Impact of Synergistic Governance of Digital Economy and Green Finance on Urban Carbon Total Factor Productivity: A Quasi-Natural Experiment from China’s Dual Pilot Programs

Abstract

The deep integration of the digital economy and green finance is a key strategic arrangement for promoting high-quality industrial development in the new stage of development. This paper uses China’s dual pilot program—comprising both Big Data Comprehensive Pilot Zones and Green Finance Reform and Innovation Pilot Zones—as a quasi-natural experiment. Based on panel data from 285 prefecture-level cities spanning 2010–2023, and employing a dual machine learning approach, the study investigates how the digital economy and green finance synergistically enhance urban carbon total factor productivity. The study finds that, compared to cities with single-policy pilot programs, the synergy of digital and green policies can promote an increase in urban carbon total factor productivity. This conclusion remains valid after a series of robustness tests, including changing the sample period, adjusting machine learning model settings, and introducing instrumental variables. Mechanism tests indicate that the synergy of digital and green policies can enhance urban carbon total factor productivity through three pathways: increasing government focus on green development, raising the level of urban green technological innovation, and expanding the scale of green investment. Heterogeneity analysis reveals that the synergistic effects vary across cities with different resource endowments and geographical locations. This study uncovers the underlying logic of how the digital economy and green finance synergistically drive urban development and transformation, providing empirical evidence from China for the formulation of sustainable development policies tailored to local conditions.

1. Introduction

Currently, the global economy and society are accelerating their transition toward a green and low-carbon future. As two key drivers of sustainable development, the deep integration of digital technology and green finance has become a critical strategic choice for countries seeking to enhance industrial competitiveness and achieve high-quality development. The 2024 Chinese Government Work Report explicitly states that efforts should be made to deepen the innovative development of the digital economy, vigorously develop green finance, and promote the greening and decarbonization of the economy and society. However, while maintaining rapid economic growth, China has paid a high environmental cost. Issues related to carbon and pollutant emissions are becoming increasingly prominent, and a strong coupling still exists between the processes of industrialization and urbanization and energy consumption and carbon emissions. The key to achieving a balance between high-level environmental protection and high-quality growth under the constraints of the “dual carbon” goals lies in maximizing factor productivity while controlling carbon emissions.

In response to these challenges, the Chinese government has placed high priority on ecological civilization construction and has successively introduced a series of policies aimed at guiding the green transformation of industries through institutional innovation. Among these, the Comprehensive Big Data Pilot Zones and the Green Finance Reform and Innovation Pilot Zones are two representative types of pilot initiatives. The former were established in two batches in 2015 and 2016, respectively, with the aim of promoting the aggregation, openness, and sharing of data as a production factor; leveraging the enabling role of digital technology in traditional industries; optimizing resource allocation; and enhancing production efficiency. The latter focuses on establishing a system of financial instruments such as green credit and green bonds to channel capital toward green industries—including energy conservation, environmental protection, and clean energy—and to reduce the financing costs of green projects. Theoretically, while big data pilot zones enhance information transparency and factor allocation efficiency, and green finance pilot zones lower the financing costs and risk premiums of green capital, the synergy between the two creates a combined policy effect at the level of carbon-based total factor productivity.

However, existing research has largely focused on the individual policy impacts of big data pilot zones and green finance pilot zones on green transition, primarily examining aspects such as carbon emission intensity, green technological innovation, and corporate emission reduction behavior [1,2]. Few studies have integrated the two into a unified framework to systematically evaluate the impact of their synergistic effects on carbon total factor productivity, a comprehensive efficiency indicator [3,4,5]. Some literature examines the enabling effects of big data pilot zones on green technological innovation, while other studies explore the guiding role of green finance reforms in the emission reduction behaviors of heavily polluting enterprises [6,7]; however, these studies generally overlook the synergies between the two policies at the spatial and institutional levels. In fact, the development of the digital economy enhances green finance’s ability to screen investments and price risks by reducing information asymmetry, while green finance provides stable market expectations and financial support for the scenario-based application of digital technologies. Through pathways such as optimizing factor allocation, incentivizing green innovation, and directing capital flows toward low-carbon sectors, the synergy between these two policies collectively enhances urban carbon total factor productivity. Therefore, incorporating big data pilot zones and green finance pilot zones into a unified analytical framework to systematically examine the impact of policy synergy on carbon total factor productivity not only serves as an important supplement to existing literature but also provides new empirical evidence for understanding the institutional pathways of China’s green transition.

Based on this, this paper treats China’s dual pilot programs—the Comprehensive Big Data Pilot Zones and the Green Finance Reform and Innovation Pilot Zones—as a quasi-natural experiment. Using panel data from 285 prefecture-level cities spanning 2010–2023 and employing a dual machine learning approach, the study systematically examines the enabling effects of the synergy between the digital economy and green finance on urban carbon total factor productivity. The marginal contributions of this study are primarily reflected in the following three aspects: First, in terms of research perspective, it overcomes the limitations of single-policy evaluation by incorporating both big data pilot zones and green finance pilot zones into a unified analytical framework. By focusing on the impact of the synergy between digital and green policies on urban carbon total factor productivity, it expands the boundaries of research on policy synergy effects. Second, in terms of methodology, the paper introduces a dual machine learning model to effectively control for high-dimensional control variables and nonlinear relationships, thereby reducing potential model specification biases and sample selection biases associated with traditional econometric methods and enhancing the reliability of causal identification. Third, regarding mechanism and heterogeneity analysis, this study reveals the transmission mechanisms of synergy effects through three pathways: government focus on green development, urban green technological innovation, and green investment. Furthermore, by examining heterogeneity based on differences in resource endowments and geographic locations, it provides a theoretical basis and policy references for promoting urban green transformation tailored to local conditions.

2. Institutional Context and Theoretical Analysis

2.1. Institutional Context

Against the backdrop of global efforts to address climate change and promote a green, low-carbon transition, the Chinese government places great emphasis on the strategic role of digital technology and green finance in advancing sustainable development. In recent years, the central government has successively introduced a series of pilot policies aimed at exploring pathways for green transition through institutional innovation. Among these, the Comprehensive Big Data Pilot Zones and the Green Finance Reform and Innovation Pilot Zones represent key policy initiatives in the fields of the digital economy and green finance, forming a dual-pilot framework in certain cities.

