Carbon markets promote environmental justice in China

Abstract

Environmental markets may influence environmental justice by disproportionately affecting economically disadvantaged or socially vulnerable communities. We investigate the impact of China’s Certified Emission Reduction (CCER) program on air pollution in disadvantaged and non-disadvantaged counties from 2000 to 2019. We find that the CCER program effectively reduces carbon dioxide (CO₂) emissions and fine particulate matter (PM_2.5) concentrations. We further examine the distributional impacts and find that the program promotes environmental justice by reducing disparities in air quality between economically disadvantaged and more affluent counties, as well as between ethnic minority and majority (Han) counties. In particular, we observe greater improvements in air quality and more equitable outcomes from renewable energy projects. Our findings offer policy-relevant insights for designing environmental markets that advance sustainable development and social equity in China and other regions.

Introduction

Low-income groups and ethnic minorities often face higher pollution exposure¹. Environmental justice aims to correct the unequal impact of environmental pollution between advantaged and disadvantaged regions, ensuring equal rights to environmental protection and benefits. Environmental justice is closely linked to the United Nations Sustainable Development Goals (SDGs), such as clean water and sanitation (SDG 6), affordable and clean energy (SDG 7), reduced inequality (SDG 10), and climate action (SDG 13). Disadvantaged areas of developed countries bear the burden of pollution transfer from advantaged regions, often resulting in poor air and water quality². Socioeconomically disadvantaged areas in China experienced less pollution activities due to slow economic growth and enjoyed a better environment in the past. Relocation of high-polluting industries to western China suggests that the resource-rich yet economically disadvantaged western regions with more ethnic minority residents may bear more environmental pollution³. Disadvantaged regions are less likely to adopt similar technologies as advantaged regions in pollution controls. Traditional, high-pollution industries and technologies tend to be more economically effective, which may lead to excessive pollution in disadvantaged regions and damage regional environmental justice. Understanding how disadvantaged regions can mitigate disparately larger pollution exposure remains an important research and policy-relevant question.

We investigate the impact of carbon market development on environmental justice issues in China. Carbon markets consist of a quota-based market, including the Emissions Trading Scheme that trades carbon quotas or permits, and a project-based market, such as the Clean Development Mechanism (CDM), that trades certified emission reductions⁴ and is used by firms in the cap-and-trade market to cover emissions. Through the CDM, firms obligated to satisfy emission targets in the cap-and-trade system are likely to invest in carbon-reducing projects in developing countries to offset their emissions. Under the Regional Greenhouse Gas Initiative, a regional US cap and trade market, for instance, firms could use CO₂ offset allowances to meet around 3.3% of their emission reduction obligations⁵. Existing studies explore the impact of market-based policies on environmental justice from the perspective of quota markets^6,7,8. Specifically, Hernandez-Cortes and Meng examine the impact of carbon markets on environmental justice by comparing the gap in pollutant concentrations between disadvantaged communities and others before and after the introduction of the California carbon market. Results show that the program narrowed environmental justice gaps in PM_2.5, PM₁₀, and NO_x concentrations from participating industrial facilities by 6–10% annually⁸, while Fowlie, Holland, and Mansur and Shapiro and Walker (2021) do not find any effects of quota-based carbon markets on the equity of environmental outcomes^6,7. While existing research has explored the environmental justice impacts of quota-based markets with mixed results, the distributional effects of project-based carbon markets remain a significant research gap. This is a critical omission, as project-based mechanisms are often targeted toward lower-income regions, positioning them to potentially directly affect environmental equity⁹. We focus on the environmental justice impact of China’s Certified Emission Reduction (CCER). CCER is a project-based market that allows regulated companies to use certified emission reduction to offset a certain percentage of their carbon allowance clearances. The CCER system was established in 2012 to replace the international CDM with a domestic framework. It creates certified carbon offsets that companies in regional pilot markets can use to meet their emission quotas. The program was suspended in 2017 to align with the development of a national carbon market. After significant reforms, the CCER system was officially relaunched in April 2024.

We quantify the impact of CCER programs on the environmental justice gap, as measured by county-level CO₂ emissions and PM_2.5 concentrations, between relatively economically disadvantaged and more affluent counties, as well as between ethnic minority and Han counties. We first employ the difference-in-differences approach proposed by Callaway and Sant’Anna (CSDID) to estimate how the CCER projects reduced the emissions of CO₂ and average ambient PM_2.5 concentrations using county panel data from 2000 to 2019¹⁰. We then employ triple difference (DDD) models to examine whether the introduction of the CCER program changed the pollutant concentration gap between economically disadvantaged counties and others and between ethnic minority and majority (Han) counties, respectively. Our estimation results show that the CCER program lowers the emission level of CO₂ and PM_2.5 concentration annually by 2.89% and 2.13%, respectively, for the average treated county with CCER projects. For the environmental justice gap, we find that CCER projects significantly reduce PM_2.5 concentrations by 2.05% on average for economically disadvantaged counties when compared to more affluent counties, and significantly reduce the PM_2.5 gap between ethnic minority and other counties by 3.34% after the introduction of a CCER project.

