The impact of pilot policy for innovative industrial clusters on green innovation efficiency

Abstract

Green transformation has become a central goal of China’s development strategy in response to mounting environmental pressure and long-term growth needs. Improving green innovation efficiency (GIE) is essential to achieving this transformation while sustaining economic momentum. This study evaluates the impact of the Pilot Policy for Innovative Industrial Clusters on GIE across Chinese cities. Using panel data from 280 prefecture-level cities between 2007 and 2021, we apply difference-in-differences and spatial difference-in-differences (SDID) models to estimate policy effects, spatial spillovers, and transmission mechanisms. The results are as follows: (1) The pilot policy significantly improves GIE in the pilot cities, with robust results after various tests. (2) The policy enhances urban green innovation through four main channels: reducing energy consumption intensity, upgrading industrial structure, fostering green technological innovation, and accelerating digital infrastructure development. (3) In addition to its direct impact on pilot cities, the policy also boosts the GIE of neighboring cities via spatial spillover effects. (4) Heterogeneity analysis reveals that the policy’s effect is more pronounced in central cities, non-resource cities, and cities with a strong environmental protection focus. This study contributes to the understanding of innovative industrial cluster policies in enhancing GIE and offers valuable policy insights for promoting urban green development.

Introduction

As global environmental challenges intensify, green innovation is increasingly seen as a key solution to the conflict between environmental pollution and economic growth, becoming central to sustainable development strategies¹. While China has made progress in green innovation, its overall level of green technological advancement remains relatively low (Xie, 2020). Data from the World Intellectual Property Organization (WIPO) reveals that China has only 21,337 PCT patent applications in green technology, far below the leading number from the United States. Improving the efficiency of green technological innovation to address the tension between environmental pollution and economic development has become an urgent challenge for China.

GIE is influenced by a range of factors, including economic, social, and policy-related elements^2,3,4. Among these, the policy supporting innovative industrial clusters plays a key role. In recent years, the government has introduced various policies to foster industrial clusters, aiming to enhance regional innovation capabilities and facilitate the transformation and concentration of technological advancements. The pilot program for innovative industrial clusters, approved by the Ministry of Science and Technology, seeks to drive industrial development through innovation, strengthen technological services, and serve as a pioneering model with broader impacts. By geographically concentrating industries and integrating supply chains, these clusters have created efficient innovation ecosystems, promoting knowledge flow and technological spillovers, thereby significantly boosting urban innovation efficiency^5,6. However, whether and how the Pilot Policy for Innovative Industrial Clusters affects urban GIE, whether there are spatial spillover effects, and the mechanisms and heterogeneity of its impact are currently lacking systematic research.

This study aims to explore the impact of the Pilot Policy for Innovative Industrial Clusters on urban GIE, analyze its underlying mechanisms, and investigate the spatial spillover effects and heterogeneity of the policy. To achieve these objectives, this study uses panel data from 280 prefecture-level cities in China between 2007 and 2021, leveraging the pilot policy of innovative industrial clusters as a quasi-natural experiment. DID model is employed to assess the policy’s impact on urban GIE, while a Spatial SDID model investigates its spatial spillover effects. A mediation model explores the roles of energy consumption intensity, industrial structure, green technological innovation and digital infrastructure. Additionally, the study analyzes the policy’s heterogeneous effects across cities with different geographic locations, resource endowments, and administrative levels.

Literature review

Measurement of GIE

Green innovation efficiency (GIE) reflects how effectively innovation inputs are transformed into outputs under environmental and resource constraints. It serves as a core metric for assessing the performance of green innovation systems. As a bridge between conceptual research and empirical application, GIE measurement has attracted growing academic interest.

The existing literature mainly adopts Stochastic Frontier Analysis (SFA)^7,8 and Data Envelopment Analysis (DEA)^9,10,11 to measure GIE. SFA estimates efficiency through econometric modeling of frontier production functions, but its outcomes are sensitive to functional form assumptions, which may introduce subjectivity. Traditional DEA models, which are typically radial or angular, cannot account for all input and output slacks, thereby limiting the precision of efficiency estimates^12,13. The SBM model improves efficiency assessment by accounting for slack variables and incorporating undesirable outputs, making it well-suited for evaluating GIE. Building on this advantage, we adopt a Super-SBM model to more precisely assess city-level GIE.

Factors affecting GIE

Green innovation efficiency (GIE) depends on how effectively a city provides incentives, absorbs green innovation inputs, and delivers outcomes that support environmental goals. These processes vary due to differences in regulatory design, institutional strength, and market structure.

Policy differences remain a primary source of variation. Environmental regulations can promote green innovation by increasing the cost of pollution, but overly rigid policies may constrain firms’ ability to respond^14,15,16. This tension suggests that the effectiveness of regulation depends not only on its stringency but also on how it is designed. Evidence shows that when regulatory pressure is combined with public R&D support, firms are more likely to pursue high-quality green innovation¹⁷. In practice, although formal incentives such as public policies play a critical role, they still require support from informal tools. Mechanisms such as transparency in disclosure, credible audits, and media oversight are particularly effective in regions with weak or uneven enforcement and help strengthen green innovation efficiency¹⁸.

To make these tools effective, cities must build the capacity to implement policies effectively. Physical infrastructure reduces coordination costs and improves the flow of knowledge and talent across regions¹⁹. Digital networks further extend this capacity by enabling cross-regional resource allocation and accelerating the diffusion of green innovation²⁰. But infrastructure alone is insufficient. Cities also need institutional mechanisms to translate policy into coordinated action. Smart city reforms have improved local governments’ ability to implement and synchronize environmental strategies across sectors²¹. Industrial structure plays a similar role. Moving from high-emission sectors to low-carbon industries increases demand for green technologies and improves alignment between innovation efforts and environmental goals²². Together, these factors define a city’s capacity to turn external pressure into green innovation.

Strong market demand further improves green innovation efficiency. Cities with stronger green export capacity tend to achieve higher green innovation efficiency²³, highlighting the importance of market demand in advancing green innovation development. Clear and responsive market signals encourage investment in R&D and make policy incentives more effective. When demand and policy are aligned, their interaction creates a reinforcing cycle that accelerates green innovation. In the absence of effective demand feedback, however, green innovation often fails to connect with industrial application, resulting in limited environmental benefits.

Innovative industry agglomeration and GIE

Industrial agglomeration promotes green innovation by concentrating key resources and lowering coordination costs. Firms located in clusters benefit from shared infrastructure, access to specialized labor, and proximity to localized knowledge networks. These advantages reduce entry barriers for environmental innovation and shape the baseline conditions for green innovation efficiency^24,25.

However, geographic proximity alone does not ensure meaningful outcomes. The quality of firm-level interaction plays a central role. Clusters led by dominant firms tend to foster more coordinated innovation. These leading firms set strategic agendas and allocate resources across the network, enabling consistent progress²⁶. By contrast, fragmented clusters often lack coherence. Weak ties and poor alignment hinder sustained collaboration and reduce the impact of green innovation efforts²⁷.

Beyond shaping the structure of clusters, effective policies can also channel collective efforts toward environmental sustainability²⁸. In contrast, excessive administrative control may undermine firm autonomy and weaken market-based incentives, limiting the innovation potential of the cluster²⁶. The overall effectiveness of agglomeration therefore depends not only on spatial concentration but also on how policy engages with local industrial dynamics. When well aligned with local conditions, policies can enhance cities’ green innovation efficiency and steer it toward more substantial environmental outcomes²⁹.