2.1.1. Comprehensive Big Data Pilot Zones

With the rapid rise of the digital economy, data has gradually emerged as another critical factor of production. To fully unleash the enabling effects of data and promote the deep integration of digital technology with the real economy, the State Council issued the “Action Plan for Promoting Big Data Development” in 2015, explicitly proposing the establishment of comprehensive big data pilot zones. In September 2015, Guizhou Province was the first to be approved as the nation’s first National Comprehensive Big Data Pilot Zone. In October 2016, the National Development and Reform Commission, the Ministry of Industry and Information Technology, and the Cyberspace Administration of China jointly approved seven additional regions—the Beijing–Tianjin–Hebei region, the Pearl River Delta, Shanghai, Henan Province, Chongqing Municipality, Shenyang City, and the Inner Mongolia Autonomous Region—as National Comprehensive Big Data Pilot Zones.

2.1.2. Green Finance Reform and Innovation Pilot Zones

To guide social capital toward green industries and promote the transition of the economy toward green and low-carbon development, the People’s Bank of China, in conjunction with seven ministries and commissions—including the National Development and Reform Commission, the Ministry of Finance, and the former Ministry of Environmental Protection—jointly issued documents such as the “Overall Plan for the Green Finance Reform and Innovation Pilot Zones in Huzhou and Quzhou, Zhejiang Province” and the “Overall Plan for the Green Finance Reform and Innovation Pilot Zone in Guangzhou, Guangdong Province” in June 2017, formally launching the construction of green finance reform and innovation pilot zones. The first batch of pilot zones covered five provinces: Zhejiang, Jiangxi, Guangdong, Guizhou, and Xinjiang. Subsequently, the scope of the pilot zones gradually expanded, forming a diversified pilot framework covering regions with varying levels of development in the eastern, central, and western parts of the country.

2.2. Theoretical Analysis

As two key drivers of sustainable industrial development, the digital economy and green finance exhibit synergistic effects not only in the combined application of policy tools but also, more profoundly, in the allocation of factors of production, incentive mechanisms, and the enabling power of the institutional environment. Based on Porter’s Hypothesis and the theory of information asymmetry, this paper systematically elucidates the enabling mechanisms of the synergy between the digital economy and green finance for industrial sustainable development from the perspectives of both direct and indirect impacts.

2.2.1. Direct Impacts

Porter’s Hypothesis posits that moderate environmental regulations can compel enterprises to engage in technological innovation, thereby partially or even fully offsetting regulatory costs and enhancing production efficiency. From the perspective of factor allocation, the establishment of comprehensive big data pilot zones helps promote the aggregation, openness, and sharing of data as a production factor. By reducing information asymmetry, these zones optimize resource allocation and enhance the efficiency of urban factor allocation. In particular, pilot zones can provide data support for urban carbon emissions accounting and environmental performance monitoring, laying a data-driven foundation for improving carbon-based total factor productivity. Green finance reform and innovation pilot zones, on the other hand, focus on building a green finance toolkit. By internalizing environmental performance into financing costs, they guide capital toward low-carbon industries and improve the efficiency of green capital allocation [8,9]. Big data pilot zones reduce the screening costs of green finance through information empowerment, while green finance pilot zones provide stable financial support for the low-carbon application of digital technologies, thereby enhancing economic output efficiency per unit of carbon emissions. The two thus complement each other in terms of efficiency. From the perspective of incentives and constraints, policy synergy compels enterprises to adjust their production and innovation behaviors by strengthening the transmission mechanisms of environmental regulations and market signals. Big data technology increases the probability of detecting environmental violations and raises the cost of penalties, while green finance internalizes environmental performance into financing costs. The combined effect of these two factors subjects enterprises to stronger incentives for emissions reduction and a clearer efficiency orientation, prompting them to seek a better balance between output growth and carbon emission control [10,11]. It is critical to distinguish policy synergy from a mere additive policy combination. Synergy implies that the marginal return of the digital economy policy increases in the presence of green finance policy, and vice versa. This complementarity arises because digital technology reduces the screening and monitoring costs of green capital allocation, while green finance generates a defined price signal and capital flow that directs digital innovation towards low-carbon applications. Consequently, the combined effect of the two pilots on urban carbon total factor productivity (CTFP) should be strictly greater than the sum of their individual effects. This hypothesis of multiplicative interaction forms the core of our empirical synergy test. Consequently, this paper proposes the following hypotheses:

H1: 

The synergy between the digital economy and green finance policies generates a positive interaction effect on urban CTFP, with the joint impact exceeding the additive sum of the individual policy effects.

2.2.2. Indirect Effects

The synergistic effects of the digital economy and green finance are not only manifested in their direct impact on urban carbon total factor productivity but also exert influence through three indirect pathways: government behavior, technological innovation, and capital allocation.

(1) Shift in Government Perceptions of Green Development

Government priorities regarding green development are key determinants of regional environmental governance and industrial policy directions. By establishing a unified data platform, Big Data Comprehensive Pilot Zones enable local governments to dynamically monitor critical information such as carbon emissions and energy consumption structures, thereby reducing information asymmetry in environmental governance and shifting governance models from empirical judgment to data-driven decision-making [12]. Green Finance Pilot Zones, through institutional constraints, require local governments to promote environmental information disclosure by financial institutions and establish green credit incentive mechanisms, which objectively intensify the pressure to meet green development targets. The synergy between these two types of policies has prompted local governments to shift green governance from mere advocacy to quantifiable and assessable governance practices. When the government’s focus on green development translates into stronger environmental regulations and stricter industrial access standards, enterprises are compelled to improve the efficiency of factor allocation and reduce carbon emissions per unit of output, thereby driving the improvement of carbon-based total factor productivity [13].

