The joint impact of the carbon market on carbon emissions, energy mix, and copollutants

From a comprehensive standpoint, this paper investigates whether and to what extent the carbon market functions in the context of the developing world. Taking advantage of a unique seven-year-plant-level panel dataset (2010–2016) on Chinese power plants, we use a matched difference-in-differences strategy to identify the joint impact of China’s carbon emissions trading (CET) pilot policy on carbon emissions reduction (objective), energy mix improvement (mechanism), and air copollutant reduction (cobenefits). We find that China’s CET pilot policy effectively lowered carbon emissions by approximately 38.61%. Further analysis shows that plants reduce carbon emissions primarily by reducing coal consumption (approximately 30.79%). Most importantly, China’s CET pilot policy induces substantial air copollutant abatement benefits by reducing sulfur dioxide and nitrogen oxides by approximately 52.19% and 48.62%, respectively. State-owned plants are more affected by China’s CET pilot policy, and the policy effects show disproportionate environmental inequality. Furthermore, the effects are not affected by the rate-based allowance allocation structure that is adopted by China’s national carbon market.


Introduction
Climate change is one of the world's most pressing issues, prompting many countries to implement a variety of low-carbon policies. As one of the most important climate policies in the world, an increasing number of emissions trading schemes (ETSs) have been implemented at the regional, national, and subnational levels. By enabling the redistribution of emission burdens among agents by establishing a market for the allowance, the ETS is merited by economists as a more cost-effective option than command-and-control (CAC) regulation in emissions abatement (Montgomery 1972, Hahn and Stavins 1992, Carlson et al 2000. This tool has grown in popularity in recent years, with ETSs covering nearly 17% of global greenhouse gas (GHG) emissions by early 2022 (ICAP 2022). Faced with rising energy demand as a result of economic growth, China has been attempting to use this market-based tool to meet its commitment to reduce GHG emissions more cost-effectively (Han et al 2012).
China recently established the world's largest carbon emissions trading (CET) market, surpassing the European Union Emissions Trading System (EU-ETS) and accounting for approximately 40% of China's carbon emissions. Because China's CET is the primary policy tool for achieving its ambitious targets (carbon peak by 2030, and carbon neutrality by 2060) and serves as a model for other indecisive CETs, the question of whether China's CET is an effective policy tool has moved to the forefront of the debate. There is some doubt about whether market-based tools such as CET will work in China, with unique political-economic context and institutional background (Lo 2013(Lo , 2016. Therefore, despite the proliferation of regional carbon markets, China's practice may not easily translate into reduced emissions.
Several potential barriers to establishing emissions trading markets in China have been discussed in the literature, including a lack of market thickness due to insufficient supply and demand, administrative intervention limiting the function of a free market, incomplete regulatory infrastructure for powerful enforcement and credible credit verification, and so on (Lo 2013, 2016, Munnings et al 2016. Notwithstanding the fact that two recent studies examine the effects of China's CET on carbon emissions or coal consumption , Cui et al 2021, they fail to account for the CET pilot coverage thresholds, and the causal evidence that informs our understanding of this debate remains insufficient. More importantly, there is a lack of causal evidence in the literature for a more comprehensive assessment of China's CET, particularly the impact of synergistic air pollutant emissions. This paper aims to close these research gaps. Theoretically, CET could reduce GHG emissions by reducing fossil fuel combustion, or it could even reduce air copollutants, because fossil fuels are major sources of multiple health-damaging air pollutants, such as sulfur dioxide (SO 2 ) and nitrogen oxides (NO x ). As a result, it is expected that the implementation of a CET market will result in air quality cobenefits (synergistic/ancillary benefits or positive externalities) and thus improve public health (Tietenberg 2006, Muller 2012, West et al 2013, Thompson et al 2014, Li et al 2018. Such cobenefits can be substantial enough to offset a portion or even the entire compliance cost (Cifuentes et al 2001a, 2001b, Burtraw et al 2003, Li et al 2018, and are sometimes greater than those obtained by combining sector-specific CAC regulations with more compliance flexibility (Thompson et al 2014). This makes CET more appealing to developing countries seeking cost-effective solutions to these intertwined environmental problems while minimizing the impact on economic development. The CET tool becomes especially appealing for countries obsessed with both intensive GHG emissions and critical air quality issues, such as China (Cai et al 2016), because marketbased tools can, in some circumstances, deliver more significant benefits for public health and environmental protection due to their appealing compliance flexibility and cost-effectiveness.
