© Wuhan University 2025

0 Introduction

Carbon Emission Trading Policy (CETP) has been extensively implemented with the aim of incentivizing companies to reduce their carbon emissions by establishing carbon emission allowances, rationalizing energy consumption allocation, and facilitating more efficient utilization of energy^[1]. CETP represents an innovative energy-saving policy that employs a market mechanism to address climate change issues, thereby providing a robust framework for achieving targets related to carbon peaking and carbon neutrality^[2]. Consequently, the Chinese government identified seven provinces and province-level municipalities for implementing CETP, which was formally launched in 2013. As existing studies have found that "low-carbon city" pilot policy with similar aims to CETP can significantly improve energy efficiency^[3], CETP may also impact energy efficiency. Figure 1 shows how energy efficiency and carbon market transactions in China have evolved from 2008 to 2020. Few studies to date have emphasized the spatial spillover effects of CETP on energy efficiency in the neighboring provinces of the pilot region. Actually, the pilot provinces may change the energy consumption structure, industrial layout and spatial layout of technological advancement, thereby affecting the energy efficiency of neighboring provinces. In addition, many existing studies used the traditional difference-in-differences (DID) model to analyze the policy effects on pilot provinces without considering the impact on non-pilot provinces. Therefore, we use the spatial difference-in-differences (SDID) model, which combines the DID model with the spatial Durbin model (SDM), to assess the effect of CETP on energy efficiency in neighboring regions from a spatial perspective, and then demonstrate the necessity of expanding the implementation of CETP to all provinces in China. Furthermore, we analyze the effect mechanism of CETP to provide a more comprehensive perspective. After establishing a mediation effect model, we find that CETP improves energy efficiency through adjusting industrial structure (IS) and restraining financial efficiency (FE). To discuss the effectiveness of CETP, we analyze heterogeneity across provinces from the perspectives of geographic location, environmental law enforcement efforts, and pollution levels.

Fig. 1 Trends of China's carbon emission trading and energy efficiency, 2008-2020

The rest of this article is arranged as follows. Section 1 provides the literature review of CETP, energy efficiency and spatial spillovers. Section 2 presents theoretical analysis and hypotheses. Section 3 describes the data sources and designs the models. Section 4 discusses the empirical results. Section 5 concludes the study and provides policy recommendations.

1 Literature Review

China's carbon trading market has expanded rapidly in recent years, positioning itself as the second-largest market globally^[4]. Consequently, research on the impacts of China's CETP has received more attention, primarily focusing on carbon emissions and carbon intensity. Previous studies concluded that CETP can reduce the intensity of carbon emissions in a sustained and stable manner, improve the carbon efficiency of energy-intensive industries in pilot regions, and play a significant positive role in helping to achieve carbon neutrality^[5-6]. CETP has a positive correlation with carbon productivity, suggesting it can help the pilot cities to achieve a low-carbon economic transition and improve their carbon productivity^[7]. In changing the energy mix, CETP significantly promotes the use of non-fossil energy to reduce carbon emissions, such as hydroelectricity and photovoltaic power generation^[8].

Regarding the relative lacking of existing research on the CETP's impact on energy efficiency, it is crucial to conduct an analysis of the correlation between CETP and energy efficiency at this pivotal juncture in China's pursuit of its "dual-carbon" objectives and advancement in manufacturing. Some researchers believe CETP has significantly reduced CO₂ emissions in the industrial sector and also brought greater economic benefits to industrial enterprises^[9]. Zhou et al^[10] have accurately argued that China's CETP has a "weak Porter" effect, i.e., it can achieve the two objectives of "environmental protection" and "economic development". In the long run, CETP is also likely to promote low-carbon technological innovation, which will lead to greater reputational and economic gains^[11]. Although CETP has the potential to enhance energy efficiency within the policy pilot areas, it remains uncertain whether it exerts a similarly positive influence on energy efficiency in the surrounding regions.

CETP was initially experimented in only a few provinces in China. The spatial spillover effect of CETP has been examined from various perspectives in numerous studies. With regard to China's transport carbon emissions, some studies have found that CETP can significantly decrease them and also produce significant spatial spillover effects^[12]. Dai et al^[11] confirmed that CETP can decrease the carbon emission intensity of industries in the pilot region by promoting technological innovation of enterprises and producing spatial spillover effect to the neighboring regions, which in turn drives the region's overall innovation development and reduces the neighboring regions' carbon emission intensity^[13]. Since 2013, CETP has significantly reduced the carbon emissions of the pilot region, and the carbon emissions of the neighboring regions also show a decreasing trend, indicating that CETP has a spatial spillover effect^[2]. Strong empirical studies have revealed that CETP can effectively influence neighboring regions from a spatial perspective.