(2) Enhancement of Urban Green Technology Innovation

Green technology innovation is the core driver of improving carbon-based total factor productivity. From the supply side, big data pilot zones, through open and shared data platforms, have reduced the cost of accessing environmental data and market demand information. This promotes regional knowledge spillovers and collaborative innovation, enabling enterprises to more accurately identify green technology R&D directions while reducing information search costs and trial-and-error costs during the R&D process [14]. From the demand side, green finance pilot zones provide stable financial support for green technology R&D and commercial application through green financial instruments, thereby alleviating financing constraints on corporate innovation activities. The synergy between these two policy types can direct innovation resources toward substantive green technologies. Big data technology can identify corporate greenwashing, while green finance internalizes environmental performance into financing costs, thereby curbing low-level repetitive innovation. When enterprises concentrate their R&D investments on green invention patents with genuine emission-reduction benefits, green technological innovation transforms into an endogenous driver for enhancing carbon total factor productivity [15].

(3) Expansion of Green Investment Scale

Green investment serves as the bridge connecting green finance and green industries; its scale and structure jointly influence carbon total factor productivity. From the perspective of financing supply, green finance pilot zones broaden financing channels for urban green projects and lower the capital costs and entry barriers for urban green investment through measures such as establishing dedicated green credit quotas, innovating financing secured by environmental rights and interests, and developing green bond markets [15]. From the perspective of information screening, big data pilot zones provide information support for green investment. Enterprise environmental profiles, project carbon emission accounting, and risk early-warning systems based on big data can reduce information asymmetry and moral hazard in green investment, thereby improving the efficiency of green capital allocation. The synergy between these two types of policies facilitates the transition of green investment toward large-scale and targeted initiatives. On the one hand, this expands the balance of green credit and the issuance scale of green bonds; on the other hand, it directs funds more effectively toward sectors with genuine emission reduction benefits. As the scale of green investment expands and its structure optimizes, the improvement in carbon total factor productivity gains sustained financial support [16].

In summary, the policy synergy between the digital economy and green finance empowers the enhancement of urban carbon total factor productivity by promoting a shift in the government’s green development mindset, elevating the level of urban green technological innovation, and expanding the scale of green investment. Consequently, this paper proposes the following hypotheses:

H2a: 

The synergy between the digital economy and green finance indirectly promotes the improvement of urban carbon total factor productivity by transforming the government’s concept of green development.

H2b: 

The synergy between the digital economy and green finance indirectly promotes the improvement of urban carbon total factor productivity by enhancing the level of urban green technological innovation.

H2c: 

The synergy between the digital economy and green finance indirectly promotes the improvement of urban carbon total factor productivity by expanding the scale of green investment.

3. Research Design

3.1. Model Specification

To accurately identify the causal effect of the synergy between digital economy and green finance policies on urban carbon total factor productivity (CTFP), we implement the double/debiased machine learning (DML) estimator for the partially linear regression model (PLR). This approach is purpose-built to handle high-dimensional control variables and nonlinear functional forms while delivering valid causal inference under mild regularity conditions [17].

The structural equation of interest is the partially linear model as shown in Equation (1):

$$CTF{P}_{i,t}={\theta }_{0}Dua{l}_{i,t}+g\left({\mathbf{X}}_{i,t}\right)+U_{i,t}, E\left(U_{i,t}\right|Dua{l}_{i,t},{\mathbf{X}}_{i,t})=0$$

(1)

Here, i denotes a city, and t denotes a year; $CTF{P}_{i,t}$ is the dependent variable, representing the carbon total factor productivity of city i in year t; $Dua{l}_{i,t}$ is the core independent variable, indicating whether city i is designated as a dual pilot city—serving as both a comprehensive big data pilot zone and a green finance reform and innovation pilot zone—in year t; ${\theta }_0$ is the core estimated coefficient of interest in this study, reflecting the treatment effect of policy synergy on industrial sustainable development; ${\mathbf{X}}_{i,t}$ represents a set of high-dimensional control variables; $g\left({\mathbf{X}}_{i,t}\right)$ denotes the nonlinear functional form of the control variables on industrial sustainable development, estimated using machine learning algorithms;$U_{i,t}$ is the error term, satisfying the assumption of a zero mean.

Directly plugging a machine learning estimate of g(X) into Equation (1) would yield a regularized estimator subject to two sources of bias: regularization bias from the complexity penalty of the machine learner, and overfitting bias from using the same observations for nuisance estimation and parameter inference. The DML framework overcomes both through the use of Neyman-orthogonal scores and cross-fitting.

Specifically, we re-write the model in terms of the following moment condition as shown in Equation (2):

$$\mathsf{\psi }(\mathrm{W};\mathrm{\theta },\mathrm{\eta })=[\mathrm{C}\mathrm{T}\mathrm{F}\mathrm{P}-\mathrm{l}(\mathrm{X})-\mathrm{\theta }(\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}-\mathrm{m}\left(\mathrm{X}\right)\left)\right]·(\mathrm{D}\mathrm{u}\mathrm{a}\mathrm{l}-\mathrm{m}(\mathrm{X}\left)\right)$$

(2)

where the nuisance functions are:

l(X) = E[CTFP | X]: the conditional expectation of the outcome given controls;

m(X) = E[Dual | X]: the propensity score, i.e., the conditional probability of receiving the dual-pilot treatment given controls.

The moment condition (2) is Neyman-orthogonal with respect to the nuisance parameters η = (l(X), m(X)). This property ensures that small estimation errors in the nuisance functions do not contaminate the estimate of θ₀ to first order, thereby eliminating the regularization bias that would plague a naive estimator.

To eliminate overfitting bias, we employ K-fold cross-fitting with K = 5. The procedure operates as follows:

Partition the sample. The full panel of 3790 observations is randomly split into 5 equally sized folds, stratified by city to preserve the panel structure within each fold.

Auxiliary estimation. For each fold k = 1, …, 5:

Using observations from the other 4 folds (the auxiliary sample), we train two separate machine learning models:

$\widehat{I}\left(X\right)$ predicting CTFP from X; $\widehat{m}\left(X\right)$ predicting Dual from X. These models produce out-of-sample predictions for the held-out fold k.