We focus on the power sector for several reasons. First, it is not only a large carbon emitter 3 but also contributes significantly to the decarbonization of other sectors. Energy conservation and emission reductions in the power sector can help to reduce emissions in other sectors, making it critical for China to meet its carbon-peak and carbon-neutral targets. Second, although the seven pilots independently established their local carbon markets and determined the regulation rules of industry enterprises (Zhang et al 2017), the thermal power industry is covered by all seven of China's CET pilots. Since we mainly intend to identify the average treatment effects on the treated of China's CET pilot policy, inconsistency in coverage would result in estimate bias. Third, when compared to other industries in China, the power industry has relatively better data richness and accuracy. Finally, the power industry has been chosen as the first and only sector to fully participate in China's national carbon market (Goulder et al 2017, Jotzo et al 2018 and it deserves more attention from both academicians and policy-makers.
Exploiting China's CET market as a quasiexperiment 4 , we detect the comprehensive effects of CET on carbon emissions, energy mix, and air copollutant emissions. By comparing the outcome variables for plants in pilot regions to those for nonpilots before and after the CET pilot policy implementation, the policy effects can be identified. To estimate the causal impact of CET, we use econometrically adjusted ex-post observed plant-level emissions that were not regulated by CET pilot regulation over the same period, rather than relying on ex-ante expectations about what aggregate emissions trajectories would have been in the absence of CET. We specifically take advantage of the distinct characteristics of regulated CET pilots and construct a better counterfactual or benchmark.
Using confidential plant-level data from a 7 year panel (derived from China's Environmental Statistics Database, CESD), we find that, in contrast to the EU-ETS trial period (Phase I, 2005(Phase I, -2007, China's CET pilot induced a significant (38.61%) carbon reduction. Furthermore, policy effects are primarily driven by reduced coal consumption. More importantly, we find that, interestingly, the plants involved in this original market-based cost-effective regulation decrease SO 2 and NO x by 52.19% and 48.62%, respectively. To summarize, we contend that China's CET market achieves two goals with a single action. These findings withstand a series of robustness tests on the identifying assumption or other econometric concerns.
In addition, we conducted extensive heterogeneous analyses from the perspective of plant-level characteristics and spatial distributions. First, given the plant-level scale effect, we examine whether the pilot policy has a different effect depending on plant size. During our study period, we found no significant difference between smaller and larger plants. Second, given China's unique circumstances, our heterogeneity analysis indicates that the significant effect of the CET pilot policy is primarily due to stateowned and top-five power corporations 5 . Third, we investigate the heterogeneous characteristics of spatial distributions in light of the sharing of development results. Our findings suggest that there is disproportionate environmental inequality in the context of China's CET pilot. Finally, because the rate-based allowance allocation structure is primarily used in the national carbon market, we further investigate the heterogeneity between the rate-and mass-based methods. However, we find no correlation between the policy effects and the rate-based structure.
