Does carbon emission trading policy promote energy structure transition? Evidence from China

Introduction: China’s Peak Carbon and Carbon Neutral Program is a key national policy to accelerate green development and establish a foundation for high-quality, sustainable growth. Transitioning toward a low-carbon economy requires gradually reducing coal consumption while ensuring energy security. This study examines how carbon emission trading policy influence energy consumption structure and their underlying mechanisms, in order to provide theoretical and policy support for China’s green transition and high-quality development.

Methods: Leveraging provincial-level panel data covering 30 jurisdictions in China from 2004 to 2021, we first quantify the average causal impact of the Carbon Emission Trading Policy (CETP) on energy-structure decarbonization through a two-way fixed-effects difference-in-differences (DID) estimator. Recognizing that linear identification strategies are ill-suited to recover heterogeneous treatment thresholds, regime-dependent mediation channels, or discrete structural breaks, we complement the DID backbone with non-parametric Classification and Regression Trees (CART) to reveal latent regime-specific policy functions and higher-order interaction effects.

Results: (1) The CETP significantly reduces carbon emission intensity and optimises the energy structure, demonstrating its dual effectiveness in emission reduction and transformation. (2) Technological innovation and industrial restructuring are the main transmission channels, through which the policy fosters a synergistic “technology–industry” dynamic by promoting green R&D (research and development) and curbing high-carbon sectors. (3) Urbanisation level critically shapes policy effectiveness, with highly urbanised regions exhibiting stronger transition outcomes due to economies of scale and institutional advantages. Notably, municipalities outperform other pilot provinces, indicating substantial interregional heterogeneity.

Innovation: This study breaks through the traditional single-path analysis framework by integrating multi-dimensional transmission mechanisms and constructing an urbanisation regulation model to reveal the economic geography logic of regional heterogeneity, while integrating econometrics and machine learning methods to enhance the robustness of the conclusions. The study provides theoretical and empirical support for the optimisation of carbon market design, the formulation of differentiated regional policies, and the acceleration of the ‘dual-carbon’ goal.

1 Introduction

In recent years, China has accelerated its pursuit of green development, positioning it as a crucial pathway toward achieving high-quality economic growth. Following the 19th National Congress of the Communist Party of China (CPC), where it was emphasized that China’s economy has shifted from a phase of rapid growth to one of high-quality development, green development has become increasingly central. In October 2012, the State Council issued the Action Plan for Carbon Peaking by 2030, advocating for strict controls on fossil fuel consumption and an accelerated reduction in coal usage. During the 14th Five-Year Plan period, the growth of coal consumption must be strictly curtailed, with the goal of achieving a gradual decline during the 15th Five-Year Plan period. Additionally, the Opinions of the CPC Central Committee and the State Council on Fully and Accurately Implementing the New Development Concept and Doing a Good Job in Carbon Peaking and Carbon Neutrality (2021) emphasized measures such as optimizing and upgrading industrial structures, increasing the share of renewable energy, and improving the carbon market mechanism. The issuance of these policies and documents demonstrates the state’s strong commitment to green development and its determination to transition toward a low-carbon economy. Collectively, these initiatives offer vital policy guarantees for achieving the” dual-carbon” goals (carbon peaking and carbon neutrality), thereby laying a solid foundation for China’s sustainable and high-quality economic development.

Transitioning to a low-carbon energy structure represents a fundamental strategy for achieving green economic growth. Reducing reliance on coal, while safeguarding national energy security, is a central element of this transition. China’s energy structure remains heavily coal-dependent, significantly exceeding the global average, and the carbon emission associated with coal mining and utilization are substantially higher than those from oil and natural gas. This structural characteristic presents a formidable challenge for China in achieving its dual-carbon objectives. To meet these goals and promote green, low-carbon development, it is imperative for China to expedite the transformation of its energy structure through effective environmental regulations.

Historically, China’s environmental governance has relied heavily on command-and-control mechanisms. Although such approaches have achieved certain emission reduction outcomes, their high degree of rigidity has also generated practical challenges. For instance, the “one-size-fits-all” enforcement of certain environmental policies has adversely affected social and economic vitality, occasionally undermining firms’ incentives to innovate. Consequently, future environmental regulatory frameworks must evolve toward greater flexibility and scientific rigor, promoting sustainable economic and social development while pursuing environmental objectives.

As a market-based environmental governance instrument, Carbon emission trading policy (CETP) has demonstrated significant effectiveness in reducing emissions in developed economies and holds critical potential for China’s achievement of its carbon neutrality targets. In 2013, Shanghai pioneered China’s pilot carbon emission trading scheme, subsequently followed by Beijing, Tianjin, Guangdong, Hubei, and Chongqing. On 25 June2021, the national carbon emissions trading market was officially launched. By the end of 2023, cumulative trading volume had reached 442 million tons of carbon allowances, with a total transaction value of Renminbi 24.919 billion, reflecting a trend of expanding trading volume, rising prices, and increasingly active market participation.

Against this backdrop, this study investigates the impact of the carbon emission trading policy on the energy consumption structure and explores its underlying mechanisms. The aim is to provide theoretical insights and policy recommendations for China’s green economic transformation and high-quality development. On the one hand, the green and low-carbon transition not only mitigates carbon emissions but also reduces other pollutant emissions, contributing to improved environmental quality and enhancing public welfare. On the other hand, the low-carbon transformation of the energy structure plays an irreplaceable role in safeguarding national energy security, maintaining economic stability, and fostering sustainable development.

This paper empirically verifies the emission reduction effects of the carbon emission trading policy and examines its transmission mechanisms influencing the low-carbon transition of the energy structure. In doing so, it offers a novel perspective and empirical evidence on the effectiveness of market-based environmental regulation. The findings aim to inform the further refinement and promotion of carbon trading systems and to serve as a scientific basis for formulating and optimizing policies related to China’s green and low-carbon development agenda.

2 Literature review

As a pivotal market-based environmental regulatory instrument, Carbon Emission Trading Policy (CETP) has attracted substantial scholarly attention for its potential to facilitate the transformation of the energy structure. Existing studies primarily examine the issue from four dimensions.