Spatial correlation

Green innovation tends to exhibit spatial interdependence rather than being confined to individual cities. This interdependence arises from factor mobility, policy diffusion, and competition among cities. Tobler’s first law of geography suggests that neighboring regions tend to influence one another through shared markets, mobile labor, and institutional spillovers³⁰. As a result, a city’s green innovation performance is often shaped by the conditions of surrounding areas. Empirical studies confirm this pattern, showing that GIE levels vary across regions and display clear spatial clustering³¹. Neighboring technological environments play a critical role in shaping a city’s green innovation outcomes^32,33. This effect is especially pronounced in urban agglomerations, where cross-city industrial chains and research networks foster deep integration. These spatial dynamics lead to uneven diffusion of green innovation across regions, highlighting the need to incorporate spatial effects into policy evaluation³⁴.

Despite their importance, spatial dependencies are often overlooked in empirical strategies. Conventional approaches, such as the standard DID model, assume independent observations and fail to capture cross-regional policy spillovers. If treated cities affect neighboring ones, either positively or negatively, the resulting spillovers violate the stable unit treatment value assumption (SUTVA). To address this issue, we adopt a spatial difference-in-differences (SDID) framework that incorporates spatial lags to capture cross-regional interactions and disentangle direct effects from spillovers, improving estimation accuracy in contexts where policies diffuse across neighboring areas.

While the literature on GIE has expanded considerably, several critical research gaps remain. First, although GIE has received increasing academic attention, research has concentrated on firms and industries, offering limited insight into how city-level policies influence GIE through industrial clustering and regional coordination. The effects of such policies in urban contexts, including their underlying mechanisms and associated spatial spillovers, remain insufficiently examined.

Second, conventional evaluation methods such as difference-in-differences (DID) typically rely on the assumption of independent observations and no interference between treated and untreated units. In practice, however, cities are linked by interregional spillovers and policy diffusion. Ignoring spatial interdependence can bias estimation and distort the identification of causal policy impacts. Therefore, there is an urgent need for causal frameworks that can jointly identify both direct impacts and spatial spillover effects. This is essential for accurately evaluating policy outcomes in spatially interdependent urban systems.

Third, many studies still construct spatial weight matrices using only geographic proximity and overlook intercity economic linkages. This narrow approach may misrepresent the underlying spatial interactions among cities. In addition, most studies do not validate their weight specifications or compare alternative matrix designs, which undermines the robustness and credibility of estimated spillover effects.

The marginal contributions of this study are as follows: first, this study advances theory by introducing a city-level perspective on green innovation efficiency. It shifts the analytical focus from macro- or firm-level studies to the city level, examining how policy-driven industrial clustering affects green innovation efficiency through four intermediary mechanisms. This perspective enriches existing research on policy diffusion and regional green development. Second, this study makes a methodological contribution by developing a spatial difference-in-differences (SDID) model. By incorporating spatial lags into the traditional DID framework, the model captures both direct effects and spatial spillovers from policy interventions, improving identification precision in the presence of spatial dependence. Third, it improves spatial data modeling by constructing a geo-economic weight matrix that integrates geographic distance and economic similarity. The use of AIC and BIC criteria to select optimal weights enhances estimation robustness compared to adjacency-based or arbitrary weighting methods, increasing the reliability of spillover effect identification.

Policy background and theoretical mechanism

Policy background

The pilot policy for innovative industry agglomeration, approved by the Ministry of Science and Technology, aims to promote innovation-driven industrial development and strengthen technological services to the economy through pilot experiences and replicable demonstration effects. It encompasses both strategic emerging industry clusters and traditional industry clusters. The development of innovative industry clusters in China follows an exploratory, phased approach, progressing from shallow to deep stages (see Fig. 1). In July 2011, the Ministry of Science and Technology initiated planning for these clusters. In February 2013, management measures for pilot recognition were introduced, and by June 2013, the first 10 pilot units were approved, marking the policy’s official launch. The second and third batches, comprising 22 and 29 pilot units respectively, were approved in 2014 and 2017, bringing the total to 61 across 55 prefecture-level and higher cities, as shown in Fig. 2. Most pilots are in high-tech industrial development zones, serving as key platforms for regional industrial expansion.

Fig. 1

The development process of pilot policy for innovative industrial clusters.

Fig. 2

Approval process of pilot cities over time.

Since the pilot policy was implemented, it has yielded significant results. According to the "China Torch Yearbook 2021," by 2020, these clusters encompassed 26,000 enterprises, including 12,000 high-tech firms. Corporate R&D investment reached 316.15 billion yuan, supporting 1.16 million scientific and technological professionals. The clusters generated 6.26 trillion yuan in revenue, 4.70 trillion yuan in industrial output, 571.18 billion yuan in net profits, 299.5 billion yuan in taxes, and 124.98 billion USD in exports. Technological achievements included 40,000 authorized invention patents, 185,000 registered trademarks, 1,071 national or industry standards, and a technology contract turnover of 187.08 billion yuan. In April 2020, the Ministry of Science and Technology issued the "Opinions on Deepening the High-Quality Development of Innovative Industry Clusters", advocating pilot expansion, enhanced monitoring, and targeted policy guidance. Therefore, the policy’s economic impact is essential for optimizing its cycle of pilot construction, experience summary, and replication.

Theoretical mechanism

Direct effect of the pilot policy for innovative industrial clusters on GIE

The Pilot Policy for Innovative Industrial Clusters directly improves urban GIE through three primary channels: reshaping institutional environments, concentrating innovation resources, and reinforcing administrative competition. First, the policy sends a clear signal of commitment to green innovation, encouraging local governments to adjust regulatory frameworks, allocate fiscal resources, and strengthen innovation support systems. These institutional adjustments reduce systemic barriers to green R&D and create a more supportive environment for urban green innovation. Second, by designating pilot zones, the policy attracts high-quality enterprises, research institutions, and support services into strategically planned innovation districts. This spatial agglomeration facilitates more efficient allocation of resources and promotes collaboration among innovative firms and institutions. Beyond attracting resources, such spatial concentration strengthens green innovation ecosystems by accelerating knowledge flows and fostering integrated innovation networks, further improving GIE. Third, the policy intensifies intercity competition and sets higher performance benchmarks for cities in the pilot regions. Competitive pressure to deliver green innovation outcomes encourages cities to pursue more ambitious green innovation goals by strengthening administrative capacity and simplifying implementation procedures. Taken together, these regulatory, spatial, and competitive mechanisms constitute the core pathway through which the policy directly enhances urban GIE. Based on this, the following hypothesis is proposed:

Hypothesis 1: The Pilot Policy for Innovative Industrial Clusters significantly improves urban GIE.

Mediating effect of the pilot policy for innovative industrial clusters on GIE

(1) Energy intensity reduction.

The pilot policy reduces city-level energy intensity by strengthening environmental performance evaluations and promoting the green transition. Stricter environmental performance assessments impose stronger accountability pressure on local governments to achieve energy efficiency targets³⁵. In response, local governments prioritize investments in green infrastructure, including clean energy systems and low-carbon industrial parks. Many pilot cities also phase out outdated, energy-intensive production capacity through stricter environmental regulation. Taken together, performance accountability, fiscal prioritization, and capacity elimination contribute to improved energy efficiency at the city level.

Lower energy intensity creates more favorable conditions for urban green innovation. Lower energy intensity creates more favorable conditions for urban green innovation. By reducing energy use per unit of output, it reduces energy use per unit of output and lowers production costs, enabling cities to direct more resources toward innovation. At the same time, stricter energy efficiency standards prompt cities to promote cleaner technologies and support green industrial upgrading. These changes not only enhance compliance with environmental standards but also accelerate green technology adoption.Overall, resource reallocation mobilization and technological upgrading jointly enhance urban GIE. Based on this, the following hypothesis is proposed:

Hypothesis 2

The Pilot Policy for Innovative Industrial Clusters can indirectly enhance urban GIE by reducing energy consumption intensity.