To further explain how CETP realizes its impact on energy efficiency, this paper explores the influencing mechanism from two aspects. First, China introduced a policy to abolish the restriction on loan-to-deposit ratio in 2015, which released positive signals of improving FE. Financial institutions may optimize their credit structure in response to the policy orientation of CETP and increase their lending to green and high-efficiency projects. Better allocative capacity has helped to increase financial support for new energy and environmentally friendly technologies^[14]. Therefore, we examine the potential impact of FE on energy efficiency by considering it as an intermediary variable. Second, promoting the transformation of industries from production-based to service-based is China's strategic plan because it needs to optimize the IS of its national economy to realize the goal of carbon reduction. Previous studies concluded that low-carbon city policy has a positive effect on optimizing IS of both the local and the neighboring regions^[15-16], and the optimization of IS can significantly reduce carbon intensity and promote energy efficiency^[17]. CETP belongs to the low-carbon and energy-saving policy. Therefore, we posit that IS could potentially serve as a mediating factor within CETP's influencing mechanism.

2 Theoretical Analysis

Porter hypothesis suggested that well-designed environmental regulations can promote companies to facilitate innovation to enhance competitiveness. The regulations can incentivize companies to develop new technologies, processes and products that are both profitable and environmentally friendly, pushing enterprises to improve energy utilization efficiency and reduce pollution emissions^[18]. Therefore, this paper argues that CETP may reduce carbon emissions while improving enterprises' low-carbon production technology and energy efficiency.

The core idea of CETP is to set annual carbon emission quotas for enterprises based on their historical emission levels or industry standards, allow enterprises to emit carbon within the prescribed limits, and then trade the quotas in the market to make profits. Enterprises that exceed the limits will be subject to huge fines^[19]. Therefore, CETP have two ways to promote energy efficiency. The first is the cost of carbon emission payed by high-polluting enterprises to purchase the required carbon emission rights, which will prompt high-polluting enterprises to take carbon emission reduction measures, such as eliminating outdated production technologies and equipment, and increasing their output value while controlling carbon emissions. The second is the profit-driven effect which encourages enterprises adopt energy-saving measures to reduce carbon emissions so that they can sell excess carbon emission allowances for financial gain^[20]. Furthermore, enterprises will extend to invest in low-carbon production technologies and energy efficiency improvement projects, ultimately reducing their dependence on fossil fuels and shaping a virtuous circle of low-carbon production and promotion of energy efficiency. In general, enterprises in the CETP pilot provinces need to maintain or increase their output value without increasing carbon emissions, and the policy puts apparent pressure on enterprises to promote energy efficiency. Hypothesis 1 is thus proposed.

H1. CETP can improve energy efficiency in pilot provinces.

In addition, firms in the CETP pilot areas are constrained by the policy, but whether this policy will have spatial spillover effects on neighboring provinces is an issue of concern. The high-carbon-emitting firms, under the pressure of higher carbon emission costs, may shift their high-polluting production activities to neighboring non-pilot provinces, resulting in a negative spatial spillover effect on energy efficiency. Moreover, as CETP encourages firms to adopt more cleaner technologies in the pilot provinces to reduce carbon emissions and gain more economic benefits^[9], firms in neighboring areas may perceive the benefits of these technologies, resulting in technology diffusion and imitation effects, and start to adopt more energy-efficient technologies to remain competitive. CETP may also affect the supply chains in the neighboring regions.

In order to meet the demand for greener and more environmentally friendly products from enterprises in the pilot provinces, suppliers and enterprises may adopt more energy-efficient production methods, which will help to improve the energy efficiency of the holistic supply chain and generate spatial spillover effects. Scholars have used panel spatial Durbin models to empirically test the impact of environmental regulations on industrial total factor productivity in China at the provincial level^[21]. Considering the spatial correlation and better understanding of the impact of spatial layout on energy efficiency changes, it is therefore essential to combine the spatial econometric model with the DID model to analyze the spatial spillover effect of CETP on energy efficiency of neighboring provinces. Hypothesis 2 is thus proposed:

H2. CETP can improve energy efficiency not only in the pilot provinces but also in the neighboring non-pilot provinces.

Important financial support from financial markets meets the financial needs of firms and ensures the liquidity and effective functioning of the carbon trading market. Financial development is generally regarded as the main driver of economic growth, so it is also likely to have an impact on energy demand^[22]. The development of the financial sector can lead to technological advances, particularly in environmental technology, which can reduce environmental pollution throughout the process^[23]. Yu-xiang and Chen^[24] thought pollution intensity is curbed by financial development, and financial development in China has played a great role in improving environmental performance.

