (Bartik) IV, constructed by interacting the historical telecommunication infrastructure (the number of fixed-line telephones per 10,000 people in each city in 1994) with a linear time trend. This instrument satisfies the relevance condition because historical fixed-line networks provided the essential physical foundation and technical path dependence for subsequent digital infrastructure and mobile internet expansion. Crucially, it also meets the exclusion restriction: as a historical technological indicator from over two decades prior to our sample period, the 1994 telephone density is highly unlikely to directly affect urban economic growth during 2011–2019 through any channel other than its influence on the development of modern digital financial networks. Furthermore, after controlling for city fixed effects, all time-invariant city-level confounders—such as local governance quality and industrial base—are absorbed, isolating the variation in digital infrastructure buildout as the sole operative channel.

The supplementary instrument is a Policy-induced IV, constructed by interacting the 1994 telephone density with a policy dummy (equal to 1 for years ≥ 2015 and 0 otherwise), leveraging the 2015 Internet Finance Rectification as an exogenous institutional shock. This instrument captures how national regulatory tightening differentially affected cities based on their historical infrastructure foundations, satisfying both the relevance and exogeneity conditions by anchoring the policy variation in pre-determined city characteristics.

Table 6. IV (2SLS) Estimation Results: Shift-Share and Policy-induced Approaches.

	(1)	(2)	(3)	(4)
Variables	DFI	Growth	DFI	Growth
IV (Shift-Share)	−0.015 ***
	(−24.306)
IV (Policy-induced)			−0.072 ***
			(−20.524)
DFI		0.128 *		0.056
		(1.837)		(0.700)
Controls	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes
KP rk Wald F-statistic	590.76		421.22
R-squared	0.987	0.248	0.986	0.248
F-statistic	165.749	129.871	128.691	129.431
Observations	2232	2232	2232	2232

Note: t-statistics are reported in parentheses. *** and * indicate significance at 1% and 10% levels respectively.

presents the results of the Two-Stage Least Squares (2SLS) estimation. Columns (1) and (2) report the first-stage and second-stage results for the Shift-Share IV, while Columns (3) and (4) report the same for the Policy-induced IV.

To ensure the reliability of our identification, we conduct several diagnostic tests. The Kleibergen–Paap rk Wald F-statistics for the Shift-Share IV and the Policy-induced IV are 590.76 and 421.22 respectively, well above the Stock-Yogo critical value of 16.38, thus dismissing weak instrument concerns. As each specification employs a single instrument for one endogenous variable, the system is exactly identified and an overidentification test is not applicable. The exogeneity of the instruments is supported by the theoretical arguments above and the high first-stage F-statistics.

As shown in Table 6, both instruments exhibit high relevance. The first-stage F-statistics for the Shift-Share IV (590.76) and Policy-induced IV (421.22) are both well above the empirical threshold of 10, ruling out weak instrument concerns. In the second stage, the coefficient on DFI in Column (2) remains significantly positive (0.128, p < 0.1), confirming that digital finance continues to stimulate economic growth after addressing endogeneity. The insignificance of the Policy-induced IV in Column (4) reflects a fundamentally different Local Average Treatment Effect (LATE). While the Bartik IV identifies the growth effect of DFI driven by organic infrastructure expansion, the Policy-induced IV identifies the effect among cities whose DFI was disrupted by regulatory tightening. The 2015 rectification simultaneously suppressed non-compliant activities and constrained legitimate digital finance services, generating offsetting effects that attenuate the net growth impact. This divergence is therefore theoretically expected rather than indicative of inconsistency. The directional consistency across both specifications nonetheless supports the validity of our baseline findings.

3.5. Regional Heterogeneity Analysis

Given the vast disparities in economic development and financial infrastructure across China, the economic impact of digital financial inclusion may exhibit significant regional heterogeneity. To address this, we follow the classification standards of the National Bureau of Statistics of China and divide the full sample into Eastern and Central-Western city sub-samples.

Table 7. Heterogeneous effects of digital financial inclusion across regions.

