How does climate policy uncertainty influence energy consumption transition in China: evidence from 277 cities

Energy consumption transition is fundamental to achieving carbon neutrality without sacrificing development in developing economies. China’s experience under its dual-carbon targets and evolving climate governance provides a valuable case. Using panel data from 277 Chinese prefecture-level cities from 2007 to 2023, this study applies two-way fixed effects models and spatial Durbin models to assess the impact of climate policy uncertainty on urban energy consumption transition. The results show that higher climate policy uncertainty significantly impedes energy consumption transition, particularly in resource-based, highly marketized, and rapidly developing eastern cities. This inhibitory effect operates through both supply and demand channels, weakening industrial chain resilience, suppressing corporate green innovation disclosure, and dampening public environmental awareness. Spatial analysis further reveals substantial negative spillovers, with the effect of climate policy uncertainty on neighboring energy consumption transition roughly six times stronger than its local impact. The findings highlight that proactively managing climate policy uncertainty is essential for enabling a stable and effective energy consumption transition, offering a practically relevant reference for developing countries to anticipate and manage governance challenges on the path to sustainable growth.

1 Introduction

Sustainable development has become a central global objective, requiring a balance between economic growth and environmental stewardship. A key aspect of this challenge is the transition in energy consumption from fossil fuels to clean, low-carbon systems (Oberschelp et al., 2019). For developing economies, this transition poses a dilemma: reconciling immediate economic needs with long-term environmental sustainability (Markard, 2018). In China, this challenge is heightened by its coal-rich, oil-poor, and gas-scarce resource endowment, which has historically produced a coal-dominated energy structure that constrains energy consumption patterns and impedes decoupling economic growth from carbon emissions (Chen et al., 2022; Wang et al., 2018). As the world’s largest developing economy and energy consumer, China’s pursuit of its “Dual Carbon” goals (carbon peaking by 2030 and carbon neutrality by 2060) offers a critical test case. The success of energy consumption transition (ECT) in China is thus crucial not only for national welfare but also for achieving global climate goals (Mercure et al., 2021).

ECT is not a simple linear substitution involving the reduction of traditional fossil fuel energy shares and the increase of renewable energy shares. Rather, it is a complex socio-technical evolution driven by the combined forces of economic structure, technological innovation, social cognition, and governance capacity. Existing research has identified key drivers of ECT, including levels of economic development, industrial restructuring, energy mix optimisation, human capital accumulation, and urbanisation processes (Gao et al., 2025; Hao et al., 2023; Hou et al., 2025; Li et al., 2024). These studies reveal the multidimensional forces propelling ECT. However, as the “dual carbon” goals advance, the stability and predictability of the policy environment are increasingly becoming significant institutional constraints influencing energy consumption behaviour and transformation pathways (Wu et al., 2025).

In contrast, research on climate policy uncertainty (CPU) has predominantly focused on its impact on micro-level subject behaviour or macroeconomic fluctuations. At the micro level, CPU has been demonstrated to reduce firms’ total factor productivity (Ren et al., 2022), compress corporate value (Ongsakul et al., 2023), and inhibit strategic actions such as digital transformation. This stems from firms’ difficulty in discerning future policy directions and implementation intensity, thereby facing heightened expectation risk (Lin and Zhao, 2023). At the macro level, CPU disturbs energy markets (Shang et al., 2022), stock markets (Xu J. et al., 2023), and financial markets (Zhang X. et al., 2025). Although research has begun to examine the impact of CPU on energy systems, discussions remain largely confined to singular dimensions such as energy pricing, energy efficiency, or renewable energy development. In contrast, far less attention has been devoted to the energy consumption structure itself—a systemic process characterised by strong path dependency and long-term persistence.

However, the ECT is exposed to systemic risks, and rising CPU has the potential to influence the transition in complex and yet underexplored ways. The complexity of climate systems, combined with incomplete information, global political volatility, and domestic economic restructuring, often results in misaligned policy targets and frequent implementation deviations, thereby intensifying CPU (Magnan et al., 2021). Ambiguous climate policy signals deter long-term investment in green technologies, causing firms to postpone clean energy adoption (Iqbal et al., 2024) and local governments to delay phasing out conventional energy assets, constraining low-carbon energy supply (Wang and Wang, 2020). Fluctuating policy environments also erode public confidence and environmental awareness, reducing the demand for clean energy (Zheng et al., 2023). Consequently, although climate policies are designed to accelerate the transition, the pervasive uncertainty surrounding them creates a paradoxical effect that may ultimately slow the pace of ECT (Lin and Zhao, 2023). Yet, despite these plausible channels, existing studies have not provided systematic evidence on how CPU affects ECT, leaving its overall impact and underlying mechanisms largely unclear. This raises key questions: Does CPU fundamentally inhibit or promote ECT across Chinese cities, and how does this effect vary with city characteristics? What are the underlying mechanisms through supply- and demand-side channels? Furthermore, does the CPU generate spatial spillovers affecting neighboring cities’ ECT, given the interconnected nature of regional economies?

Within China’s institutional framework for advancing its dual carbon goals, climate policies typically follow a “pilot-first, gradual expansion” implementation pattern. Localities pursue diverse pathways, with significant variations in enforcement intensity and policy pacing. Rapid policy iteration, regional inconsistencies, and frequent short-term target adjustments render CPU particularly pronounced at the municipal level (Ma et al., 2023). The emergence of phenomena such as “campaign-style carbon reduction” and “power rationing” in some regions of China in 2021 demonstrates that when policy implementation lacks stability and predictability, urban energy supply and demand structures become vulnerable to systemic shocks¹. Concurrently, China’s resource endowment exhibits marked regional disparities characterised by “abundant coal, ample wind, and scarce water” (Chen et al., 2022). This results in variations among cities in energy structures, industrial foundations, and green technology capabilities, leading to significant heterogeneity in their sensitivity to CPU and response strategies. Furthermore, ECT itself exhibits characteristics of “high upfront investment, long payback periods, and strong path dependency.” Should the CPU rise, it may simultaneously suppress technological investment and industrial upgrading incentives on the supply side while weakening public green consumption preferences on the demand side. This creates a structural mismatch risk where clean energy demand expansion coexists with traditional energy supply inertia (Wu and Li, 2025). These characteristics indicate that, within the Chinese context, CPU is not merely policy noise but rather a key variable influencing ECT (Wu et al., 2025).

In light of these considerations, this study systematically examines the impact of CPU on ECT using a panel of 277 Chinese prefecture-level cities from 2007 to 2023. First, a two-way fixed effects model controlling for city and year characteristics shows that CPU significantly inhibits local ECT. Second, heterogeneity analysis indicates that this negative effect is stronger in eastern cities and in cities with higher resource dependence, greater marketization, or more advanced economic development. Third, a “dual-suppression” mechanism is identified: CPU constrains industrial chain resilience and green innovation disclosure on the supply side while dampening public environmental awareness on the demand side. Fourth, a spatial Durbin model reveals that CPU generates negative spillovers on neighboring cities’ ECT, with the spatial effect roughly six times greater than the local inhibitory effect.

The marginal contributions of this paper lie in three main aspects:

Firstly, the methodology for measuring ECT has been refined. Unlike existing research, which typically relies on single indicators, this paper builds upon Shen et al. (2023)’s Energy System Performance Index to construct a comprehensive ECT index capable of simultaneously capturing optimisation of energy consumption structure, efficiency gains, and environmental benefits. This index integrates both process-oriented and outcome-oriented dimensions, thereby more comprehensively reflecting the multidimensional characteristics of ECT. Secondly, the theoretical mechanism through which the CPU influences ECT has been enhanced. This paper proposes a “dual suppression of supply and demand” framework, systematically identifying three transmission pathways: industrial chain resilience, green innovation disclosure, and public environmental awareness. This addresses the limitation of single-mechanism approaches in existing research, providing a more systematic theoretical logic and empirical support for understanding the complex relationship between CPU and ECT. Thirdly, the study broadens the research perspective on CPU. It identifies heterogeneous effects at the city level under varying resource endowments, marketisation, and economic development, revealing significant negative spatial spillovers of CPU. This provides a basis for formulating climate policies that balance local adaptation with regional coordination.

The remainder of this paper is structured as follows. Section 2 develops the theoretical framework and presents the research hypotheses. Section 3 describes the empirical methodology, variable definitions, and data sources. Section 4 presents the baseline results and conducts robustness tests, and examines the underlying mechanisms through mediation analysis. Section 5 investigates the spatial spillover effects. Section 6 concludes with findings and policy implications.

2 Theoretical analysis and research hypotheses

2.1 Direct effects of CPU on ECT

Promoting the green and low-carbon transition of urban energy consumption is a key pathway to achieving the “dual carbon” goals. As a central guiding instrument, the CPU can profoundly affect the energy transition by altering the expectations and decision-making logic of economic agents. Grounded in rational expectations theory (Muth, 1961) and prospect theory (Kahneman and Tversky, 1979), CPU inhibits ECT by distorting the decision-making calculus of both firms and households. Expected utility theory further indicates that, under budget constraints, agents choose optimal options based on maximizing expected utility. Firms and households, as the main actors of energy consumption, make decisions constrained by both the policy environment and expected returns. On the supply side, firms facing heightened CPU perceive increased risks in long-term green investments (Iqbal et al., 2024), given ambiguities in future regulatory standards and carbon prices (Morão, 2025). This risk aversion leads to postponed or reduced low-carbon investments as firms prioritize short-term returns from existing high-carbon modes (Xu X. et al., 2023). On the demand side, in household energy choices, the decision between renewable and traditional energy is essentially a cost–benefit trade-off. Households exhibit similar loss aversion, focusing on short-term cost increases rather than long-term environmental benefits, thus preferring conventional energy sources (Shang et al., 2022). This dual-suppression mechanism constrains ECT from both supply and demand perspectives, leading to our first hypothesis.