Parameter estimation. For observations in fold k, we construct the orthogonalized residuals as shown in Equation (3):

$$\tilde{Y}=CTFP-\widehat{I}\left(X\right),\tilde{D}=Dual-\widehat{m}\left(X\right)$$

(3)

and estimate θ₀ by regressing $\tilde{Y}$ on $\tilde{D}$ without an intercept. The final estimate $\widehat{\mathrm{\theta }}$ is the average of the fold-specific coefficients. This procedure ensures that each observation’s nuisance prediction comes from a model trained on an independent subset of the data, fully decoupling the machine learning step from the inference step.

To fully exploit the flexibility of DML, the control vector X is augmented with quadratic terms, selected two-way interactions, and pre-treatment city characteristics interacted with a linear time trend, allowing the model to capture complex nonlinear confounding. In the specific estimation process, this paper employs sample splitting and cross-fitting techniques, dividing the sample into several folds and alternately using one fold for estimation and the remaining folds for prediction to mitigate the risk of overfitting. Following existing studies, this paper sets the sample splitting ratio to 1:5. Meanwhile, random forest is adopted as the baseline machine learning algorithm.

3.2. Variable Selection

3.2.1. Dependent Variable (Urban Carbon Total Factor Productivity)

The core dependent variable is urban carbon total factor productivity (CTFP). Compared with single indicators such as carbon emission intensity or green total factor productivity, CTFP integrates the dual objectives of economic growth and carbon emission reduction, thus more accurately reflecting a city’s green efficiency performance under resource constraints. Following Cheng Z. and Chen W. & Yao L., we employ the slack-based measure (SBM) directional distance function incorporating undesirable outputs, combined with the global Malmquist–Luenberger (GML) index, to measure CTFP [18,19].

Specifically, each city is treated as a decision-making unit (DMU). In each period, three inputs—capital stock, labor, and total energy consumption—are used to produce the desirable output (real urban GDP) and the undesirable output (urban carbon emissions). Among these, labor is directly measured by year-end employment in each prefecture-level city from statistical yearbooks. Capital stock and total energy consumption, however, are not directly available at the prefecture level and must be estimated. The estimation procedures are detailed below [19].

Capital Stock Estimation. Prefecture-level capital stock is estimated using the perpetual inventory method with a base year of 2000. The capital stock of city i in year t is given by:

$$K(i,t)=K(i,t-1)(1-\delta )+\mathrm{I}(i,t)/\mathrm{P}(i,t)$$

(4)

where $K(i,t)$ is the real capital stock, $\mathrm{I}(i,t)$ is nominal fixed asset investment, $\mathrm{P}(i,t)$ is the province-level fixed asset investment price index. The initial capital stock in 2000, K(i,2000), is derived by allocating the province-level capital stock to cities proportional to each city’s share of fixed asset investment in that year:

$$K(i,2000) = \left(I\right(i,2000)/I(prov,2000\left)\right) \times K(prov,2000)$$

(5)

For cities established after 2000, the initial capital stock is set equal to the first-year fixed asset investment divided by the sum of the depreciation rate and the average growth rate of investment over the subsequent five years. All nominal values are deflated to constant 2010 prices.

Energy Consumption Estimation. Because prefecture-level energy consumption statistics are not published for most Chinese cities, we estimate city-level energy consumption (E(i,t)) by downscaling provincial total energy consumption using a composite allocation index. For each city i in province prov and year t, the allocation weight ω(i,t) is defined as:

$$\begin{gathered} \omega (i,t)=\left(\right(GDP(i,t)/GDP(prov,t))\times 0.3+(IndusVA(i,t)/IndusVA(prov,t))\times 0.5 \\ +(Light(i,t)/Light(prov,t))\times 0.2) \end{gathered}$$

(6)

where $\mathrm{G}\mathrm{D}\mathrm{P}(\mathrm{i},\mathrm{t})$ is gross regional product, $\mathrm{I}\mathrm{n}\mathrm{d}\mathrm{u}\mathrm{s}\mathrm{V}\mathrm{A}(\mathrm{i},\mathrm{t})$ is industrial value-added, and $\mathrm{L}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}(\mathrm{i},\mathrm{t})$ is the average nighttime light digital number from calibrated DMSP-OLS and VIIRS satellite data. The weights 0.3, 0.5, and 0.2 are chosen to reflect the relative importance of industrial activity, overall economic scale, and satellite-observed luminosity in shaping energy demand. The resulting $\mathsf{\omega }(\mathrm{i},\mathrm{t})$ is normalized so that the sum across all cities within a province equals 1. City-level energy consumption is then obtained as:

$$\mathrm{E}(i,t)=\omega (i,t)\times \mathrm{E}(prov,t)$$

(7)

where $\mathrm{E}(i,t)$ is the total provincial energy consumption (in tons of standard coal equivalent) from the China Energy Statistical Yearbook.

CO₂ Emissions Calculation. Urban carbon emissions, serving as the undesirable output, are computed using the IPCC sectoral approach based on the estimated energy consumption and provincial fuel mix.

With the three inputs and two outputs (desirable and undesirable) thus constructed, the GML index captures changes in CTFP and can be further decomposed into an efficiency change index and a technological change index. We accumulate the annual GML indices multiplicatively to obtain the relative level of CTFP for each city in each year, denoted as $CTF{P}_{i,t}$.

3.2.2. Key Independent Variable (Dual)

The core explanatory variable of this paper is a dummy variable for the synergy of dual-pilot policies. Following existing studies, if a city is designated both as a Big Data Comprehensive Pilot Zone and a Green Finance Reform and Innovation Pilot Zone, it is considered a dual-pilot city, taking the value 1 from the year of policy implementation onward, and 0 otherwise. For pilot zones established in batches, the policy implementation time is based on the earlier batch. For provinces fully covered by a pilot program, all prefecture-level cities under that province are treated as pilot cities.

