China’s carbon emission trading and income inequality: evidence from factor substitution Authors: Shixiong He, Lei Jiang, Mengxu Li, Jianghua Liu Published: 2026 ABSTRACT As carbon pricing mechanisms gain global prominence, understanding their distributional impacts, particularly in developing economies, becomes crucial for the design of equitable climate policies. We explore the effect of carbon emission trading (CET) policy on income inequality in China through integrated theoretical and empirical analysis. We find that the CET policy significantly reduces income inequality, and this conclusion passed the robustness tests. Theoretical analysis shows that if carbon emission allowances are treated as an input factor of production, when carbon prices are relatively low, firms tend to substitute labor for carbon emission allowances, thereby increasing the labor income share (LIS). Empirical testing also confirms that the mitigation of income inequality through CET is attributable to the increase in the LIS. In addition, the equalizing effect is more pronounced in cities with higher initial income inequality and greater marketization. Government environmental expenditure and smooth labor market mobility also have positive synergistic effects. Our research provides policy insights on balancing emissions reduction with social equity. Introduction Climate action and reduced inequalities are two of the Sustainable Development Goals (SDGs) proposed by the United Nations1. Against the backdrop of rising climate change, the carbon emission trading (CET) policy is demonstrated to be effective in mitigating carbon emissions and has become one pillar of climate mitigation [3, 25, 30]. However, the impact of the CET policy on income inequality remains limited and ambiguous. On the one hand, the implementation of the CET policy will generate an environmental regulation effect [2, 49]. The increased regulatory intensity may adversely affect the labor market through industrial contraction and structural adjustment, particularly increasing unemployment risks for low-skilled workers employed in carbon-intensive sectors [26] and enlarging income inequality. On the other hand, the CET policy is supposed to drive low-carbon transition and expand the renewable energy and environmental protection industries [38], thus creating substantial employment opportunities and reallocating the low-skilled workers. The redistributive effect is supposed to mitigate income inequality. In our view, the opinion division in the existing studies partly stems from an insufficiently explicit articulation of the factor substitution effect of the CET policy. As a market-based environmental regulation, China’s CET policy is still in its initial stage, with relatively low carbon prices and a regulatory intensity lower than that of typical command-and-control policies. Consequently, firms’ responses may differ. In theory, if the implementation of the CET policy increases the labor income share (LIS), particularly among low-skilled workers, income inequality is likely to reduce. Conversely, the CET policy may exacerbate income inequality. Therefore, we focus on the causal effect of the CET policy on income inequality from the perspective of factor substitution in this paper. In detail, we leverage the CET policy in China and conduct causal inference based on the difference-in-differences (DID) framework. China is the largest carbon emitter worldwide and has the biggest CET market within a country. The wide coverage of the CET scheme in one economy with a unified institution and regional heterogeneity provides an ideal scenario for identifying the effect of the CET scheme. It is shown that the implementation of the CET policy can mitigate the income inequality by 3.53%, which is statistically and economically significant. This finding is verified as credible by a series of robustness tests, including conducting placebo tests, adopting alternative estimation methods and variables, and eliminating the interference of relevant policies. The mechanism analysis results reveal that the CET policy mitigates income inequality by increasing the share of labor income. The factor substitution effect of the CET policy is demonstrated to be beneficial in narrowing the income gap. Heterogeneity analyses identify the impact of initial income inequality and the role of government, market, and citizens. The results indicate that the equity effect of the CET policy is more pronounced in cities with higher initial income inequality. In addition, greater marketization, government environmental expenditure, and labor mobility contribute to inequality mitigation. The spatial analysis observes the unintended equity consequences in neighboring regions due to the CET policy. This study provides three marginal contributions to the existing literature. First, we enrich the literature on the income distributional effects of CET by identifying a new mechanism from the perspective of labor-oriented factor substitution under an early-stage carbon market. Existing studies have shown that CET may affect income inequality through revenue recycling schemes [52], urban-rural heterogeneity [51], the distribution of policy benefits [20], energy structure optimization [48], innovation gap reduction and factor allocation improvement [58]. Unlike these studies, our study demonstrates that labor-oriented factor substitution is an important channel through which CET reduces income inequality in China’s early-stage carbon market, providing a new explanation for the distributional effects of carbon pricing. Second, we contribute to the literature