Hypothesis 1

Hypothesis 1. The CPU hinders the urban ECT.

2.2 Mechanism effects of CPU on ECT

Climate policy uncertainty not only affects the expectations and judgements of businesses and households, but also systematically alters the supply-demand structure of energy markets by disrupting three core mechanisms: coordinated investment, technology diffusion, and consumption preferences. This, in turn, hinders the green transition of energy consumption. Accordingly, this paper selects “industrial chain resilience” and “green innovation disclosure” from the supply side to represent structural adaptability and technological diffusion efficiency, respectively. From the demand side, “public environmental awareness” is chosen to represent the driving force behind green consumption behaviour. These three mechanisms correspond to the three key dimensions of “structure-technology-behaviour” in ECT, enabling a comprehensive characterisation of the institutional and behavioural pathways through which CPU influences ECT.

2.2.1 Supply-side mechanism: Industrial chain resilience (ICR)

The supply-side adjustment of energy consumption transition depends not only on firms’ green investment or technological innovation but, more fundamentally, on the stability and coordination capacity of the entire energy industry chain. According to production network theory, a firm’s production and operational activities simultaneously rely on upstream output supply and downstream input demand. The complex industrial-chain network formed through input–output linkages transmits and amplifies external shocks (Acemoglu et al., 2016). The energy system, likewise, consists of multiple tightly connected segments with high interdependence and pronounced upstream–downstream risk transmission characteristics (Dohale et al., 2023), making it highly sensitive to changes in the policy environment. From the perspective of supply chain management theory, policy uncertainty disrupts the expectations of upstream and downstream firms, leading to supply–demand mismatches and rising coordination costs, thereby weakening supply chain resilience (Katsaliaki et al., 2022). As a systemic risk, CPU propagates and amplifies through the industry chain, making it difficult for firms to recover promptly from shocks and ultimately undermining green production efficiency and the release of green capacity. Meanwhile, fluctuations in upstream costs or disruptions in key segments can rapidly spread along the chain and generate systemic supply bottlenecks (Wu et al., 2024). Compared with firm-level mechanisms, industrial chain resilience more comprehensively captures the systemic and cross-segment impacts of CPU on the energy supply system. Therefore, it constitutes a key mediating mechanism for understanding how CPU causes the supply side of the energy consumption system to become “unwilling or unable to transition.”

Industrial chain resilience serves as an essential buffer and enabler for ECT on the supply side. A resilient industrial chain can flexibly adapt to technological changes and resource reallocation during low-carbon upgrading, smoothing capacity fluctuations, and facilitating the shift from resource-intensive to technology-intensive production (Yang et al., 2021). Furthermore, it resists input fluctuations and market risks in developing low-carbon industries, ensuring stable progress from R&D to large-scale application (Song et al., 2024). Moreover, evidence on climate governance suggests that coherent and stable policy signals play a supportive role in promoting industrial ecological upgrading (Yan, 2025), implying that uncertainty in policy orientation may have the opposite effect. In this context, rising CPU systematically weakens ICR. CPU disrupts collaborative expectations among upstream and downstream firms, heightening concerns about potential policy shifts. To avoid the risk of cooperation losses under uncertain policy environments, firms become more cautious about engaging in long-term, relationship-specific investments in low-carbon cooperation. This reduction in commitment undermines the industrial chain’s ability to buffer shocks and reorganize effectively (Chowdhury and Quaddus, 2017). This erosion of collaborative stability and adaptive capacity hinders the industrial chain’s upgrade from “carbon-dependent” to “low-carbon adaptive.” Consequently, CPU disrupts low-carbon collaboration expectations and weakens collective resilience, thereby cutting off a crucial supply-side support for ECT. Based on this, this paper proposes.

Hypothesis 2

Hypothesis 2. CPU hinders the ECT by weakening ICR.

2.2.2 Supply-side mechanism: green innovation disclosure (GID)

Green technological innovation serves as the core driver for advancing the low-carbon transformation of energy consumption structures. Its successful implementation hinges significantly on enterprises’ capacity to secure capital, technology, and collaborative resources. According to information asymmetry theory (Akerlof, 1970), external investors and partners often find it difficult to accurately assess a firm’s green innovation capability and its commitment to low-carbon transition, which increases financing costs and reduces willingness to collaborate. Existing research also shows that environmental and innovation-related disclosures can mitigate information risk and enhance corporate credibility, thereby facilitating resource acquisition and cooperation (Fan and Ma, 2024). As a vital non-financial disclosure mechanism, green innovation reporting conveys an organisation’s technical capabilities and long-term commitment to green transition to investors and partners. This mitigates information asymmetry, reduces financing and collaboration costs, thereby fostering the research, development, and diffusion of green technologies (Healy and Palepu, 2001). Amid rising climate policy uncertainty, market expectations become more susceptible to disruption, exposing corporate green innovation activities to heightened risks. In such circumstances, disclosure mechanisms prove particularly crucial for stabilising expectations and attracting resource allocation. Consequently, green innovation disclosure effectively reflects how CPU influences green technology supply through expectation and resource allocation channels.

GID acts as both a key signal carrier for attracting resources and a vital channel for technology diffusion in the ECT. Through annual or CSR reports, firms disclose green innovation information to reduce information asymmetry with investors and partners, thereby securing more support for green R&D (Xie et al., 2019). Such disclosure also generates knowledge spillovers, promoting the large-scale application of green technologies and improving clean energy supply (Huang et al., 2025). Nevertheless, CPU systematically undermines this mechanism across motivation, quality, and effectiveness. At the motivation stage, CPU blurs the expected return of disclosure, making firms hesitant to commit for fear of inconsistency with future policies (Xiang et al., 2020; Zhao et al., 2023). At the quality stage, heightened risk aversion leads firms to disclose vague, non-quantitative information to avoid future accountability, lowering the informativeness of disclosures (Badia et al., 2021). At the effectiveness stage, CPU undermines the credibility of disclosure signals, causing investors to reduce financing for green R&D and peers to delay technology adoption (Niu et al., 2023). In sum, CPU suppresses disclosure across the entire chain, breaking the critical link between green technologies and market resources. Based on this, this paper proposes.

Hypothesis 3

Hypothesis 3. CPU hinders ECT by suppressing firms’ GID.

2.2.3 Demand-side mechanism: public environmental awareness (PEA)

The demand-side drivers of ECT stem from residents’ awareness, attitudes, and behavioural choices regarding low-carbon lifestyles, all of which are significantly influenced by public environmental consciousness. According to behavioural economics theory, the allocation of attention resources constitutes a crucial factor influencing economic agents’ decision-making (Sims, 2006). Consequently, heightened public attention to climate change increases cognitive investment in related issues, thereby strengthening perceptions of climate risks, enhancing awareness of low-carbon responsibilities, and fostering greater inclinations towards energy conservation and green consumption. This, in turn, drives improvements in household-level energy consumption structures. Thus, public environmental awareness effectively delineates how CPU influences energy demand through social cognition and behavioural preference channels.

PEA constitutes the social foundation and a critical demand-side driver for advancing ECT. Informed by the theory of planned behavior (Ajzen, 1991) and the value-belief-norm theory (Stern et al., 1999), individual energy consumption behaviors are shaped by environmental knowledge, attitudes, and social norms. Enhanced public awareness drives demand for green products and low-carbon services, imposing market constraints on corporate environmental behavior and compelling firms to accelerate their green transformation. However, an elevated CPU can disrupt these positive dynamics. First, CPU weakens the impetus for governments and NGOs to promote low-carbon concepts, slowing the dissemination of environmental knowledge and undermining the cognitive foundation for green consumption (Ding et al., 2018). Second, CPU reduces residents’ trust in the stability and effectiveness of government policies, such as subsidies and tax incentives, thereby diminishing their enthusiasm for purchasing low-carbon products (Dong et al., 2020). Moreover, frequent policy adjustments signal instability, which may foster public perception that “individual actions have limited effects, discouraging active participation in pro-environmental behaviors (Nowlin, 2024). Thus, by suppressing information transmission, weakening incentive expectations, and eroding trust, CPU dampens PEA and its translation into green consumption. Based on this reasoning, we propose the following hypothesis.

Hypothesis 4

Hypothesis 4. CPU hinders the ECT by reducing PEA.

2.2.4 Theoretical analytical framework

This study develops an integrated framework in which CPU affects ECT through a direct inhibitory effect (Hypothesis 1) and indirect supply (Hypothesis 2, Hypothesis 3) and demand-side (Hypothesis 4) mechanisms. The framework, combining the direct effect and the three mediating pathways, is illustrated in Figure 1.

Figure 1

Figure 1. Framework diagram.