3.2.3. Control Variables

To mitigate omitted variable bias [20,21,22,23], this study draws on existing literature and selects the following control variables: ① Population density (Popd): measured as the ratio of year-end total population to administrative area. ② Economic development level (Pgdp): measured as the natural logarithm of real GDP per capita. ③ Government intervention (Gove): measured as the ratio of local general public budget expenditure to regional GDP. ④ Openness (Open): measured as the ratio of total imports and exports to regional GDP. ⑤ Financial development (Fin): measured as the ratio of year-end deposits of financial institutions to regional GDP. ⑥ Urban economic density (Ued): measured as the ratio of GDP in built-up areas to urban area. ⑦ Human capital (Hcap): measured as the natural logarithm of the number of students enrolled in regular higher education institutions per 10,000 persons. ⑧ Urbanization level (Urban): measured as the ratio of urban population to year-end total population. ⑨ Environmental regulation intensity (Er): measured as the removal rate of industrial SO₂ (removed amount/generated amount) to control for the impact of environmental policy differences on carbon efficiency. ⑩ Industrial structure upgrading index (Indus): measured as the ratio of value added of the tertiary industry to that of the secondary industry, capturing the effect of industrial structure transformation on carbon efficiency.

3.3. Data Sources and Sample Description

The research sample of this paper is a panel dataset of 285 prefecture-level cities in China from 2010 to 2023, as shown in

Table 1. Variable Categories and Definitions.

Variable Category	Name and Symbol	Sample Size	Mean	Standard Deviation	Minimum	Mediation	Maximum
Dependent Variable	CTFP	3790	0.986	0.124	0.623	0.981	1.542
Core Explanatory Variable	Dual	3790	0.042	0.201	0.000	0.000	1.000
Control Variable	Popd	3790	5.739	0.974	1.733	5.799	9.089
	Pgdp	3790	10.768	0.592	8.618	10.755	12.576
	Gove	3790	0.205	0.120	0.044	0.177	2.352
	Open	3790	0.191	0.535	−0.227	0.082	28.368
	Fin	3790	2.606	1.281	0.504	2.315	21.297
	Ued	3790	7.259	1.346	1.522	7.277	12.066
	Hcap	3790	0.021	0.025	−0.005	0.011	0.147
	Urban	3790	0.563	0.152	0.181	0.548	1.082
	Er	3790	0.674	0.218	0.112	0.701	0.982

. The city sample covers all prefecture-level and above cities nationwide except Hong Kong, Macao, and Taiwan. After removing cities with severe missing data, 285 cities are retained, yielding 3790 observations. The data sources include the China City Statistical Yearbook, China Environmental Statistical Yearbook, China Energy Statistical Yearbook, and provincial and municipal statistical yearbooks. Missing data for individual years are filled using linear interpolation.

4. Empirical Results

4.1. Benchmark Regression

This paper employs the double machine learning method to examine the impact of the synergy between the digital economy and green finance policies on urban carbon total factor productivity.

Table 2. Benchmark Regression Analysis.

Variable	(1) CTFP	(2) CTFP	(3) CTFP
Dual	0.036 ***	0.039 ***	0.009 ***
	(0.005)	(0.006)	(0.001)
Low-Carbon City Pilot			0.001 *
			(0.000)
Carbon Emissions Trading Pilot			0.005 **
			(0.003)
Broadband China Demonstration City			0.143
			(0.130)
Control Variables (linear terms)	NO	Yes	Yes
Control Variables (quadratic terms)	NO	Yes	Yes
Observed Values	3790	3790	3790

Note: Robust standard errors are indicated in parentheses; ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. The same applies to subsequent tables.

reports the baseline regression results. Model (1) controls only for city and year fixed effects, while Model (2) further adds linear and quadratic terms of control variables to capture the nonlinear effects of city characteristics. As shown in columns (1) and (2) of Table 2, the estimated coefficient of the core explanatory variable Dual is positive and significant at the 1% level, indicating that, compared with non-dual-pilot cities, the simultaneous implementation of the Big Data Comprehensive Pilot Zone and the Green Finance Reform and Innovation Pilot Zone effectively promotes urban carbon total factor productivity. Our estimated effect sits comfortably within these ranges, suggesting that the synergy between digital economy and green finance policies generates an economically meaningful efficiency gain comparable to that of established market-based environmental regulations, thus validating research hypothesis H1.

Furthermore, Column (3) incorporates binary indicators for three concurrent national pilot programs—the Low-Carbon City Pilot, the Carbon Emissions Trading Pilot, and the Broadband China Demonstration City. These three policies are selected because they directly target carbon regulation or digital infrastructure, the two core dimensions of our synergy framework, and thus constitute the most plausible confounders among contemporaneous initiatives. In Column (3), which additionally controls for the three major contemporaneous policies, the coefficient of Dual is statistically indistinguishable from the estimate in Column (2), as the two coefficients lie well within each other’s 95% confidence intervals. The coefficients on the three policy indicators themselves are small and partially insignificant, suggesting that their independent effects are largely absorbed by the comprehensive set of economic and environmental controls already present in X.

The stability of the Dual coefficient across specifications—from the parsimonious specification in Column (1) through the fully controlled specification in Column (3)—provides robust evidence that the identified synergy effect is not driven by the compounding influence of other major national pilot programs. This consistent pattern lends strong empirical support to research hypothesis H1, confirming that the joint implementation of the Big Data Comprehensive Pilot Zone and the Green Finance Reform and Innovation Pilot Zone significantly enhances urban carbon total factor productivity, and that this enhancement reflects genuine policy synergy rather than confounded policy overlap.

4.2. Robustness Tests

To ensure the reliability of the baseline results, this paper conducts robustness checks from the following five aspects: accounting for province–year interactive fixed effects, adjusting the research sample, resetting the double machine learning model, introducing instrumental variables, and changing the measurement of the dependent variable.

4.2.1. Incorporating Province–Year Interaction Fixed Effects

In China’s government governance system, provincial governments play a crucial bridging role, and cities within the same province share strong commonalities in policy implementation, geographical location, and historical culture. To control for the interference of time-varying unobservable factors at the provincial level on the estimation results, this paper adds province–year interactive fixed effects to the baseline model. Column (1) of

Table 3. Results of Robustness Tests.