(2) Industrial structure upgrading.

The pilot policy promotes urban industrial structure upgrading by encouraging the concentration of high-tech industries, redistributing resources, and phasing out outdated production capacity. Driven by green technology incentives and infrastructure development in innovation zones, the policy encourages the relocation of capital and human resources to high-tech and green industries³⁶. At the same time, stricter environmental regulations eliminate outdated, inefficient production capacity, making room for the emergence of more sustainable industries³⁷. Through these mechanisms, the policy promotes industrial upgrading toward green, innovation-driven industries.

Industrial upgrading encourages R&D investment, creating a more favorable environment for green technological advancement. Meanwhile, the growth of green industries enhances collaboration among firms, research institutions, and government within cities, forming a more cohesive innovation ecosystem. Overall, industrial upgrading drives green technological advancement and fosters collaboration among innovation actors, thereby improving urban GIE. Based on this, the following hypothesis is proposed:

Hypothesis 3

The Pilot Policy for Innovative Industrial Clusters can enhance urban GIE by promoting industrial structure upgrading.

(3) Green Technological Innovation.

The pilot policy enhances green technological innovation by creating institutional and spatial conditions that foster collaborative R&D. First, the clustering of innovative firms in industrial zones reduces information asymmetries and facilitates knowledge spillovers, allowing firms to share technical expertise, benchmark green practices, and jointly pursue green technological breakthroughs^38,39. Second, the policy provides innovation platforms, research subsidies, and environmental financing to reduce the costs and risks associated with green R&D. Reduced barriers stimulate greater investment in green technology development, which strengthens cities’ capacity to advance and deploy green technologies in key areas such as energy efficiency, pollution control, and clean production.

As green technologies become more widely adopted, cities gain access to validated green innovation outputs with demonstrated environmental and technological value. By building on proven green technologies, cities can streamline innovation processes and focus resources on scaling adoption and refining implementation strategies, which enhance urban GIE. Based on this, the following hypothesis is proposed:

Hypothesis 4

The Pilot Policy for Innovative Industrial Clusters can significantly enhance urban GIE by increasing green technological innovation.

(4) Digital Infrastructure Development.

The pilot policy promotes digital infrastructure development by mobilizing public investment and supporting the construction of collaborative innovation platforms. These efforts improve the availability of advanced digital infrastructure technologies, including data centers, edge computing and AI-driven analytics platforms, enabling cities to integrate them into green innovation systems to enhance information flow and accelerate innovation diffusion⁴⁰. The integration allows cities to monitor green innovation in real time and quickly identify inefficiencies. Accordingly, cities can refine innovation processes to reduce unnecessary resource consumption and shorten the green technology development cycle, ultimately improving urban GIE. Based on this, the following hypothesis is proposed:

Hypothesis 5

The Pilot Policy for Innovative Industrial Clusters can significantly improve urban GIE by enhancing the level of digital infrastructure development.

Spatial spillover effects of the pilot policy for innovative industrial clusters on GIE

In addition to its direct impact, the pilot policy is expected to generate substantial spatial spillover effects on urban GIE through three distinct mechanisms. First, the policy induces a policy competition effect among neighboring cities. As pilot cities achieve measurable improvements in green innovation, nearby municipalities respond strategically by adopting similar incentives and R&D support schemes to attract innovation resources and remain competitive⁴¹. Second, the policy gives rise to a demonstration effect. Successful experiences in industrial agglomeration, technological upgrading, and ecological innovation in pilot cities spread to neighboring cities through regional collaboration, institutional learning, and policy imitation. This enables neighboring cities to adopt effective instruments and organizational practices despite not being formally included in the pilot program. Third, the policy facilitates technological diffusion across city boundaries. Through supply chain integration, talent mobility, and policy collaboration, the policy fosters spatial concentration of R&D activities, innovation networks, and high-skilled labor in pilot cities. This concentration strengthens cross-regional knowledge flows and facilitates the diffusion of green innovation.

Geographic proximity and intercity linkages across economic, industrial, and ecological spheres amplify the three spillover effects above, which is consistent with Tobler’s first law of geography. As a result, neighboring non-pilot cities also benefit from the policy through intensified spatial interaction and institutional learning. Based on this, the following hypothesis is proposed:

Hypothesis 6

The Pilot Policy for Innovative Industrial Clusters has spatial spillover effects on urban GIE.

The theoretical framework is shown in Fig. 3:

Fig. 3

Theoretical framework.

Model specification and variable selection

Data sources and descriptive statistics

To more accurately assess the effect of the pilot policy, this paper excludes cities with innovation-oriented industrial clusters in the incubation stage that have not yet met the pilot recognition standards. Cities that underwent administrative boundary adjustments or have significant data gaps during the study period are also excluded. Ultimately, a panel dataset of 280 prefecture-level and above cities in China from 2007 to 2021 is constructed, comprising 4200 observations. Among these, 55 cities are pilot cities, while 225 are non-pilot cities. The list of pilot cities for innovation-oriented industrial agglomeration is obtained from the official website of the Ministry of Science and Technology of China. Patent data are sourced from the China National Intellectual Property Administration, and green patent classification numbers are derived from the WIPO Green Patent List. Other data are collected from the "China City Statistical Yearbook," EPS Database, and WIND Database. Descriptive statistics of variables are shown in Table 1.

Table 1 Descriptive statistics.

Variable selection

Dependent variable

The dependent variable in this study is GIE. According to the study of Li and Zeng²², this paper adopts the Super-SBM model proposed by Tone to measure urban GIE⁴², overcoming the limitations of traditional DEA methods.

Common methods for measuring GIE include stochastic frontier analysis (SFA) and data envelopment analysis (DEA), but both have limitations. SFA depends on the choice of functional form, making results sensitive to model specification bias. Traditional DEA models often ignore slack variables and fail to incorporate undesirable outputs, which are essential for green innovation. To address these issues, this study uses the Super-SBM model. It is non-radial and non-angular, captures slack in inputs and outputs, and directly includes undesirable outputs. It avoids the functional form assumptions required by SFA and addresses the limitations of traditional DEA models by capturing slack and incorporating undesirable outputs. Therefore, this study adopts the Super-SBM model to measure urban GIE.