3 Data and methodology

3.1 Model design

3.1.1 Baseline regression model

This study constructs a two-way fixed effects model incorporating both time and city dimensions to assess how CPU influences the urban energy consumption transition. The model is specified as follows:

$EC{T}_{it}={\beta }_0+{\beta }_{1}CP{U}_{it}+{\beta }_2{Z}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}(1)$

Where represents energy consumption transition in the city $i$ during the year $t$. $CP{U}_{it}$ captures the extent of climate policy uncertainty. $Z_{it}$ refers to a vector of relevant control variables. The term ${\epsilon }_{it}$ is the error. ${\beta }_0$ is the intercept. Coefficient ${\beta }_1$ reflects the marginal impact of CPU and ${\beta }_2$ indicates the effects of the control variables. To account for unobserved heterogeneity linked to both urban characteristics and macro-level temporal factors, the model incorporates ${\mu }_i$ and ${\delta }_t$ as city- and time-fixed impact, respectively. In Equation 1, the coefficient of central interest is ${\beta }_1$. A statistically significant negative value of ${\beta }_1$ would imply that elevated CPU suppresses the advancement of energy transition in urban areas.

3.1.2 Mediating effect model

To further investigate the mechanisms through which CPU affects the transition of energy consumption transition, according to the research Hypothesis 2–4, this paper explores three key pathways: (1) the effect of CPU on the supply side through weakening industrial chain resilience; (2) the influence of CPU on the supply side through suppressing green innovation disclosure; and (3) the impact of CPU on the demand side through reducing public environmental awareness.

The traditional three-step approach for testing mediation effects has been subject to criticism due to its inherent limitations. In particular, when the parameter estimates across the three regression equations are inconsistent—especially in the presence of endogeneity—the method may yield biased or misleading conclusions, weakening its inferential validity. To address these concerns, this study employs the two-step approach introduced by (Jiang, 2022), which offers improved robustness. Specifically, we first analyze the theoretical causal link between the mediating variable and the transition of energy consumption. We subsequently conducted an empirical test of how CPU influences the intermediary variable, thereby providing a more reliable identification of the mediation pathway. Based on these considerations, we construct a mediating effect model to empirically test the conduction mechanisms by which CPU influences energy consumption transition. The model is specified as follows:

$ME{D}_{it}={\beta }_0+{\beta }_{1}CP{U}_{it}+{\beta }_2{Z}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}(2)$

$EC{T}_{it}={\alpha }_0+{\alpha }_{1}CP{U}_{it}+{\alpha }_{2}ME{D}_{it}+{\alpha }_3{Z}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}(3)$

Here, $i$ and $t$ represents the cities and times of observation, respectively. $ME{D}_{it}$ of Equation 2 refers to the mediating factors, specifically Industrial Chain Resilience (ICR)、Green Innovation Disclosure (GID), and public environmental awareness (PEA). In Equation 3, the dependent variable is ECT, the mediating variables, and CPU serve as the core explanatory variables. ${\alpha }_0$ represent the constant term. ${\alpha }_1$, ${\alpha }_2$, ${\alpha }_3$ represents the coefficients of interest corresponding to both CPU and the mediating factors. $Z_{it}$ represents control variables. In addition, ${\mu }_i$ and ${\delta }_t$ to account for fixed effects for cities and years, ${\epsilon }_{it}$ captures the unexplained variance.

3.2 Variable descriptions

3.2.1 Dependent variable: ECT

The essence of ECT is a systemic process in which the terminal energy system dynamically evolves from “high-carbon fossil fuel dominance with inefficient use” to “low-carbon renewable energy priority with efficient allocation”. Its core lies in continuously optimizing the structure and efficiency of energy use, achieving a synergy between low carbonization and high efficiency, and thereby supporting the coordinated development of the economy and the environment. The systemic and dynamic nature of ECT makes it difficult to capture with a single indicator, such as the share of renewable energy or coal consumption. A single indicator often reflects only structural optimization while ignoring efficiency improvements, or captures short-term outcomes while missing long-term value, which can lead to biased assessments. Therefore, this study adopts the Energy System Performance Index developed by Shen et al. (2023) as the core proxy variable to measure ECT at the prefecture-level city scale. The index consists of two equally weighted dimensions: energy system structure and environmental sustainability (each accounting for 50%).

The energy system structure dimension includes four indicators: energy structure (share of coal in primary energy), power structure (share of local coal-fired power consumption), energy intensity (energy use per unit of GDP), and per capita energy consumption. These indicators directly capture the transition of energy consumption from high-carbon dependence to low-carbon substitution, while addressing both structural optimization and efficiency improvement. In practice, cities scoring higher on this dimension are usually characterized by a lower share of high-carbon energy, higher penetration of renewable energy, and lower energy consumption per unit of economic output. These features help to avoid “pseudo-transition” that focuses only on structural change while neglecting efficiency, aligning closely with the dual goal of both quantity and quality in ECT.

The environmental sustainability dimension includes two indicators: carbon emissions per unit of GDP and per capita carbon emissions. The former captures the ecological benefits of ECT by quantifying the environmental externalities of fossil energy and assessing its role in reducing the carbon intensity of economic growth. The latter reflects equity concerns, preventing biases that arise from focusing only on aggregate values while ignoring individual differences.

It should be noted that the original dataset of the Energy System Performance Index developed by Shen et al. (2023) covers the period 2003–2019. This study extends the data to 2023 using the same calculation method and applies a normalization process to convert all indicators into scores ranging from 0 to 100. As shown in Equation 4, the formulas for positive and negative indicators are presented below:

$\begin{array}{l}{x}^{\prime }=\frac{x-\min \left(x\right)}{\max \left(x\right)-\min \left(x\right)}\\ x^{\prime }=\frac{\max \left(x\right)-x}{\max \left(x\right)-\min \left(x\right)}\end{array}(4)$

After standardizing each indicator, the index is constructed based on a hierarchical aggregation from indicators to dimensions to the overall index, using an arithmetic mean with equal weights. The index incorporates both process-oriented and outcome-oriented measures, providing a comprehensive and accurate reflection of ECT at the prefecture-city level and offering a solid foundation for the baseline regression in this study.

3.2.2 Independent variables

The key independent variable in this study is city-level CPU, Ma et al. (2023) constructed CPU indices at the national, provincial, and municipal levels in China using the MacBERT deep learning model. This method offers stronger semantic understanding and better generalization capability than traditional text analysis strategies that rely on dictionaries or rule-based approaches. It can automatically extract features related to policy uncertainty from large-scale Chinese corpora, significantly enhancing the sensitivity and accuracy of the resulting index. The study selects six mainstream newspapers in China with high credibility as data sources, collecting approximately 1.75 million news articles. The study constructs a dynamic and regionally differentiated CPU index through a systematic text mining and modeling process. A manual verification mechanism is introduced to ensure the index’s credibility, involving multiple rounds of screening and revision of the model output. This process provides both the index’s technical rigor and quality control, thereby reinforcing its scientific validity and practical applicability. Based on this research, the present study adopts the prefecture-level CPU index developed as the primary indicator to measure climate policy uncertainty across Chinese cities, aiming to capture the uncertainty characteristics manifested in local governments’ climate policy formulation and implementation processes.

3.2.3 Mediating variables

3.2.3.1 Industrial chain resilience

To balance data availability and scientific rigor, this study follows Sun and Zhu (2023) and constructs a composite index of industrial chain resilience at the prefecture-city level. The index is built from two dimensions—resistance capacity and renewal capacity—using the entropy weight method. Resistance capacity is measured by the industrial diversification index (Indiv), which is calculated using the Herfindahl–Hirschman Index (HHI). A smaller HHI indicates lower industrial concentration and higher diversification, thereby enhancing the ability of the industrial chain to withstand external shocks and maintain stable operations. Renewal capacity (Renew) is measured by the number of invention patents granted in a city. A higher value reflects stronger capabilities in new product development and industrial upgrading, indicating a greater renewal capacity of the industrial chain for future development. The specific calculation formula for resistance capacity is as follows:

$Indiv=\frac{1}{HHI}=\frac{1}{{\sum }_{i=1}^n{S}_i^2}(5)$

In Equation 5, $S_i$ represents the share of industry $i$ in regional GDP.

3.2.3.2 Green innovation disclosure

Drawing upon the analytical frameworks of Xie et al. (2019) and Huang et al. (2025), and considering the specific characteristics of corporate information disclosure in China, this paper measures the level of corporate green innovation disclosure using a content analysis approach. The underlying logic is that firms voluntarily disclose information on green innovation through annual reports and Corporate Social Responsibility (CSR) reports. The breadth and depth of such disclosure reflect firms’ actual actions in green transition and their technological reserves, thereby providing signaling support and a basis for technology diffusion in the process of ECT.

Specifically, we examine the sections in annual and CSR reports of sample firms located in each city that relate to environmental protection, sustainable development, and technological innovation. The analysis covers five dimensions: improvements in resource and energy efficiency, pollution control and emission reduction, green product research and development, low-carbon management systems, and green innovation cooperation. Based on predefined keywords (e.g., energy-saving technologies, carbon capture, green products), we employ word frequency statistics and semantic intensity scoring to encode the text. An entropy-weighting method is then applied to construct a composite index of green innovation disclosure. At the city level, the index is obtained by calculating the revenue-weighted average of all firms’ green innovation disclosure scores within the city, which mitigates the potential bias from large firms dominating the results. The index ranges from 0 to 100, with higher values indicating greater breadth and depth of disclosure.