In financially advanced countries, well-functioning financial markets channel resources to R&D (Research and Development)-driven industries^[25]. On the contrary, some scholars have found negative impacts of FE, which may cause industrial pollution and environmental degradation while promoting economic growth. Over the long term, FE has a negative effect on carbon emissions; financial development reduces regional carbon emissions and boosts carbon emissions in neighboring regions through spillover effects^[26]. Some researchers conclude that FE might attract new investors into heavy industries^[27]. Due to the country's emphasis on green and low-carbon development, CETP may affect the lending of banks and other financial institutions to high-carbon emitting enterprises because CETP increases their environmental and legal risks, making it more difficult to apply for finance loans and thus restricting the expanding of high-carbon-emitting production to reach the target of raising energy efficiency. Based on these theories, Hypothesis 3 is proposed:

H3. CETP improves energy efficiency by controlling financial efficiency.

CETP can have a positive impact on the optimization of IS by promoting the development of cleaner industries and services, eliminating high-carbon emission industries and improving the comprehensive energy utilization rate. In the advancement of the real economy, the optimization of IS is closely related to driving energy efficiency. To fulfill the purpose of energy saving and emission reduction through source management, it is imperative to decrease the proportion of high-energy-consuming industries^[28]. Through the optimization of IS and energy structure, carbon emission reduction technology can be upgraded, which is conducive to the reduction of carbon emissions and has an important supporting role in the realization of low-carbon cities in China^[29-30]. The optimization of IS implies the brisk development of the service industry and the control of high-polluting industries, which is of great significance to the goal of a low-carbon and high-quality economy. Relying on highly developed information technology, electronic equipment and innovations, the service sector can generate higher economic benefits while consuming less fossil energy. By encouraging low-carbon production and increasing the cost of carbon emissions for enterprises, CETP may prompt enterprises to a transition away from over-reliance on industrial production and invest in expanding service sector production. Therefore, CETP may be able to achieve energy efficiency in the process of optimizing IS by controlling production activities with high carbon emissions. Hypothesis 4 is proposed based on these theories.

H4. CETP improves energy efficiency by optimizing industrial structure.

Figure 2 shows the theoretical mechanism for setting up the impact of CETP on energy efficiency.

Fig. 2 Mechanism of CETP's impact on energy efficiency

3 Methods and Modelling

3.1 SDID Model

3.1.1 Description of variables and data sources

Explained variable: energy efficiency ($\mathrm{E}{\mathrm{E}}_{it}$). Explanatory variables: $\mathrm{d}\mathrm{i}\mathrm{d}$ is the interaction term of CETP dummy and time dummy. $\mathrm{t}\mathrm{r}\mathrm{e}\mathrm{a}{\mathrm{t}}_i$ is the policy dummy variable, if $i$ province is a CETP pilot province or province-level municipality then $\mathrm{t}\mathrm{r}\mathrm{e}\mathrm{a}{\mathrm{t}}_i=\mathrm{1}$, otherwise $\mathrm{t}\mathrm{r}\mathrm{e}\mathrm{a}{\mathrm{t}}_i=\mathrm{0}$; $\mathrm{p}\mathrm{o}\mathrm{s}{\mathrm{t}}_t$ is the time dummy variable, this study set 2013 as the start time of CETP, if the year $t\ge \mathrm{2013}$ then $\mathrm{p}\mathrm{o}\mathrm{s}{\mathrm{t}}_t=\mathrm{1}$, otherwise $\mathrm{p}\mathrm{o}\mathrm{s}{\mathrm{t}}_t=\mathrm{0}$. All variables are defined in Table 1.

The data were obtained from the National Bureau of Statistics, the EPS database, the China Energy Statistical Yearbook and the provincial statistical yearbooks. Table 2 demonstrates the descriptive statistics of the above variables.