Our research sheds new light on the carbon mitigation and energy mix improvement benefits, as well as the often overlooked air quality cobenefits anticipated to be achieved through CET. To begin, this is, to the best of our knowledge, the first comprehensive study that provides plant-level empirical evidence on the joint impact of CET on carbon emissions, energy mix, and copollutants. Substantial simulation works have shown significant improvement in carbon emissions, energy mix, air quality, or human health benefits realized by this market-based tool both globally and regionally, but empirical evidence is still lacking. Second, we present new evidence on the understudied cobenefits of China's CET pilot. Exploring the synergistic effects of climate policy on air pollution and human health is critical for climate policy evaluation. When these synergies are overlooked, they can lead to unintended welfare distortions in policy design and have an impact on the social acceptability of climate policies. Third, our empirical findings can provide not only experience and policy implications for China's national carbon market construction but also a policy option for other developing countries exploring low-carbon development paths. Finally, by carefully addressing the endogeneity problem associated with decarbonization regulations, this paper contributes to the literature on the evaluation of climate policies. Meanwhile, our use of plant-level microdata from a developing country (China) complements the existing relevant studies from developed countries particularly Europe and the United States.

Strategy for selecting control groups
Previous studies on the plant-level effects of China's CET pilot policy are conditional on, and highly sensitive to, contentious assumptions about the, e.g. emissions and performance of a plant would have been in the absence of the program. In this study, we use the unique design features (i.e. inclusion thresholds) of China's CET pilot program to build more tenable and transparent estimates of counterfactual emissions. In this section, we will discuss our identification strategy and empirical framework in detail.
China's CET pilot policy was in effect during our study period and applies to a subset of the plants we observed (detailed in section 1 of SI). The regional and temporal variations in CET pilot policy adoption naturally motivate us to run a classic difference-in-differences (DiDs) model to estimate a plant-level regression (Fowlie et al 2012, Calel andDechezleprêtre 2016). The treatment groups include plants regulated by the CET pilot policy, while the control groups include plants not regulated by the CET pilot policy.
We identify thermal power plants by using the plant's unique industry code. They are identified in our dataset by the four-digit industry code '4411' according to the Classification and Code of National Economy Industry (GB/4754-2011). As a result, the thermal power plants in the pilot regions comprise our treatment group. The caveat is that some current studies simply identify regulated plants based on pilot regions, ignoring the fact that not all plants in pilot regions are regulated, resulting in skewed estimates (e.g. Zhu et al (2020)). To identify the causal effect, however, a proper counterfactual that is observably similar to the treatment group must be constructed. Whether conducting research at the province/ city level (e.g. (Hu et al 2020)) or plant level (e.g. Cui et al (2021)), several previous empirical papers on China's CET pilot policy primarily constructed the counterfactual using a matching approach. These studies primarily employ propensity score matching (PSM) or coarsened exact matching (CEM) methods. However, whether they use the PSM or CEM matching technique, they fail to account for the CET pilot coverage thresholds (table S1), which is a critical part of developing a proper counterfactual. Because the logic behind China's carbon trading policy pilot (illustrated in figure S2) is that seven pilot regions are assigned first, and then each region assigns its own inclusion thresholds based on the carbon emissions or energy consumption levels of firms within that region, as well as other regulating rules.
Considering this, our matching counterfactual strategy is built in two steps based on the plant inclusion thresholds of the CET pilots. First, we choose regions that are sufficiently similar to the pilot regions to serve as control group regions. Second, within each control group region, we apply the same plant inclusion threshold as that applied in the corresponding pilot region, and control groups are formed by power plants that exceed the thresholds. Specifically, we chose four indicators that characterize socioeconomic development, industrial structure, and carbon emissions (detailed in table S2) to compare the differences between all other regions in China and the regions in which the treatment groups are located.
A generally comparable between treatment and control units is required for reliable causal inference. Hence, regions that are sufficiently similar to the regions in which the treatment groups are located are then chosen to be the regions from which the control groups are selected (two in our baseline model and three for the robustness check). In addition, our selection process ensures that the control and treatment group locations are in the same economic region of China.