3.2.3.3 Public environmental awareness

Following the approaches of Zhou and Ding (2023) and Li and Jin (2024), this study constructs a city-level index of public environmental awareness based on the annual Baidu Search Index for environment-related keywords. Specifically, we select keywords such as “pollution,” “environmental protection,” “carbon neutrality,” “carbon peaking,” “carbon emissions,” and collect the annual search frequency data from both PC and mobile platforms across cities. The search data are log-transformed to capture the intensity of public attention to environmental issues.

This construction method is consistent with El Ouadghiri et al. (2021), who employed the Google Search Index to measure public environmental concern in the United States. As the dominant search engine in China, Baidu provides extensive data on user search behavior. By quantifying the frequency of keyword searches, the Baidu Search Index offers a comprehensive and dynamic measure of regional public concern for environmental issues, thereby exhibiting strong representativeness and sensitivity.

3.2.4 Control variables

Following established literature, this study controls for key urban-level factors across four dimensions: (1) Economic factors include economic development level (LNP: log GDP per capita) and trade openness (FDR: FDI-to-GDP ratio), which shape energy demand and structural upgrading; (2) Institutional factors incorporate government intervention intensity (GOV: fiscal expenditure share of GDP), reflecting policy guidance in transition processes; (3) Social factors comprise urbanization rate (Urban: urban population share) and population density (POP: log population per km²), capturing agglomeration effects on energy patterns; (4) Educational factors account for educational expenditure (EDE: share of GDP) and human capital structure (EBS: higher education proportion), supporting transition through talent development.

3.2.5 Data sources

This paper explores how the CPU influences ECT. Given data availability, the research sample comprises 277 Chinese cities over the period 2007–2023. ECT is sourced from Shen et al. (2023), while city-level CPU measurements are obtained from Ma et al. (2023). Data for the remaining variables are primarily drawn from the CSMAR database, the China Statistical Yearbook, and the China City Statistical Yearbook. To address potential heteroscedasticity, a logarithmic transformation is applied to all variables. For years, with incomplete data, linear interpolation has been used to fill in gaps. Summary statistics for all variables are reported in Table 1.

Table 1

Table 1. Description statistics.

4 Estimation results and analysis

4.1 Benchmark regression results

Building upon the preceding theoretical analysis and model specification, this section empirically investigates the quantitative influence of CPU on ECT. The baseline regression results are presented in Table 2. Columns (1) and (2) report the regression results without and with control variables. As shown, there exists a significantly negative correlation between CPU and ECT. Specifically, after controlling for economic, institutional, social, and educational factors, the coefficient of CPU in Column (2) is −0.346 and statistically significant at the 1% level. This indicates that, ceteris paribus, a 1% increase in climate policy uncertainty is associated with a 0.346% slowdown in urban energy consumption transition. These findings suggest that the CPU hinders the progress of ECT in cities, thereby supporting Hypothesis 1.

Table 2

Table 2. Benchmark regression results.

4.2 Endogenous discussions

In empirical analysis, measurement errors in the explanatory variable and the omission of important control variables may cause endogeneity issues. To mitigate potential estimation bias, this study employs multiple strategies, including the instrumental variable (IV) approach, propensity score matching (PSM), and the inclusion of city-specific characteristics and environmental policy variables, to identify and address potential endogeneity concerns.

4.2.1 Instrumental variable approach

Although the CPU exhibits a degree of exogeneity, government behavior in the actual policy-making process may still be influenced by various urban-level factors, including economic, social, and environmental conditions. Therefore, potential endogeneity cannot be ruled out. To address this issue, we adopt the 2SLS estimation approach, using as an instrumental variable (IV) the absolute difference between the current year’s average land surface temperature and its 3-year moving average as an instrumental variable (IV). The rationale for selecting this IV is as follows: on the one hand, abnormal fluctuations in land surface temperature reflect the intensity and frequency of climate change, which may prompt local governments to adjust existing policy agendas, thereby increasing climate policy uncertainty. On the other hand, natural temperature variations do not directly affect the evolution of urban energy consumption structures, which are primarily shaped by institutional design, industrial composition, and social preferences. Hence, the chosen variable fulfills the relevance and exogeneity criteria necessary for a valid instrumental variable.

The 2SLS estimation outcomes are reported in Table 3. As shown in Column (1), a greater discrepancy between a city’s current land surface temperature and its preceding 3-year average is significantly associated with higher CPU levels, with the coefficient being statistically significant at the 1% level. Additionally, the identification and weak instrument tests in the first stage reject the null hypotheses of under-identification and instrument weakness, respectively, indicating that the selected IV demonstrates strong explanatory power and statistical validity. In addition, we employ extreme heat and extreme cold events as instrumental variables to conduct robustness checks. The regression results reported in Supplementary Table SA1 show that The IV estimates remain consistent with the baseline results in both sign and significance, indicating that the negative effect of CPU is unlikely to be driven by potential endogeneity concerns and that the core findings are robust.

Table 3

Table 3. Two-stage least squares estimation results.

4.2.2 Propensity score matching method

Given the substantial heterogeneity across cities in economic development, policy implementation capacity, and industrial structure, such differences may influence the CPU level, leading to sample selection bias and undermining the credibility of the estimation results. This study utilizes the PSM technique to adjust the sample to mitigate potential self-selection bias and enhance the validity of causal inference. Specifically, all sample cities are categorized into treatment and control groups depending on whether their CPU index is above or below the national average. Municipalities with CPU values above the mean are assigned to the treatment group (1), while those below the average fall into the control group (0). Using all control variables defined in previous sections as covariates, this study applies the nearest neighbor matching approach to pair cities from the two groups, ensuring comparability in observable characteristics. Subsequently, regression analysis is conducted based on the matched sample. The results, presented in Table 4, indicate that the influence of CPU on ECT remains significantly negative at the 1% level, thereby confirming the stability of the baseline estimation outcomes.

Table 4

Table 4. Results after propensity score matching.

4.2.3 Controlling for city-specific characteristics

This study incorporates city-specific characteristics into the regression model as additional control variables to mitigate estimation bias arising from potential confounding variables. These characteristics may simultaneously influence CPU and energy ECT. Specifically, three types of city attributes are considered: First, provincial capital cities often enjoy institutional advantages in terms of political resource allocation, fiscal support, and policy implementation, which may significantly affect the stability and formulation pace of climate policies, thereby influencing CPU. Second, special economic zones (SEZs), due to their higher degree of policy autonomy and institutional flexibility, may exhibit greater volatility in climate-related policymaking, resulting in higher levels of policy uncertainty. Third, cities located within acid rain control zones or sulfur dioxide emission control areas—so-called “dual control zones” (Dzone)—are subject to stricter and more standardized environmental regulatory frameworks, which typically result in more stable and consistent policy implementation, thereby affecting the level of CPU.

Based on the above considerations, three dummy variables are constructed: Capital, indicating whether a city is a provincial capital; Ezone, whether a city is a special economic zone; and Dzone, whether a town falls within dual control zones. Each dummy variable interacts with a time trend and is included in the baseline regression model. As presented in Table 5, even after adjusting for city-specific characteristics, the influence of CPU on energy consumption transition remains significantly negative and statistically robust, further confirming the validity of the baseline findings. It is worth noting that the coefficient of Dzone is smaller than the average treatment effect (ATE = −0.346), indicating a weaker negative impact of CPU on ECT in dual-control zones. This is consistent with China’s environmental governance context, where stricter and more standardized regulatory frameworks in these zones lead to more stable policy implementation and thus lower levels of policy uncertainty, reducing CPU’s disruptive effect on ECT.

Table 5

Table 5. Results after controlling for City-Specific characteristics.

4.2.4 Controlling for environmental policy variables

Given that specific environmental policies may simultaneously affect both the level of CPU and the process of ECT in cities, failure to control for these factors could result in biased regression estimates. This study incorporates environmental policy variables into the empirical model to mitigate potential confounding effects. On the one hand, ecological regulations directly influence the adjustment of energy structures and consumption patterns, thereby significantly influencing ECT (Shen et al., 2024). On the other hand, discrepancies between local environmental and climate policies—in either goal setting or implementation approaches—may exacerbate policy inconsistency and coordination problems, thus increasing climate policy uncertainty. Following the framework proposed by Zeng et al. (2025), environmental regulation is divided into two categories: formal and informal. For formal environmental regulation, and considering the availability of city-level data in China, five pollution control indicators are selected: the treatment of industrial wastewater, sulfur dioxide, soot, solid waste, and discarded raw materials. These indicators are integrated into a comprehensive measure of formal environmental regulation via the entropy weighting method, with monetary values adjusted using the provincial-level GDP deflator (base year = 2006). For informal environmental regulation, three indicators are employed: per capita income, average educational attainment, and population density. These are also synthesized into an index using the entropy weighting method. Both indices are constructed based on the dataset (Zeng et al., 2025) developed and included in the baseline regression model as control variables. As reported in Table 6, after controlling for environmental policy factors, the coefficient of CPU remains negative and statistically significant. This result further confirms the suppressive effect of CPU on ECT. It also provides additional evidence supporting the robustness of our primary outcomes.

Table 6

Table 6. Results after controlling for environmental policies.

4.3 Robustness checks

To ensure the robustness and dependability of our conclusions, we implement a comprehensive set of robustness tests from multiple perspectives. Specifically, we perform regressions using alternative dependent variables, introduce lagged terms of the explanatory variable, incorporate interaction fixed effects, adjust the clustering level of standard errors, and exclude special sample observations. These robustness checks, conducted through cross-validation using diverse approaches, aim to reinforce the consistency and credibility of the main findings.