	(1)	(2)
Variables	Growth (East)	Growth (Central-Western)
DFI	−0.123 **	0.442 ***
	(−1.980)	(8.305)
CapInten	−0.300 ***	−0.521 ***
	(−9.144)	(−15.500)
Labour	−0.002 ***	−0.005 ***
	(−4.061)	(−12.088)
Edu	−0.006	−0.159 ***
	(−0.300)	(−6.175)
Open	1.072 ***	1.540 **
	(3.869)	(1.975)
Constant	−0.808 **	−1.323 ***
	(−2.041)	(−3.695)
Year FE	Yes	Yes
City FE	Yes	Yes
R-squared	0.987	0.879
F-statistic	22.101	100.495
Observations	900	1602

Notes: t statistics in parentheses. *** p < 0.01, ** p < 0.05.

presents the regional heterogeneity estimation results. Interestingly, the impact of DFI displays a stark divergence between regions. In the Central-Western cities (Column 2), the coefficient on DFI is positive (0.442) and highly significant at the 1% level. Conversely, in the Eastern cities (Column 1), the coefficient turns significantly negative (−0.123, p < 0.05).

This contrasting result highlights the fundamental “inclusive” characteristic of digital finance. Central-Western regions are typically characterised by underdeveloped traditional financial systems and higher geographical barriers. Here, DFI acts as a critical substitute, overcoming physical constraints to provide “long-tail” users and small enterprises with accessible credit, thereby strongly catalysing local economic growth. In contrast, Eastern regions already possess highly mature and saturated traditional financial markets. In such contexts, the excessive expansion of digital finance—particularly in the form of speculative internet credit and peer-to-peer (P2P) lending before the regulatory tightening—exhibits diminishing marginal returns. It may even induce the financialisation of the real economy and crowd out productive investments, leading to a negative impact on regional GDP growth. This regional divergence perfectly aligns with our findings in Section 3.3 regarding the structural risks associated with excessive financial usage depth.

To ensure the negative impact of usage depth is not driven by specific policy shocks (e.g., the 2015 Rectification or 2018 P2P crisis), we perform sub-sample robustness tests. As reported in Appendix A 

Table A1. Sub-Sample Robustness Tests.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Variables	Growth	Growth	Growth	Growth	Growth	Growth	Growth	Growth
DFI	0.219 ***	0.131 ***
	(5.216)	(4.003)
CapInten	−0.484 ***	−0.552 ***	−0.484 ***	−0.551 ***	−0.446 ***	−0.517 ***	−0.457 ***	−0.527 ***
	(−16.652)	(−18.892)	(−17.079)	(−19.383)	(−15.269)	(−17.623)	(−15.800)	(−18.311)
Labour	−0.003 ***	−0.003 ***	−0.004 ***	−0.003 ***	−0.003 ***	−0.003 ***	−0.003 ***	−0.003 ***
	(−8.213)	(−8.421)	(−8.447)	(−8.639)	(−8.090)	(−8.353)	(−8.053)	(−8.321)
Edu	−0.108 ***	−0.086 ***	−0.107 ***	−0.085 ***	−0.117 ***	−0.089***	−0.117 ***	−0.090 ***
	(−5.055)	(−5.431)	(−5.055)	(−5.404)	(−5.422)	(−5.591)	(−5.405)	(−5.646)
Open	1.284 ***	0.998 ***	1.193 ***	0.914 ***	1.620 ***	1.187 ***	1.577 ***	1.167 ***
	(3.359)	(3.195)	(3.181)	(2.971)	(4.252)	(3.819)	(4.134)	(3.755)
Breadth			0.162 ***	0.105 ***
			(8.918)	(7.442)
Depth					−0.074 **	−0.040 *
					(−2.483)	(−1.714)
Digit							0.002	0.005
							(0.148)	(0.438)
Constant	−0.961 ***	−0.716 ***	−0.681 ***	−0.596 ***	0.508 **	0.096	0.148	−0.100
	(−3.208)	(−3.256)	(−3.008)	(−3.614)	(2.009)	(0.530)	(0.667)	(−0.621)
R-squared	0.957	0.975	0.958	0.975	0.956	0.974	0.956	0.974
F-statistic	74.486	94.133	87.097	104.147	69.414	90.806	67.934	90.107
Observations	1946	1946	1946	1946	1946	1946	1946	1946

Notes: t statistics are reported in parentheses. Columns (1), (3), (5), and (7) present robustness results excluding the years 2015 and 2018 to mitigate the impact of the Internet Finance Rectification and the 2018 P2P crisis. Columns (2), (4), (6), and (8) utilise the pre-2018 sub-sample to isolate organic growth effects from the systemic platform failures observed in 2018–2019. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

, the coefficients on Depth remain significantly negative when excluding major shock years (Column 5) or restricting the sample to the pre-2018 period (Column 6). This confirms that the adverse growth effect is a systemic structural issue rather than a reaction to isolated regulatory events, further validating our core findings.