This study makes several contributions to the previous research on the environmental inequity effects of carbon markets. First, it is one of the first to empirically examine the distributional effects of a project-based carbon market, contrasting with the existing literature focused on quota-based systems. Second, it provides timely evidence on the CCER’s impact, offering highly policy-relevant insights as China revitalizes its national carbon offset mechanism. Third, this study offers empirical evidence that CCER projects enhance environmental justice by stimulating regional green investment and green innovation. Our results bridge an important gap between the carbon market and environmental equity, providing policymakers with practical insights for promoting both emission reduction and equitable development. Finally, our micro-level data allows us to identify which types of CCER projects are most effective in promoting a more equitable outcome.

Results

CCER offers co-benefits by reducing the concentration levels of carbon dioxide and PM2.5

To evaluate the impact of the CCER introduction on county-level CO₂ emissions and PM_2.5 concentrations, we employ Callaway and Sant’Anna difference-in-differences (CSDID) based on annual PM_2.5 concentration, CO₂ emission, and county-level social and economic data merged with the CCER project data. We define a county as treated after the implementation of the first CCER project. The control group consists of the counties adjacent to the treatment group that did not have CCER programs during the same year.

We compare changes in CO₂ emissions and ambient air pollutant concentrations between treated and non-treated counties before and after the CCER intervention based on CSDID. We regress the natural log of the CO₂ emissions and the natural log of the PM_2.5 concentration on the dummy variable indicating the establishment of a CCER project. A set of time-varying controls that potentially affect the emissions of CO₂ are also included, such as county gross domestic product, secondary industry value added, loans from financial institutions, public budget revenue, number of employees in the secondary industry, urbanization rate, and total power of agricultural machinery, annual precipitation, annual temperature, and environmental policies. Our regression further controls for year and county fixed effects, and city-level linear time trends. The fixed effects address unobserved time-invariant determinants of CO₂ emissions and PM_2.5 concentration by region, as well as common shocks in counties across years, such as national environmental regulations. We test the parallel trend assumption of the CSDID approach by plotting the event-study coefficients (Supplementary Fig. 1). Our results indicate that there is no statistically significant difference in the emission of CO₂ and concentration of PM_2.5 between control and treatment groups before the introduction of the CCER project.

The coefficient of CCER establishment measures the average treatment effect on the treated. Our results show a negative and statistically significant relationship between introducing CCER projects and local CO₂ emissions and PM_2.5 concentration (see Fig. 1). Specifically, the implementation of the CCER projects lowers the county-level ambient CO₂ emissions and PM_2.5 concentrations by 2.89% and 2.13%, respectively (Supplementary Table 1). We also investigate the effect of CCER capacity on CO₂ emissions and air pollution concentration (Supplementary Table 2). Our results show that a one-million-ton increase in the carbon abatement capacity of the CCER projects could reduce the county-level ambient CO₂ emissions and PM_2.5 concentrations by 0.41% and 0.8%, respectively (see Fig. 1).

Fig. 1: The Impact of CCER on CO₂ emissions and PM_2.5 concentrations.

CCER (China’s Certified Emission Reduction) with control indicates that the CSDID (Callaway and Sant’Anna difference-in-differences) approach includes county gross domestic product, secondary industry value added, loans from financial institutions, public budget revenue, number of employees in the secondary industry, urbanization rate, total power of agricultural machinery, annual precipitation, annual temperature, and environmental policies. Capacity includes all control variables, year, county fixed effects and city-specific time trend and is estimated by the two-way fixed effects model. The dependent variable of A is the natural log of CO₂ emission level. The dependent variable of B is the natural log of PM_2.5 concentration level (particulate matter with aerodynamic diameter ≤2.5 µm). The results of CCER without control and CCER with control represent the effects of the implementation of the CCER projects, while the results of capacity represent the influence of one-million-ton increase in the carbon abatement capacity of the CCER projects. The central estimates shown with circles are regression point estimates, dark-colored lines denote two-sided 90% confidence intervals and light-colored lines represent two-sided 95% confidence intervals. Standard errors are clustered at the county level. Sample sizes (n) for each specification are as follows: A CCER without control (n = 7952), CCER with control (n = 7949), Capacity (n = 30,582). B CCER without control (n = 7952), CCER with control (n = 7949), Capacity (n = 33,980). Statistical inference is based on two-sided t tests of the null hypothesis that the coefficients equal zero. P values are reported as exact values where possible. No adjustments were made for multiple comparisons. The full results are presented in Supplementary Tables 1 and 2. Each observation represents an independent county-year unit. Counties without CCER projects serve as the control group.

CCER promotes environmental justice

To examine the effect of CCER on air quality disparities between economically disadvantaged and more affluent counties and between ethnic minority and non-minority counties, we employ a DDD approach by including the interaction term of the treatment variable and the dummy variable indicating the economic status or the ethnical composition of the county. We follow the 2014 Chinese State Council’s definitions of 832 economically disadvantaged counties, and follow public information and the National Ethnic Affairs Commissions of China’s definitions of 127 minority counties. There are 56 ethnic groups in China, of which the Han ethnicity accounts for 91%. Other ethnicity except the Han is called minorities. The Chinese State Council lists 86 of the 127 counties with dominant minority ethnic groups as economically disadvantaged counties. The detailed definition of a disadvantaged county can be found in the Data. Panel A of Fig. 2 shows the geographical distribution of CCER projects and disadvantaged counties (economically disadvantaged counties and ethnic minority counties), while panels B and C present the proportion of each project type. The control variables are the same as in the DID specifications. The coefficient of interest is the interaction term of CCER and Poor when evaluating the environmental justice gaps between economically disadvantaged counties and others. Similarly, the coefficient of interest is the interaction term of CCER and Minority when evaluating the environmental justice gaps between ethnic minority counties and Han counties.