Variables	(1) CTFP	(2) CTFP	(3) CTFP	(4) CTFP	(5) CTFP	(6) CTFP	(7) CTFP	(8) CTFP
Dual	0.034 ***	0.036 ***	0.038 ***	0.040 ***	0.037 ***	0.077 ***	0.048 **	0.503 ***
	(0.007)	(0.006)	(0.006)	(0.007)	(0.007)	(0.015)	(0.019)	(0.071)
Control Variables (linear terms)	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Control Variables (quadratic terms)	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Kleibergen–Paap F-statistic							18.30
Hansen J p-value							0.40
Observed Values	3790	3790	3790	3790	3790	3790	3790	3790

Note: Robust standard errors are indicated in parentheses; *** and ** denote significance at the 1% 5% levels, respectively. The same applies to subsequent tables.

reports the corresponding results. The estimated coefficient of Dual remains positive and significant at the 1% level, consistent with the baseline regression, indicating that the baseline results are not affected by time-varying provincial-level factors.

4.2.2. Adjusting the Sample

China’s four municipalities directly under the central government—Beijing, Tianjin, Shanghai, and Chongqing—differ substantially from ordinary prefecture-level cities in terms of administrative level, resource endowment, and policy authority, which may interfere with the estimation results. Therefore, this paper excludes the municipality samples and re-runs the regression using only prefecture-level cities. Column (2) of Table 3 shows that the estimated coefficient of Dual remains significant at the 1% level, consistent with the baseline results, indicating that the baseline conclusion is not driven by the municipality samples.

4.2.3. Re-Parameterizing the Dual Machine Learning Model

To test the influence of machine learning model specifications on the baseline results, this paper first adjusts the sample splitting ratio, changing the baseline ratio of 1:5 to 1:3 and 1:7 respectively. Second, it replaces the baseline machine learning algorithm (random forest) with gradient boosting. Columns (3) to (5) of Table 3 report the corresponding results. Regardless of the sample splitting ratio or machine learning algorithm used, the estimated coefficient of Dual remains significantly positive, and the magnitude is close to the baseline result, fully demonstrating the robustness of the baseline conclusion.

4.2.4. Introduction of Instrumental Variables

Although the DML framework with high-dimensional controls substantially mitigates concerns about observable confounders, the designation of dual-pilot cities may still be endogenous. Provincial and central governments may select cities for pilot programs based on unobserved characteristics—such as political connections, administrative bargaining power, or anticipated economic potential—that also influence CTFP trajectories. To address this, we implement three complementary identification strategies.

We first augment the control vector X with pre-determined city characteristics interacted with a cubic time polynomial. Specifically, we include interactions of baseline (2010) GDP per capita, fiscal capacity (the ratio of local fiscal revenue to expenditure), and industrial SO₂ emission intensity with t, t², and t³. These terms flexibly absorb systematic differences in growth trajectories between cities that were later selected as pilots and those that were not, thereby reducing the risk that omitted trends drive the estimated treatment effect. Column (6) of Table 3 reports the result. The coefficient on Dual is 0.077, virtually identical to the baseline estimate. Following existing studies, we employ two historical instruments for dual-pilot designation: the city’s average altitude and the number of post offices in 1984. The relevance condition is satisfied because altitude is correlated with pilot selection probability: mountainous cities face greater pressure for ecological protection and green development, making them more likely to be included in green finance and digital economy pilot zones. Similarly, the number of post offices in 1984 reflects historical communication infrastructure that is correlated with subsequent digital infrastructure development, yet predates the contemporary policy discourse.

Regarding the exclusion restriction, we take several steps to bolster credibility. First, we directly control for geographic and historical confounders in the first- and second-stage regressions, including terrain ruggedness, distance to major seaports, and population density in 1984. By conditioning on these observables, we make the assumption that the instruments affect CTFP only through pilot selection more plausible. Second, we apply the plausibly exogenous, which relaxes the strict exclusion restriction and constructs union of confidence intervals under the assumption that the instruments may exert a small direct effect on CTFP. The 95% confidence interval for the Dual coefficient remains strictly above zero even when allowing a non-zero direct effect of realistic magnitude. Third, we construct a Bartik-style instrument by interacting a city’s initial internet penetration rate with the national-level growth rate of big data policy intensity across pilot zones. This instrument leverages a common aggregate trend and is less susceptible to local endogeneity. Column (7) of Table 3 reports the two-stage least squares (2SLS) results using altitude and historical post offices. The Kleibergen–Paap F-statistic is 18.3, exceeding the Stock–Yogo critical value of 11.59 at the 10% maximal IV size, thus rejecting weak identification. The Hansen J-statistic p-value is 0.40, failing to reject the overidentifying restrictions. The estimated coefficient of Dual is 0.048, larger than the baseline DML estimate, which is consistent with the expectation that DML already partially addresses selection on observables, whereas the IV strategy additionally corrects for selection on unobservables.

4.2.5. Changing the Measurement Method of the Dependent Variable

To test the robustness of the baseline results to the measurement method of carbon total factor productivity, this paper re-estimates urban carbon total factor productivity using the super-efficiency SBM model, replacing the baseline SBM-DDF and global ML index method. The super-efficiency SBM model allows further comparison among efficient decision-making units and can more accurately characterize efficiency differences [24,25]. Column (8) of Table 3 reports the corresponding results. The estimated coefficient of Dual remains significantly positive at the 1% level, indicating that the baseline conclusion is not affected by the measurement method of carbon total factor productivity.

4.3. Mechanism Testing

The baseline regression results show that the synergy between the digital economy and green finance policies significantly promotes sustainable industrial development. To further reveal the underlying transmission channels, this paper employs causal mediation analysis and uses a two-step method to test whether the mechanism variables mediate the relationship between policy synergy and carbon total factor productivity. Based on the theoretical analysis above, policy synergy may enhance carbon total factor productivity through three pathways: government green development attention, green technology innovation, and the scale and structure of green investment. This paper selects government green development concepts, urban green technology innovation level, and green investment scale as mechanism variables. The measurement of these mechanism variables is as follows: government green development concepts are measured by the frequency share of keywords such as “green,” “environmental protection,” “low-carbon,” and “sustainable development” in local government work reports; urban green technology innovation level is measured by the number of green invention patent applications (logarithmically transformed), which, compared to green utility model patents, better reflects substantive technological innovation capability; green investment scale is measured by the ratio of urban environmental pollution control investment to regional GDP [26,27,28].