To further address the heterogeneity across DFI sub-dimensions, Appendix A 

Table A2. Regional Heterogeneity of DFI Sub-dimensions.

	(1)	(2)	(3)	(4)	(5)	(6)
Variables	Growth (East)	Growth (Central-Western)	Growth (East)	Growth (Central-Western)	Growth (East)	Growth (Central-Western)
Breadth	0.046	0.215 ***
	(1.299)	(10.162)
Depth			−0.338 ***	−0.013
			(−6.224)	(−0.358)
Digit					−0.002	0.011
					(−0.109)	(0.562)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes
R-squared	0.987	0.882	0.987	0.873	0.986	0.873
Observations	900	1602	900	1602	900	1602

Notes: t statistics in parentheses. *** p < 0.01. All specifications include year and city fixed effects. Eastern cities are defined according to the National Bureau of Statistics of China classification.

reports the regional heterogeneity results for coverage breadth, usage depth, and digitisation level separately. The results reveal a clear asymmetric pattern: coverage breadth exerts a significantly positive effect only in Central-Western regions (0.215, p < 0.01), while usage depth exhibits a significantly negative effect only in Eastern regions (−0.338, p < 0.01). Digitisation level remains insignificant in both regions, consistent with the main regression results. This asymmetric pattern confirms that the inclusive growth effect of DFI operates primarily through expanding financial access in underserved regions, while the structural risks of excessive usage depth are concentrated in financially mature markets.

4. Discussion

This study provides empirical evidence that DFI exerts heterogeneous effects on economic growth depending on which dimension is considered, a finding that carries both theoretical and policy implications.

The significantly positive effect of coverage breadth is consistent with the broader financial inclusion literature (Chinoda & Kapingura, 2023; Sha’ban et al., 2021; R. Huang et al., 2021; Khan et al., 2022) and aligns with the view that expanding access to formal financial services lowers transaction costs, reduces information asymmetry, and broadens the pool of productive investment. Our mechanism analysis further reveals that this growth effect operates primarily through capital deepening and TFP improvement, suggesting that wider account ownership facilitates capital accumulation among previously underserved households and firms, and improves the efficiency with which resources are allocated across the economy (Sethi & Acharya, 2018; X. Huang & Meng, 2023). The robustness of this finding across both OLS and 2SLS specifications strengthens confidence that coverage expansion represents a genuine driver of city-level growth rather than a statistical artefact.

The negative and significant effect of usage depth is arguably the most policy-relevant finding of this study. While deeper engagement with digital financial services—such as internet credit, online insurance, and digital wealth management—might be expected to deepen financial intermediation and stimulate productive activity, our results suggest the opposite holds in the Chinese context during 2011–2019. This period coincided with the rapid and largely unregulated expansion of China’s peer-to-peer (P2P) lending sector and other fintech credit platforms, which attracted high-risk borrowers and accumulated systemic vulnerabilities that ultimately triggered widespread platform failures between 2018 and 2019 (Tang, 2019). Furthermore, the negative effects on TFP and capital deepening further suggest that usage depth may have misallocated capital toward speculative or consumption-oriented activities rather than productive investment, undermining rather than enhancing long-run growth. These findings echo concerns raised in the literature that the pace of fintech innovation can outstrip the capacity of regulatory frameworks to manage associated risks (Tay et al., 2022) and underscore the importance of sequencing financial deepening with appropriate supervisory infrastructure. The comprehensive clean-up of the P2P lending sector and the subsequent regulatory interventions targeting major fintech platforms such as Ant Group after 2020 provide real-world corroboration of the risks associated with usage depth identified in this study. These post-sample developments are broadly consistent with our empirical finding that excessive usage depth was associated with negative growth outcomes during 2011–2019, lending external validity to our results.

The positive but statistically insignificant effect of the digitisation level points to a different kind of limitation. Digital infrastructure—encompassing the convenience, cost-efficiency, and credit transformation dimensions of DFI—may require a longer gestation period before its productivity benefits materialise. The 2011–2019 window likely captures a phase in which significant investments were being made in digital financial infrastructure, but the complementary conditions necessary to translate that infrastructure into growth-enhancing outcomes—such as digital literacy, interoperable payment systems, and regulatory clarity—had not yet fully developed. This interpretation is consistent with Khera et al. (2021), who find that the growth effects of digital financial inclusion are conditional on institutional quality and human capital. The counterintuitive negative impact of the digitisation level on technological innovation (Table 5) warrants a more nuanced theoretical interpretation. This suggests a structural crowding-out dynamic prevalent during the sample period. As digital financial infrastructure rapidly outpaced regulatory maturity, platforms potentially prioritised high-frequency, algorithm-driven consumption lending, which offers quicker returns but diverts patient capital away from high-risk, long-cycle technological R&D. Consequently, while digitisation lowers transaction costs, it may simultaneously reallocate entrepreneurial resources toward speculative financial activities, thereby hindering substantive innovation outputs.