Fig. 2: Information of CCER projects and disadvantaged counties.

A shows the distribution of CCER projects and disadvantaged counties. In the legend, the forward slash represents economically disadvantaged counties, and the backslash represents minority counties. Different colors indicate the carbon reduction capacity of CCER within the county, measured by million tons (Mt). The column represents the number of CCER projects in the county. The length of the columns in the legend is a reference ratio. B represents the proportion of different CCER project types. C represents the proportion of carbon reduction capacities of different CCER projects. The base map used in this study is the 2023 China map, which has been reviewed and approved with the map review number GS (2023)2767 by the Standard Map Service System of the Ministry of Natural Resources.

We find that the introduction of CCER projects narrows the gap in PM_2.5 concentration between economically disadvantaged and more affluent counties. Specifically, CCER projects reduce PM_2.5 concentrations by an additional 2.05% on average for economically disadvantaged counties when compared to other counties (see Fig. 3, panel A). The concentration of PM_2.5 in ethnic minority counties is reduced by an additional 3.34%, when compared to other counties after the introduction of a CCER project, as shown in Fig. 3, panel B (also see Supplementary Table 3). We test the parallel trend assumption of the DDD approach by plotting the event-study coefficients (shown in Supplementary Figs. 2 and 3). The results indicate that there are no significant pre-trends, and the impact of CCER after the policy implementation is significant. When using the CCER capacity as the explanatory variable, we find that a one-million-ton increase in the carbon abatement capacity of the CCER projects can reduce county-level ambient PM_2.5 concentrations by 0.48% between economically disadvantaged and more affluent counties (see Fig. 3, panel A). We also find a 0.74% reduction in county-level ambient PM_2.5 concentrations between ethnic minority counties and Han counties (see Fig. 3, panel B). The results based on the CCER capacity are consistent with and further support our main conclusions.

Fig. 3: The impact of CCER on environmental justice gaps.

A shows the influences of China’s Certified Emission Reduction projects (CCER) on environmental justice gaps between economically disadvantaged counties and others, and B shows the influences on environmental justice gaps between minority and Han counties. CCER with FE 1 includes year and county fixed effects. CCER with FE 2 and capacity with FE include year, county fixed effects and city-specific time trend. The dependent variable is the natural log of PM_2.5 concentration level (particulate matter with aerodynamic diameter ≤2.5 µm). The results of CCER with FE 1 and CCER with FE 2 represent the effects of the implementation of the CCER projects, while the results of capacity with FE represent the influences of a one-million-ton increase in the carbon abatement capacity of the CCER projects. The central estimates shown with circles are regression point estimates, dark-colored lines denote two-sided 90% confidence intervals and light-colored lines represent two-sided 95% confidence intervals. Standard errors are clustered at the county level. Sample size (n) is 33,980 for all regression specifications. Statistical inference is based on two-sided t-tests of the null hypothesis that the coefficients equal zero. P values are reported as exact values where possible. No adjustments were made for multiple comparisons. The full results are presented in Supplementary Table 3. Each observation represents an independent county-year unit. A contrasts economically disadvantaged (n = 514) and more affluent (n = 1185) counties, while B contrasts minority (n = 97) and Han (n = 1602) counties.

Our results survive a series of robustness checks (see Fig. 4). First, we use the CCER project announcement time as an alternative treatment time (see Supplementary Table 4 for results). The coefficients of CSDID and DDD specifications are not statistically significantly different from zero. Next, we only include counties ranked between 10–90% in terms of night lights to mitigate the influence of economic activities on emissions and to improve the comparability between treatment and control groups. Supplementary Table 5 indicates that after enhancing the comparability between the control and treatment groups, the model results remain robust. Third, we use alternative CO₂ and PM_2.5 data sources to assess the robustness of our results. We employ satellite imagery data from the Emissions Database for Global Atmospheric Research (EDGAR) to construct a CO₂ data source at the county level¹¹. Besides, the alternative PM_2.5 data is sourced from Wei et al.¹². The model results remain consistent based on alternative data sources (Supplementary Table 6). Finally, only 59 counties have a dominant wind direction, and we do not find that wind direction has any effect on the results, as shown in Supplementary Table 7.

Fig. 4: Robustness tests.

The x-axis represents the significance of the regression results, while the y-axis represents the regression results. Besides, the left dashed line indicates the significance of 0.01, the middle-dashed line indicates the significance of 0.05, and the right dashed line indicates the significance of 0.1. All robustness tests remain consistent with the main results. The colors represent the groups for the main regression and robustness checks, while the shapes indicate different estimation contents and methods. CSDID-CO2 and CSDID-PM25 refer to the robustness tests of the influences of CCER projects on the emission level of CO₂ and on the concentration level of PM_2.5, respectively. Similarly, DDD-POOR means the robustness test of the influences of CCER on the environmental justice gap between economically disadvantaged counties and others. DDD-MIN means the robustness test results of the influences of CCER on the environmental justice gap between minority and Han counties. Dark red refers to the main regression results. Light purple represents the robustness tests when using alternative treatment time (see Supplementary Table 4 for results). Light gray means the robustness tests when only including counties ranked between 10 and 90% in terms of night lights (see Supplementary Table 5). Light grayish green represents the robustness tests when using alternative CO₂ and PM_2.5 data sources (see Supplementary Table 6). Amber means the robustness tests when considering the wind direction (see Supplementary Table 7).