4.3.1 Replacing the dependent variable and lagging the explanatory variable

First, recognizing that energy consumption transition is reflected in current transition outcomes and rooted in the institutional and infrastructural groundwork laid in earlier periods, we replace the original dependent variable—the Energy Consumption Transition Index (ECT)—with the Transition Readiness Index (TR). The TR index captures the institutional environment and supporting capacity that underpin a city’s readiness for energy system transformation. As shown in Table 7, whether control variables are omitted in column 1 or accounted for in column 2, CPU continues to exhibit a significantly adverse effect on TR, with significance at the 1% level. This outcome suggests that the suppressive impact of CPU on energy transition remains robust even when the dependent variable is altered. Second, considering that the formulation and implementation of climate policies typically entail time lags, and their influence on energy systems often materializes with some delay, we further incorporate a one-period CPU lag into the benchmark model. This approach also helps to mitigate potential contemporaneous confounding effects. If the lagged CPU variable still demonstrates a significant influence, it would indicate that the impact of CPU is not a short-term fluctuation or the result of random shocks. Additionally, to some extent, lagged explanatory variables can function as a special instrumental variable, helping alleviate potential endogeneity concerns.

Table 7

Table 7. Robustness test results.

The estimation outcomes are shown in columns (3) and (4) of Table 7, corresponding to the models without and with control variables, respectively. The findings reveal that the lagged CPU variable maintains a significantly adverse effect on ECT, with significance at the 1% level. This further reinforces the robustness of our core conclusions.

4.3.2 Additional robustness tests

To further validate the robustness of the benchmark regression results, we conduct additional robustness checks by incorporating interaction fixed effects, adjusting the clustering level of standard errors, and excluding special samples. The corresponding results are reported in Table 9. First, considering that provincial and temporal factors may jointly influence climate policy formulation and the energy consumption transition process, we introduce two-way interaction fixed effects between provinces and years to account for unobservable region-time-specific factors. As shown in column (1) of Table 8, the estimated coefficient of CPU remains significantly negative after the inclusion of interaction fixed effects, suggesting that the negative influence of CPU on ECT is unlikely to stem from unobserved regional or temporal heterogeneity. Second, we re-cluster standard errors at different administrative levels to relax the assumption of independently and identically distributed error terms and improve estimation accuracy. Specifically, standard errors are clustered at the municipal level (column 2) and the provincial level (column 3), respectively. As shown in Table 8, the CPU coefficients remain statistically significant at the 1% level across different clustering dimensions, suggesting that the estimated relationship is robust to changes in the clustering strategy. Lastly, recognizing that China’s four centrally administered municipalities (Beijing, Shanghai, Tianjin, and Chongqing) differ significantly from regular prefecture-level cities in terms of governance capacity, data quality, and climate policy enforcement, we exclude these cities and re-estimate the model to avoid potential distortions. As shown in column (4) of Table 8, the adverse effect of CPU on ECT persists with statistical significance even after removing these special samples. These robustness checks collectively reinforce our core finding: CPU significantly hinders urban energy consumption systems’ structural transformation and upgrading.

Table 8

Table 8. Additional robustness test results.

4.4 Heterogeneity analysis

4.4.1 Resource endowment heterogeneity

One of the key channels through which the CPU affects ECT lies in its potential to suppress the shift toward a low-carbon energy structure. Given the substantial differences between resource-based and non-resource-based cities in terms of energy composition, industrial foundation, and path dependency (Li et al., 2024), it is plausible that these two types of cities respond differently to CPU. Following the classification outlined in the Sustainable Development Plan for Resource-Based Cities (2013–2020) issued by the State Council of China, we divide the sample into resource-based and non-resource-based cities to conduct a heterogeneity analysis. As evidenced in columns (1) and (2) of Table 9. The coefficient on CPU is significantly more negative for resource-based cities, indicating that such cities face greater barriers to transformation under policy uncertainty, and resource-based cities are more vulnerable to CPU and experience more substantial adverse effects during the energy consumption transition process.

Table 9

Table 9. Resource-based city and Marketization degree heterogeneity test results.

4.4.2 Degree of marketization

Variations in marketization levels shape local governments’ capacity to implement policies and influence how firms respond to and cope with policy uncertainty. We adopt the China Marketization Index to categorize the sample into high- and low-marketization cities and perform separate regressions to investigate the heterogeneous impact of CPU. The results (see Table 9, columns 3 and 4) suggest that CPU exerts a more severe adverse effect on ECT, which is more pronounced in highly marketized urban areas.

Two possible mechanisms may drive this result. First, firms in highly marketized regions generally face more intense market competition. To control costs and hedge against the risks arising from CPU, they may reduce green investments and technological innovation, thereby slowing the pace of energy transition. Second, governments in highly marketized areas tend to rely more on market mechanisms for resource allocation. In such contexts, environmental regulations may be less strictly enforced, making it easier for firms to adopt carbon-intensive production practices, exacerbating the challenges of optimizing the energy structure.

4.4.3 Regional heterogeneity

Considering the significant regional variation in economic development, energy structure, industrial base, and policy implementation capacity across China, the influence of CPU on ECT may also vary by region. Based on the National Bureau of Statistics’ classification, we categorize cities into eastern, central, and western areas and perform region-specific regression.

As presented in Table 10, CPU exerts a significantly negative influence on ECT primarily in the eastern region, with no statistically meaningful effects observed in the central and western regions. This regional variation may reflect differences in policy responsiveness. In the eastern region, which is characterized by advanced economic development, active market dynamics, and stricter environmental constraints, the disruptive effects of CPU on both governmental and corporate behavior are more direct and pronounced. In contrast, the western region exhibits higher dependence on natural resources and lower sensitivity to policy signals, leading to a more delayed or muted response to climate policy uncertainty.

Table 10

Table 10. Regional heterogeneity test results.

4.5 Mechanism analysis

The preceding empirical analyses have demonstrated that CPU exerts a significant suppressive influence on the ECT. Building on this finding, we further explore the underlying mechanisms by which the CPU influences the transition process. To this end, we adopt a two-step mediation analysis approach to systematically examine three potential transmission channels: (1) industrial chain resilience, (2) green innovation disclosure, and (3) public environmental awareness. These dimensions capture supply-side and demand-side dynamics that may mediate the relationship between CPU and ECT. The detailed empirical results are presented in Table 11.

Table 11

Table 11. Mechanism test results.

According to the regression results in column (1) of Table 11, the coefficient of CPU on IRC is significantly negative at the 1% level, indicating that an increase in CPU markedly undermines urban industrial chain resilience, which is consistent with Hypothesis 2. The underlying logic is that intensified policy fluctuations strengthen firms’ risk-averse tendencies, prompting them to maintain reliance on high-carbon industries to secure short-term returns. This not only crowds out the development space of low-carbon sectors such as renewable energy but also reinforces the path dependence of high-carbon industries on traditional energy and technologies. As a result, the industrial chain finds it difficult to achieve structural optimization through “substitution of high-carbon links with low-carbon links and cultivation of low-carbon segments”, thereby reducing its capacity to withstand external shocks. Ultimately, CPU weakens industrial chain resilience and impedes the supply-side foundation for the transition of energy consumption.

The regression results in column (2) of Table 11 show that CPU has a significantly negative effect on GID, suggesting that CPU impedes the process of urban ECT by discouraging firms’ disclosure of green innovation, thereby supporting Hypothesis 3. The mechanism lies in the fact that heightened policy uncertainty reinforces firms’ short-sighted behavior. Facing unknown risks such as policy adjustments, firms find it difficult to form stable expectations regarding the long-term returns of green technology R&D. To avoid potential sunk costs associated with early-stage investment, they tend to favor short-term, low-cost, and high-emission production modes, while reducing investment in energy-saving and emission-reduction technologies. Such short-sighted behavior directly weakens firms’ incentives to disclose green innovation outcomes: on the one hand, the lack of substantial green technological progress limits the content available for disclosure; on the other, firms are reluctant to disclose for fear of additional compliance costs and market pressure. Consequently, the CPU disrupts the technological support pathway for ECT.

The regression results in column (3) of Table 11 indicate that CPU has a significantly negative impact on PEA, thereby validating Hypothesis 4. The underlying mechanism can be explained as follows: frequent fluctuations in climate policies reduce the efficiency of green information transmission, making it difficult for the public to capture stable low-carbon signals. At the same time, such instability undermines the credibility of climate policies, weakening public recognition and acceptance of low-carbon policies. This information disorder and lack of trust directly dampen the public’s green behavioral responses, discouraging them from actively engaging in green consumption practices, such as adopting new energy products or reducing high-carbon energy use. As a result, weakened environmental awareness and actions on the demand side hinder the progress of ECT.