5. Conclusions

This study examines the relationship between digital financial inclusion and economic growth using panel data from 278 Chinese prefecture-level cities over 2011–2019. By decomposing DFI into three sub-dimensions and jointly testing three transmission mechanisms, we contribute to the literature in two respects. Empirically, we demonstrate that DFI’s growth effects are highly dimension-specific: coverage breadth exerts a robust positive effect operating through capital deepening, TFP, and technological innovation, while usage depth exerts a negative effect, likely reflecting regulatory deficiencies in China’s rapidly expanding digital credit markets during the sample period. Digitisation level shows no significant direct growth effect, suggesting that infrastructure investment alone is insufficient without the complementary conditions needed to generate productivity gains. Moreover, we find a profound regional divergence, demonstrating that DFI powerfully stimulates growth in Mid-Western regions but negatively impacts Eastern regions due to diminishing marginal returns and capital misallocation. Theoretically, these findings challenge the treatment of DFI as a homogeneous construct and highlight the risk of drawing policy conclusions from aggregate indices that mask opposing sub-dimensional effects.

These findings carry concrete policy implications. First, policymakers should continue to prioritise the expansion of coverage breadth—particularly in underserved urban peripheries and lower-tier cities—as the returns to access expansion remain robust and operate through well-identified productive channels. Second, the negative growth effects of usage depth underscore the urgent need for more actionable and proactive regulation of digital credit markets. Rather than generic regulatory tightening, policymakers should implement specific regulatory tools, such as dynamic capital requirements for digital lenders to mitigate systemic vulnerabilities. Furthermore, it is essential to establish robust consumer protection frameworks tailored to high-frequency digital credit users, thereby ensuring that the pace of fintech innovation does not outstrip structural financial stability. Third, the underperformance of digitisation-level infrastructure suggests that investment in digital financial technology should be accompanied by parallel efforts to build digital literacy, strengthen regulatory capacity, and ensure interoperability across platforms, so that the productivity potential of digital infrastructure can be more fully realised.

The three structural shifts that reshaped China’s digital financial landscape after 2019 lend additional significance to these findings. The aggressive shutdown and restructuring of the P2P lending industry is consistent with the study’s identification of regulatory lag as the mechanism behind usage depth’s negative growth effect. The sweeping regulatory crackdown on major fintech platforms, including Ant Group and Tencent, targeted precisely the category of systemic risks identified here. Meanwhile, the COVID-19 pandemic’s exogenous shock to digital adoption makes a pre-pandemic baseline more necessary, not less: without establishing how DFI dimensions operate under organic market conditions, it is impossible to disentangle pandemic-driven forced digitalisation from DFI’s intrinsic growth contributions. The present study thus provides an indispensable reference point for understanding and governing China’s current digital financial landscape.

While providing robust evidence, this study identifies several directions for future research. Firstly, regarding data granularity, the use of provincial-level proxies for human capital was necessitated by the current absence of consistent, uniform education statistics at the prefectural level across the sample period. While this approach provides a robust approximation of regional human capital, future studies could further refine these estimations as more disaggregated city-level datasets become accessible. Secondly, as the analysis is situated within the unique institutional context of China, the generalisability of these findings—particularly the observed negative impact of usage depth—to other emerging or developed economies warrants further verification. Thirdly, while the Peking University DFI is a widely recognised proxy, it may not fully capture the complexity of the latest fintech innovations; thus, future studies could utilise alternative metrics to refine these insights. Fourthly, although we employ a dual IV strategy, residual bias from unobserved factors cannot be entirely ruled out. To address these issues, future research could usefully extend this framework to incorporate cross-country comparisons across diverse economic systems to test the international validity of these dimensional effects. Furthermore, it would be valuable to examine regional heterogeneity across different city tiers, assess whether the negative effects persist following China’s post-2019 regulatory tightening, and explore how the relationship between DFI dimensions and growth evolves globally as digital financial markets continue to mature.