Treatment and control units must be broadly comparable for reliable causal inference. As a result, regions that are sufficiently similar to the locations of the treatment groups are chosen to be the regions from which the control groups are drawn (two in our baseline model and three for the robustness check). Furthermore, our selection procedure ensures that the control and treatment groups are located in the same economic region of China. Three pilot CET regions, in particular, are megacities with very different political, social, and economic development than the rest of China, making it difficult to match them with sufficiently similar control group regions. Nevertheless, we argue that this has no effect on our estimates because we perform a robustness check (SI section 7.7) on the national sample with all seven pilot CET-regulated power plants as treatment groups (using the PSM-DiD estimation method) to demonstrate that the baseline results are not sensitive to the matching strategy (Zhu et al 2022). Finally, figure S3 depicts the geographical distribution of the control and treatment groups in the baseline model.

Difference-in-differences (DiDs) specification
Then, with the control groups well matched, we can compare our outcomes of interest in CETregulated plants before and after the implementation of the CET pilot policy to the corresponding changes in non-CET-regulated plants over the same period. To accomplish this, we estimate the following specification, where i, p, and t denote the plant, province, and year, respectively. The dependent variable Y o ipt for plant i in province p in year t consists of three strands of outcomes, namely, carbon emissions (comprising carbon emissions and carbon intensity, in logarithmic form), energy mix (comprising coal consumption, total energy consumption, coal consumption ratio, in logarithmic form) 6 , and copollutant emissions (comprising sulfur dioxide and nitrogen oxide emissions, in logarithmic form), which are represented by o equal to 1, 2, 3. More specifically, CO 2 emissions are calculated at the plant-year level by multiplying the five major energy consumption factors by their respective emission factors (as shown in table S4) and then adding the results. CO 2 intensity equals the calculated plant-year CO 2 emissions divided by the output values, where the output values are obtained directly from the CESD. Total energy consumption is computed in the same manner as carbon emissions. The coal consumption ratio is calculated by dividing coal consumption by total energy consumption.
CET i is a policy dummy variable that is set to 1 if plant i is regulated by the CET pilot and 0 otherwise. Post t denotes the time dummy variable. As shown in table S1, all seven regional pilot carbon markets began trading in the second half of 2013 (and almost all near the end) or the first half of 2014. Given this, as well as the lag in policy effects, we set 2014 as the uniform policy shock point in our plant-year data structure (Guo et al 2019, Tan et al 2022 . As a result, Post t equals 1 for all years after 2013 and 0 otherwise. CET i × Post t is the interaction term between CET i and Post t , whose coefficient β o 1 (o ∈ {1, 2, 3}) captures the average differential change in three effects (i.e. carbon reduction, energy mix improvement, and air pollution reduction cobenefits) in the regulated plants of the CET pilots relative to those in the control group during the policy period. More specifically, if β 1 1 is significantly negative, we can infer that the policy was effective at reducing carbon emissions, which was the starting point and the ultimate goal of China's CET pilot policy. To determine the underlying mechanism, we can infer the plant-level energy mix improvement strategy if β 2 1 is significantly negative; finally, if coefficient β 3 1 is significantly negative, we can infer that the pilot policy resulted in cobenefits of reduced air emissions.
We take advantage of the panel-data nature of our dataset to include both plant fixed effects (γ o i ) and year fixed effects (λ o t ) in our DiD specification. Specifically, plant fixed effects γ o i can control for all unobserved plant-specific time-invariant characteristics that influence the dependent variables, whereas year fixed effects λ o t can control for all general macroeconomic factors affecting all plants over the year. 6 To rule out other possible explanations, such as increased energy/production efficiency, we estimate the model in section 3.2 with energy efficiency (total energy consumption divided by power generation) as the dependent variable. 7 It is worth noting that a few studies that use 2011 as policy shock time (e.g. Cui et al (2018)  The CET policy, in particular, varies from province to province. Hence, we further include a set of provinceyear interactions (ω o pt ) to account for other unobserved province-specific time trends (e.g. provincial time-variant shocks and pressure). To account for potential heteroscedasticity and serial correlation, a random error term ε o it is included, which is clustered at the plant level.