2.1 Direct impacts of carbon emission trading on the energy consumption structure

International empirical studies have demonstrated that CETP directly curtails high-carbon energy consumption and optimizes the energy structure through market mechanisms. For instance, Mansanet-Bataller and Pardo (2008) found that the European Union Emissions Trading System (EU ETS) significantly reduced the share of coal in the power sector while promoting the substitufengtion of natural gas and renewable energy sources. Similarly, Murray and Maniloff (2015) showed that the U.S.Regional Greenhouse Gas Initiative (RGGI) accelerated the transition in energy structures by investing auction proceeds into energy efficiency programs. Kenneth et al. (2020) confirmed the long-term positive effects of the EU carbon market, demonstrating CETP’s capability to reduce reliance on high-carbon energy sources. Focusing on China, Jia et al. (2024) argued that CETP compels enterprises to shift from fossil fuels to non-fossil energy sources through dynamic pricing and demand-response mechanisms (Xu, 2021; Zhao et al., 2018),thus significantly optimizing the energy structure in pilot regions.

2.2 Indirect impacts of carbon emission trading via technological innovation

Technological innovation constitutes a critical transmission pathway through which CETP fosters the energy transition. While early debates questioned the innovation effect of environmental regulation (Jaffe and Palmer, 1997; Lai and Lorne, 2015; Medema, 2020), subsequent scholarship, grounded in Porter’s Hypothesis and agency theory, generally supports the view that well-designed CETP schemes incentivize breakthroughs in green technology (Schmutzler, 2001; Bhatia and Kumar Jakhar, 2021). Aghion et al. (2016) demonstrated that the EU ETS significantly boosted patent applications for solar and wind technologies,highlighting the catalytic role of carbon markets in clean energy innovation. Similarly, Fowlie and Mar, (2022) found that California’s carbon market facilitated the adoption of Carbon Capture and Storage (CCS) technologies, indirectly reducing fossil fuel dependency in the power sector. Cross-national research has further established a significant positive correlation between carbon pricing mechanisms and the diffusion of green technologies, underscoring the role of policy stability (Popp, 2019). Guo and Hu (2024), Wang et al. (2025a), and others have explored the impact of financial markets and artificial intelligence on carbon emission policies, such as blockchain, have also enhanced carbon market transparency and expedited the globalization of low-carbon innovation networks (Sipthorpe et al., 2022),creating a notable” innovation compensation effect” (Klaus et al., 2024).

2.3 Synergies between carbon emission trading and industrial restructuring

CETP can also promote the low-carbon transformation of the energy structure by driving industrial restructuring. Branger et al. (2016) highlighted that the EU ETS significantly reduced emissions from heavy industries through technological upgrades (e.g., electric arc furnace steelmaking) while encouraging the expansion of renewable energy sectors. Ohlendorf et al. (2022) showed that Germany’s carbon price floor mechanisms accelerated industrial restructuring by raising operational costs for carbon-intensive industries and incentivizing low-carbon investments. Quantitative analyses, such as Cantone et al. (2023), found that a $1 increase in carbon price leads to a 1.4%rise in low-carbon technology patent applications. Focusing on China, Herman and Jun (2022) demonstrated that CETP reallocates production factors toward low-carbon industries through quota allocation, while Zhang and Bi, (2023) emphasized the moderating role of carbon price signals in shaping corporate behavior. Moreover, policy synergies—such as integrating artificial intelligence and carbon-neutral technologies—further enhance the transition effects (Hussain et al., 2025).

2.4 Spillover effects of carbon emission trading policy

The spillover effects of CETP exhibit complex spatiotemporal dynamics. Shi and Zhou (2024) observed that CETP indirectly stimulate residential research and development activities by intensifying firms’ compliance pressures, creating a positive feedback loop of “regulation–innovation.” Conversely, Qi et al. (2022) warned that excessively stringent environmental regulations (e.g., central inspections) could suppress innovation in neighboring regions, underscoring the importance of coordinated policy design. On the international front, Cui et al. (2024) quantified the spillover effects between the China and EU carbon markets, finding that EU carbon price volatility positively influences China’s new energy sector (elasticity coefficient = 0.35). However, Jiang et al. (2023) cautioned that increasing interlinkages among global carbon markets could exacerbate price volatility risks. Wang et al. (2025a), Wang et al. (2025b) suggests re-examining the dynamic impact of carbon emissions trading on the crude oil market and argues that geopolitical and economic policy factors should be considered when discussing the volatility of carbon futures. Furthermore, spillovers are moderated by regional innovation capacities: initially hindered by technical barriers, they eventually transition into facilitation as knowledge accumulates (Li and Du, 2021; Hwang et al., 2023). Meanwhile, Guo and Hu (2024) conducted an in-depth exploration of how energy consumption rights trading and the urban innovation status effect promote ecological civilization and high-quality development. Specifically, regarding relevant policies, Tayyab et al. (2020) provided a sustainability framework for clean textile production systems (wet processing category).

2.5 Research gaps and innovations

While prior research has substantially enriched our understanding of CETP’s impacts, three critical gaps remain. First, most studies concentrate on emission reduction outcomes, with limited systematic empirical testing of CETP’s direct effects on energy structure transitions. Second, existing analyses typically isolate a single transmission pathway (e.g., technological innovation or industrial upgrading),neglecting the synergistic effects among multiple mechanisms. Third, few studies have addressed the moderating role of structural factors such as urbanization levels or regional economic disparities. Finally, although scholars have begun exploring the application of cutting-edge models in such policy evaluations—such as Guo and Hu (2024), Feng et al. (2024) employing time-varying difference-of-two-models to analyze the dynamic impact of environmental information disclosure on corporate investment value, (Moon et al., 2022; Lv and Guo 2025) employing geometric programming to optimize carbon emission production systems—the vast majority of research in this field remains methodologically constrained by traditional paradigms, with limited innovation in modeling frameworks.