To ensure the robustness of the identified mediation mechanisms, this study conducts two sets of robustness checks. First, we perform alternative variable tests. Given that GID and PEA may be subject to measurement errors due to data sources, media attention, or differences in city-level firm composition, we use the number of green patents as an alternative measure for GID and residents’ green electricity consumption as an alternative measure for PEA. The regression results reported in Supplementary Table SA2 show that CPU remains significantly negative for both alternative indicators, demonstrating the stability of our mediation variable measurements. Second, we test the potential “reverse logic.” To examine whether CPU may, in a precautionary manner, motivate firms and the public to enhance IRC, increase GID, or improve PEA, we conduct three empirical analyses: (1) Regression tests with mediation variables as dependent variables. We regress IRC, GID, and PEA on CPU. If precautionary motives existed, CPU should show a positive effect. Supplementary Table SA3 indicates that the coefficients of CPU are significantly negative across all models, even after adding control variables, providing no support for the reverse logic. (2) Lagged CPU tests. Considering that behavioral adjustments may take time, we include one-period lagged CPU in the regressions. If reverse logic holds, the lagged term should be positive. Supplementary Table SA4 shows that lagged CPU remains significantly negative, rejecting the possibility of delayed precautionary responses. (3) Panel threshold model tests. Using CPU as the threshold variable, we construct a panel threshold model to detect potential structural changes across different uncertainty regimes. Supplementary Table SA5 shows that neither single-threshold nor double-threshold specifications are statistically significant, and CPU exhibits no sign reversal across uncertainty intervals, indicating no evidence for the reverse logic.

In summary, CPU affects the ECT through three major mechanisms: on the supply side, it suppresses the capacity for renewable energy development by constraining industrial structure optimization and green technology innovation; on the demand side, it reduces renewable energy consumption by weakening public environmental awareness. Together, these supply- and demand-side effects constitute the intrinsic mechanism through which CPU exerts a negative influence on the ECT.

5 Further analysis: spatial spillover effects

China’s regional energy consumption structure shows significant heterogeneity. The eastern region, supported by technological and industrial advantages, has achieved relatively rapid energy structure optimization, with the share of non-fossil energy consumption notably higher than the national average. In contrast, western regions such as Xinjiang and Inner Mongolia, constrained by resource endowments and industrial structures, remain highly dependent on fossil fuels such as coal and electricity, with coal-fired power consumption exceeding 70% in some areas (Sahu, 2018). This regional divergence not only reinforces the path dependence of high-carbon energy consumption but also leads to a “spatially unbalanced” pattern in the ECT (Zhang Z. et al., 2025). Against this background, the ECT across cities may exhibit spatial correlation. Ignoring this feature may result in estimation bias due to omitted spatially related variables. However, spatial heterogeneity alone does not fully capture the dynamics of ECT. Cities are embedded in interconnected regional energy systems, where energy production, consumption, and industrial supply chains often extend across administrative boundaries. Clean energy development in one region can affect neighboring regions through power transmission networks, technology diffusion, factor mobility, and policy imitation effects. Moreover, climate and environmental regulations frequently generate cross-regional spillovers, making one city’s transition progress non-independent of nearby cities.

Given these interaction channels, ECT across cities is likely to exhibit spatial correlation. Ignoring this spatial dependence may lead to biased estimates due to omitted spatially linked factors. Therefore, this paper employs spatial econometric methods to systematically examine the spatial spillover effects of CPU on the ECT.

5.1 Construction of the spatial weight matrix

Considering that geographical distance is a key factor influencing policy transmission and factor flows between cities, this study follows mainstream approaches to construct an inverse-distance spatial weight matrix to characterize the spatial correlation strength between cities. The specific form is as follows:

$W_{ij}=\left\{\begin{array}{l}\frac{1}{d_{ij}},\left(i\ne j\right)\\ 0,\left(i=j\right)\end{array}\right.(6)$

In Equation 6, $d_{ij}$ denotes the geographical straight-line distance between city $i$ and city $j$. The inverse form not only reflects the distance decay effect (where spatial influence diminishes as distance increases), but also amplifies the correlation weight of nearby cities, making it more consistent with the practical characteristics of regional policy transmission.

5.2 Selection of spatial econometric models

The heterogeneous responses across regions and city types indicate that CPU’s effects are not spatially independent. The clustering of policy-sensitive cities in the east and resource-dependent cities in the west suggests potential spillovers, whereby CPU in one city may affect neighboring cities. Hence, spatial econometric models are employed to identify these spillover effects and reveal CPU’s transmission mechanisms in the ECT process.

Spatial econometric analysis commonly employs the Spatial Autoregressive Model (SAR), the Spatial Error Model (SEM), and the Spatial Durbin Model (SDM). Their general forms are as follows:

$EC{T}_{it}=\rho WEC{T}_{it}+{\beta }_{1}CP{U}_{it}+{\beta }_2{Z}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}(7)$

$EC{T}_{it}={\beta }_{1}CP{U}_{it}+{\beta }_2{Z}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it},{\epsilon }_{it}=\lambda W_{ij}{\epsilon }_{jt}+{\delta }_{it}(8)$

$EC{T}_{it}=\rho W_{ij}EC{T}_{jt}+\theta W_{ij}CP{U}_{jt}+{\beta }_{1}CP{U}_{it}+{\beta }_2{Z}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}(9)$

In Equations 7–9, theThe dependent variable is ECT. The coefficient of the spatial lag term of the dependent variable $\rho$ measures the spatial dependence of ECT. The parameter $\lambda$ denotes the spatial dependence of the error term, while $\theta$ captures the spatial spillover effect of CPU on ECT.

To determine the optimal model specification, this paper follows the spatial econometric model selection procedure proposed by Elhorst (2014), with results summarized in Table 12. First, both LM and robust LM tests are significant at the 1% level, rejecting the null of no spatial dependence and supporting the use of a spatial econometric framework. Second, LR tests indicate that individual and time fixed effects are jointly significant, validating the adoption of a two-way fixed effects specification to account for cross-sectional and temporal heterogeneity. Furthermore, Wald and LR tests both reject the null that the SDM can be reduced to either the SAR or SEM at the 1% level, confirming the SDM as the most appropriate model. Finally, the Hausman test is significant at the 1% level, further supporting the fixed-effects specification over the random-effects alternative.

Table 12

Table 12. Spatial econometric model selection.

Based on the diagnostic tests, the two-way fixed effects SDM is selected for empirical estimation, and the results are presented in Table 13. Two main findings emerge. (1) Local inhibitory effect: The direct coefficient of CPU is −0.368, significant at the 1% level, indicating that, after controlling for other factors, CPU significantly suppresses local ECT. Specifically, a one-unit increase in CPU reduces the ECT index by an average of 0.368 units, consistent with the baseline results. (2) Spatial spillover effect: The coefficient of the spatial lag of CPU is −1.166, also significant at the 1% level, suggesting that higher CPU not only constrains local ECT but also exerts a substantial negative spillover effect on neighboring cities through spatial linkages. The magnitude of this spillover (1.166) exceeds the local effect (0.368), underscoring the necessity of regional coordination in mitigating CPU impacts.

Table 13

Table 13. Regression results of the SDM.

To further elucidate the spatial spillover effects of CPU on ECT, the impacts are decomposed into direct, indirect, and total effects based on the SDM estimation results. As reported in Table 14, the direct effect of CPU is −0.365 and significant at the 1% level, representing its average impact on local ECT. This estimate is nearly identical to the baseline coefficient (−0.368), reaffirming the robustness of the local inhibitory effect. The indirect effect is −2.215 and highly significant, indicating a pronounced negative spillover of CPU on neighboring cities through spatial linkages. Its magnitude is roughly six times that of the direct effect, suggesting that the CPU’s adverse influence is primarily transmitted across regions. The total effect is −1.850 and significant at the 1% level, reflecting the overall suppressive impact of CPU on both local and neighboring ECT, underscoring its systemic negative influence within the regional energy transition network.

Table 14

Table 14. Decomposition of spatial effects.

These findings suggest that CPU not only significantly inhibits local ECT but also generates even stronger negative spillover effects on neighboring cities, with cross-regional transmission serving as an important complement to the local effect. ECT is thus not an isolated urban process but a systemic transformation shaped by spatial interdependencies among regions. The inhibitory effect of CPU on ECT tends to transmit across regions through a “local inhibition–neighboring drag” mechanism, undermining the overall coordination of regional energy transitions. Hence, climate policy design should be optimized through enhanced regional coordination and integrated governance frameworks.

6 Conclusion

6.1 Conclusion and policy implications

The global climate imperative requires large developing economies to identify viable pathways for decoupling economic growth from carbon emissions. As the world’s largest developing economy and a major emitter, China faces this challenge with particular urgency, making its “Dual Carbon” goals a high-profile test of ECT. This study moves beyond documenting China’s progress to examine a central paradox in its ECT: climate policies intended to promote ECT can, through the CPU they generate, introduce systemic risks that impede it. By analyzing the impact, mechanisms, and spatial spillovers of CPU across 277 Chinese cities, this research offers critical insights into managing complex ECT processes worldwide.

Our findings show that the inhibitory effect of CPU is not uniform but operates through a dual-channel mechanism and exhibits distinct spatial patterns. First, CPU significantly constrains ECT, a result robust to endogeneity and multiple tests. Second, the effect operates on both supply and demand sides: it weakens industrial chain resilience and suppresses green innovation disclosure, while reducing public environmental awareness. Third, the impact is heterogeneous, being stronger in resource-based cities, highly market-oriented cities, and the economically advanced eastern region. Fourth, the CPU generates negative spatial spillovers, suppressing ECT in neighboring cities and creating regional drag, highlighting the limitations of isolated local policies.