Data
We compile a unique dataset at the plant level in China from 2010 to 2016, primarily derived from three major sources. First, the primary plant-level panel data on energy consumption and pollution are obtained from CESD, which is maintained by Ministry of Ecological Environment of the People's Republic of China. Second, the regulatory status and detailed rules of China's seven CET pilots are provided by provincial or municipal Development and Reform Commissions, each of which regulates its own local carbon market. We manually extract the list of all plants covered in the CET pilots. Finally, the city-level socioeconomic and demographic data are primarily derived from the China City Statistic Yearbook (2011)(2012)(2013)(2014)(2015)(2016)(2017). Table 1 lists a brief description of key variables, its acronyms we use in our analysis, main summary statistics, as well as the number of observations for both the treatment and matched control groups. SI sections 3 goes into great detail about data processing and descriptive statistics analysis. Parallel trend tests for our DiD models are reported in SI section 4.

Objectives: have carbon emissions been reduced?
The primary goal of the carbon market is to reduce carbon emissions. As a result, our first effectiveness assessment is to determine whether and to what extent the CET policy has reduced firm-level carbon emissions, which is also the issue of greatest concern for policy-makers in China and other developing countries planning to implement a national or subnational CET policy. Therefore, in this subsection, we will do our best to provide the causal effect on carbon emissions of China's CET using firm-level microdata. More specifically, we estimate equation (1) with o = 1, as follows: where ln (CO 2 ) ipt and ln(CO 2 Intensity) ipt are the logarithms of CO 2 emissions and CO 2 intensity for plant i in province p at year t, respectively. The estimation results are shown in table 2. Column (1) shows the log change in carbon emissions that can be attributed to the CET pilot policy; Column (2) shows the causal effect in terms of the log change in carbon intensity. Not surprisingly, the coefficient of CET i × Post t in Column (1) is significantly negative, indicating that compared to the con-  (1) is e −0.488 − 1 = −0.3861.) However, carbon intensity did not appear to be affected during our study period, which is consistent with the results of the EU ETS during its trial period (Petrick and Wagner 2014). Meanwhile, assuming that all regulatory plants are affected on an average level, we can plug the estimate (38.61%) into our plant-year-level dataset, and the result shows that the absolute value of carbon mitigation is 536.48 million tons for the three years (2014-2016) following CET implementation (i.e. 178.83 million tons per year). It is worth noting that because we did not cover all the regulated plants, the actual carbon mitigation effects of the CET pilot policy are greater than our estimates. That is, our estimated value of carbon mitigation serves as the floor. Furthermore, Chongqing, Hubei, Shenzhen, and Guangdong have total control targets (or caps) of 130, 320, 30, and 350 million tons per year, respectively. The power sector alone has contributed approximately 21.55% (178.83 divided by 830) of the total control target. Furthermore, based on Nordhaus (2017)'s estimate of the social cost of carbon (SCC) ($37.3 per ton of CO 2 ), the annual monetary value of the carbon reductions achieved by China's CET pilot would be over $6.6 billion, which is close to Liu et al (2021)'s estimate of $8.6 billion 8 .

Mechanism analysis: improving the energy mix?
Carbon mitigation policies that are strict but flexible, such as CET, may induce a variety of active compliance responses at various levels and on various subjects. The CET can provide enough incentives for plants to comply, in addition to having an impact at the industry level (e.g. industrial structure upgrading). Positive CET compliance responses at the plant level include but are not limited to increasing energy efficiency, reducing carbon-intensive production, and adjusting the energy structure (so-called fuel substitution, which involves either decreasing the proportion of carbon-intensive fuels or increasing/introducing clean energy). Given the limitations of our data, we can only investigate whether plantlevel energy mix adjustments occurred in the pilots. We specifically estimate equation (1) with o = 2, as follows: where ln (Coal) ipt , ln(Energy) ipt , and ln(CoalRatio) ipt are the logarithms of coal consumption, total energy consumption, and coal consumption ratio, respectively. As shown in table 3, the coefficients of CET i × Post t in Columns (1) and (2) are significantly negative, indicating that after the effective implementation of China's CET pilot policy, the regulated plants' coal consumption and total energy consumption decreased by approximately 30.79% and 37.44%, respectively, when compared to the control groups. However, the coefficient in Column (3) is not statistically significant, indicating that China's CET pilot policy has no effect on carbon intensity. The absolute integrated reductions of plant-level coal consumption (2014-2016) are 185.98 million tons (or 61.99 million tons per year) using the method described above.