Assume there are n decision-making units (DMUs). Each DMU has m types of input factors, s1 types of desirable output factors, and s2 types of undesirable output factors, represented in vector form as $x\in {R}_{m},{y}^{g}\in {R}_{{s}_{1}},{y}^{b}\in {R}_{{s}_{2}}$, respectively. The matrix forms are $X,{Y}^{g}$,${Y}^{b}$, where: $X=[{x}_{1},\cdots ,{x}_{n}]\in {R}_{m\times n},{Y}^{g}=[{y}_{1}^{g},\cdots ,{y}_{n}^{g}]\in {R}_{{s}_{1}\times n},{Y}^{b}=[{y}_{1}^{b},\cdots ,{y}_{n}^{b}]\in {R}_{{s}_{2}\times n}$. The production possibility set excluding the DMU $DMUs({x}_{0},{y}_{0}^{g},{y}_{0}^{b})$ is given as:

$$\begin{array}{c}p\left({x}_{0},{y}_{0}^{g},{y}_{0}^{b}\right)=\left\{\left(\overline{x },{\overline{y} }^{g},{\overline{y} }^{b}\right)|\begin{array}{c}\overline{x}\ge {\sum }_{j=1}^{n}{\lambda }_{j}{x}_{j},{\overline{y} }^{g}\le {\sum }_{j=1}^{n}{\lambda }_{j}{y}_{j}^{g},\\ {\overline{y} }^{b}\ge {\sum }_{j=1}^{n}{\lambda }_{j}{y}_{j}^{b},\lambda \ge 0\end{array}\right\}\end{array}$$

(1)

The linear formulation of Super-SBM model considering undesirable outputs can be expressed as follows:

$${\rho }^{*}=min\frac{\frac{1}{m}{\sum }_{i=1}^{m}\frac{{\overline{x} }_{i}}{{x}_{i0}}}{\frac{1}{{s}_{1}+{s}_{2}}\left({\sum }_{r=1}^{{s}_{1}}\frac{{\overline{y} }_{r}^{g}}{{y}_{r0}^{g}}+{\sum }_{u=1}^{{s}_{2}}\frac{{\overline{y} }_{l}^{b}}{{y}_{l0}^{b}}\right)}$$

(2)

$$\begin{array}{c}s.t. \left\{\begin{array}{c}\overline{x}\ge {\sum }_{j=1,\ne 0}^{n}{\lambda }_{j}{x}_{j},j=1,\cdots ,m\\ {\overline{y} }^{g}\le {\sum }_{j=1,\ne 0}^{n}{\lambda }_{j}{y}_{j}^{g},r=1,\cdots ,{s}_{1}\\ \overline{y}\ge {\sum }_{j=1,\ne 0}^{n}{\lambda }_{j}{y}_{j}^{b},l=1,\cdots ,{s}_{2}\\ \overline{x}\ge {x }_{0},{\overline{y} }^{g}\le {y}_{0}^{g},{\overline{y} }^{b}\ge {y}_{0}^{b}\\ \lambda \ge 0,{\sum }_{j=1,\ne 0}^{n}{\lambda }_{j}=1\\ {s}^{-},{s}^{g},{s}^{b}\ge 0\\ \end{array}\right.\end{array}$$

(3)

where ${\rho }^{*}$ is the efficiency value, $x$ represents the inputs, ${y}^{g}$ represents the desirable outputs, and ${y}^{b}$ represents the undesirable outputs. $\lambda$ is the weight, and ${s}^{-},{s}^{g},{s}^{b}$ are the slack variables for inputs, desirable outputs, and undesirable outputs, respectively. When $\lambda \ge 0$, it meets the condition of constant returns to scale. When $\lambda \ge 0$ and ${\sum }_{j=1,\ne 0}^{n}{\lambda }_{j}=1$, it meets the condition of variable returns to scale. Given a fixed level of input, if the desirable output is larger and the undesirable output is smaller, the efficiency value ${\rho }^{*}$ will be greater, indicating a higher level of urban GIE.

The selection of indicators is crucial when measuring GIE. In constructing the input factors, expected outputs, and undesired outputs, this study carefully considers data availability, alignment with research objectives, and the measurement standards established in existing literature. First, concerning input indicators, while enterprise-level data could offer a more micro-level perspective, nationwide enterprise-level green innovation data is scarce, and the long-term consistency of such data is poor. This makes it challenging to integrate this data into the spatial weight matrix analysis framework of the study. Furthermore, the externalities of green innovation typically involve city-level factors such as policy support, the innovation environment, and resource allocation. As a result, this study selects more comparable city-level input indicators, including government education expenditure, the number of scientific researchers, and energy consumption, to comprehensively measure the resource inputs for green innovation.

Second, regarding undesired outputs, although carbon dioxide emissions are a key indicator of environmental pollution, long-term, stable publicly available statistical data at the prefecture-level city level are currently unavailable. In contrast, industrial sulfur dioxide emissions, industrial wastewater discharge, and industrial dust emissions are widely used to assess urban environmental pollution levels and have better continuity and comparability in China’s official statistical data. Therefore, this study adopts these pollution emission indicators to represent undesired outputs. Finally, for expected output indicators, while new product outputs reflect the innovation achievements of enterprises, the core goal of green innovation is ecological benefits and sustainable technological progress. This study, therefore, uses the number of green patent applications as the primary indicator of green technological innovation, to more accurately capture the outcomes of green innovation activities. The selected input factors, expected outputs, and undesired outputs are summarized in Table 2.

Table 2 Input and output indicators.

The concept of GIE emphasizes the ecological aspect of innovation activities, aiming to maximize resource savings and minimize environmental pollution. In measuring GIE, this study incorporates both energy and environmental factors, selecting the following categories of indicators:

Input Indicators: ① Capital Input: This includes total expenditures on government education, government scientific undertakings, and environmental governance as the capital input indicators. ② Labor Input: Labor input is measured by the total number of employees in scientific research and technical services, as well as those working in water conservancy, environmental, and public facility management industries. ③ Resource Input: The entropy method is used to integrate the total water supply, total electricity consumption, and total liquefied petroleum gas supply to calculate a city’s resource consumption index, representing energy and resource input.

Desirable Outputs: ① Desirable Outputs: Urban per capita GDP, urban green coverage area, and green patent applications are selected to measure the economic, technological, and ecological outputs, respectively, within the desirable outputs. ② Undesirable Outputs: The undesirable outputs consist of industrial sulfur dioxide emissions, industrial wastewater discharge, and industrial smoke emissions. The entropy method is also applied to derive the urban environmental pollution index to measure the undesirable outputs.

Core explanatory variable

The core explanatory variable is $\text{did}$, which represents the dummy variable for the Pilot Policy for Innovative Industrial Clusters. Its coefficient reflects the treatment effect of policy. Specifically, $\text{did}$ is defined as $did={treat}_{i}\times {post}_{t}$, where ${treat}_{i}$ is the policy dummy variable. Pilot cities for innovative industrial clusters are assigned a value of 1 as the treatment group, while other cities are assigned a value of 0 as the control group. $treat$ is the time dummy variable.

Control variables

Five key control variables are included in the model. Urban Economic Density (Eco): Cities with higher economic density tend to have more developed infrastructure and greater market vitality, which facilitates the agglomeration of innovative resources and knowledge spillovers, thereby enhancing GIE⁴³. This study measures economic density as the ratio of regional Gross Domestic Product (GDP) to regional area. Financial Development (Fin): Adequate financial support can reduce the financial constraints on enterprises’ green innovation, promote R&D investment, and facilitate technological upgrades, thereby improving GIE⁴⁴. This study uses the ratio of the sum of financial institutions’ deposit balances and loan balances to GDP. Urbanization Level (Urban): The process of urbanization affects labor and resource allocation. A reasonable level of urbanization can optimize the environment for applying green technological innovations, but overly rapid urbanization may increase environmental pressures, thereby impacting GIE⁴⁵. This study measures urbanization as the ratio of urban permanent population to the total permanent population (urban and rural). Government Intervention (Gov): Government fiscal expenditure can support green innovation through policy guidance, environmental regulation, and financial subsidies. However, excessive intervention may reduce market efficiency⁴⁶. This study measures government intervention as the ratio of general government fiscal expenditure to regional GDP. Industrial Structure Upgrading (Industry): the upgrading of industrial structure (such as the transformation from pollution-intensive industries to high-tech industries) can reduce pollution emissions and promote green technological innovation, thus enhancing GIE³³. This study measures industrial upgrading as the ratio of the output value of the secondary and tertiary industries to the total output value.