on income inequality mitigation by identifying the institutional conditions under which CET is more effective. Rather than simply documenting the average policy effect, we show that government environmental expenditure, market development, and labor mobility jointly strengthen the equity-enhancing effect of CET, providing a more comprehensive understanding of the conditions required for effective policy implementation. Third, we contribute to the literature on environmental justice in developing economies by providing evidence from China’s early-stage carbon market. Existing studies mainly focus on mature carbon markets such as the EU ETS [3, 9]. Different from these evidence, we show that, in the absence of a strong Porter effect, firms tend to rely on labor-oriented adjustments rather than innovation to comply with emission constraints. Our findings suggest that gradual carbon market development with relatively low initial carbon prices can simultaneously promote carbon reduction and social equity, offering practical implications for developing economies. The research framework of the paper is shown in Fig. 1. And the remainder of this paper is arranged as follows. Section "Institutional background and research hypotheses" presents the institutional background and proposes research hypotheses. Section "Methodology and data" introduces the methodology, variables, and data. Section "Empirical results" exhibits the empirical results. And Section "Concluding remarks" provides concluding remarks. Fig. 1Analytical framework of the research Analytical framework of the research Institutional background and research hypotheses Institutional background Global climate change has profoundly affected the economies, societies, and ecosystems. To achieve emission reduction targets, countries have been actively employing market-based mechanisms [29]. As a core instrument, the carbon pricing mechanism internalizes the external costs of carbon emissions and provides clear price signals to emitters, thereby effectively encouraging carbon reduction. Carbon pricing mechanisms have expanded rapidly across the globe. By 2025, 37 emissions trading systems and 43 carbon taxes are in force, covering nearly 28% of global greenhouse gas emissions [40]. As the largest carbon emitter around the world, China has consistently placed carbon reduction in a strategic position and is committed to fulfilling its due responsibilities [18, 21, 28]. To achieve efficient carbon reduction through market-based policies, China approved the establishment of CET pilots as early as 2011. Thereafter, seven CET pilots were successively launched between 2013 and 2014. Building on the positive progress of the pilot program, Fujian and Sichuan joined the pilot scheme in 2016, which further expanded the coverage and influence of the CET market. In 2020, the Chinese government formally announced its strategic targets of “Dual Carbon Goals”. To accelerate the progress toward “Dual Carbon Goals”, China launched the national CET market in 2021. As a key industry for green transition, the power sector was the first to be covered by the national CET market. In 2024, the cement, steel, and electrolytic aluminum industries were also included in the national CET market. With the expansion of sectoral coverage, CET is becoming an important policy tool for carbon reduction. While the primary objective of the CET is to reduce carbon emissions, the policy may also have important distributional implications. As China pursues both the “Dual Carbon Goals” and the objective of common prosperity, understanding whether carbon pricing policies influence income inequality has become an increasingly important policy concern. Evaluating the distributional effects of the CET therefore not only contributes to assessing its overall effectiveness but also provides evidence for coordinating environmental sustainability with equitable economic development. Research hypotheses To explain the relationship among CET, LIS, and income inequality, this paper starts from an initial equilibrium and examines how firms adjust their input structures when faced with a carbon price shock, eventually reaching a new equilibrium. Since the analysis begins from an initial equilibrium, the theoretical framework in this paper is most applicable to the early stage of CET, when carbon prices are relatively low [14]. This setting is also consistent with the current stage of development of China’s CET market. To understand firms’ responses when facing the carbon costs generated by CET, we treat carbon emission allowances as an input for carbon-intensive sectors. In traditional production processes, environmental capacity is treated as a freely accessible public resource. However, the CET policy imposes a total emissions cap, thereby transforming environmental capacity into a scarce resource with clearly defined property rights [13]. Firms must hold sufficient emission allowances to cover their emissions in order to carry out production activities. This implies that carbon emission allowances become a necessary input and a cost constraint in the production process. When CET is implemented and a carbon price is established, the production costs of carbon-intensive firms increase. Firms therefore need to adjust the structure of capital, labor, and carbon emission allowances to absorb the carbon cost. In the initial term following the implementation of CET, carbon prices are relatively low, and firms typically do not immediately undertake large-scale green technological investments or capital equipment upgrades. This is because, in many carbon-intensive