Two birds with one stone? Cobenefits of reducing core air pollutants
The combustion of fossil fuels emits CO 2 as well as air copollutants (Li et al 2019, Zwickl et al 2021. In other words, GHG emissions and air pollutants stem from the same source (i.e. fossil fuels). Now that we know that, as a result of China's CET pilot policy, one of the primary means for regulating plants to meet their carbon emission reduction targets is to reduce coal utilization, the natural question is whether and to what extent this will result in the cobenefit of reduced air pollution 9 . Research on the cobenefits in these two directions is critical in evaluating specific environmental policies.
Although we have detected the significant effectiveness of carbon mitigation and lower coal consumption of China's CET pilot policy, it is too early  to conduct a direct benefit-cost analysis (BCA) or simply draw conclusions about the cost-effectiveness of the carbon abatement strategy. We cannot ignore coemitted pollutants, which have become a growing focus in the BCA procedure for policy evaluation and are increasingly a concern for researchers and policymakers (Karlsson et al 2020). Because cobenefits can account for a significant portion of the monetized benefits, not considering them is extremely dangerous, especially when it is critical to the quantified net benefit calculation. In other words, disregarding the cobenefits would lead to the opposite conclusion (Aldy 2020). We perform the following regression (i.e. equation (1) with o = 3): where the dependent variables are the logged SO 2 and NO x emissions. The results for the cobenefit effects of the CET pilot on air pollutant reduction are presented in table 4. All the CET i × Post t estimates are statistically significantly negative, implying that this pilot policy has a negative effect on air copollutants. Specifically, the estimates from Column (1) to Column (2) indicate that, following the implementation of the CET pilot policy, SO 2 and NO x emissions from regulated plants were reduced by approximately 52.19% and 48.62%, respectively, compared to the control groups 10 . Three years after China's CET pilot policy was implemented, the estimated (floor) absolute reductions in SO 2 and NO x emissions were approximately 0.8497 and 0.7786 million tons, respectively.

Sizes of plants
In some dimensions, estimating the homogeneous effect may overlook treatment effect heterogeneity. First, we can examine the potential heterogeneous effects of different plant sizes. We investigate whether CET regulates larger plants differently than smaller plants by dividing them into two subgroups based on output values. Plants above the median (545.498 million CNY) are classified as larger after removing plants with no output values (only seven observations), while those below the median are classified as smaller.
In figure 1, the plots of the regression results are the first groups (the left 'Size' group) of the first two rows. (We combined the coefficients of all heterogeneity analyses with 95% confidence intervals into a single figure, and tabulated forms are available from the authors.) The coefficients of CET i × Post t for nearly all of our outcomes of interest are insignificantly negative in both subsamples with lower and higher sizes, indicating that the effect of China's CET pilot policy is not significantly different for largersized or smaller-sized plants during our study period.

Ownership of plants
We investigate whether the plant response to CET is affected by nation ownership in this subsection. Stateowned enterprises (SOEs), a legacy of the planned economy era, are critical to China's economic development. Furthermore, both non-SOEs and SOEs are participating in China's CET pilots. We identify the types of ownership for our sample using the 10 The possible reasons for the smaller reduction in air copollutants compared to carbon dioxide may be related to coal quality and the development of clean energy technologies. We appreciate the constructive comment from an anonymous reviewer. CESD plant ownership information, divide them into two subgroups (non-SOEs and SOEs), and then reestimate DiD models. The second group of the first two rows in figure 1 shows that the CET pilot policy has significant negative effects on carbon emissions and coal and total energy consumption for SOEs but the effect is not significant for non-SOEs. We believe that one plausible explanation for the findings is that because China's carbon market is largely policy-driven, especially in its early stages, SOEs will most likely take the lead in responding to the CET pilot policy.