Spatial weight matrix

The spatial weight matrix is essential for estimating spatial models. Most studies construct these matrices based on adjacency, geographic distance, or economic distance^47,48, but few conduct comprehensive screening. Given the strong link between urban economic development and green innovation, this study integrates both geographic distance and economic connections to construct a composite spatial weight matrix.

Unlike previous studies, this research leverages 15 years (2007–2021) of economic data, covering China’s economic restructuring and green innovation transformation. By averaging annual city-level GDP, it provides a more stable measure of economic development, minimizing annual fluctuations. Additionally, rather than relying on provincial or regional data, this study refines geographic information to the prefecture level, using latitude and longitude data for precise distance calculations. This approach enhances the analysis of green innovation diffusion and policy spillover effects.

To quantify spatial relationships, this study constructs a geographic distance matrix (${w}_{g}$) and an economic distance matrix (${w}_{e}$) and then calculates their weighted average to form a composite matrix. The process for constructing economic-geographic matrix is as follows:

$$\begin{array}{c}{w}_{g\left(ij\right)}=\left\{\begin{array}{cc}\frac{\frac{1}{{d}_{ij}^{2}}}{{\sum }_{j=1}^{N}\frac{1}{{d}_{ij}^{2}}},& i\ne j\\ 0,& i=j\end{array}\right.\end{array}$$

(4)

In Eq. (4), ${d}_{ij}$ represents the average geographic distance between cities over the years. The constructed economic distance matrix (${w}_{e}$) is as follows:

$$\begin{array}{c}{w}_{e\left(ij\right)}=\left\{\begin{array}{cc}\frac{\frac{1}{\left|{\dot{y}}_{i}-{\dot{y}}_{j}\right|}}{{\sum }_{j=1}^{N}\frac{1}{\left|{\dot{y}}_{i}-{\dot{y}}_{j}\right|}},& i\ne j\\ 0,& i=j\end{array}\right.\end{array}$$

(5)

In Eq. (5), $\dot{y}$ represents the average per capita regional GDP of the cities over the years. The composite spatial weight matrix is constructed by taking the weighted average o ${w}_{g}$ and ${w}_{e}$ as follows:

$$\begin{array}{c}W=a{W}_{g}+\left(1-a\right){W}_{e}\end{array}$$

(6)

The formula (6) indicates that 0 ≤ a ≤ 1. The optimal weight coefficient a^* is determined based on the AIC and BIC information criteria. Specifically, set a = 0, 0.1,…, 1, estimate the spatial model for each value of a, and calculate the corresponding AIC and BIC values for the model. The weight coefficient that minimizes both the AIC and BIC values is chosen as a^*, and the final composite spatial weight matrix is obtained accordingly.

Figure 4 compares the traditional spatial weighting and the composite spatial weight matrix:

Fig. 4

Comparison between traditional spatial weighting and the composite spatial weight matrix.

Model specification

DID model

At the current stage, the pilot policy is steadily expanding, providing an ideal quasi-natural experiment to assess its impact on high-tech industry development. First, pilot city selection is government-driven, ensuring a degree of exogeneity and mitigating endogeneity concerns. Second, the policy aims to upgrade traditional industries and foster emerging sectors, with implementation primarily at the prefecture level. Analyzing its effects at this level helps minimize policy spillover bias. Moreover, although the number of pilot cities has increased over time, most cities remain non-pilot, naturally forming treatment and control groups. This study employs a DID model to compare high-tech industry development between pilot and non-pilot cities, isolating the policy’s impact. Since implementation occurred in phases, a traditional DID model would only capture effects at a single point in time. To provide a more comprehensive evaluation, this study constructs a multi-period DID model with two-way fixed effects, as shown in Eq. (7).

$$\begin{array}{c}{GIE}_{it} = \alpha + \beta did + \delta {X}_{it} + {\mu }_{i} + {\lambda }_{t} + {\varepsilon }_{it}\end{array}$$

(7)

In the model, i and t represent the city and year, respectively. The dependent variable ${GIE}_{it}$ indicates the GIE of the city. $did$ is a dummy variable representing the pilot policy, ${X}_{it}$ denotes the control variables, ${\mu }_{i}$ 和 ${\lambda }_{t}$ represent city fixed effects and year fixed effects, respectively, and ${\varepsilon }_{it}$ is the random error term.

SDID model

The traditional DID model relies on the Stable Unit Treatment Value Assumption (SUTVA), which requires that one unit’s treatment status does not affect the outcomes of others⁴⁹. However, in spatial contexts, this assumption often fails because policy interventions can influence neighboring units through economic linkages, technology diffusion, or institutional learning. In this study, the pilot policy is likely to generate such spillovers, affecting nearby non-pilot cities and thus violating SUTVA. Thus, the analysis requires spatial models that explicitly account for interunit dependence. Traditional spatial econometric models such as the Spatial Autoregressive Model (SAR), Spatial Error Model (SEM), and Spatial Durbin Model (SDM) effectively capture static spatial dependence. However, these models focus on estimating spatial correlations through lagged outcomes or spatially correlated error terms rather than identifying the causal effects of policy interventions. They do not distinguish between exogenous treatment effects and endogenous spatial interactions, which makes them unsuitable for identifying policy-induced spatial spillovers. To address these limitations, researchers incorporated spatial lags into the DID framework and developed the SDID model.^50,51. This approach embeds treatment indicators into the spatial DID specification, enabling researchers to identify both direct effects and spatial spillovers, making it particularly well suited for evaluating policies with staggered treatment adoption across regions. Building on this approach, this study extends Eq. (8) to construct an SDID model:

$$\begin{array}{c}{GIE}_{it}=\rho W\times {GIE}_{it}+{\alpha }_{1}did+{\alpha }_{2}W\times did+{\gamma }_{1}{X}_{it}+{\gamma }_{2}W\times {X}_{it}+{u}_{i}+{v}_{t}\\ +{\left(1-\lambda W\right)}^{-1}{\varepsilon }_{it}\end{array}$$

(8)

In the model, ${GIE}_{it}$ represents carbon dioxide emissions; $did$ is the core explanatory variable, indicating the interaction term of the policy; ${X}_{it}$ represents other control variables; ${\mu }_{i}$, ${\nu }_{t}$, ${\varepsilon }_{it}$ represent regional fixed effects, time fixed effects, and residuals, respectively.

W is the economic distance matrix generated by the latitude and longitude of prefecture-level city government locations and average GDP from 2007 to 2021. $\rho$ is the spatial autocorrelation coefficient of the dependent variable; ${\alpha }_{2}$ represents the policy spillover effect; ${\gamma }_{2}$ is the spillover effect of the control variables; and $\lambda$ is the spatial autocorrelation coefficient of the random error. Notably, the spillover effect ${\alpha }_{2}$ occurs not only between pilot and non-pilot cities but also between pilot cities. Under SDID model, the total effect of the policy on node cities is ${\alpha }_{1}+{\alpha }_{2}$. Equation (8) represents the general form of SDID model. Depending on whether the relevant coefficients are zero, it can be classified into the Spatial Error DID Model (SEM-DID), the Spatial Lag DID Model (SLM-DID), and the Spatial Durbin DID Model (SDM-DID). In the analysis of spatial spillover effects, the appropriate model among the three will be further selected based on correlation tests.

Results analysis

Spatiotemporal distribution of GIE

The spatiotemporal distribution map visually captures the evolution of innovation across Chinese prefecture-level cities, offering insights into the impact of innovation policies and regional development imbalances. Figure 4 shows GIE distribution in 2007, 2012, 2017, and 2021, with darker shades indicating higher GIE and lighter shades representing lower levels.