This study addresses these gaps through several key innovations:

Utilizing a Difference-in-Differences (DID) model and decision tree analysis to comprehensively evaluate the net effects of CETP on China’s low-carbon energy transition. Uncovering the dual transmission pathways—scientific and technological inputs, and industrial restructuring—through mediation analysis, thereby moving beyond traditional single-path frameworks. Introducing urbanization as a critical moderating variable and integrating regional heterogeneity within the theoretical framework to better elucidate the economic-geographic mechanisms underlying policy effectiveness, thus providing theoretical guidance for the design of differentiated policies.

3 Theoretical analysis and research hypotheses

Based on Coase’s theorem, the carbon emission trading mechanism, under aggregate emissions control, commodifies carbon emission rights, thereby enabling their market trading. Through market pricing, the negative externalities associated with environmental pollution are internalized into firms’ direct costs and investment opportunity costs. Enterprises with lower abatement costs can undertake more emissions reduction activities to gain economic benefits, while those with higher abatement costs must purchase additional allowances to comply. This mechanism not only regulates emissions at the enterprise level but also exerts broader effects on regional industrial structures by promoting the transformation of high-carbon industries and encouraging the development of emerging low-carbon sectors. Simultaneously, it influences regional fiscal expenditures, motivating governments to increase investments in low-carbon industries and energy-saving technologies. Consequently, carbon emissions trading ultimately fosters the upgrading of regional energy consumption structures, reduces dependence on high-carbon energy sources, and facilitates a shift toward low-carbon, green, and sustainable development.

The theoretical framework of this study is illustrated in Figure 1.

Figure 1

Figure 1. Theoretical analysis architecture.

3.1 Impact of carbon emission trading policy on the energy consumption structure

Under the carbon emission trading pilot framework, the marginal cost of coal consumption rises significantly. Assuming economically rational behavior, market participants aim to optimize their cost-effectiveness by dynamically adjusting their energy consumption portfolios: reducing reliance on high-carbon energy sources (primarily coal) while progressively increasing the share of lower-carbon alternatives, such as oil and natural gas. Furthermore, enterprises, driven by long-term strategic considerations, expand investments in and applications of renewable energy technologies, including wind and solar power.

This behavioral adjustment leads to a gradual decline in coal’s proportion within the overall energy structure. Two internal mechanisms underlie this process:

First, the policy framework often favors enterprises that meet low-carbon standards by allocating quotas preferentially, allowing such firms to generate surplus emissions reductions and realize economic returns through trading, thus incentivizing continuous investment in low-carbon technologies.

Second, the iterative innovation and large-scale deployment of low-carbon technologies progressively squeeze the market space for high-carbon energy sources, forming a positive feedback loop of “technological innovation - cost reduction - substitution enhancement.”

Hypothesis 1. Carbon emission trading policy promote the low-carbon transition of the energy consumption structure.

3.2 Mediating effects of carbon emission trading policy

First, the institutional incentives embedded within carbon emission trading policy substantially enhance local governments’ financial support for technological research and development (R&D), thereby promoting energy system decarbonization. Numerous empirical studies suggest that once carbon markets surpass critical thresholds in trading volume, local governments’ investments in clean technology R&D experience exponential growth. Through the “emission reduction performance — financial gain” transmission mechanism, local governments optimize fiscal resource allocation structures, transforming environmental compliance pressures into intrinsic motivations for technological innovation. Specifically, governments derive sustainable financial resources from carbon market liquidity cultivation and central fiscal rewards for emission target achievements, thereby establishing a self-reinforcing investment cycle supporting low-carbon technology innovation, which ultimately underpins the low-carbon restructuring of regional energy systems. Figure 2 illustrates the specific channels through which government fiscal expenditures influence the CETP.

Figure 2

Figure 2. The impact of local government financial investment.

Hypothesis 2. Carbon emission trading policy promote the low-carbon transition of the energy structure by incentivizing local governments to increase financial investments in technological research and development.

Second, carbon emission trading policy play a pivotal role in driving industrial restructuring, serving as a key pathway to achieving carbon emission reduction targets. Through well-designed market mechanisms, CETP simultaneously suppresses the disorderly expansion of high-carbon industries and creates opportunities for the development of low-carbon, high-value-added industries, thereby accelerating the transition and rationalization of industrial structures toward low-carbon trajectories.

Hypothesis 3. Carbon emission trading policy promote the low-carbon transformation of the energy structure by facilitating industrial restructuring.

4 Research design

4.1 Model setup

To assess the impact of the carbon emission trading policy on the low-carbon transition of the energy structure, this study adopts a difference-in-differences (DID) model, treating the CETP policy pilot as a quasi-natural experiment. Given that carbon trading policies commenced pilot programs in 2013, provincial panel data from 2004 to 2021 provide both a sufficient pre-policy baseline trend and a post-policy observation window for the DID model. Furthermore, the relatively comprehensive energy balance sheets and carbon emission accounting information ensure estimation accuracy. This period also features a stable macroeconomic institutional environment and high sample homogeneity, effectively controlling for confounding factors. Furthermore, the carbon market mechanism exhibits path dependency and persistent spillover effects. Consequently, the estimation results from this period can be extrapolated beyond 2025, providing a robust historical reference and policy benchmark for quota allocation, sectoral expansion, and regionally differentiated governance under the current “dual carbon” goals. Considering data gaps and other issues, samples from Tibet and Hong Kong, Macao, and Taiwan were excluded. Consequently, data from 30 provinces (municipalities and autonomous regions) spanning 2004–2021 were ultimately selected as the sample. Among these, Beijing, Shanghai, Tianjin, Chongqing, Hubei, and Guangdong are designated as the treatment group (pilot regions), while the remaining provinces serve as the control group.