The heterogeneity analysis

We conduct heterogeneity analysis based on CCER project types for emission and concentration reduction effects, and environmental justice, respectively. The dataset contains six types of CCER projects, including methane recovery and utilization (MRU), waste incineration power generation (WIPG), renewable energy (RE), fuel substitution (FS), energy conservation and efficiency improvement (ECEI), and forestry, which allows us to estimate the heterogeneous effects of each type of CCER project. MRU includes gas power generation, biogas power generation, wastewater biogas utilization, etc. WIPG includes landfill power generation and WIPG, etc. RE includes wind power generation, photovoltaic power generation, hydroelectric power generation, biomass power generation, etc. FS includes oil-to-gas, biomass energy conversion, and coal-to-gas conversion, etc. ECEI includes energy-saving retrofits, waste heat recovery, cogeneration, and liquefied natural gas cold energy used for air separation, etc. Forestry includes afforestation and reforestation. Supplementary Table 8 provides the proportion of each type of project.

In terms of the reduction effects, only the coefficient for RE is negative and statistically significant for CO₂, indicating that RE contributes to the reduction of greenhouse gas emissions. Coefficients for MRU, FS, and Forestry are consistently negative but not significant for CO₂. Similarly, the coefficients for FS and RE are negative and statistically significant for PM_2.5. Coefficients for WIPG and Forestry are consistently negative but not significant for PM_2.5. Panel A of Fig. 5 presents the estimated results of the CSDID approach, and detailed information can also be found in Supplementary Table 9. These suggest that RE projects are particularly effective in both cutting greenhouse gas emissions and improving air quality.

Fig. 5: Heterogeneous treatment effects of CCER.

The dependent variable of A.1 is the natural log of CO₂ emission level. The dependent variable of A.2, B.1, and B.2 is the natural log of PM_2.5 concentration level. Our study dataset contains six types of CCER projects: MRU (methane recovery and utilization), WIPG (waste incineration power generation), RE (renewable energy), FS (fuel substitution), ECEI (energy conservation and efficiency improvement), and forestry, as well as two regions: eastern and central-western. Each dot represents a regression point estimate of the treatment effect, and the horizontal lines denote the corresponding two-sided 95% confidence intervals. Standard errors are clustered at the county level. Sample sizes vary across specifications due to differences in data availability for different project types and regions. Detailed n values for each regression are reported in Supplementary Table 9–13. Statistical inference is based on two-sided t-tests of the null hypothesis that coefficients equal zero. P values are reported as exact values where possible. No adjustments were made for multiple comparisons. ***p < 0.01, **p < 0.05, *p < 0.1.

With respect to the environmental justice gap, our estimation results show that each CCER project contributes significantly to closing the air pollution gap between economically disadvantaged and more affluent counties and between minority and Han counties, as shown in panel B of Fig. 5 (or Supplementary Table 10 and 11). Furthermore, Forestry CCER projects have the most significant effect in reducing environmental justice.

Besides, we divide the samples into eastern and central-western regions to test the robustness of the model results, as shown in panel A and B of Fig. 5 (or Supplementary Tables 12 and 13). We find that the CCER projects reduce local CO₂ emissions in both eastern and central-western regions, and only reduce the PM_2.5 concentration in central-western region. In terms of the environmental justice gap, CCER projects narrow the concentration of PM_2.5 between economically disadvantaged and more affluent counties in the central-western region, and between ethnic minority counties and Han counties. The results indicate regional disparities, with CCER projects having a more significant impact in central-western region. This may be because eastern region implemented pollution control measures earlier, reducing the marginal effect of CCER, while central-western region is still in the early stages, leading to a greater impact.