We further consider the response to the CET pilot policy from China's top-five corporations. The top five corporations, which are all central SOEs, play a critical strategic role in the emissions trajectory of China's power sector. Based on the names of the plants from the CESD, we divide our data into two subgroups: Top five and non-top five. The regression results are depicted in the third group of the first two rows in figure 1. Overall, the pattern of the results is strikingly similar to that of the ownership grouping. The SOEs and top five corporations are primarily responsible for the CET pilot policy's significant effectiveness.

Spatial distribution
Since the implementation of the Clean Air Act of 1970, environmental justice (EJ) has been a focus of attention not only for policy-makers but also for industry representatives and the general public (Ikeme 2003, Ringquist 2005, Mohai et al 2009, Banzhaf et al 2019. In recent studies, there has been growing public concern about the effect of the spatial reallocation of local pollutants caused by ETSs on EJ (Anderson et al 2018, Cushing et al 2018, Grainger and Ruangmas 2018, Mansur and Sheriff 2021, Hernandez-Cortes and Meng 2023, particularly since Fowlie et al (2012). This appears to be critical for large countries, particularly developing countries such as China, with a significant development gap. The expected air quality cobenefits are further complicated by the spatial transition of carbon credits. Because the impact of air pollutants is localized, trading carbon credits alters CO 2 emission patterns and causes a geographical redistribution of air pollution. This redistribution can have different effects on public health risks depending on the compliance strategies of emitting firms (purchasing credits vs. abating CO 2 emissions) and the characteristics of surrounding communities. If emitting firms located near more populated communities choose to purchase credits, the resulting air pollution may result in some loss of social welfare; on the other hand, if these firms decide to realize additional abatement, the improved air quality may result in a larger benefit for public health (Cai et al 2016). Furthermore, by changing the exposure levels of economically disadvantaged communities, which are less capable of selfprotection, this redistribution can either exacerbate or alleviate environmental injustice issues (Burtraw et al 2005, Kaswan 2008). However, almost none of these dimensions have been investigated in the context of China's ETS.
We investigate the spatial distributional heterogeneities of cobenefits by matching city-level demographic and socioeconomic data. The first group of the last row in figure 1 depicts the heterogeneous results obtained by dividing our sample into higher and lower subgroups based on the gross domestic product (GDP) per capita of the cities where plants are located. SO 2 is statistically significant in areas with lower economic development (generally with lower-income populations), whereas NO x is statistically insignificant in both subgroups. As a result, we cannot find convincing evidence that China's CET pilot policy caused environmental inequity at the level of economic development during our study period. The heterogeneous results obtained by categorizing our data into higher and lower subgroups based on citylevel population density are depicted in the second group of the last row of figure 1. While the pattern for SO 2 remains the same, for NO x , the CET pilot policy has a significant negative impact only in areas of high population density. Therefore, we find weak evidence that China's CET pilot policy caused environmental inequalities at the population distribution level during our study period. Overall, our spatial distribution analysis results indicate that there is some disproportionate environmental inequality in the context of China's CET pilot.
We also conduct two additional heterogeneity analyses based on the allowance allocation structure and combined heat and power (CHP) plants (see SI section 6), demonstrating that the impact of China's CET pilot is unrelated to the rate-based allowance allocation structure and CHP plants. We perform a battery of robustness checks and one falsification test to alleviate concerns in areas such as counterfactual construction, identifying assumptions, and other econometric issues, which are detailed in sections 7 and 8 of SI.