The DID model is defined as shown in Equation 1,

${Es}_{it}={\delta }_1+{\beta }_1{DID}_{it}+{\beta }_x{X}_{it}+v_t+{\mu }_i+{\epsilon }_{it}(1)$

Among them, ${DID}_{it}$ is the policy variable (${DID}_{it}={treat}_{it}\times {post}_{it}$); the explanatory variable ${es}_{it}$ is the energy structure, which is used to portray the low-carbon transition of the energy structure; ${treat}_{it}$ is the regional grouping variable, if it is the experimental region,${treat}_{it}$ takes the value of 1, otherwise, ${treat}_{it}$ takes the value of 0; ${post}_{it}$ is the policy point variable (The policy pilot program commenced in 2014.) referring to the setting of John Guo, if the policy pilot is in the first 6 months of a certain year, then from that year onwards, ${post}_{it}$ takes the value of 1; on the contrary, from the latter year onwards ${post}_{it}t$ akes the value of 1, and the year prior to the implementation of the policy takes the value of 0; is the set of control variables. In the year before the implementation of the policy,${post}_{it}$ takes the value of 0;$X_{it}$ is the set of control variables. In model (1), ${\beta }_1$ is the effect variable of policy implementation, ${\beta }_x$ is the regression coefficient of the control variables, and $v_t$ and ${\mu }_i$ are the year fixed effect and province fixed effect respectively.

The energy structure of each region is calculated on the basis of the consumption of different energy sources, and standard coal conversion factors for different energy sources. The specific equation for calculating the energy structure is shown in Equation 2.

${Es}_{it}=\frac{{energy}_{it1}\times {\gamma }_1}{\sum k{energy}_{itk}\times {\gamma }_k}(2)$

Where ${energy}_{it1}$ is the consumption of coal (10,000 tons) in region $i$ in year, ${energy}_{itk}$ is the consumption of $k$ energy in region $i$ in year$t$ , ${\gamma }_1$ is the coal conversion factor, ${\gamma }_k$ is the $k$ energy conversion factor.

4.2 Variables and data

The explanatory variable in this paper is energy structure ($es$). In this paper, the calculation of energy structure draws on relevant literature (energy structure is expressed as the share of coal’s CO₂ emissions in total emissions). The policy variable in this paper is the carbon emission trading policy pilot ($DID$).

In order to accurately analyze the policy effects, avoid omitting variables and ensure the robustness of the empirical results, the regression model introduces the following control variables: urbanization rate ($urban$), expressed as the proportion of urban population to total population. The level of economic development ($\left.levelofeconomicdevelopment\right)$, measured by the value of real per capita income in each province ($GNIpercapita$). Degree of openness to the outside world ($open$), measured by the proportion of total import and export to GDP.

In order to further analyze the transmission mechanism of the impact of carbon emission trading policy on the low-carbon transition of energy structure, this paper adopts the following mechanism variable: science and technology input ($R\&Dinvestment$), which is measured by the ratio of the local financial expenditure on science and technology to the total local financial expenditure.

The data on carbon emission and energy structure used in this paper come from the China Carbon Accounting Database ($CEAD$), and all other data are sourced from the statistical yearbooks published annually on the official website of the National Bureau of Statistics ($NBS$). The descriptive statistics of each variable are shown in Table 1.

Table 1

Table 1. Statistical description.

5 Empirical results

5.1 Baseline regression results

Table 2 presents the baseline regression results. Column (1) reports estimates without control variables, while Column (2) adds all control variables. In all specifications, region and year fixed effects are controlled, and standard errors are clustered at the provincial level.

Table 2

Table 2. Impact of carbon trading policies on the energy structure.

The key findings are:

In Column (1), the estimated coefficient on the policy interaction term (DID) is −0.079-0.079–0.079 and is statistically significant at the 1% level, indicating that carbon emission trading policy significantly promote the low-carbon transition of the energy structure.

In Column (2), after introducing control variables, the DID coefficient remains significantly negative (−0.042-0.042–0.042) at the 1% level, confirming the robustness of the baseline result.

Thus, the empirical evidence supports Hypothesis 1, verifying that CETP contribute meaningfully to optimizing China’s regional energy structures.

5.2 Robustness checks

5.2.1 Parallel trend test

A key assumption of DID analysis is the parallel trend between the treatment and control groups before policy implementation. Figure 3 depicts the dynamic effects of the policy over time.

Figure 3

Figure 3. Dynamic effects of the parallel trend hypothesis.

Prior to the policy introduction, there were no significant differences in trends between the two groups, supporting the parallel trend assumption. Post-policy, treated regions exhibit a clear divergence toward a lower-carbon energy structure compared to the control group, reinforcing the credibility of the main findings.

5.2.2 Placebo test

To further validate the results, we conducted a placebo test by randomly assigning treatment status to regions 1,000 times and re-estimating the model. Figure 4 shows that the distribution of the placebo DID coefficients centers around zero, with very few significant results, indicating that the observed treatment effects are unlikely to be driven by random factors.

Figure 4

Figure 4. Placebo test.

This evidence strengthens confidence in the causal interpretation of the policy effect.

5.3 Further analysis

In order to further explore the effect of control variables on the experimental results, the decision tree model in machine learning was used to analyze the available data.

5.3.1 Decision tree principle

The core of the decision tree algorithm lies in constructing decision trees with high accuracy and small size. A decision tree generation algorithm proposed by Breiman, the CART (classification and regression tree) algorithm, is used, the basic idea of which is to analyze the training samples consisting of feature variables and directory variables in a continuous loop and decompose them into binary tree The basic idea of the algorithm is to analyze the training samples consisting of feature variables and catalog variables and loop continuously and decompose them into the form of binary tree. The GiniIndex is a key parameter in the CART algorithm, which determines the optimal detection variables and the division threshold. Intuitively, the GiniIndex is the result obtained by selecting 2 samples from group D with different categories. The Gini coefficient expression for sample D, Gini(D), is given by Equation 3:

$Gini\left(D\right)={\displaystyle \sum _{k=1}^{\left|y\right|}}{\displaystyle \sum _{k^{\prime }\ne k}}{P}_k{P}_{k^{\prime }}=1-{\displaystyle \sum _{k=1}^{\left|y\right|}}{P_k}^2(3)$

Where: $P_k=\frac{C^k}{D}$, $D$ is the number of all samples, $C^k$ is the number of samples in the $k$ the category; $k$ is an index variable to traverse all possible categories in the dataset $D$ ; and $\left|y\right|$ denotes the total number of categories in the dataset $D$.