3.1.2 Model construction

We collect balanced panel data for 30 provinces and province-level municipalities of China from 2008 to 2020 and sets 2008-2012 as the time before the introduction of CETP, and 2013-2020 as the time when the policy is implemented on a pilot basis, with the sample of the experimental group comprising six provinces and province-level municipalities of China: Beijing, Shanghai, Chongqing, Tianjin, Hubei and Guangdong. Most of the studies use DID model to study CETP, but this model only evaluates the effect of the policy on a few pilot provinces and is unable to measure the spatial spillovers of the policy, which reduces the robustness of the results. To analyze CETP's effect more comprehensively, we further use the SDID model to measure the spatial spillover effect of CETP on non-pilot regions. The formula is set as follows:

$\mathrm{E}{\mathrm{E}}_{it}=\alpha +{\beta }_{\mathrm{1}}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+{\beta }_{\mathrm{2}}\mathrm{W}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+\rho \mathrm{W}\mathrm{E}{\mathrm{E}}_{it}+{\eta }_i{\displaystyle \sum _i^k}{\mathit{X}}_{it}+{\gamma }_{i}W{\displaystyle \sum _i^k}{\mathit{X}}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$(1)

where $\mathit{X}$ represents a vector of all control variables, ${\lambda }_t$ represents time-fixed effects, ${\mu }_i$ represents province-fixed effects and ${\epsilon }_{it}$ represents a stochastic disturbance term. ${\beta }_{\mathrm{1}}>\mathrm{0}$ means that CETP promotes energy efficiency in the pilot area. The interaction term $\mathrm{W}\mathrm{d}\mathrm{i}\mathrm{d}$ denotes the degree of influence of CETP in pilot provinces on energy efficiency in neighboring provinces, $\mathrm{W}\mathrm{E}\mathrm{E}$ denotes the spatial lag term of $\mathrm{E}\mathrm{E}$, and $\rho$ is the spatial regression coefficient, which denotes the influence of energy efficiency in neighboring provinces on this province.

3.2 Modelling of Spatial Mediating Effects

Existing studies analyzed the mechanism of CETP on energy efficiency from the viewpoint of technological progress or green technology innovation and ignored the intermediary role of FE and IS in it. Based on the SDID model, we combined the three-step mediation effect method to establish a spatial mediation effect model to research impact mechanisms^[31], we set formulas (2-4) as follows to represent the total effect, direct effect and indirect effect models, respectively.

$\mathrm{E}{\mathrm{E}}_{it}={\alpha }_{\mathrm{0}}+{\alpha }_{\mathrm{1}}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+{\alpha }_{\mathrm{2}}\mathrm{W}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+\rho \mathrm{W}\mathrm{E}{\mathrm{E}}_{it}+{\beta }_i\sum {\mathit{X}}_{it}+{\gamma }_{i}W\sum {\mathit{X}}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$(2)

$\mathrm{E}{\mathrm{E}}_{it}={{\alpha }_{\mathrm{0}}}^{\mathrm{\text{'}}}+{{\alpha }_{\mathrm{1}}}^{\mathrm{\text{'}}}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+{{\alpha }_{\mathrm{2}}}^{\mathrm{\text{'}}}\mathrm{W}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+{\rho }^{\text{'}}\mathrm{W}\mathrm{E}{\mathrm{E}}_{it}+{{\beta }_i}^{\mathrm{\text{'}}}\sum {\mathit{X}}_{it}+{{\beta }_{\mathrm{6}}}^{\mathrm{\text{'}}}\mathrm{M}\mathrm{e}{\mathrm{d}}_{it}+{{\gamma }_i}^{\mathrm{\text{'}}}W\sum {\mathit{X}}_{it}+{{\gamma }_{\mathrm{6}}}^{\mathrm{\text{'}}}\mathrm{W}\mathrm{M}\mathrm{e}{\mathrm{d}}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$(3)

$\mathrm{M}\mathrm{e}{\mathrm{d}}_{it}={{\alpha }_{\mathrm{0}}}^{″}+{{\alpha }_{\mathrm{1}}}^{″}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+{{\alpha }_{\mathrm{2}}}^{″}\mathrm{W}\mathrm{d}\mathrm{i}{\mathrm{d}}_{it}+{\rho }^{″}\mathrm{W}\mathrm{M}\mathrm{e}{\mathrm{d}}_{it}+{{\beta }_i}^{″}\sum {\mathit{X}}_{it}+{{\gamma }_i}^{″}W\sum {\mathit{X}}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$(4)

where $\mathrm{M}\mathrm{e}\mathrm{d}$ in formula (3) represents the intermediary variable, which stands for the FE and IS. We firstly conduct the regression analysis on formula (2) to observe whether the coefficient ${\alpha }_{\mathrm{2}}$ is significant or not; if it is significant, then further run the formula (4) to observe whether ${{\alpha }_{\mathrm{2}}}^{″}$ is significant or not; if ${{\alpha }_{\mathrm{2}}}^{″}$ is significant, then it means that CETP has a significant effect on $\mathrm{M}\mathrm{e}\mathrm{d}$; finally, we run formula (3) to observe whether the coefficient ${{\alpha }_{\mathrm{2}}}^{\mathrm{\text{'}}}$ is significant or not; if ${{\alpha }_{\mathrm{2}}}^{\mathrm{\text{'}}}$ is significant, then $\mathrm{M}\mathrm{e}\mathrm{d}$ has an incomplete mediation effect.