The mechanism analysis

We conduct mechanism analyses based on regional green investment and green innovation levels for CO₂ emission reduction, pollution concentration reduction, and environmental justice, respectively. Green investment reflects local efforts in environmental governance and industrial upgrading and is a key channel through which CCER projects generate their impacts. We replace CO₂ emissions and pollutant concentration in the main identification strategy with the proportion of city-level environmental pollution control investment to GDP. These data are sourced from the city-level statistical yearbooks. Panel A of Fig. 6 presents the mechanism analysis results, indicating that the introduction of CCER can promote local green investment, with a greater driving force in disadvantaged areas (Supplementary Table 14). Furthermore, we divide the sample into high green investment and low green investment groups based on the median of the ratio of city-level environmental pollution control investment to GDP. Panel B of Fig. 6 presents the model results of the CSDID approach, with detailed information provided in Supplementary Table 15. The high green investment group has a greater impact on reducing CO₂ emissions and lowering PM_2.5 concentrations. The introduction of CCER is an important initiative by local governments to pursue green development, resulting in larger pollution reduction and carbon abatement outcomes. The results of environmental justice are also shown in Fig. 6 and Supplementary Table 16. With respect to the environmental justice gap, our estimation results indicate that the low green investment group has made a greater contribution to narrowing the air pollution gap between economically disadvantaged and more affluent counties, as well as between ethnic minority and Han counties. Considering that disadvantaged regions in China typically have better environmental conditions, this narrowing gap suggests that these regions may rely on environmental degradation for development, rather than pursuing green development. Beyond government-led green investment, technological progress also plays a critical role. To capture this channel, we further examine green innovation as a potential mechanism. We replace CO₂ emissions and pollutant concentration in the main identification strategy with city-level green patent applications. These data are sourced from China National Intellectual Property Administration. Panel B.1 of Fig. 6 reports the mechanism analysis, showing that the introduction of CCER stimulates local green innovation, with stronger effects observed in disadvantaged regions (see Supplementary Table 14). To further examine heterogeneity, we split the sample into high and low green innovation groups according to the median of the ratio of city-level green patent applications. Panel B.2 of Fig. 6 presents the CSDID estimation results (see Supplementary Table 15). The high green innovation group demonstrates larger reductions in CO₂ emissions and PM_2.5 concentrations. Evidence on environmental justice is presented in Fig. 6 and Supplementary Table 16. The results indicate that the low green innovation group contributes more to reducing the air pollution gap between economically disadvantaged and more affluent counties, as well as between ethnic minority and Han counties. These results confirm that CCER promotes green innovation and are broadly consistent with the effects observed for green investment.

Fig. 6: Mechanism analysis of CCER.

A presents the model results with green investment as the mechanism, while B shows the results with green innovation as the mechanism. The dependent variable of A.1 is the natural log of the proportion of city-level environmental pollution control investment to GDP. The dependent variable of B.1 is the natural log of the proportion of city-level green patent application. The dependent variable of CSDID-CO2 in A.2 and B.2 is the natural log of CO₂ emission level. The dependent variable of remaining models in panel A.2 and B.2 is the natural log of PM_2.5 concentration level. In A, Low and High indicate low and high green investment groups, respectively; in B, they indicate low and high green innovation groups. Bars are the values of the coefficients, which represent point estimates from the regressions and indicate the average effects of CCER. Two-sided 95% confidence intervals are plotted vertically. Standard errors are clustered at the county level. Sample sizes (n) for each specification are as follows: A.1 CSDID (n = 7776); POOR and MIN (n = 33,460 each). A.2 CSDID-CO₂, Low (n = 4637) and High (n = 3310); CSDID-PM25, Low (n = 4,639) and High (n = 3,310); DDD-POOR and MIN-POOR, Low (n = 18,413) and High (n = 15,567). B.1 CSDID (n = 6830); POOR and MIN (n = 30,694 each). B.2 CSDID-CO₂, Low (n = 4368) and High (n = 3581); CSDID-PM25, Low (n = 4351) and High (n = 3569); DDD-POOR and MIN-POOR, Low (n = 15,559) and High (n = 18,421). Statistical inference is based on two-sided t-tests of the null hypothesis that coefficients equal zero. P values are reported as exact values where possible. No adjustments were made for multiple comparisons. The full results are presented in Supplementary Tables 14–16. Each observation represents an independent county-year unit.

Discussion

Environmental justice contributes substantially to SDGs by addressing inequalities, promoting inclusivity, and ensuring sustainable and equitable development. We examine the potential impact of CCER on environmental justice based on a county-level panel dataset. We first investigate the effect of the introduction of a CCER program on air pollution concentrations and find a significantly negative effect. We then estimate the impacts of the program on inequity in environmental exposure along income and ethnic lines. The results suggest that CCER programs promote equity by narrowing the air pollution concentration gap between economically disadvantaged and more affluent counties and between ethnic minority and Han counties, respectively. Heterogeneity analysis reveals that among all CCER projects, RE projects are the most effective at boosting environmental justice. Mechanism analysis results indicate that green investment is a key channel through which CCER exerts its impact.

Sustainable environmental development requires both environmental justice and environmental risk mitigation. Coordination and balanced development of various regions are needed to prevent the transfer of pollutants from more affluent regions to impoverished ones. To actively combat climate change and advance carbon peak and carbon neutrality goals, China relaunches the CCER market in January of 2024 after suspending in 2017¹³. The Ministry of Ecology and Environment of China collaborated with the General Administration of Market Supervision and Administration to revise the “Interim Measures for the Management of Greenhouse Gas Voluntary Emission Reduction Transactions” and formulate the “Management Measures for Greenhouse Gas Voluntary Emission Reduction Transactions (Trial)” in July, 2023. “The Provisional Regulations on Carbon Emission Trading Management” implemented by China in May 2024 further promoted the development of CCER market. We firstly document empirical evidence of the potential positive impact of CCER on environmental inequity to offer policy implications for the revival of CCER markets in China and other regions. We rely on county-level analyses, and unlike Hernandez-Cortes and Meng⁸, we cannot leverage the spatial location of these projects in identification due to data limitations. Future research could extend studies to examine the policy effects on pollution emissions at the facility level, enabling a more comprehensive understanding of changes in pollution concentration.