In CART decision tree, by calculating the Gini coefficient for each feature, the smaller the Gini value, the higher the purity of the data set is represented. The feature with the smallest Gini coefficient is selected as the division node of the decision tree. This will result in the highest purity of the sub-data set after each division, thus improving the accuracy of the decision tree. If the sample set D is partitioned according to attribute a, then the formula is as shown in Equation 4:

$Gini\_index={\displaystyle \sum _{\epsilon =1}^v}\frac{\left|D^{\epsilon }\right|}{D}Gini\left(D^{\epsilon }\right)(4)$

Where: $a$ is the attribute; $D$ is the total number of samples; $\epsilon$ is an index variable to iterate through all possible values of the attribute $a$; $v$ is the attribute $a$ of the number of different values.

The set with the lowest Gini coefficient is used as the categorical attribute, and the rest of the attributes are called recursively until all the sampled attributes in all the subdatasets are the same, or there are no more attributes to be divided within the subdatasets, and then the bifurcation of the decision tree can be stopped.

5.3.2 Model runs

After pre-processing the raw data in order to ensure that the model has enough data to be trained and its generalization performance can be effectively evaluated on unseen data, the model is trained to get the best combination of parameters, again based on the characteristics of the data set and the needs of the actual research problem of this study. 70% of the data was divided into training set and the other 30% was used as test set, using SPSS software, the results of the run are shown in Figure 5.

Figure 5

Figure 5. Decision tree model output by CART algorithm.

In Figure 5, Sample represents the number of samples under the layer; Value indicates how many samples belong to each of the three classes among the samples at the current node; Gini is the Gini value; Class is the category. The specific rules extracted from the decision tree diagram are as follows.

1. If the degree of urbanization is ranked less than or equal to 1.5, the GDP per capita is ranked less than or equal to 1.5, and the degree of openness is ranked less than or equal to 1.5, then the impact of the carbon trading policy on the energy structure of this pilot region belongs to rank 1.

2. If the rank of urbanization is less than or equal to 1.5, the rank of GDP per capita is greater than 1.5, and the rank of openness is greater than 1.5, then the impact of carbon emission trading policy on the energy structure in this pilot region belongs to rank 1.

3. If the rank of urbanization degree is greater than 1.5 and the degree of openness is greater than 1.5, the impact of carbon emission trading policy on energy structure in this pilot region belongs to rank 2.

5.3.3 Model review

In this study, Receiver Operating Characteristic (ROC) curve graphs are obtained by training result measurements. ROC curve is a common visualization used in machine learning for binary classification problems. The prediction result is used as the judgment threshold, and the Area Under the Curve (AUC) is used as the assessment index of prediction accuracy, with a range of [0, 1]. The larger the value of the AUC, and the closer the curve is to the upper-left corner, indicates the better performance of the model.

After applying the SPSS program to run, the ROC curves of the training set and test set data were obtained, as shown in Figure 6. Table 3 shows that the decision tree model has excellent performance on the training set, with an AUC index of 0.862; it also performs well on the independent test set, with an AUC value of 0.82, indicating that the model has strong classification ability for both training and test data. This indicates that the decision tree model, after rigorous training and testing, has excellent performance indicators and good classification accuracy, stability and generalization ability. Comprehensively analyzing the key result indicators, it can be judged that the decision tree classification model is highly effective.

Figure 6

Figure 6. Roc curve.

Table 3

Table 3. Model evaluation results.

5.4 Analysis of results

The CART algorithm recursively bisects the data by selecting the Gini coefficient as the feature criterion. As can be seen from this decision tree, the features selected are numerical variables, which are recursively divided using numerical bisection points. The decision tree model constructed in this study has 4 layers and 13 nodes.

The analysis of factors influencing the effects of carbon emission trading policy based on the decision tree model shows that the degree of regional urbanization presents a significant core explanatory power in the system of explanatory variables (characteristic importance 76.9%). By setting an urbanization class division threshold of 1.5 (Gini coefficient of 0.416), the sample is effectively differentiated into two groups (230 high-urbanization samples versus 138 low-urbanization samples), and the high-urbanization group demonstrates a more significant effect of carbon emission trading policy implementation. It is worth noting that although the second level decision node reveals the auxiliary explanatory role of regional GDP per capita and openness to the outside world, its explanatory effectiveness decreases by 61.9%–69.9% compared with the degree of urbanization, which confirms the dominant role of the urbanization level in the mechanism of the policy effect from the structural level of the model.

It is speculated that the strengthening mechanism of urbanization degree on the effect of carbon emission trading policy mainly stems from the scale effect and factor agglomeration, and the optimization of the institutional environment in two dimensions. Highly urbanized areas form economies of scale through industrial concentration and population agglomeration, which reduces the unit cost of promoting low-carbon technologies, and at the same time, the completeness of the infrastructure such as smart grids and carbon capture is greatly improved compared with that of the low-urbanized areas, which provides a physical At the same time, the government regulatory system in highly urbanized areas tends to be more perfect and strict, which will effectively curb policy arbitrage.

6 Conduction mechanism test

Based on the normative framework of mediation effect analysis, this study selects industrial structure and Science and Technology (S&T) input as the core transmission mechanism variables for empirical testing. Among them, industrial structure is characterized by the degree of high industrial structure ($is$), and S&T investment ($R\&Dinvestment$) is measured by the ratio of local fiscal expenditure on S&T to total local fiscal expenditure. Although the stepwise regression method (i.e., Wen Zhonglin’s three-step method) of Baron and Kenny is commonly used for traditional mediation effect tests, this method has been questioned by academics due to the possible endogeneity of the key equations (e.g., the correlation between the mediating variable and the error term is not dealt with). To overcome this measurement shortcoming, this study draws on the causal mediation analysis framework of Imai et al. and Mechanism testing was conducted using an improved two-stage model, with the econometric model system constructed as shown in Equations 5, 6:

${Es}_{it}={\delta }_2+{\beta }_2{DID}_{it}+{\beta }_x{X}_{it}+v_t+{\mu }_i+{\epsilon }_{it}(5)$

${Es}_{it}={\delta }_3+{\gamma }_1{DID}_{it}+{\alpha }_1{\theta }_1{Med}_{it}+{\beta }_x{X}_{it}+v_t+{\mu }_i+{\epsilon }_{it}(6)$

Where ${Med}_{it}$ is the mechanism variable (i.e., $is$ and $R\&D$), and the rest of the variable definitions remain consistent with the baseline model. Considering the strict assumption of normal distribution by the Sobel test and its inherent limitation of insufficient statistical power, this study employs the nonparametric $bootstrap$ method for robustness inference. Specifically, 95% deviation-corrected confidence intervals for the product term of ${\alpha }_1{\theta }_1$ were constructed by repeated sampling with put-back 1,000 times, and a significant mediation effect was determined to hold if the confidence interval did not contain a zero value. The test results in Table 4 indicate that the transmission mechanism passes the significance test, and the specific statistical inference process is detailed in the subsequent analysis.