In 2007, most regions had low GIE (0.0001–0.0529), with only a few cities, such as Wuhan and Tangshan, exhibiting higher levels. High-efficiency areas were sparse. By 2012, GIE had moderately improved, particularly in northern cities like Beijing, Tianjin, and Shenyang, and central cities like Wuhan, where some regions exceeded 0.1495. However, many areas still had low GIE, with most below 0.0199. By 2017, GIE continued to rise, particularly in northern and central China, with significant increases in Beijing, Tianjin, Hebei, and Wuhan. Southeastern coastal cities like Fuzhou also showed improvement. By 2021, northern and central regions maintained high GIE, while cities like Shenzhen and Changsha experienced substantial growth, signaling a more balanced distribution and expansion of high-efficiency areas.

From 2007 to 2021, northern China experienced significant GIE improvements, this might be largely driven by pilot policies. Initiatives such as coordinated innovation in the Beijing-Tianjin-Hebei region, national high-tech zones, and green manufacturing demonstration parks fostered green technology development, industrial upgrading, and regional cooperation, accelerating green innovation commercialization. Meanwhile, the policy also attracted key scientific resources from institutions like the Chinese Academy of Sciences, universities, and research institutes. Technology transfer and industrial collaboration enhanced spillover effects, boosting regional efficiency. Meanwhile, traditional heavy industries increasingly shifted to high-tech and green sectors, as seen in Tianjin Binhai New Area and Hebei Xiong’ an New Area, improving GIE, reducing pollution, and advancing green technology adoption. Additionally, fiscal support and policy incentives—including subsidies, tax breaks, and green R&D funds—further accelerated regional green innovation, while green manufacturing zones and smart manufacturing pilots played a critical role.

Overall, from 2007 to 2021, GIE in China’s prefecture-level cities exhibited a clear upward trend. Initially concentrated in a few cities, high-efficiency regions expanded over time, reflecting the positive impact of national policies, economic growth, and increased innovation investment.

Baseline regression results

The baseline regression results in Table 3 show that the policy has a significant positive impact on GIE. Column (1) reports results without additional control variables. The coefficient of the policy variable (did) is 0.0173 and statistically significant at the 1% level, indicating that the policy significantly increased GIE in pilot cities compared to non-pilot ones. Columns (2) to (6) sequentially incorporate control variables. The did coefficient remains consistently positive and significant, with a value of 0.0143 in the fully specified model. This confirms the robustness of the estimated policy effect. These results support Hypothesis 1, which posits that the Pilot Policy for Innovative Industrial Clusters improves urban GIE.

Table 3 Baseline regression results.

Parallel trend test

To ensure the validity of the DID estimation strategy, this study tests the parallel trend assumption, which requires that GIE trajectories in pilot and non-pilot cities follow similar patterns prior to the policy intervention. Figure 5 reports the event-study estimates over a symmetric window around the policy year. The year of implementation is labeled as “current”, while years before and after are denoted as “pre5” through “pre2” and “post1” through “post5”, respectively. All pre-treatment coefficients are small and statistically insignificant, with confidence intervals overlapping zero. This confirms that treated and control cities exhibit parallel pre-trends in GIE, validating the identification strategy of the DID model. After the policy implementation, the estimated effects increase over time, indicating a growing influence of the policy on GIE. These results reinforce the validity of the identification strategy and suggest that the estimated policy effect on GIE is not confounded by pre-existing trends. Overall, the evidence supports Hypothesis 1 and identifies a causal impact of the policy on urban GIE.

Fig. 5

Spatiotemporal Distribution of GIE (Map created using ArcGIS 10.2, http://www.esri.com/software/arcgis).

Robustness test

PSM-DID

Although pilot policy is issued by the National Development and Reform Commission, the selection of cities may have been influenced by factors such as economic development conditions, resource endowments, and industrial foundations. To reduce sample selection bias, this paper re-examines the regression results of Eq. (1) using the PSM-DID method. We employ control variables as covariates in a logistic regression model to calculate the propensity scores of each covariate based on the Logit model. A1-to-4 nearest neighbor matching method was used to match the experimental group and the reference group. Next, the balance of the matched samples is tested. The results are shown in Fig. 6. The black solid dots represent the standardized bias percentage before matching, where the bias of the covariates was relatively large, indicating significant differences between the treatment and control groups on these variables. After matching, the bias is significantly reduced, with nearly all variables controlled at a low level. This demonstrates that PSM effectively balanced the covariates between the groups, minimizing differences. Finally, using the newly matched samples, we apply Eq. (1) for regression. Column (1) in Table 4 shows that the estimated coefficient of pilot policy dummy variable is 0.016, which remains significantly positive at the 1% level.

Fig. 6

Parallel trends test.

Table 4 PSM-DID and replace the explained variable.

Substituting dependent variable

Empirical studies have shown a significant positive correlation between the improvement of GIE and Green Total Factor Productivity (GTFP)^44,52. Panel data analysis of Chinese cities indicates that cities with higher GIE also tend to exhibit superior GTFP performance. Through green technological innovation, cities can achieve sustainable development goals by reducing resource consumption and environmental pollution while promoting economic growth. Therefore, this paper uses GTFP of cities as the dependent variable and incorporates it into Model (1) for regression analysis. The regression results are shown in Column (2) of Table 4, where the coefficient of the did variable is 0.0238, significantly positive at the 1% level. This indicates that the pilot policy significantly improves GTFP and has a positive impact on urban green innovation activities, confirming the robustness of results.

Placebo test

To ensure that the DID regression results are not influenced by unobservable city characteristics or other factors, a placebo test is conducted. Specifically, 500 samples are drawn from all 280 cities, with 55 randomly selected as the virtual treatment group in each sample, while the remaining cities serve as the control group for the DID regression. Figure 7 shows the kernel density distribution of the estimated coefficients in the absence of actual policy effects. The density is concentrated around zero and exhibits a symmetrical shape, indicating that, in the virtual experiment, the estimated coefficients do not significantly deviate from zero, confirming the validity of the model specification. In contrast, the actual estimated value does not fall within the density distribution of the virtual experiment, further demonstrating the robustness of the results. This suggests that model misspecification or random errors do not affect the conclusions. Overall, the placebo test supports the model’s robustness and reinforces the reliability of the policy effect estimates (Fig. 8).

Fig. 7

PSM-DID test results.

Fig. 8

Placebo test.

Excluding other policy interference

Several policies implemented during the study period are closely related to this paper, including the three batches of 81 low-carbon city pilots (2010–2017), the “Broadband China” strategy (2013), and the smart city pilot programs initiated after 2012. The low-carbon city policy promoted the adoption of green technologies and environmental measures, encouraging R&D and industrial adjustments to improve GIE. The “Broadband China” policy enhanced urban green innovation by developing communication infrastructure and digital applications, boosting overall GIE. The smart city policy optimized resource management, improved energy efficiency, and fostered technological innovation, further advancing green innovation.

To control for the potential impact of these policies, this paper sequentially adds dummy variables representing the years of implementation of the three policies into the baseline regression model. The results, shown in columns (1), (2), and (3) of Table 5, reveal that after controlling the three policies, the coefficient of the did variable is 0.0125, 0.0138 and 0.0142 respectively, significantly positive at the 1% level. The results indicate that the pilot policy has a significant green innovation effect, and the results remain robust.

Table 5 Excluding other policy interferences.