Methods

CCER data

Our analysis draws on restricted data from the SinoCarbon database, which provides detailed information on all CCER projects in China up to 2017, including project names, construction periods, capacity, and locations. Supplementary Table 17 shows the number of counties participating and not participating in the CCER program from 2011 to 2019. This dataset also includes information on each project type, including MRU, WIPG, RE, FS, ECEI, and forestry, which allows us to estimate the heterogeneous effects of each type of CCER project.

Disadvantaged counties data

To define disadvantaged groups, we follow the United States (US) Environmental Protection Agency’s definition for “overburdened” community in the US as a broad definition for a disadvantaged community. Previous environmental justice literature mostly uses the median income of the region and/or the percentage of minorities as proxies for disadvantaged neighborhoods⁶. We divide counties into two categories along income and ethnic dimensions, which include the economically disadvantaged and more affluent counties, as well as minority and Majority (Han) counties.

The classification of economically disadvantaged and more affluent counties leverages a list of 832 national-level economically disadvantaged counties published by the Leading Group of Poverty Alleviation and Development of the State Council of China in 2014. The poverty counties depend on the average annual net income of local residents, primarily relying on pre-2014 indicator constructs, which mitigate concerns about direct impacts of the CCER project on the identification of vulnerable community designations. We differentiate minority counties and non-minority counties based on the list of 117 autonomous counties published by the National Ethnic Affairs Commission and the list of 10 counties treated as autonomous counties based on public information. China is a multi-ethnic country, with the vast majority being Han Chinese. Ethnic minorities may face numerous disadvantages stemming from various factors. For example, they often inhabit resource-deprived mountainous regions from the origin, limiting their access to essential resources and hindering economic growth and employment opportunities. Moreover, minority groups’ preferences for ethnic languages over Mandarin may pose communication challenges, creating barriers to effective interaction and understanding with other ethnic groups. To improve the status and development conditions of ethnic minorities, the Chinese government implements a system of regional national autonomy. There are 117 autonomous counties determined by the central government and 10 ethnic minority support policy counties determined by local governments based on the proportion of ethnic minority populations. The central government provides these minority counties with additional policies to support development. Supplementary Table 18 shows the number of economically disadvantaged and minority counties affected by CCER, and Supplementary Table 19 shows the sampling statistics by county type.

Pollutant, economic and rainfall data

We derive annual county-level CO₂ emissions based on a particle swarm optimization-back propagation methodology to harmonize the scales of DMSP/OLS and NPP/VIIRS satellite pictures¹⁴. This approach yields improved fitting outcomes compared to prior studies that rely on original models and conventional econometric methods. Our county-level CO₂ emissions data consist of 2800 counties nationwide from 2000 to 2017. Supplementary Fig. 4 plots the spatial distribution of the emission level of CO₂ and per capita GDP on average during our study period. We also employ satellite imagery data from the EDGAR to construct a CO₂ data source at the county level to assess the robustness of our results¹¹.

The annual ground-level PM_2.5 is derived by combining Aerosol Optical Depth retrievals with the GEOS-Chem chemical transport model and subsequently calibrated to global ground-based observations using a geographically weighted regression, following Van Donkelaar et al.¹⁵. The ground-calibrated PM_2.5 data have a high grid cell resolution of 0.01°. We then use ArcGIS to obtain PM_2.5 concentration data at the county level. Supplementary Fig. 5 plots the spatial distribution of the concentration level of PM_2.5 and per capita GDP on average during our study period. We also use data from Wei et al.¹² as an alternative data source for PM_2.5 for the robustness test¹². Compared to other pollutants, such as sulfur dioxide (SO₂), nitrogen oxides (NO_x), ozone (O₃), PM_2.5 monitoring technology is more mature, with a wide range of data coverage and high frequency of data updates.

We gather annual data from the China County Statistical Yearbook on various control variables, including county gross product, secondary industry value added, loans from financial institutions, public budget revenue, number of employees in the secondary industry, urbanization rate, and total power of agricultural machinery. In addition to economic variables, our control variables include rainfall data obtained from China’s National Meteorological Science Data Sharing Service platform and temperature data obtained from Peng¹⁶. Supplementary Table 20 provides detailed definitions of the variables used in our analysis.

Difference-in-differences

We first investigate the impact of CCER programs on CO₂ emissions using the following two-way fixed effects (or the generalized DID) model:

$${{{\mathrm{ln}}}{CE}}_{{it}}=\theta\, * \,{{CCER}}_{{it}}+{X}_{{it}}^{{\prime} }\, * \,\beta+{\delta }_{t}+{\gamma }_{i}+{\mu }_{{kt}}+{\varepsilon }_{{it}},$$

(1)

where ${{{\mathrm{ln}}}{CE}}_{{it}}$ represents the natural log of the emissions of CO₂ in county i in year t; ${{CCER}}_{{it}}$ is a dummy variable that equals 1 in the years after a CCER project establishment in county i. We classify a county as treated after the first CCER project is implemented. The control group is composed of neighboring counties to the treatment group that did not have CCER programs. ${X}_{{it}}^{{\prime} }$ denotes a set of time-varying control variables that may both affect the emissions of CO₂ and CCER project constructions, including county gross domestic product, secondary industry value added, loans from financial institutions, public budget revenue, number of employees in the secondary industry, urbanization rate, and total power of agricultural machinery, annual precipitation, annual temperature, and environmental policies. The model also contains interactions between these control variables and year fixed effects to account for time-varying unobservables¹⁷. We further use fixed effects to control for time-invariant and time-variant unobservables, where ${\delta }_{t}$ represents year fixed effects; ${\gamma }_{i}$ denotes county fixed effects; ${\mu }_{{kt}}$ represents city-specific linear time trends; and ${\varepsilon }_{{it}}$ is the random error term. The coefficient θ measures the DID estimate of the differential effect of a CCER program on CO₂ emissions for counties with CCER projects versus those without.