Table 4

Table 4. Scientific and technological inputs (1,000 times).

6.1 Scientific and technological input

The results of the mediation effect analysis based on 1,000 $bootstrap$ samples in Table 4 indicate that science and technology inputs ($R\&Dinvestment$), as a mediating variable, play a significant mediating role between carbon emission trading policy and low-carbon transformation of regional energy structure. Specifically, the estimated value of the indirect effect (i.e., the mediating effect) is −0.104, and its 95% confidence interval lies completely below zero and does not cross the zero point, which indicates that the mediating effect is significant. This result reveals that the carbon emission trading policy promotes the optimization of the regional energy structure and low-carbon transition by promoting scientific and technological inputs, which in turn promotes the optimization of the regional energy structure and low-carbon transition. Further analyzing the direct effect, its estimated value is −0.136, and the 95% confidence interval is also completely below zero, without crossing the zero point, which indicates that the direct effect is also significant. In addition, the mediating effect accounts for 43.1% of the total effect, which indicates that the effect is an incomplete mediating effect. This finding verifies hypothesis H2, that is, science and technology inputs have a significant mediating role between carbon emission trading policy and low-carbon transformation of regional energy structure.

6.2 Industrial structure

Table 5 demonstrates the results of mediating effect analysis based on 1,000 times $bootstrap$ ampling. In this study, industrial structure ($is$) is set as the mediating variable to explore the mechanism of the impact of carbon emission trading policy on the low-carbon transition of regional energy structure. Specifically, the estimated value of the indirect effect (in this paper, the mediating effect) is −0.079, and its 95% confidence intervals are all negative (without passing through 0), which indicates that the mediating effect significantly exists. This result reveals that the carbon emission trading policy promotes the optimization of the regional industrial structure by promoting the low-carbon transformation of the regional energy structure. Further analyzing the direct effect, its estimated value is −0.1613, and the 95% confidence intervals are all negative as well, indicating that the direct effect is significant. In addition, the mediating effect accounts for 32.9% of the total effect, which indicates that the effect is an incomplete mediating effect. This finding verifies Hypothesis H3, that is, industrial restructuring has a significant mediating role between carbon emission trading policy and low-carbon transformation of regional energy structure.

Table 5

Table 5. Industry structure (1,000 times).

7 Heterogeneity test

7.1 Regional heterogeneity

The results of the benchmark regression model indicate that the carbon emission trading policy has generally contributed significantly to the low-carbon transition of the regional energy structure. However, given the large differences in the carbon emissions trading markets in the pilot regions, the impact of the policy on the low-carbon transition of the energy structure in the pilot regions may be heterogeneous. Therefore, it is necessary to analyze the heterogeneity of the impact of the carbon emission trading policy in each pilot region. Table 6 demonstrates the regression results of each pilot province and city. The analysis results show that the policy effect of Beijing and Tianjin is significantly negative, which indicates that in these two regions, the carbon emission trading policy can effectively promote the low-carbon transition of the energy structure, and the policy effect of Beijing is the most significant; the policy effect of Shanghai, Hubei, and Chongqing is negative but not significant, which means that in these regions, the carbon emission trading policy does not have a significant effect on the promotion of low-carbon transition of the energy structure; and Guangdong’s policy effect is positive, indicating that in this region, the carbon emission trading policy fails to significantly promote the low-carbon transition of energy structure. To summarize, the policy effect of carbon emission trading in Beijing and Tianjin is significant, while the policy effect in other provinces and cities is not obvious. Table 6 provides specific parameters.

Table 6

Table 6. Regression results by pilot provinces and cities.

7.2 Heterogeneity analysis of the municipality sub-sample and the province sub-sample

According to administrative divisions, the pilot areas of carbon emission trading policy can be categorized into four municipalities (Beijing Municipality, Tianjin Municipality, Shanghai Municipality and Chongqing Municipality) and two economically powerful provinces (Hubei Province and Guangdong Province). This study further examines whether there are significant differences in the impacts of the carbon emission trading policy on the low-carbon transformation of the energy structure of the above four municipalities and two economic powerhouse provinces. Table 7 demonstrates the results of the regression analysis for the municipality sub-sample and the province sub-sample, where column (1) shows the regression results for the municipality sub-sample and column (2) shows the regression results for the province sub-sample.

Table 7

Table 7. Regression results of directly governed municipalities and economically strong provinces.

The results of the empirical analysis show that the regression coefficients of the carbon emission trading policy are negative for both municipalities directly under the central government and pilot provinces. Specifically, the carbon emission trading policy in municipalities can significantly promote the low-carbon transformation of the regional energy structure; in contrast, the promotion effect of the carbon emission trading policy in provinces is not significant. This suggests that the carbon emission trading policy in municipalities has a more obvious role in promoting the low-carbon transition of energy structure, while the policy effect in provinces is relatively weak. Table 7 provides specific parameters.

8 Policy implications

To effectively advance carbon emission trading policy and facilitate the transition to a low-carbon energy structure, this paper proposes the following recommendations:

First.Deepen Market-Oriented Reforms.

Utilize carbon emission trading policy pilot programs to strengthen market mechanisms, impose strict controls on the expansion of high-energy-consumption and high-emission projects, and accelerate the phase-out of outdated production capacity to support a green economic transition.

Second.Promote Green Technology Innovation.

Provide tax incentives and R&D subsidies to encourage corporate investment in green technology. Enhance budget autonomy for universities and research institutions to stimulate innovation and support the long-term achievement of dual carbon goals.

Third.Optimize Regional Policy Design.