Heterogeneity analysis

Administrative level

Central cities concentrate key innovation resources, including research institutions, specialized talent, and capital, which drive green technological innovation. They also benefit from superior infrastructure, policy frameworks, and market demand, facilitating the adoption and diffusion of green technologies. To assess whether the policy effect varies with administrative hierarchy, we classify municipalities, sub-provincial cities, and provincial capitals as central cities, and treat others as peripheral. As shown in Table 6, columns (1) and (2), the policy significantly improves GIE in central cities, with a coefficient of 0.0607 at the 1% level, while its effect in peripheral cities is not statistically significant.

Table 6 Results of heterogeneity analysis.

The disparate reflects differences in resource allocation and technological adoption. Central cities are better at aligning incentives with their strengths, enabling more targeted investment and stronger green innovation outcomes. In addition, technological advancement is faster in central cities due to stronger innovation infrastructure, better knowledge absorption, and closer academic–industry ties. As a result, the policy tends to generate greater improvements in green innovation in these cities. In contrast, peripheral cities often lack the capacity to apply policy support effectively to green innovation, which limits the policy impact.

Resource endowment

Non-resource-based cities drive growth through technological progress, industrial upgrading, and innovation, while resource-based cities rely on resource extraction, which may limit innovation incentives. Additionally, non-resource-based cities face greater environmental pressures, motivating them to prioritize green technological innovation to balance economic growth and sustainability. Based on the National Sustainable Development Plan for Resource-based Cities (2013–2020), this study classifies cities into resource-based and non-resource-based categories and conducts grouped regression, with results presented in Table 6, columns (3) and (4).

In resource-based cities, the estimated coefficient for did is not statistically significant, whereas in non-resource-based cities, it is 0.0192 and significantly positive at the 1% level. This indicates that the pilot policy significantly enhances GIE in non-resource-based cities but has a limited effect on resource-based ones. Non-resource-based cities, which emphasize technological advancement and high-tech industries, benefit from stronger innovation capacity and policy support. Additionally, their lower dependence on resource extraction reduces environmental constraints, facilitating green technology adoption. In contrast, resource-based cities often prioritize economic growth over environmental protection, leading to lower GIE.

Key environmental protection cities

Key environmental protection cities enforce stricter environmental policies, compelling businesses and governments to prioritize green innovation. These cities receive greater policy and financial support, fostering green industry development, while higher public awareness drives demand for green technologies. This study examines the impact of the policy on GIE from an environmental perspective. Based on the National Environmental Protection 11th Five-Year Plan (2007), cities are classified as key or non-key environmental protection cities, with grouped regression results presented in Table 6, columns (5) and (6).

In key environmental protection cities, the coefficient for did is 0.0190 and significantly positive at the 1% level, whereas in non-key cities, it is not statistically significant. This suggests that the policy significantly enhances GIE in key environmental cities but has little effect elsewhere. The disparity likely stems from stricter regulations, stronger oversight, and greater funding and policy incentives in key cities, which drive R&D and green technology adoption. Additionally, higher public environmental awareness reinforces sustainability priorities, further improving GIE. In contrast, non-key cities face weaker environmental pressures, resulting in lower incentives for green innovation.

Mechanism analysis

To further explore the relationship between the pilot policy and GIE, it is necessary to analyze the mechanisms. Based on the previous discussion, the pilot policy enhances urban GIE by reducing energy consumption intensity, promoting industrial structure upgrading, improving green technology innovation levels, and enhancing the development of digital infrastructure. We establish the following mediation effect model⁵³:

$$\begin{array}{c}{GIE}_{it}={\eta }_{0}+{\eta }_{1}did+{\eta }_{2}\sum X{^{\prime}}_{it}+{\mu }_{i}+{\gamma }_{t}+{\epsilon }_{it}\end{array}$$

(9)

$$\begin{array}{c}{M}_{it}={\theta }_{0}+{\theta }_{1}did+{\theta }_{2}\sum X{^{\prime}}_{it}+{\mu }_{i}+{\gamma }_{t}+{\epsilon }_{it}\end{array}$$

(10)

$$\begin{array}{c}{GIE}_{it}={\lambda }_{0}+{\lambda }_{1}did+{\lambda }_{2}{M}_{it}+{\lambda }_{3}\sum X{^{\prime}}_{it}+{\mu }_{i}+{\gamma }_{t}+{\epsilon }_{it}\end{array}$$

(11)

Among them, ${M}_{it}$ represents the mediator variables, including energy consumption intensity (Energy), industrial structure upgrading (Stru), green technology innovation level (Innov), and digital infrastructure development level (Digital_infra). Energy consumption intensity (Energy) is measured by energy consumption per unit of GDP. Industrial structure upgrading (Stru) is calculated as: the proportion of value added by the primary industry to GDP × 1 + the proportion of value added by the secondary industry to GDP × 2 + the proportion of value added by the tertiary industry to GDP × 3. The level of green technological innovation (Innov) is measured by the urban innovation index of each city. The level of digital infrastructure (Digital_infra) is measured based on China’s Classification of Strategic Emerging Industries (2018), which includes sectors such as next-generation information technology, high-end equipment manufacturing, internet services, cloud computing, big data, artificial intelligence, and smart terminals. This study assesses digital infrastructure development by analyzing the proportion of related terms in government reports from provincial, municipal, and autonomous regional governments⁵⁴. The methodology involves collecting government work reports from 2004 to 2018, identifying keywords related to new digital infrastructure, and using Python for word segmentation to count total words and digital infrastructure-related terms. The proportion of these terms is then calculated. To refine the measure, the study incorporates the ratio of IT employment to the total city population, multiplying this factor with keyword frequency to assess each city’s digital infrastructure development level. All variables are logarithmically transformed. If the coefficients ${\eta }_{1}, {\theta }_{1}$ and ${\lambda }_{2}$ are significant, and the coefficient and significance of ${\lambda }_{1}$ are lower than those of ${\eta }_{1}$, it indicates that the mediating variable plays a mediating role in the relationship between the pilot policy and GIE.

This study first examines the mediating effect of energy consumption. As shown in Column (1) of Panel A in Table 7, the policy significantly reduces energy consumption intensity (coefficient = − 0.2155, significant at 1% level), indicating that it effectively lowers energy consumption. Column (2) of Panel A in Table 7 shows that energy consumption intensity negatively affects GIE (coefficient = − 0.0010, significant at 1% level), suggesting that energy reductions enhance GIE. Additionally, the did coefficient remains significantly positive at the 1% level, indicating that energy consumption intensity mediates the relationship between the policy and GIE. These results are consistent with Hypothesis 2, which suggests that the policy improves GIE partly by reducing energy intensity.

Table 7 Results of mechanism analysis.

Next, this study examines the mediating effect of industrial structure upgrading. As shown in Column (3) of Panel A in Table 7, did has a coefficient of 0.0240, significant at 10% level. This suggests that the pilot policy promotes industrial structure. Column (4) of Panel A in Table 7 indicates that industrial structure upgrading positively impacts GTI (coefficient = 0.0019, significant at 1% level). Additionally, the did coefficient is 0.0145, remains significant at 1% level, suggesting that industrial structure upgrading mediates this effect. This is consistent with Hypothesis 3, where industrial structure upgrading contributes to the policy’s effect on GIE.

The mediating effect of green technological innovation also be examined. As shown in Column (1) of Panel B in Table 7, the policy’s impact on green technological innovation is significant, with a coefficient of 0.0942, significant at the 1% level. This suggests that the policy promotes green technological innovation in cities. Column (2) of Panel B in Table 7 shows that green technological innovation positively influences GIE (coefficient = 0.0059, significant at 1% level), suggesting that improvements in green technologies can enhance urban GIE. Moreover, the DID coefficient remains significant after adding the green technological innovation variable, indicating a mediating effect. The result aligns with Hypothesis 4, highlighting green technological innovation as a transmission channel.