3.3 Spatial Correlation

To calculate the spatial correlation of energy efficiency in different provinces, we use Moran's I index to conduct spatial correlation test to observe whether there is a significant spatial correlation. The formula for calculating Moran's I index is as follows:

$I=\frac{{\displaystyle \sum _{i=\mathrm{1}}^n}{\displaystyle \sum _{j=\mathrm{1}}^n}{\mathit{W}}_{ij}(\mathrm{E}{\mathrm{E}}_i-\overline{\mathrm{E}\mathrm{E}})(\mathrm{E}{\mathrm{E}}_j-\overline{\mathrm{E}\mathrm{E}})}{S^{\mathrm{2}}{\displaystyle \sum _{i=\mathrm{1}}^n}{\displaystyle \sum _{j=\mathrm{1}}^n}{\mathit{W}}_{ij}}$(5)

where $S^{\mathrm{2}}=\frac{\mathrm{1}}{n}{\displaystyle \sum _{i=\mathrm{1}}^n}(\mathrm{E}{\mathrm{E}}_i{-\overline{\mathrm{E}\mathrm{E}})}^{\mathrm{2}}$, $\overline{\mathrm{E}\mathrm{E}}=\frac{\mathrm{1}}{n}{\displaystyle \sum _{i=\mathrm{1}}^n}\mathrm{E}{\mathrm{E}}_i$. $\mathrm{E}{\mathrm{E}}_i$ denotes the energy efficiency of i province, and $y_i$ denotes the average GDP of i province from 2008 to 2020. We use the economic distance weight matrix to analyze the spatial correlation of energy efficiency. ${\mathit{W}}_{ij}=(w_{ij}{)}_{n\times n}$ denotes the spatial weight matrix, where $w_{ij}=\frac{\mathrm{1}}{\left|y_i-y_j\right|}$, and $n$ represents the number of provinces. Moran's I index takes the value range within $[-\mathrm{1,1}]$, if the index is greater than zero, it means that energy efficiency has a positive spatial correlation, otherwise it indicates a negative spatial correlation.

3.4 Selection of Spatial Measurement Models

From the Moran's I index, we can learn that energy efficiency of different provinces has significant spatial correlation, so it is vital to use SDID model for empirical analysis, which can be used for testing the spatial correlation. We need to select the most appropriate spatial econometric model through several tests, and Hausman tests are required to determine whether the fixed effect model is selected.

Table 3 shows that the P-value under Lagrange Multiplier (LM)-error test is significant at 1% level, meaning that there is a spatial error effect, so SDM or spatial error model (SEM) can be selected for regression, and it is necessary to carry out the Likelihood Ratio (LR) test and Wald test further to observe whether SDM degrades to SEM. LR test and Wald test are significant at 1% level. The results indicate that SDM should be used in the model of this paper. Hausman test can determine whether SDM applies to random effects or fixed effects, and the result is significant positive, indicating that SDM should be selected for fixed effects. After the above tests, we adopt SDM with fixed effects for regression analysis.

4 Empirical Results and Discussion

4.1 Spatial Features of Energy Efficiency in Chinese Provinces

Using Arcgis 10.8 software, we analyzed the changes of energy efficiency in 30 provinces from 2008 to 2020 to explore the spatial distribution characteristics and the evolutionary process of energy efficiency in China's provinces. As can be seen from Fig. 3, there are only a few provinces with high energy efficiency in 2008. By 2020, the high-efficiency zones have been extended from the east coast to most of the southern and eastern part of China. The seven pilot provinces of CETP have improved the energy efficiency of neighboring provinces through the spatial spillover effect.

Fig.3 Spatial distribution of energy efficiency in Chinese provinces

4.2 Moran’s I Index

Table 4 shows the results of Moran's I index. The Moran's I index of energy efficiency of 30 provinces from 2008 to 2020 are all significantly greater than 0.3, which indicates that there is a significant positive spatial correlation between energy efficiencies of various provinces, and the energy efficiency of one province is affected by other provinces while affecting the energy efficiency of neighboring provinces. The Moran's I index of energy efficiency in 2020 is a little lower than before, indicating that the spatial correlation of energy efficiency has decreased, which may be because the energy efficiency of each province has been significantly raised in the past few years, and the gap among different provinces has been narrowed.