As shown in

Table 4. Results of Mechanism Tests.

Variables	(1) Government’s Green Development Outlook	(2) Green Technological Innovation	(3) Scale of Green Investment
Dual	2.003 ***	0.179 ***	0.135 **
	(0.400)	(0.030)	(0.060)
Control Variables (linear terms)	Yes	Yes	Yes
Control Variables (quadratic terms)	Yes	Yes	Yes
Observed Values	3790	3790	3790

Note: Robust standard errors are indicated in parentheses; *** and ** denote significance at the 1% 5% levels, respectively. The same applies to subsequent tables.

, the synergy between the digital economy and green finance policies has a significant positive impact on all three mediating pathways. In column (1), the estimated coefficient of Dual is significant at the 1% level, indicating that policy synergy significantly increases the frequency share of green development keywords in local government work reports—i.e., government green development concepts have undergone a substantive shift. The Big Data Pilot Zone provides local governments with precise environmental data support, while the Green Finance Pilot Zone strengthens local governments’ green governance responsibilities. The synergy of the two policies encourages governments to translate green development from concept to governance practice. Once government green development concepts undergo a substantive shift, governments use planning guidance, fiscal support, and project approval administrative measures to create a favorable policy environment for carbon total factor productivity. Therefore, government green development concepts are an important transmission channel through which policy synergy enhances carbon total factor productivity. In column (2), the estimated coefficient of Dual is significant at the 1% level, indicating that policy synergy significantly promotes green technology innovation. The environmental monitoring and information tracing functions of digital technology can effectively identify corporate greenwashing behavior, while green finance internalizes environmental performance into financing costs, enabling firms to obtain better financing conditions for green innovation. When firms concentrate R&D investment in green technology fields, the increase in green technology innovation translates into an enhancement of carbon total factor productivity. In column (3), the estimated coefficient of Dual is significant at the 5% level, indicating that policy synergy significantly expands the scale of urban green investment. The synergy of the two policies promotes the expansion and structural optimization of green investment, providing sustained financial support for carbon total factor productivity. Therefore, green investment scale is an important mediating channel through which policy synergy exerts its enabling effect. These results reveal that policy synergy enhances carbon total factor productivity by transforming government green development concepts, raising the level of urban green technology innovation, and expanding the scale of green investment. Thus, research hypotheses H2a, H2b, and H2c are verified.

4.4. Heterogeneity Analysis

The policy synergy effect may vary depending on differences in urban resource endowments, geographical location, and environmental protection area characteristics. Accordingly, this paper conducts heterogeneity analysis from the following dimensions to reveal the differentiated effects of policy synergy.

4.4.1. Heterogeneity in Resource Endowments

Resource-based cities and non-resource-based cities differ significantly in industrial structure, innovation drivers, and transition paths. Referring to the National Plan for the Sustainable Development of Resource-Based Cities (2013–2020), this paper divides the sample cities into resource-based and non-resource-based cities and runs regressions separately. Columns (1) and (2) of

Table 5. Results of Heterogeneity Analysis.

Variables	(1) Resource-Based	(2) Non-Resource-Based	(3) Eastern	(4) Central	(5) Western	(6) Environmental Protection Zones	(7) Non-Environmental Protection Zones
Dual	0.019 *	0.045 ***	0.050 ***	0.030	0.015	0.017 *	0.043 ***
	(0.011)	(0.007)	(0.011)	(0.021)	(0.013)	(0.010)	(0.007)
Control Variables (linear terms)	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Control Variables (quadratic terms)	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observed Values	1696	2094	1606	1108	1076	1722	2068

Note: Robust standard errors are indicated in parentheses; *** and * denote significance at the 1% and 10% levels, respectively. The same applies to subsequent tables.

report the results. The policy synergy effect on carbon total factor productivity is more pronounced in non-resource-based cities, while the effect on resource-based cities is relatively smaller. This difference may be because resource-based cities have long relied on traditional resource industries, exhibit strong industrial rigidity, and face high sunk costs and path dependence in green transformation, which limits the room for improving green efficiency.

4.4.2. Geographical Heterogeneity

China’s eastern, central, and western regions differ considerably in economic development level, marketization degree, and infrastructure conditions. This paper divides the sample cities into eastern, central, and western regions and runs regressions separately. Columns (3) to (5) of Table 5 report the results. The policy synergy effect is most pronounced in eastern cities, and there is no statistical evidence of differentiated effects across central and western regions. Eastern regions have well-developed digital infrastructure and advanced financial markets, providing a favorable institutional environment and factor support for policy synergy, whereas western regions are constrained by factor endowments and market conditions, limiting the enabling effect of policy synergy.

4.4.3. Heterogeneity of Environmental Protection Zones

Environmental protection areas differ from non-environmental protection areas in environmental regulation intensity and green development pressure. This paper defines cities where national nature reserves or key ecological function zones are located as environmental protection areas, and the rest as non-environmental protection areas, and runs regressions separately. Columns (6) and (7) of Table 5 report the results. The policy synergy effect on carbon total factor productivity is more pronounced in non-environmental protection areas, while the effect in environmental protection areas is relatively smaller. This difference may be because environmental protection areas face stricter environmental regulations and have limited room for industrial transformation, resulting in a lower marginal effect of policy synergy, whereas non-environmental protection areas have greater flexibility in environmental governance and industrial transformation, making it easier to release the potential of policy synergy.

4.5. Further Analysis

The baseline regression and mechanism analysis above have confirmed that the synergy between the digital economy and green finance policies significantly enhances urban carbon total factor productivity. However, do the two policies generate synergistic effects during their implementation? To answer this question rigorously, we move beyond the single Dual dummy and instead examine the interaction between the two pilot policies directly. Specifically, we define BigData as a binary indicator equal to 1 if city i is a Big Data Comprehensive Pilot Zone in year t, and GreenFinance analogously for the Green Finance Reform and Innovation Pilot Zone. We then include both main effects and their interaction term BigData × GreenFinance in the DML model. A positive and statistically significant coefficient on the interaction term constitutes direct evidence of synergy, indicating that the combined effect of the two policies exceeds the sum of their independent contributions.