Account for regional differences in economic development and industrial structure when designing policies. Highly urbanized regions should pioneer reforms such as quota auctions and carbon derivatives pilots to leverage urbanization-market synergies.

Fourth.Enhance Public Awareness and Participation.

Strengthen low-carbon education and publicity to boost public understanding and involvement in energy transition, fostering multi-stakeholder governance for green development.

9 Conclusion

Based on panel data from 30 provinces (municipalities and autonomous regions) in China from 2004 to 2021, this study employs a difference-in-differences model and decision tree model to systematically evaluate the impact of carbon emission trading policies on carbon emissions and energy structure transformation, while delving into their underlying mechanisms. The findings reveal:

1. Carbon emission trading policy yield significant carbon reduction and energy structure optimization effects. This policy not only effectively lowers carbon emissions but also drives regional energy structures toward low-carbon transformation. This conclusion remains valid after incorporating control variables and undergoing a series of robustness tests, providing solid empirical evidence for advancing the national carbon market and deepening market-based reforms.

2. Technological investment serves as a critical pathway. By incentivizing local governments to increase scientific and technological expenditures, the carbon emission trading policy accelerates energy structure transformation through promoting green technological innovation. This further validates the necessity of the earlier recommendation to “promote green technological innovation,” demonstrating that enhancing research investment and refining incentive mechanisms play a central supporting role in achieving the “dual carbon” goals.

3. Industrial restructuring plays a vital intermediary role. By curbing high-carbon industries and fostering low-carbon sectors, the policy optimizes industrial structures, indirectly advancing low-carbon energy transitions. Findings support the policy recommendation to “strictly restrict the blind expansion of high-energy-consuming and high-emission projects,” highlighting industrial restructuring as the pivotal link between policy regulation and energy transformation.

4. Regional development disparities influence policy outcomes. Decision tree analysis reveals that urbanization levels are a key factor in policy effectiveness differentiation, reflecting that China’s uneven regional development remains a critical context for policy implementation. This provides theoretical support for “optimizing regional policy deployment and implementing differentiated policy designs,” while also indicating that regional development stages and urbanization levels must be fully considered when advancing deep reforms such as quota auctions and carbon finance.

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

ZZ: Writing – original draft, Writing – review and editing. YL: Formal analysis, Writing – review and editing. YC: Conceptualization, Writing – review and editing. YW: Funding acquisition, Writing – review and editing. BG: Data curation, Writing – review and editing.

Funding

The author(s) declare that no financial support was received for the research and/or publication of this article.

Conflict of interest

The authors declare that the research 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) declare that no Generative AI was 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.

References

Aghion, P., Dechezlepretre, A., Hemous, D., Martin, R., Van Reenen, J., Taxes, C., et al. (2016). Carbon taxes, path dependency, and directed technical change: evidence from the auto industry. J. Political Econ. 124 (1), 1–51. doi:10.1086/684581

CrossRef Full Text | Google Scholar

Bhatia, M. S., and Kumar Jakhar, S. (2021). The effect of environmental regulations, top management commitment, and organizational learning on green product innovation: evidence from automobile industry. Bus. Strategy Environ. 30 (8), 3907–3918. doi:10.1002/bse.2848

CrossRef Full Text | Google Scholar

Branger, F., Quirion, P., and Chevallier, J. (2016). Carbon leakage and competitiveness of cement and steel industries under the Eu ets: much ado about nothing. Energy J. 37 (3), 109–136. doi:10.5547/01956574.37.3.fbra

CrossRef Full Text | Google Scholar

Cantone, B., Evans, D., and Reeson, A. (2023). The effect of carbon price on low carbon innovation. Sci. Rep. 13 (1), 9525. doi:10.1038/s41598-023-36750-9

PubMed Abstract | CrossRef Full Text | Google Scholar

Cui, H., Fan, Y., and Juchao, Li. (2024). Research on the risk spillover effect of carbon emission trading market and various industry markets in China and the european union. Energy Sci. and Eng. 12 (10), 4680–4702. doi:10.1002/ese3.1907

CrossRef Full Text | Google Scholar

Feng, Y., Guo, B. N., Wang, X., and Hu, F. (2024). Facilitating or inhibiting? The impact of environmental information disclosure on enterprise investment value. Environ. Sci. Pollut. Res. 31 (5), 7793–7805. doi:10.1007/s11356-023-31583-5

PubMed Abstract | CrossRef Full Text | Google Scholar

Fowlie, M. L., and Mar, R. (2022). Mitigating emissions leakage in incomplete carbon markets. J. Assoc. Environ. Resour. Econ. 9 (2), 307–343. doi:10.1086/716765

CrossRef Full Text | Google Scholar

Guo, B. N., and Hu, P. J. (2024). Investigating the relationship between energy-consuming rights trading and urban innovation quality. Sustain. Dev. 32 (4), 3248–3270. doi:10.1002/sd.2820

CrossRef Full Text | Google Scholar

Herman, K. S., and Jun, X. (2022). Channeled through trade: how foreign environmental regulations induce domestic renewable energy innovation. Energy Res. and Soc. Sci. 89, 102629. doi:10.1016/j.erss.2022.102629

CrossRef Full Text | Google Scholar

Hussain, J., Lee, C. C., and Lev, B. (2025). Optimizing Ai-Based emission reduction efficiency and subsidies in supply chain management: a sensitivity-based approach with duopoly game dynamics. J. Clean. Prod. 494, 144991. doi:10.1016/j.jclepro.2025.144991

CrossRef Full Text | Google Scholar

Hwang, Q., Yao, M., Li, S. G., Wang, F., Luo, Z. M., Li, Z., et al. (2023). Risk spillovers between China's carbon and energy markets. Energies 16 (19), 6820. doi:10.3390/en16196820

CrossRef Full Text | Google Scholar

Jaffe, A. B., and Palmer, K. (1997). Environmental regulation and innovation: a panel data study. Rev. Econ. and Statistics 79, 610–619. doi:10.1162/003465397557196

CrossRef Full Text | Google Scholar

Jia, S. B., Zhu, X. W., Gao, X., and Yang, X. T. (2024). The influence of carbon emission trading on the optimization of regional energy structure. Heliyon 10 (11), e31706. doi:10.1016/j.heliyon.2024.e31706