Figure 4 plots the local Moran scatter plots of energy efficiency for four different years, with the provinces mainly distributed in the first and third quadrants. China's current energy efficiency is mostly dominated by low-low and high-high agglomeration, indicating that the energy efficiency of different provinces has significant local spatial aggregation characteristics in space.

Fig. 4 Scatterplot of localized Moran's I index of energy efficiency over the past few years

4.3 SDID

Table 5 shows the effect of CETP on energy efficiency, with all regressions controlling for province-fixed effects and year-fixed effects. The results of DID regressions are included for comparison with the SDID regression results. We added the regression results for DID as a control in the first two columns, where the coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ is 0.526 and significant without adding control variables in column (1), and the model with control variables added in column (2) has a regression coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ which is 0.159 and significant at 1% level. The results show that after the implementation of CETP, the energy efficiency in the pilot area has significantly improved compared to other non-pilot areas^[32], proving that hypothesis H1 is correct. Columns (3) and (4) are the regression results of SDID model, in which the control variables are not added in column (3), and the coefficients of $\mathrm{d}\mathrm{i}\mathrm{d}$ and $\mathrm{W}\mathrm{d}\mathrm{i}\mathrm{d}$ are both significant at 1% level, which indicates significant promotion and spatial spillover effect of CETP on energy efficiency. It proves that hypothesis H2 is correct. The regression results of SDID are similar to the DID model, but it can be observed that the coefficients of $\mathrm{d}\mathrm{i}\mathrm{d}$ in columns (3)-(4) are greater than those in columns (1)-(2), which indicates that effect of CETP is underrated in the DID model. This fully demonstrates the advantage of using the SDID model in this study, which can more accurately evaluate the impact of CETP by considering spatial factors. Concerning the control variables, the coefficients of $\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{h}$ and $\mathrm{P}\mathrm{g}\mathrm{d}\mathrm{p}$ are both significant at 1% level, indicating that the improvement of technological level, economic development level and people's affluence are beneficial to energy efficiency. The improvement of regional technology will help enterprises to reduce the waste of fossil energy and adopt cleaner technology to increase the utilization of clean energy to gradually replace fossil energy. The rising economic level means that enterprises have stronger economic capacity to carry out technological research and development and equipment renewal. The coefficients of $\mathrm{G}\mathrm{O}\mathrm{V}$ and $\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{O}\mathrm{P}$ are significantly negative, representing that increased government regulatory capacity and the expansion of urban population size will not be helpful to improve energy efficiency. Government administrative intervention in the market affects the efficiency of energy allocation, while population growth increases the consumption of energy for electricity, heating, transportation fuels, etc., leading to a higher concentration of energy demand in time and space, and a greater waste of resources in the absence of effective energy management measures. In addition, the coefficient of $\mathrm{l}\mathrm{n}\mathrm{I}\mathrm{n}\mathrm{v}$ is not significant, indicating that changes in foreign investment do not have a significant effect on energy efficiency. The government can increase the openness of the industrial sector and gradually lift foreign investment restrictions, allowing more qualified foreign investors to participate in enterprise management, thus introducing advanced clean production technologies and energy management models to improve energy efficiency.

Based on column (4) in Table 5, the effect of CETP on energy efficiency is dissolved into direct, indirect and total effects (Table 6). Column (1) represents the effect of CETP on energy efficiency in the pilot provinces; column (2) shows the spatial spillover effect, represents the effect of CETP on energy efficiency in the neighboring provinces; and the total effect represents the impact of CETP on the overall energy efficiency in all provinces. It can be seen that the coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ in column (1) is 0.208, which indicates that CETP has increased the energy efficiency of the pilot provinces by 20.8% after its implementation, the coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ in column (2) is 0.448 and it means that CETP in the pilot provinces raises the neighboring provinces' energy efficiency by 44.8%. And the coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ in column (3) is 0.656, which means that CETP makes the overall energy efficiency of all provinces significantly improved by 65.6%, i.e., the average energy efficiency of 30 provinces in the sample is improved by 2.19%. The policy effects of CETP are widespread, which is in line with hypothesis H2.

4.4 Robustness

4.4.1 PSM-DID

The 30 provinces and province-level municipalities in China have large differences in terms of economy, geographic location and technology level, and these differences may cause selection bias in the empirical results. To solve the above problems, we adopt the propensity score matching (PSM) method to match the variables according to the sample characteristics and performs DID analysis on the matching results. As the regression results in column (1) of Table 7, CETP has significant improvement on energy efficiency.