Columns (1) and (2) of

Table 6. Results of Synergy Analysis.

Variables	(1) CTFP	(2) CTFP	(3) CTFP	(4) CTFP	(5) CTFP
Big Data	0.016 **
	(0.008)
Green Finance		0.019 **
		(0.009)
Dual			0.039 ***
			(0.006)
Big Data × Green Finance			0.070 **
			(0.034)
Dual (Big Data first, then Green Finance)				0.030 **
				(0.013)
Dual (Green Finance first, then Big Data)					0.010 *
					(0.005)
Control Variables (linear terms)	Yes	Yes	Yes	Yes	Yes
Control Variables (quadratic terms)	Yes	Yes	Yes	Yes	Yes
Observed Values	3790	3790	3790	706	619

Note: Robust standard errors are indicated in parentheses; ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. The same applies to subsequent tables.

report the results. In Column (1), where only the main effects are included, the coefficient of BigData is 0.016 and that of GreenFinance is 0.019. Both single-pilot policies individually enhance carbon total factor productivity, consistent with the findings reported earlier. Column (3) introduces the interaction term. The coefficient on BigData × GreenFinance is 0.070, significant at the 5% level. This positive and statistically significant interaction effect provides direct evidence that the digital economy and green finance policies reinforce each other: the marginal impact of one policy on CTFP is significantly larger when the other policy is also in place. The total effect of dual-pilot designation is therefore the sum of the two main effects plus the interaction effect, substantially larger than the baseline single-dummy estimate, reflecting the fact that the interaction term fully captures the synergy dividend. Digital technology improves the screening and monitoring capacity of green finance, while green finance channels capital toward the low-carbon application of digital technologies; the interaction between the two generates a mutually reinforcing dynamic that amplifies their individual impacts on urban carbon total factor productivity [29].

Furthermore, this paper examines the impact of different pilot implementation orders on the synergy effect. This paper divides dual-pilot cities into two categories: those implementing the Big Data Pilot first and the Green Finance Pilot later, and those implementing the Green Finance Pilot first and the Big Data Pilot later, and examines their policy effects separately [30]. Columns (4) and (5) of Table 6 show that the policy effect is significantly larger for cities that implemented the Big Data Pilot first and the Green Finance Pilot later than for those that implemented the Green Finance Pilot first and the Big Data Pilot later. This finding reveals a timing-dependent feature of the policy synergy effect: digital infrastructure, as a support for information flow and factor allocation, when built first, can lay the data foundation and institutional environment for subsequent green finance policy implementation, thereby releasing the synergy dividend more effectively.

5. Conclusions and Policy Implications

5.1. Research Findings

Against the backdrop of deep integration between the digital economy and green finance, this paper takes the dual-pilot program of China’s Big Data Comprehensive Pilot Zone and Green Finance Reform and Innovation Pilot Zone as a quasi-natural experiment. Using panel data of 285 prefecture-level cities from 2010 to 2023 and employing the double machine learning method, this paper systematically examines the enabling effect of the synergy between digital economy and green finance policies on urban carbon total factor productivity (CTFP). The main research conclusions are as follows:

First, the synergy between digital economy and green finance policies can enhance urban CTFP. This conclusion remains robust after a series of robustness checks, including accounting for province–year interactive fixed effects, excluding municipality samples, resetting the double machine learning model specification, introducing instrumental variables, and changing the measurement of the dependent variable.

Second, the synergy between the digital economy and green finance enhances urban CTFP by raising local governments’ attention to green development concepts, improving the level of urban green technology innovation, and expanding the scale of green investment.

Third, the policy synergy effect exhibits significant regional heterogeneity. From the perspective of resource endowment, the promoting effect of policy synergy is stronger in non-resource-based cities than in resource-based cities. From the perspective of geographical location, the synergy effect is most prominent in the eastern region, followed by the central region, while no significant effect is found in the western region. From the perspective of environmental protection zoning, the synergy effect is stronger in non-environmental protection areas than in environmental protection areas.

5.2. Policy Implications

Based on the above research conclusions, this paper proposes the following policy implications:

First, strengthen the policy synergy between the digital economy and green finance by coordinating the layout of digital infrastructure with the allocation of green financial resources, and promoting the interconnection between the data element market and the green finance market, thereby forming a combined force of policies. Specifically, building on the existing dual-pilot cities, a policy synergy evaluation mechanism should be established, and CTFP should be incorporated into the policy performance evaluation system to avoid resource misallocation during policy implementation [31]. At the same time, the successful experiences of existing dual-pilot cities should be summarized, and the policy synergy model should be promoted in regions with suitable conditions to amplify the synergy effect.

Second, policy strategies should be differentiated. For non-resource-based cities with strong synergy effects, dual-pilot expansion can serve as a primary lever for CTFP enhancement. For resource-based cities, where the synergy effect is muted, the dual-pilot program should be coupled with dedicated green transition funds and digital platforms that specifically target high-carbon industrial value chains, in order to break structural lock-in [32]. In environmental protection zones, given the weaker synergy, a more tailored approach integrating strict protection with technology-enabled ecological asset management is warranted, rather than generic industrial green transformation.

Third, implement differentiated policy strategies to enhance urban CTFP according to local conditions. For resource-based cities, support for the green transformation of traditional industries should be strengthened, and measures such as establishing special green transformation funds and introducing digital energy consumption monitoring platforms should be adopted to break the constraints of carbon lock-in effects and industrial rigidity on efficiency improvement. For the central and western regions, the construction of digital infrastructure and the green finance system should be accelerated to compensate for the lack of factor endowments, while green technology transfer and industrial collaboration between the eastern region and the central and western regions should be enhanced to promote the cross-regional diffusion of synergy effects. For environmental protection areas, on the basis of strictly adhering to ecological red lines, mechanisms for realizing the value of ecological products should be explored to transform ecological advantages into green development momentum. At the same time, policy restrictions on green technology application and green investment should be moderately relaxed to avoid excessive regulation inhibiting efficiency improvement.