Discussion and conclusion
This study examines the joint effects of China's carbon market on carbon emissions, the energy mix, and air copollutants. The CET market is one of the most cost-effective policy tools in place in China to address the challenges of climate change and environmental degradation and is critical for meeting the 2030 carbon-peak and 2060 carbon-neutral targets. Policy-makers around the world have expressed concerns about whether market-based instruments such as CET will work in a country such as China with a unique political-economic context and institutional background, especially with a low carbon price and low liquidity. Several recent studies examine the effects of China's CET pilot policy on carbon reductions and energy consumption, but they do not take advantage of the unique coverage thresholds for CET-regulated pilots, and the causal evidence that informs our understanding of this debate remains insufficient. The literature lacks causal evidence for a more comprehensive assessment of China's CET policy. Furthermore, it is unclear whether and to what extent China's CET policy has resulted in synergistic benefits regarding lowering air pollutant emissions.
Using China's CET pilots as a quasi-natural experiment, we utilize a comprehensive unique plant-level dataset from 2010 to 2016 to empirically identify the joint effects of the CET policy. We demonstrate that China's CET pilot policy is effective in reducing plant CO 2 emissions in the early trading stage (2014-2016), despite a low carbon price and low liquidity (Bayer andAklin 2020, Cui et al 2021). However, this is more likely to be caused by government regulation , especially in the early stages; thus, better market-oriented rules are required to obtain better CET results . We find that the CET pilots have no effect on the carbon intensity or energy efficiency of the power plants regulated by the CET policy. Furthermore, the policy effect of carbon reduction is driven primarily by reduced coal consumption from plants that are required to participate in the CET, while there is no effect on the share of coal in total energy consumption.
More importantly, one of the marginal contributions of this study to the literature is that, based on microplant-level data, we provide the first causal evidence of synergistic emission reductions in CO 2 and conventional air pollutants from China's CET pilots. We find that the CET pilots achieve significant cobenefits in terms of reducing SO 2 and NO x . This provides an opportunity for policy-makers to make better use of CET to address climate change and environmental issues, but it also poses a challenge to coordinating different policies. During our study period, our heterogeneity analysis revealed no significant differences in the effects on smaller and larger plants, with the policy effects affecting SOEs and top five corporations almost exclusively. Furthermore, the policy effects demonstrate disproportionate environmental inequality. Therefore, how to better balance regional environmental equity in the function of carbon markets, especially the national uniform carbon market, is an important challenge for policy-makers. Another important finding is that the policy effects are unrelated to the rate-based allowance allocation structure that is applied by China's national uniform carbon market in the power sector. Finally, our findings withstand a battery of robustness tests on the identifying assumption and other econometric concerns.
In the end, we argue that China's CET pilot policy achieved multiple goals in terms of carbon reduction, coal consumption reduction, and air copollutant reduction and that it provides a good blueprint for China to achieve its mid-century carbon neutrality target more cost-effectively. Our paper contributes to the literature on climate policy evaluation by carefully addressing the endogeneity issue associated with decarbon regulations. Meanwhile, our use of plantlevel microdata from a developing country (China) extends previous studies on ex post causal assessment of the CET policy. The empirical evidence of synergistic emission reductions can motivate other hesitant developing countries to accelerate the pace of carbon trading structuring.
The following four major aspects could be fruitful fields for future related research: (1) how will China's national carbon market impact the economy, industry, and businesses? This necessitates simulation analysis as well as an empirical evaluation using microdata.
(2) Whether and to what extent China's carbon market has had unintended consequences, particularly in terms of overlapping policies (Goulder and Stavins 2011). (3) In addition to traditional CAC policies, comparative studies of CET with other policies, such as climate change mitigation, environmental protection, and energy conservation policies, are urgently needed. Only recently has preliminary research on these two topics begun to emerge (Gugler et al 2021, Greenstone et al 2022. (4) What is the long-term impact of China's CET policy, given the orderly promotion of China's carbon market and the accumulation of microdata?

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.