4.4.2 Adjusting the timing of policy initiation

The year before the launch of CETP in China is chosen as the start time of the policy for the robustness test. This study adjusts the time of CETP to 2012 for SDID regression analysis. Column (2) of Table 7 demonstrates the results and the coefficient of did2012 is basically consistent with those of did in Table 5, which shows that the results are robust.

4.4.3 Replacement of spatial weight matrices

The economic distance weight matrix in model (4) of Table 5 is replaced by the geographic distance weight matrix and 0-1 neighborhood matrix before SDID regression to achieve the robustness test, and the results are obtained as shown in columns (3)-(4) of Table 7, with W₁ denoting the geographic distance weight matrix and W₂ denoting the 0-1 neighborhood matrix. The coefficients of most of the key variables are significantly positive, which is close to the results in Table 5, indicating that CETP does have a spatial spillover effect on energy efficiency, proving the accuracy of the regression results.

4.5 Heterogeneity

4.5.1 Geographical

We study the differences in the effectiveness of CETP implementation in 30 provinces and province-level municipalities in China. The samples are divided into eastern, central and western provinces according to their geographic position and the the results of SDID regression are shown in Table 8. The $\mathrm{d}\mathrm{i}\mathrm{d}$ coefficients of the three regions are significantly positive at 1% level, with the largest impact on western region. This suggests that the effect of CETP is significant across the country. The reason may be that the eastern region is more economically developed, the production technology of industrial enterprises is more advanced, the efficiency of energy use is already higher, and the impact of CETP is relatively weaker. Still, the economic level and low-carbon production technology of the central and western regions are comparatively backward, and their dependence on fossil energy is larger, so the progress of the energy efficiency of the two regions is more obvious under the implementation of CETP.

4.5.2 Environmental law enforcement efforts

In the presence of stricter environmental regulations, firms prefer to focus on energy-efficient technologies to significantly reduce energy intensity^[33]. The strength of enforcement of environmental regulations greatly influences whether industrial firms comply with them or not, and provinces with stronger enforcement are likely to be better able to ensure that firms comply with CETP, which can help to prevent emissions cheating and ensure that the carbon trading market operates efficiently^[34]. Provinces with sufficiently strong enforcement can impose penalties for non-compliance, prompting firms to comply with CETP and adopt emission reduction measures. Therefore, this paper distinguishes the strength of environmental enforcement in different provinces by the quantity of environmental penalty cases, and divides them into strong and weak enforcement provinces using the average number of environmental penalty cases as the dividing line, with data derived from the China Environmental Yearbook. Columns (4)-(5) of Table 8 show that the coefficients of $\mathrm{d}\mathrm{i}\mathrm{d}$ are positive and significant in provinces with different enforcement strengths, and the coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ in provinces with high enforcement strengths is 0.320, which is greater than that of 0.246 in provinces with low enforcement strengths, which suggests that the enhancement of energy efficiency in provinces with high enforcement strengths by CETP is higher than that in provinces with low enforcement strengths. This may be because high-intensity environmental enforcement has made carbon-emitting enterprises more cautious, limiting their illegal carbon emission behavior, adopting greener and low-carbon production methods, and improving energy efficiency.

4.5.3 Pollution levels

To examine the impact of CETP under different pollution levels on energy efficiency, this paper distinguishes between high-polluting and low-polluting provinces based on the average sulfur dioxide emissions of all the provinces in the period of 2008-2020 as the cut-off line. Columns (6) and (7) of Table 8 give the results of the SDID regression. The coefficients of $\mathrm{d}\mathrm{i}\mathrm{d}$ for the samples of high- and low-pollution provinces are all significantly positive, but the coefficients of $\mathrm{d}\mathrm{i}\mathrm{d}$ for high-pollution provinces are larger than those for low-pollution provinces, which suggests that CETP plays a better policy effect in high-pollution provinces; the coefficients of $\mathrm{W}\mathrm{d}\mathrm{i}\mathrm{d}$ are also all significantly positive, which suggests no matter in the high or low-pollution areas, CETP has significant spatial spillover effect on the neighboring provinces.

4.6 Mechanisms

4.6.1 Financial efficiency

To test hypothesis H3, we establish a spatial mediation effect model with FE as the intermediary variables to study the intermediary mechanism of CETP on energy efficiency, and the first three columns of Table 9 show the regression results of the total effect, direct effect and indirect effect models, respectively. We found the coefficient of FE in the direct effect model is -0.345 significant at the 1% level, which indicates that the progress of FE has a significant negative effect on energy efficiency; in the indirect effect model, the coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ is significantly negative, so it means the implementation of CETP has an inhibitory effect on FE, while the improvement of FE significantly reduces the energy efficiency, in other words, CETP improves energy efficiency by reducing FE which indicates that hypothesis H3 is correct.