PubMed Abstract | CrossRef Full Text | Google Scholar

Jiang, H., Wang, J., Yu, S., Zheng, X., and Wang, C. (2023). Exploring global carbon market link mechanism: efficiency evaluation in the context of carbon neutrality. J. Clean. Prod. 420, 138474. doi:10.1016/j.jclepro.2023.138474

CrossRef Full Text | Google Scholar

Kenneth, B., Marten, O., and Erik, D. (2020). The long-term impact of the market stability reserve on the Eu emission trading system. Energy Econ. 89, 104746. doi:10.1016/j.eneco.2020.104746

CrossRef Full Text | Google Scholar

Klaus, G., Florian, S., and Thomas, W. (2024). Environmental policies and directed technological change. J. Environ. Econ. Manag. 124, 102916. doi:10.1016/j.jeem.2023.102916

CrossRef Full Text | Google Scholar

Lai, L. W. C., and Lorne, F. T. (2015). The fourth coase theorem: state planning rules and spontaneity in action. Plan. Theory 14 (1), 44–69. doi:10.1177/1473095213510430

CrossRef Full Text | Google Scholar

Li, J., and Du, Y. X. (2021). Spatial effect of environmental regulation on green innovation efficiency: evidence from prefectural-level cities in China. J. Clean. Prod. 286, 125032. doi:10.1016/j.jclepro.2020.125032

CrossRef Full Text | Google Scholar

Lv, L., and Guo, B. N. (2025). Do pilot zones for green finance reform and innovation policy enhance China's energy resilience? Sustainability 17 (13), 5757. doi:10.3390/su17135757

CrossRef Full Text | Google Scholar

Mansanet-Bataller, M., and Pardo, A. (2008). What you should know about carbon markets. Energies 1 (3), 120–153. doi:10.3390/en1030120

CrossRef Full Text | Google Scholar

Medema, S. G. (2020). The coase theorem at sixty. J. Econ. Literature 58 (4), 1045–1128. doi:10.1257/jel.20191060

CrossRef Full Text | Google Scholar

Moon, I., Yun, W. Y., and Sarkar, B. (2022). Effects of variable setup cost, reliability, and production costs under controlled carbon emissions in a reliable production system. Eur. J. Industrial Eng. 16 (4), 371–397. doi:10.1504/ejie.2022.123748

CrossRef Full Text | Google Scholar

Murray, B. C., and Maniloff, P. T. (2015). Why have greenhouse emissions in Rggi states declined? An econometric attribution to economic, energy market, and policy factors. Energy Econ. 51, 581–589. doi:10.1016/j.eneco.2015.07.013

CrossRef Full Text | Google Scholar

Ohlendorf, N., Flachsland, C., Nemet, G. F., and Steckel, J. C. (2022). Carbon price floors and low-carbon investment: a survey of German firms. Energy Policy 169, 113187. doi:10.1016/j.enpol.2022.113187

CrossRef Full Text | Google Scholar

Popp, D. (2019). Environmental policy and innovation: a decade of research. Cambridge, United States: National Bureau of Economic Research. doi:10.3386/w25631

CrossRef Full Text | Google Scholar

Qi, Y., Bai, T., and Tang, Y. (2022). Central environmental protection inspection and green technology innovation: empirical analysis based on the mechanism and spatial spillover effects. Environ. Sci. Pollut. Res. 29 (57), 86616–86633. doi:10.1007/s11356-022-21833-3

PubMed Abstract | CrossRef Full Text | Google Scholar

Schmutzler, A. (2001). Environmental regulations and managerial myopia. Environ. Resour. Econ. 18 (1), 87–100. doi:10.1023/A:1011113106055

CrossRef Full Text | Google Scholar

Shi, H., and Zhou, Q. (2024). Government R&D subsidies, environmental regulation and corporate green innovation performance. Finance Res. Lett. 69, 106088. doi:10.1016/j.frl.2024.106088

CrossRef Full Text | Google Scholar

Sipthorpe, A., Brink, S., Van Leeuwen, T., and Staffell, I. (2022). Blockchain solutions for carbon markets are nearing maturity. One Earth 5 (7), 779–791. doi:10.1016/j.oneear.2022.06.004

CrossRef Full Text | Google Scholar

Tayyab, M., Jemai, J., Lim, H., and Sarkar, B. (2020). A sustainable development framework for a cleaner multi-item multi-stage textile production system with a process improvement initiative. J. Clean. Prod. 246, 119055. doi:10.1016/j.jclepro.2019.119055

CrossRef Full Text | Google Scholar

Wang, X. Q., Jin, W. X., and Stan, S. E. (2025a). Empowering energy transition: revisiting the dynamic impacts of carbon emissions trading and the crude oil market. Econ. Analysis Policy 86, 988–1001. doi:10.1016/j.eap.2025.04.020

CrossRef Full Text | Google Scholar

Wang, X. Q., Jin, W. X., Xu, B. C., and Wang, K. H. (2025b). Volatility in carbon futures amid uncertainties: considering geopolitical and economic policy factors. J. Futur. Mark. 45 (4), 308–325. doi:10.1002/fut.22565

CrossRef Full Text | Google Scholar

Xu, W. Z. (2021). The impact and influencing path of the pilot carbon emission trading Market--Evidence from China. Front. Environ. Sci. 9, 787655. doi:10.3389/fenvs.2021.787655

CrossRef Full Text | Google Scholar

Zhang, G., and Bi, S. L. (2023). Inhibition or promotion: the impact of carbon emission trading on market structure: evidence from China. Front. Energy Res. 11, 1238416. doi:10.3389/fenrg.2023.1238416

CrossRef Full Text | Google Scholar

Zhao, X. X., Zhang, Y. Z., Liang, J., Li, Y. B., Jia, R. D., and Wang, L. (2018). The sustainable development of the economic-energy-environment (3e) system under the carbon trading (Ct) mechanism: a Chinese case. Sustainability 10 (1), 98. doi:10.3390/su10010098

CrossRef Full Text | Google Scholar