One of the potential concerns is that the two-way fixed effects may generate bias resulting from heterogeneous treatment timing¹⁸. Goodman–Bacon demonstrates that the estimator in Eq. (1) is a weighted average of all possible 2 × 2 DID estimators that compare timing groups to each other¹⁹. The variance-weighted treatment effect parameters sometimes occur with negative weights when using the already-treated units as controls, which leads to biased estimates. Since CCER projects are deployed at different times, we conduct CSDID proposed by Callaway and Sant’ Anna by calculating the cohort-specific average treatment effects on the treated (CATT) and then weight and sum across group and time dimensions to prevent the inclusion of “bad control groups”¹⁰. CSDID is an improved difference-in-differences method designed for staggered treatment adoption, effectively avoiding the bias of traditional DID under treatment timing heterogeneity and providing robust causal effect estimation. Callaway and Sant’Anna proposed to use groups that have never been treated before period t as the control group, applicable in cases with multiple time periods and variations in treatment timing¹⁰. Thus, we define the indicator variable ${E}_{e}$, which takes the value of 1 when the county implemented the first CCER project at time point e. Define the dummy variable C, which takes the value of 1 if counties have never been deployed by CCERs. For each county in the sample, in the dummy variable set $\left\{{E}_{1},\cdots,{E}_{T},C\right\}$, one of the dummy variables must take the value of 1. We denote ${P}_{e}\left(X\right)=P({E}_{e}=1{|X},\,{E}_{e}+C=1)$ as the probability of county are deployed by CCERs at time point e, given the covariates X and the treatment condition in which CCER is deployed for the first time at time e or has never been deployed by CCER. Based on these settings, we can express ${ATT}(e,t)$ as:

$${ATT}(e,t)=E\left[\left(\frac{{E}_{e}}{E\left[{E}_{e}\right]}-\frac{\frac{{P}_{e}\left(X\right)C}{1-{P}_{e}\left(X\right)}}{E\left[\frac{{P}_{e}\left(X\right)C}{1-{P}_{e}\left(X\right)}\right]}\right)\left({{{\mathrm{lnCE}}}}_{t}-{{{\mathrm{lnCE}}}}_{e-1}\right)\right]$$

(2)

The average treatment effect θ of CCERs can be obtained by performing a weighted average:

$$\theta={\sum }_{e}{\sum }_{t=2}^{T}w(e,t)\times {ATT}(e,t)$$

(3)

Where $w(e,t)$ is the weight used for the weighted average. Callaway and Sant’Anna (2021) suggest that the proportion of the group that accepts the treatment time e, denoted as $P(E={e|E}\le T)$, can be used as the weight, thereby defining the parameter corresponding to the average treatment effect¹⁰:

$${\theta }_{w}^{0}=\frac{1}{K}{\sum }_{e}{\sum }_{t=2}^{T}1\left\{t\ge e\right\}{ATT}(e,t)P\left(E={e|E}\le T\,\right)$$

(4)

where $K={\sum }_{e}{\sum }_{t=2}^{T}1\left\{t\ge e\right\}P\left(E={e|E}\le T\right)$.

We use the annual ambient PM_2.5 concentration levels for each county as a measure of local pollution. To estimate the impact of CCER programs on pollution levels, we apply a DID method based on the following model:

$${{{\mathrm{ln}}}E}_{{it}}=\theta\, * \,{{CCER}}_{{it}}+{X}_{{it}}^{{\prime} }\, * \,\beta+{\delta }_{t}+{\gamma }_{i}+{\mu }_{{kt}}+{\varepsilon }_{{it}}$$

(5)

where lnE_it represents the natural log of the ambient concentration level of PM_2.5 in county i in year t. Other variables follow the same definition as in Eq. (1). We also adopt CSDID as the main identification strategy.

Recent literature suggests that wind direction affects air pollution concentrations^8,20. Due to the annual aggregation of county-level data, we first note that a single wind direction does not generally influence counties on an annual basis because wind directions change across seasons, although there are a few counties that have dominant wind directions. We address the impact of wind direction on pollutant concentration by excluding observations associated with these dominant wind directions. To conduct our analysis, we obtain daily wind direction data for all counties from the China Meteorological Data Network and then calculate the frequency and proportion of each wind direction across all counties. We find that only 59 out of 1699 counties have a dominant wind direction that prevails for at least 40% of days. For the remaining counties, wind directions vary over time, and a single prevailing direction does not influence the counties on an annual basis. Supplementary Tables 21–23 show detailed wind direction information.