4.6.2 Industrial structure

To test hypothesis H4, we take the IS as the mediating variable to test whether CETP improves energy efficiency by optimizing IS. Column (4) of Table 9 is the direct effect model, with the coefficient of IS is 0.251 and significant at 1% level, the optimization of IS can significantly improve energy efficiency; the coefficient of $\mathrm{d}\mathrm{i}\mathrm{d}$ in the indirect effect model in column (5) is also significantly positive at 1% level, which indicates that CETP plays a significant role in facilitating the optimization of IS. Overall CETP improves the energy efficiency by promoting the optimization of their IS and making the proportion of their secondary industry relatively contracted, which indicates that hypothesis H4 is correct.

5 Conclusion and Policy Implications

This paper analyzes the impact and spatial spillover effects of CETP on energy efficiency in China from 2008-2020. Firstly, we analyze the Moran's I index of energy efficiency in each province and determines that energy efficiency has a positive spatial correlation. Secondly, we collect the panel data of 30 provinces and province-level municipalities, and establish the SDID model to analyze the impact of CETP on energy efficiency, including the decomposition of the spatial total utility, and the analysis of heterogeneity in geographic regions, environmental enforcement efforts and the degree of pollution. Finally, a spatial mediation model is built to test the role of FE and IS in the intermediary mechanism of the impact of CETP on energy efficiency. After some robustness tests, the main conclusions are robust.

The conclusions are as follows: 1) The energy efficiency has a significant spatial agglomeration, and the improvement of energy efficiency in one province has a positive driving effect on its neighboring provinces. 2) CETP has achieved remarkable achievements and significantly improved the energy efficiency of the pilot provinces, while this policy also has a positive spatial spillover effect and positively influences the energy efficiency of the neighboring provinces, which indicates that CETP is an effective tool to enhance environmental protection, reduce carbon emissions and improve energy efficiency. 3) FE has a significant negative impact on energy efficiency, but CETP has a significant negative impact on FE, then it means that CETP improves energy efficiency by restraining FE. 4) The optimization of IS has a significant contributing effect on energy efficiency improvement. CETP achieves lower carbon emissions, higher production output and thus higher energy efficiency by promoting the optimization of IS, facilitating the transformation of enterprises, increasing the investment in low-carbon emission industries and services, and decreasing the proportion of secondary industry in China. 5) Heterogeneity analysis shows that CETP has a better performance in western provinces, the regions with strong environmental law enforcement and high pollution areas.

After analyzing the above conclusions, we put forward three suggestions. 1) CETP is extremely beneficial to the progress of energy efficiency in the pilot provinces and neighboring provinces. Therefore, it is appropriate for CETP to be extended to all provinces across the country to improve national energy efficiency, while the government should continue to improve CETP by placing more emphasis on energy efficiency. In addition, enforcement and monitoring of laws are key issues, and environmental regulations should be strictly enforced to ensure that companies comply with emission limits, otherwise the effectiveness of the policy will be threatened. 2) The government should control the disorderly expansion of financial capital. From the spatial mediation effect model, we found that improved FE has an adverse impact on the energy efficiency of enterprises. The banks should provide green credit support to prompt enterprises to adopt low-carbon and energy-efficient technologies and avoid disorderly expansion of high-emission industries, which will help promote the implementation of CETP. Moreover, governments could establish a green finance framework that channels funds into energy-efficient sectors through the issuance of green bonds or sustainability-linked loans. 3) IS optimization plays a crucial intermediary role in the impact of CETP on energy efficiency improvement. By reducing the proportion of energy-intensive industries, increasing investment in service industries, and promoting the growth of the service sector, it can improve energy efficiency and help China realize green and sustainable economic growth. Financial incentives from the government can also motivate firms to upgrade their IS. Subsidies, tax credits, and low-interest loans can be provided to firms that invest in energy-efficient technologies or undertake R&D aimed at industrial upgrading. These incentives would lower the financial burden of transitioning to greener practices and encourage small and medium-sized enterprises, which might otherwise lack the resources to invest in green production technologies.

References

All Tables

All Figures

	Fig. 1 Trends of China's carbon emission trading and energy efficiency, 2008-2020
In the text

	Fig. 2 Mechanism of CETP's impact on energy efficiency
In the text

	Fig.3 Spatial distribution of energy efficiency in Chinese provinces
In the text

	Fig. 4 Scatterplot of localized Moran's I index of energy efficiency over the past few years
In the text