Finally, this study verifies the mediating effect of digital infrastructure development. As shown in Column (3) of Panel B in Table 7, the pilot policy significantly promotes digital infrastructure development (coefficient = 0.0876, significant at 1% level). This indicates that the policy accelerates the adoption of digital technologies and enhances the digital economy’s capacity. Column (4) of Panel B in Table 7 demonstrates that digital infrastructure development positively impacts GIE (coefficient = 0.0196, significant at 5% level), suggesting that improved digital infrastructure enhances GIE in cities. Furthermore, did coefficient remain significant after adding the digital infrastructure variable (0.0125, significant at 1% level), indicating that digital infrastructure development mediates the relationship between policy and GIE. These results are in line with Hypothesis 5, digital infrastructure emerges as a key mechanism enhancing GIE.

Spatial effect analysis

The empirical results presented above use the DID model to identify the causal relationship between the pilot policy for innovative industrial clusters and urban GIE. However, they do not account for the spatial effects of the policy on urban GIE. Therefore, this study incorporates spatial factors and conducts further analysis using spatial econometric models.

Test of spatial correlation

Before estimating the SDID model, it is essential to verify the spatial correlation of urban GIE. Global Moran’s I is used to test the spatial correlation and spillover effects of GIE. As shown in Table 8, the Moran’s I values are positive and statistically significant at the 1% level, indicating a significant positive spatial correlation of GIE between cities.

Table 8 Global Moran’s index of GIE.

Applicability test of the spatial DID

Before estimating the spatial econometric model, it is crucial to select the appropriate model type to ensure reliable and accurate results. Initially, the LM test compares the constructed model with panel mixed regression, assessing spatial dependence and verifying whether spatial lag and error terms exhibit correlation, justifying the use of the Spatial Durbin Model (SDM). Next, the LR test determines whether the SDM can be simplified to a Spatial Autoregressive Model (SAR) or Spatial Error Model (SEM). Finally, the Hausman test identifies whether a fixed-effects or random-effects model is more suitable.

Table 9 shows that both the LM test and robust LM test statistics are significant, confirming spatial error and spatial lag effects, supporting the initial selection of SDM. The LR test rejects the null hypothesis that SDM can be simplified to SAR or SEM, further validating its choice. Additionally, the Hausman test, significant at the 1% level, favors a fixed-effects model over random effect. Based on these results, the SDM with fixed effects is used for empirical analysis.

Table 9 LM & LR & Hausman test results.

SDID regression results

Table 10 presents the SDID model estimation results based on the composite spatial weight matrix, alongside SAR and SEM regression results for robustness. The did coefficient is significantly positive across columns (1) to (3), confirming that the pilot policy significantly enhances GIE in pilot cities, even after accounting for spatial correlation. Additionally, the spatial effect coefficients of green technological innovation (W × did) are significantly positive in all three models, indicating strong spatial spillover effects and further supporting the use of the SDID model.

Table 10 Regression results of SDID model.

Using the partial differential method, the spatial effects in column (3) are decomposed. The direct effect of did is 0.0150 and significantly positive, demonstrating that the policy improves GIE in pilot cities. This aligns with baseline regression results, though the SDM-DID model yields a smaller coefficient, suggesting that the baseline regression may overestimate the direct policy impact by ignoring spatial correlation. The estimated indirect effect is 0.0219 and statistically significant, suggesting that the policy generates spillover benefits extending to neighboring cities. This is consistent with Hypothesis 6, which posits that the policy improves GIE through both direct and spatial spillover effects.

Conclusion and policy recommendations

Using panel data from 280 Chinese cities (2007–2021), this study examines the impact of the Pilot Policy for Innovative Industrial Clusters on urban GIE through a multi-method approach. The DID model estimates the direct policy effect, the SDID model captures spatial spillovers, and the mediation model identifies underlying mechanisms. The findings reveal that: (1) The policy significantly enhances GIE in pilot cities compared to non-pilot cities, demonstrating its effectiveness in promoting green innovation. (2) It operates through four channels: reducing energy intensity, upgrading industrial structure, driving green technology innovation, and accelerating digital infrastructure development. (3) The policy generates positive spatial spillovers, improving GIE in both pilot and neighboring cities. (4) Its impact varies by city type, being stronger in central and non-resource-based cities and weaker in peripheral and resource-based cities. Additionally, it is more effective in key environmental protection cities than in non-key cities.

Based on the above conclusions, the following recommendations are provided: (1) Refine the design and implementation of Pilot Policy for Innovative Industrial Clusters. Governments should develop clearer policy objectives centered on green innovation and scale up successful pilot models in other cities. They should also coordinate local implementation efforts under a unified policy framework to ensure consistency across regions. This approach can maximize the policy’s effectiveness and accelerate green transformation at the national level. (2) Strengthen core mechanisms that promote green innovation. Governments should adopt targeted fiscal incentives to ease resource constraints and encourage greater investment in green innovation. Such investment accelerates industrial upgrading and lays the foundation for larger-scale, longer-term improvements in urban green innovation. Meanwhile, digital infrastructure should be reinforced to support technological renewal and value chain upgrading throughout the green innovation process. These measures leverage key transmission mechanisms to amplify the impact of the Pilot Policy for Innovative Industrial Clusters on urban green innovation. (3) To ensure the Pilot Policy for Innovative Industrial Clusters delivers broad and equitable outcomes, governments should adopt city-specific strategies based on characteristics. Peripheral cities often require improved innovation infrastructure and coordination mechanisms to strengthen their capacity to respond effectively to policy initiatives. For example, cities such as Anshun and Yulin need stronger inter-city linkages, basic infrastructure, and collaborative platforms to engage in green innovation. In resource-dependent cities such as Fushun and Tongchuan, policy efforts should aim to reduce reliance on traditional extractive sectors and support the transition toward cleaner, innovation-oriented industries. Facing long-term dependence on coal and heavy industry, these cities require fiscal incentives for green equipment, enterprise transition subsidies, and pilot zones for low-carbon transformation. In contrast, non-resource-based cities like Suzhou and Ningbo have built strong foundations in high-end manufacturing and technological innovation. These cities have leveraged the policy to embed green standards in manufacturing and expand innovation ecosystems. Scaling their experience can help diffuse green innovation more broadly. For non-priority environmental cities, policymakers must strengthen regulatory and institutional incentives to create favorable conditions for green innovation. In cities like Handan and Xinxiang, where oversight is limited, stricter performance targets, transparency, and subsidies can stimulate green innovation. By addressing city-specific constraints, the pilot policy can enhance its effectiveness in promoting green innovation and improve the equity of outcomes across regions. (4) Align policy incentives with market demand. While policy incentives play a critical role in initiating green innovation, sustained progress requires strong and consistent market signals. Governments should enhance demand-side mechanisms such as green public procurement, eco-labeling, and export support to promote green innovation. These instruments can strengthen incentives by improving the visibility and commercial appeal of green products. Therefore, aligning policy incentives with market demand is critical to sustaining green innovation beyond the policy period. (5) Institutionalize spatial coordination in green policy implementation. Green innovation policies often generate spatial spillovers, making intercity coordination essential to fully realize their potential. Governments should align performance targets across regions, promote joint planning, and facilitate shared access to innovation resources. At the same time, evaluation systems should account for cross-city impacts rather than focusing solely on local outcomes. Overall, Spatial coordination aligns policies across cities and promotes the diffusion of green innovation. Such diffusion amplifies policy spillovers and strengthens the spatial reach of green innovation. It improves policy effectiveness over time and enhances regional green innovation efficiency.