Finally, other environmental policies or regulations may drive our results during the same study period²¹. To control for the potential confounders, we consider the impact of the pilot projects for carbon emission trading, air quality standards, and air quality information disclosure regulations. The pilot projects for carbon emission trading is a regional mandatory carbon trading system initiated by China in 2013, which covers seven province-level administrative regions in China, and aims to reduce greenhouse gas emissions and promote green development^22,23. Air quality standards refer to a new standard proposed by China in 2012, specifically aimed at reducing air pollutant concentrations, such as fine particulate matter. Information disclosure refers to China’s air quality monitoring and real-time data disclosure program launched in 2013, to enhance public access to pollution information and increase awareness of air quality^24,25. We incorporate the three policies as new variables in the main regressions to further control for time-varying unobservables.

Triple difference model (DDD)

To obtain the effects of CCER projects on environmental justice, we use a DDD approach to estimate the PM_2.5 concentration gap between economically disadvantaged and affluent counties due to program interventions. We specify the empirical regression model as follows:

$${{{\mathrm{ln}}}E}_{{it}}= {\alpha }_{1}\, * \,{{CCER}}_{{it}}\, * \,{{Poor}}_{{it}}+{\alpha }_{2}\, * \,{{CCER}}_{{it}}+{\alpha }_{3}\, * \,{{Poor}}_{{it}}+{X}_{{it}}^{{\prime} }\\ * \,\beta+{\delta }_{t}+{\gamma }_{i}+{\mu }_{{kt}}+{\varepsilon }_{{it}}$$

(6)

where the variables follow the same definition as in Equation (1) and Poor_it is a dummy variable that equals 1 if county i is categorized as an economically disadvantaged county if it is on the Poverty County List released by China’s National Rural Vitalization Administration in 2014, and zero otherwise in year t. The status of economically disadvantaged counties changes, and the variable Poor_it possesses temporary variations as indicated by the subscript t. We use counties with similar local economic activities proxied by night lights so that we are not comparing the extremely economically disadvantaged counties with extremely more affluent counties. The coefficient of ${{CCER}}_{{it}} * {{Poor}}_{{it}}$, ${\alpha }_{1},$ is of primary interest as it measures the effect of CCER programs on air pollution exposure gaps between economically disadvantaged and more affluent counties. We also investigate the relationship between the CCER program and ethnic disparities in exposure to PM_2.5 by replacing Poor_it with Minority_it in Eq. (6). The empirical regression model is specified as follows:

$${{{\mathrm{ln}}}E}_{{it}}= {\alpha }_{1}\, * \,{{CCER}}_{{it}}\, * \,{{Minority}}_{{it}}+{\alpha }_{2}\, * \,{{CCER}}_{{it}}+{\alpha }_{3}\\ * \,{{Minority}}_{{it}}+{X}_{{it}}^{{\prime} }\, * \,\beta+{\delta }_{t}+{\gamma }_{i}+{\mu }_{{kt}}+{\varepsilon }_{{it}}$$

(7)

where Minority_it is a dummy variable that equals 1 if the county i belongs to the list of minority counties, and 0 otherwise. The status of minority counties may also change, so Minority_it is time-varying. The other variables have the same definitions as in Eq. (6).

Event study analysis

We empirically test the plausibility of the parallel trend assumption by modifying Eq. (6) into a dynamic DDD model as follows²⁶:

$${{{\mathrm{ln}}}}{{{\rm{E}}}}_{{it}}= {\sum }_{m=-5}^{-2}({{\pi }_{m}{CCER}}_{{im}}\, * \,{{Poor}}_{{it}}+{\tau }_{m}{{CCER}}_{{im}}+{\in }_{m}{{Poor}}_{{im}})\\ +{\sum }_{m=0}^{5}({{\omega }_{m}{CCER}}_{{im}}\, * \,{{Poor}}_{{it}}+{\varphi }_{m}{{CCER}}_{{im}}+{\rho }_{m}{{Poor}}_{{im}})\\ +{X}_{{it}}^{{\prime} }\, * \,\beta+{\delta }_{t}+{\gamma }_{i}+{\mu }_{{kt}}+{\varepsilon }_{{it}}$$

(8)

where CCER_im is an indicator variable that equals 1 if it is m years from the year of CCER placement in county i. We omit CCER_im = -1 as the base group in Eq. (8). Similarly, we test the plausibility of the parallel trend assumption by modifying Eq. (7) into a dynamic DDD model as follows:

$${{{\mathrm{lnE}}}}_{{it}}= {\sum }_{m=-5}^{-2}({{\pi }_{m}{CCER}}_{{im}}\, * \,{{Minority}}_{{it}}+{\tau }_{m}{{CCER}}_{{im}}+{\in }_{m}{{Minority}}_{{im}})\\ +{\sum }_{m=0}^{5}({{\omega }_{m}{CCER}}_{{im}}\, * \,{{Minority}}_{{it}}+{\varphi }_{m}{{CCER}}_{{im}}+{\rho }_{m}{{Minority}}_{{im}})\\ +{X}_{{it}}^{{\prime} }\, * \,\beta+{\delta }_{t}+{\gamma }_{i}+{\mu }_{{kt}}+{\varepsilon }_{{it}}$$

(9)

We further explore the effects of CCER project capacity on CO₂ emissions, PM_2.5 concentration levels, and the environmental justice gap by substituting CCER in the above four equations with Capacity. Capacity refers to the county-level carbon reduction capacity contributed by CCER. We use CCER carbon reduction emissions approved by third-party certification agencies as the data source and construct county-level continuous variables based on CCER project establishment time for analysis.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.