For China and other major emerging economies, these findings imply a four-pronged strategy to future-proof their ECT. First, from uncertainty to credibility: institutionalize a stable climate policy pathway. This involves embedding the “Dual Carbon” goals within phased transition roadmaps, establishing cross-ministerial coordination, and leveraging real-time data monitoring to ensure policy predictability. Second, from uniformity to precision: tailor transition strategies to local contexts. Policy must be tailored. Resource-based cities require dedicated support for industrial upgrading and renewable infrastructure; market-oriented cities should use carbon markets to price and hedge CPU risks; financial hubs must channel capital via green credit and bonds, promoting cross-regional technology sharing. Third, from fragmentation to integration: govern regional commons. Spatial spillovers of CPU necessitate cooperative, multi-jurisdictional governance. Establishing regional climate governance committees and creating integrated markets for green factors are essential to transform a zero-sum dynamic into collective action. Fourth, from single-sided to synergistic: activate both supply and demand. Supply-side measures should accelerate green technology R&D and streamline renewable project deployment, while demand-side initiatives should foster low-carbon lifestyles through public engagement and innovative instruments such as “green credit–carbon account” systems.

In conclusion, China’s experience shows that the success of a national ECT depends not only on deploying appropriate technologies but also on ensuring a stable and coherent policy environment. Globally, China’s encounter with the CPU paradox highlights that effective climate transitions require governance capable of anticipating and mitigating the risks generated by policy itself, balancing climate action with economic development. It is in this nuanced governance challenge that China’s local experience achieves its global significance.

6.2 Research limitations and future expansion

While this study provides a systematic analysis of CPU’s role in the ECT, several limitations should be acknowledged and offer avenues for future research. First, the micro-foundations of transmission mechanisms remain underexplored. Although supply- and demand-side pathways are verified at the macro level, micro-level dynamics, including firm and household heterogeneity, deserve further investigation using firm-level production data and detailed household surveys. Second, the quantification of spatial transmission pathways. While the spatial Durbin model confirms the existence of negative spillover effects, it does not quantify the specific channels through which CPU propagates across regions. Third, expansion to an international comparative perspective. The current findings are rooted in the Chinese context. A promising future direction is to incorporate city-level data from other major economies to conduct a comparative analysis. Examining how different national institutional contexts, market structures, and governance models shape the impact of CPU on ECTs would yield valuable insights for global climate governance and international policy learning.

Data availability statement

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

Author contributions

YW: Conceptualization, Data curation, Formal Analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review and editing. ZX: Conceptualization, Formal Analysis, Methodology, Validation, Visualization, Writing – review and editing. LS: Conceptualization, Investigation, Project administration, Validation, Writing – review and editing.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvs.2025.1744044/full#supplementary-material

Footnotes

¹https://www.caixin.com/2021-09-27/101778577.html

References

Acemoglu, D., Akcigit, U., and Kerr, W. (2016). Networks and the macroeconomy: an empirical exploration. Nber Macroecon. Annu. 30 (1), 273–335. doi:10.1086/685961

CrossRef Full Text | Google Scholar

Ajzen, I. (1991). The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50 (2), 179–211. doi:10.1016/0749-5978(91)90020-T

CrossRef Full Text | Google Scholar

Akerlof, G. A. (1970). The market for “lemons”: quality uncertainty and the market mechanism. Q. J. Econ. 84 (3), 488–500. doi:10.2307/1879431

CrossRef Full Text | Google Scholar

Badia, M., Duro, M., Jorgensen, B. N., and Ormazabal, G. (2021). Disclosure regulation and competitive interactions: evidence from the oil and gas industry. Account. Rev. 96 (5), 1–29. doi:10.2308/tar-2018-0436

CrossRef Full Text | Google Scholar

Chen, H., Liu, K., Shi, T., and Wang, L. (2022). Coal consumption and economic growth: a chinese city-level study. Energy Econ. 109, 105940. doi:10.1016/j.eneco.2022.105940

CrossRef Full Text | Google Scholar

Chowdhury, M. M. H., and Quaddus, M. (2017). Supply chain resilience: conceptualization and scale development using dynamic capability theory. Int. J. Prod. Econ. 188, 185–204. doi:10.1016/j.ijpe.2017.03.020

CrossRef Full Text | Google Scholar

Ding, Z., Jiang, X., Liu, Z., Long, R., Xu, Z., and Cao, Q. (2018). Factors affecting low-carbon consumption behavior of urban residents: a comprehensive review. Resour. Conservation Recycl. 132, 3–15. doi:10.1016/j.resconrec.2018.01.013

CrossRef Full Text | Google Scholar

Dohale, V., Verma, P., Gunasekaran, A., and Ambilkar, P. (2023). Covid-19 and supply chain risk mitigation: a case study from India. Int. J. Logist. Manag. 34 (2), 417–442. doi:10.1108/ijlm-04-2021-0197

CrossRef Full Text | Google Scholar

Dong, X., Zhang, B., Wang, B., and Wang, Z. (2020). Urban households’ purchase intentions for pure electric vehicles under subsidy contexts in China: do cost factors matter? Transp. Res. Part a Policy Pract. 135, 183–197. doi:10.1016/j.tra.2020.03.012

CrossRef Full Text | Google Scholar

El Ouadghiri, I., Guesmi, K., Peillex, J., and Ziegler, A. (2021). Public attention to environmental issues and stock market returns. Ecol. Econ. 180, 106836. doi:10.1016/j.ecolecon.2020.106836

CrossRef Full Text | Google Scholar

Elhorst, J. P. (2014). Spatial Econometrics: from cross-sectional data to spatial panels. Heidelberg: Springer. 479, 480. doi:10.1007/978-3-642-40340-8

CrossRef Full Text | Google Scholar

Fan, Z., and Ma, R. (2024). in Corporate environmental information disclosure and green innovation level: evidence from China. Editors J. Xu, N. A. Binti Ismail, S. Dabo-Niang, M. H. Ali Hassan, and A. Hajiyev (Singapore: Springer Nature Singapore), 1300–1313.

CrossRef Full Text | Google Scholar

Gao, X., Luo, X., and Liu, X. (2025). Thresholds to rural energy transition: the multistage effects of urbanization on the rural household energy consumption structure. Energy 319, 135121. doi:10.1016/j.energy.2025.135121

CrossRef Full Text | Google Scholar

Hao, X., Li, Y., and Wu, H. (2023). Ways to improve the efficiency of clean energy utilization: does digitalization matter? Energy Strategy Rev. 50, 101257. doi:10.1016/j.esr.2023.101257

CrossRef Full Text | Google Scholar

Healy, P. M., and Palepu, K. G. (2001). Information asymmetry, corporate disclosure, and the capital markets: a review of the empirical disclosure literature. J. Account. Econ. 31 (1), 405–440. doi:10.1016/S0165-4101(01)00018-0

CrossRef Full Text | Google Scholar

Hou, X., He, Y., Liu, X., Li, D., and Zhang, S. (2025). Market segmentation and urban energy consumption: an analysis of threshold effects from an agglomeration perspective. Energy Econ. 147, 108517. doi:10.1016/j.eneco.2025.108517

CrossRef Full Text | Google Scholar

Huang, K., Chi, J., Liao, J., and Yuen, M. K. (2025). In the heat of the moment, secrets will out: oil price uncertainty and firm green innovation disclosure. Int. Rev. Econ. and Finance 98, 103934. doi:10.1016/j.iref.2025.103934

CrossRef Full Text | Google Scholar

Iqbal, N., Bouri, E., Shahzad, S. J. H., and Alsagr, N. (2024). Asymmetric impacts of chinese climate policy uncertainty on chinese asset prices. Energy Econ. 133, 107518. doi:10.1016/j.eneco.2024.107518

CrossRef Full Text | Google Scholar

Jiang, T., and Luo, Z. B. (2022). Mediating and moderating effects in empirical studies of causal inference. China Ind. Econ. 23 (5), 100–120. doi:10.3892/ol.2022.13220

PubMed Abstract | CrossRef Full Text | Google Scholar

Kahneman, D., and Tversky, A. (1979). Prospect theory: an analysis of decision under risk. Econometrica 47 (2), 263–291. doi:10.2307/1914185

CrossRef Full Text | Google Scholar

Katsaliaki, K., Galetsi, P., and Kumar, S. (2022). Supply chain disruptions and resilience: a major review and future research agenda. Ann. Operations Res. 319 (1), 965–1002. doi:10.1007/s10479-020-03912-1

PubMed Abstract | CrossRef Full Text | Google Scholar

Li, Z., and Jin, B. (2024). A breath of fresh air: coal power plant closures and health in China. Energy Econ. 129, 107235. doi:10.1016/j.eneco.2023.107235

CrossRef Full Text | Google Scholar

Li, X., Mao, H., and Fang, L. (2024). The impact of rural human capital on household energy consumption structure: evidence from shaanxi province in China. Sustain. Futur. 8, 100301. doi:10.1016/j.sftr.2024.100301

CrossRef Full Text | Google Scholar

Lin, B., and Zhao, H. (2023). Tracking policy uncertainty under climate change. Resour. Policy 83, 103699. doi:10.1016/j.resourpol.2023.103699

CrossRef Full Text | Google Scholar

Ma, Y., Liu, Z., Ma, D., Zhai, P., Guo, K., Zhang, D., et al. (2023). A news-based climate policy uncertainty index for China. Sci. Data 10 (1), 881. doi:10.1038/s41597-023-02817-5

PubMed Abstract | CrossRef Full Text | Google Scholar

Magnan, A. K., Pörtner, H., Duvat, V. K. E., Garschagen, M., Guinder, V. A., Zommers, Z., et al. (2021). Estimating the global risk of anthropogenic climate change. Nat. Clim. Change 11 (10), 879–885. doi:10.1038/s41558-021-01156-w

CrossRef Full Text | Google Scholar

Markard, J. (2018). The next phase of the energy transition and its implications for research and policy. Nat. Energy 3 (8), 628–633. doi:10.1038/s41560-018-0171-7

CrossRef Full Text | Google Scholar

Mercure, J. F., Salas, P., Vercoulen, P., Semieniuk, G., Lam, A., Pollitt, H., et al. (2021). Reframing incentives for climate policy action. Nat. Energy 6 (12), 1133–1143. doi:10.1038/s41560-021-00934-2

CrossRef Full Text | Google Scholar

Morão, H. (2025). Uncertainty in climate policy and energy industry. Energy 328, 136013. doi:10.1016/j.energy.2025.136013

CrossRef Full Text | Google Scholar

Muth, J. F. (1961). Rational expectations and the theory of price movements. Econometrica 29 (29), 315–335. doi:10.2307/1909635

CrossRef Full Text | Google Scholar

Niu, S., Zhang, J., Luo, R., and Feng, Y. (2023). How does climate policy uncertainty affect green technology innovation at the corporate level? New evidence from China. Environ. Res. 237, 117003. doi:10.1016/j.envres.2023.117003

PubMed Abstract | CrossRef Full Text | Google Scholar

Nowlin, M. C. (2024). Policy beliefs, belief uncertainty, and policy learning through the lens of the advocacy coalition framework. Policy and Polit. 52 (4), 675–695. doi:10.1332/03055736Y2023D000000002

CrossRef Full Text | Google Scholar

Oberschelp, C., Pfister, S., Raptis, C. E., and Hellweg, S. (2019). Global emission hotspots of coal power generation. Nat. Sustain. 2 (2), 113–121. doi:10.1038/s41893-019-0221-6

CrossRef Full Text | Google Scholar

Ongsakul, V., Papangkorn, S., and Jiraporn, P. (2023). Estimating the effect of climate change exposure on firm value using climate policy uncertainty: a text-based approach. J. Behav. Exp. Finance 40, 100842. doi:10.1016/j.jbef.2023.100842

CrossRef Full Text | Google Scholar

Ren, X., Zhang, X., Yan, C., and Gozgor, G. (2022). Climate policy uncertainty and firm-level total factor productivity: evidence from China. Energy Econ. 113, 106209. doi:10.1016/j.eneco.2022.106209

CrossRef Full Text | Google Scholar

Sahu, B. K. (2018). Wind energy developments and policies in China: a short review. Renew. Sustain. Energy Rev. 81, 1393–1405. doi:10.1016/j.rser.2017.05.183

CrossRef Full Text | Google Scholar

Shang, Y., Han, D., Gozgor, G., Mahalik, M. K., and Sahoo, B. K. (2022). The impact of climate policy uncertainty on renewable and non-renewable energy demand in the United States. Renew. Energy 197, 654–667. doi:10.1016/j.renene.2022.07.159

CrossRef Full Text | Google Scholar

Shen, Y., Shi, X., Zhao, Z., Sun, Y., and Shan, Y. (2023). Measuring the low-carbon energy transition in Chinese cities. Iscience 26 (1), 105803. doi:10.1016/j.isci.2022.105803

PubMed Abstract | CrossRef Full Text | Google Scholar

Shen, J., Zhang, Y., and Chen, X. (2024). Environmental regulation and energy consumption transition of rural residents: a case of China. Energy 310, 133195. doi:10.1016/j.energy.2024.133195

CrossRef Full Text | Google Scholar

Sims, C. A. (2006). Rational inattention: beyond the linear-quadratic case. Am. Econ. Rev. 96 (2), 158–163. doi:10.1257/000282806777212431

CrossRef Full Text | Google Scholar

Song, Y., Bai, W., and Zhang, Y. (2024). Resilience assessment of trade network in copper industry chain and the risk resistance capacity of core countries: based on complex network. Resour. Policy 92, 105034. doi:10.1016/j.resourpol.2024.105034

CrossRef Full Text | Google Scholar

Stern, P. C., Dietz, T., Abel, T. D., Guagnano, G., and Kalof, L. (1999). A value-belief-norm theory of support for social movements: the case of environmentalism. Huxley Coll. Peninsulas Publ. 6 (2), 81–97. Available online at: http://www.jstor.org/stable/24707060.

Google Scholar

Sun, H., and Zhu, J. (2023). Can the establishment of free trade pilot zones enhance the resilience of china’s industrial chain? — an empirical test based on the mediating mechanism of multiple innovation factor agglomeration. Mod. Econ. Res. 11, 72–84. doi:10.13891/j.cnki.mer.2023.11.002

CrossRef Full Text | Google Scholar

Wang, H., and Wang, M. (2020). Effects of technological innovation on energy efficiency in China: evidence from dynamic panel of 284 cities. Sci. Total Environ. 709, 136172. doi:10.1016/j.scitotenv.2019.136172

PubMed Abstract | CrossRef Full Text | Google Scholar

Wang, Z., Luo, D., and Liu, L. (2018). Natural gas utilization in China: development trends and prospects. Energy Rep. 4, 351–356. doi:10.1016/j.egyr.2018.05.005

CrossRef Full Text | Google Scholar

Wu, R., and Li, M. (2025). Impact of consumer green preferences on pricing decisions in dual-channel supply chains: a complex systems-based dynamic game analysis. Glob. J. Flexible Syst. Manag. 26 (4), 883–902. doi:10.1007/s40171-025-00466-8

CrossRef Full Text | Google Scholar

Wu, R., Li, M., Liu, F., Zeng, H., and Cong, X. (2024). Adjustment strategies and chaos in duopoly supply chains: the impacts of carbon trading markets and emission reduction policies. Int. Rev. Econ. and Finance 95, 103482. doi:10.1016/j.iref.2024.103482

CrossRef Full Text | Google Scholar

Wu, R., Zeng, H., Yan, J., and Işık, C. (2025). Introducing the environmental governance policy uncertainty (Egpu) for China. J. Environ. Manag. 386, 125748. doi:10.1016/j.jenvman.2025.125748

PubMed Abstract | CrossRef Full Text | Google Scholar

Xiang, X., Liu, C., Yang, M., and Zhao, X. (2020). Confession or justification: the effects of environmental disclosure on corporate green innovation in China. Corp. Soc. Responsib. Environ. Manag. 27 (6), 2735–2750. doi:10.1002/csr.1998

CrossRef Full Text | Google Scholar

Xie, X., Huo, J., and Zou, H. (2019). Green process innovation, green product innovation, and corporate financial performance: a content analysis method. J. Bus. Res. 101, 697–706. doi:10.1016/j.jbusres.2019.01.010

CrossRef Full Text | Google Scholar

Xu, J., Wang, J., Li, R., and Yang, X. (2023). Spatio-temporal effects of urbanization on co2 emissions: evidences from 268 Chinese cities. Energy Policy 177, 113569. doi:10.1016/j.enpol.2023.113569

CrossRef Full Text | Google Scholar

Xu, X., Huang, S., Lucey, B. M., and An, H. (2023). The impacts of climate policy uncertainty on stock markets: comparison between China and the us. Int. Rev. Financial Analysis 88, 102671. doi:10.1016/j.irfa.2023.102671

CrossRef Full Text | Google Scholar

Yan, J. (2025). Climate governance and industrial ecological transformation. Corp. Soc. Responsib. Environ. Manag. 1 (32), 718–728. doi:10.1002/csr.2987

CrossRef Full Text | Google Scholar

Yang, M., Fu, M., and Zhang, Z. (2021). The adoption of digital technologies in supply chains: drivers, process and impact. Technol. Forecast. Soc. Change 169, 120795. doi:10.1016/j.techfore.2021.120795

CrossRef Full Text | Google Scholar

Zeng, S., Gu, Y., Su, B., and Li, T. (2025). Energy consumption transition and green total factor productivity in chinese prefecture-level cities. Energy Econ. 142, 108156. doi:10.1016/j.eneco.2024.108156

CrossRef Full Text | Google Scholar

Zhang, X., Li, G., Wu, R., Zeng, H., and Ma, S. (2025). Impact of carbon emissions, green energy, artificial intelligence and high-tech policy uncertainty on China's financial market. Finance Res. Lett. 82, 107599. doi:10.1016/j.frl.2025.107599

CrossRef Full Text | Google Scholar

Zhang, Z., Zhang, C., Hao, Y., Zhang, L., Li, X., and Li, Y. (2025). Dynamic changes in water use patterns of coal power generation during China's energy transition. Energy Policy 198, 114460. doi:10.1016/j.enpol.2024.114460

CrossRef Full Text | Google Scholar

Zhao, J., Qu, J., Wei, J., Yin, H., and Xi, X. (2023). The effects of institutional investors on firms' green innovation. J. Prod. Innovation Manag. 40 (2), 195–230. doi:10.1111/jpim.12652

CrossRef Full Text | Google Scholar

Zheng, Q., Wan, L., Wang, S., Chen, Z., Li, J., Wu, J., et al. (2023). Will informal environmental regulation induce residents to form a green lifestyle? Evidence from China. Energy Econ. 125, 106835. doi:10.1016/j.eneco.2023.106835

CrossRef Full Text | Google Scholar

Zhou, B., and Ding, H. (2023). How public attention drives corporate environmental protection: effects and channels. Technol. Forecast. Soc. Change 191, 122486. doi:10.1016/j.techfore.2023.122486

CrossRef Full Text | Google Scholar