industries, capital equipment and energy use—and the associated emissions—are often embedded in large production facilities and existing production lines, making it difficult to achieve significant adjustments through technological innovation or equipment replacement in the short run [33]. Moreover, capital-intensive production modes are usually associated with higher energy consumption and carbon emissions [7]. Therefore, when firms face relatively weak carbon cost constraints, they are unlikely to achieve emission reductions through immediate capital substitution. By contrast, labor input is more flexible, and firms therefore tend to prioritize increasing labor input to alleviate carbon cost pressures. One typical approach is to expand producer service activities built around core manufacturing operations [34]. These activities not only help extend firms’ value chains but are also generally more labor-intensive. Through this strategy, firms can partially reduce their reliance on carbon-intensive production activities without substantially adjusting their capital equipment. Therefore, in the early stage after the introduction of carbon cost constraints through the CET policy, firms are more likely to reduce carbon-intensive inputs by increasing labor input rather than immediately achieving emission reductions through capital substitution. In other words, the CET policy may encourage firms to shift part of their input structure away from carbon-intensive production activities toward greater reliance on labor input, thereby increasing LIS. This firm-level adjustment may, at the city level, manifest as inter-sectoral resource reallocation, whereby resources shift from relatively carbon-intensive sectors to low-carbon sectors with stronger labor absorption capacity, thereby promoting a low-carbon transformation of the industrial structure. Such a low-carbon structural transformation can increase the output of labor-intensive and low-carbon sectors, such as services. As a result, labor may gradually move from high-carbon sectors to low-carbon sectors, thereby further increasing LIS [27, 37, 46]. We further explain how an increase in the LIS can help mitigate income inequality. Household income is primarily composed of labor income and capital income, with the distribution of capital income being far more concentrated than that of labor income [10]. Capital income is highly concentrated among high-income groups, while labor income is more widely distributed among residents. An increase LIS means that the overall distribution of national income will shift towards a more equitable direction, helping to alleviate income inequality caused by the concentration of capital income. Numerous empirical studies also confirm a negative association between the LIS and income inequality [8, 11]. Figure 2 summarizes the mechanism through which CET influences income inequality via factor substitution. On the basis of the theoretical analysis, the following hypotheses are proposed: H1: The implementation of the CET policy reduces income inequality. H2: The CET policy encourages firms to increase labor input, thereby raising the LIS and ultimately reducing income inequality. Fig. 2Mechanism of CET mitigating income inequality through factor substitution Mechanism of CET mitigating income inequality through factor substitution Methodology and data Methodology We apply the DID model as our main research method, as described below:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&LnInequalit{y_{it}}={\alpha _0}+{\alpha _1}Trea{t_i}*Pos{t_t}\\&+\gamma Contro{l_{it}}+{\mu _i}+{\lambda _t}+{\varepsilon _{it}}\end{aligned}$$\end{document} Inequality represents income inequality. Treat denotes a binary classification reflecting whether a city is in the CET policy. It takes the value of 1 for cities included in the pilot scheme, and 0 for those that are not. Post is a dummy time indicator representing the year before and after the policy implementation. α₁ is the parameter to be estimated, reflecting CET’s policy effect. Variables Dependent variable Income Inequality (Inequality). Urban-rural income gap exhibits strong synchronicity with aggregate inequality and consistently plays a dominant role in its evolution [39, 45, 55]. Accordingly, we use urban-rural income inequality as the dependent variable, which is defined as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Inequalit{y_{it}}=UD{I_{it}}/RD{I_{it}}$$\end{document} where Inequality denotes the degree of income inequality in prefecture-level. UDI and RDI express the per capita disposable income of urban and rural residents, respectively. Explanatory variable Treat*Post. In October 2011, the National Development and Reform Commission issued the establishment of CET markets in Beijing, Tianjin, Shanghai, Chongqing, Hubei, and Guangdong, and these CET markets were put into operation successively between late 2013 and 2014. Therefore, the period before 2014 is defined as the policy implementation pre-phase, and 2014 onward as the post-implementation phase. The treatment group consists of 36 cities located in CET pilot markets, and the other 245 cities form the control group. Control variables Our baseline regression includes a set of control variables, as detailed below: (1) Economic Development Level (GDPPC). We use GDP per capita to measure economic development [16]. (2) Urbanization (Urban), which is an important factor influencing income inequality, particularly in China, where the urban-rural dual structure is pronounced [36]. We use cities’ average nighttime light intensity as a proxy for urbanization [15]. (3) Industrialization (Industry). Industrial development significantly affects employment opportunities and income levels. Industry is proxied by the ratio of value added from the secondary sector relative to GDP. (4) Internet Development (Internet). The development of the internet is beneficial for non-agricultural employment and will broaden income sources for rural residents, thereby raising rural income levels [5]. We use internet user numbers in each city as a proxy for internet development. (5) Fiscal Expenditure (Fiscal). Government investment in social infrastructure improves the efficiency of economic activities and is beneficial for rural and less-developed regions [23, 50]. Fiscal is the ratio of public budget expenditure relative to GDP. (6) Financial Development (Finance). Urban residents have greater access to financial services such as loans, insurance, and investments, which enhances their opportunities for income growth and wealth accumulation, while rural areas typically lack comparable financial resources. Thus, financial development may exacerbate income inequality [4, 24]. Finance is proxied by the ratio of year-end total loan balances to GDP. Data This study focuses on the policy effects of China’s pilot CET markets. Compared with the national CET market, which initially covered only the power generation sector, the pilot CET markets encompass a broader range of industries, including not only high energy-consuming sectors but also locally distinctive industries. To avoid potential interference from the national CET market in identifying the policy effects of the pilot programs, this study restricts the sample period to the years prior to the official launch of the national CET market in 2021, namely 2010–2020. Specifically, the research sample consists of data from 281 cities in China over the period 2010–2020. The data are sourced from the National Bureau of Statistics of China. Descriptive statistics are summarized in Table 1. Table 1Distributional Characteristics of the VariablesTypeVariableObs.Mean.SDMin.Max.Dependent variable Inequality 3,0912.3910.5001.2234.616Explanatory variable Treat*Post 3,0910.0820.27401Control variables GDPPC 3,09151,26232,0065304256,877 Urban 3,0917.7558.9720.13059.12 Industry 3,09146.5510.9610.6889.75 Internet 3,091102.5154.53.0205,174 Fiscal 3,0910.1990.1020.04391.485 Finance 3,0910.9950.6050.1327.450 Table 1Distributional Characteristics of the VariablesTypeVariableObs.Mean.SDMin.Max.Dependent variable Inequality 3,0912.3910.5001.2234.616Explanatory variable Treat*Post 3,0910.0820.27401Control variables GDPPC 3,09151,26232,0065304256,877 Urban 3,0917.7558.9720.13059.12 Industry 3,09146.5510.9610.6889.75 Internet 3,091102.5154.53.0205,174 Fiscal 3,0910.1990.1020.04391.485 Finance 3,0910.9950.6050.1327.450 Distributional Characteristics of the Variables Empirical results Baseline results We gradually add the control variables to the baseline model. As reported in Table 2, Treat*Post consistently exhibits a significant negative coefficient. These results indicate that CET significantly reduces income inequality, thereby confirming H1. Column (7) reports the baseline regression results including all control variables. In Column (7), the coefficient on Treat*Post is − 0.036, indicating that CET policy reduces income inequality in pilot regions by approximately 3.53% on average (e− 0.036−1 ≈ − 0.0353). The baseline results are consistent with our expectations. This study’s findings are consistent with existing research using China’s CET market as the empirical context. For instance, Fang et al., [12], based on provincial-level panel data, and Xiao et al., [44], employing a difference-in-differences method, both find that CET helps alleviate income inequality in China. However, in contrast to these findings, Känzig, [22], which examines the more mature European Union CET market, reports that CET has a larger negative impact on the income of low-income groups. Compared with the EU CET market, China’s CET market started later and exhibits the typical features of “low-price initiation and gradual expansion”. The inequality-reducing effect of the CET policy reflects the Chinese government’s effort to balance efficiency and equity in the process of promoting a low-carbon transition. It is also consistent with the original intention of China’s CET policy design, which aims to avoid negative impacts on the goal of common prosperity by adopting a low initial carbon price, piloting the policy in selected regions, and gradually expanding its coverage. The cautious design of China’s CET system also provides an important reference for other developing countries. Table 2Benchmark resultsVariableDependent variable: LnInequality(1)(2)(3)(4)(5)(6)(7) Treat*Post −0.043***−0.037**−0.035**−0.035**−0.041***−0.038***−0.036***(0.015)(0.015)(0.014)(0.015)(0.014)(0.013)(0.013) LnGDPPC −0.102***−0.091***−0.088***−0.072***−0.098***−0.091***(0.017)(0.018)(0.019)(0.018)(0.020)(0.020) LnCity −0.038***−0.038***−0.034**−0.036***−0.037***(0.015)(0.015)(0.014)(0.014)(0.014) Industry −0.0001−0.0003−0.001−0.001(0.001)(0.001)(0.001)(0.001) LnInternet −0.038***−0.038***−0.039***(0.007)(0.008)(0.008) Fiscal −0.261***−0.325***(0.087)(0.078) Finance 0.029***(0.011) Constant 0.971***2.015***1.945***1.928***1.898***2.226***2.139***(0.005)(0.176)(0.182)(0.184)(0.179)(0.211)(0.198)City FE√√√√√√√Year FE√√√√√√√Observations3,0913,0913,0913,0913,0913,0913,091R20.5550.5820.5870.5870.5980.6100.616***, **, * indicate the statistical significance at the 1%, 5%, and 10% levels, respectively. Robust standard errors are shown in parentheses. This note applies to all subsequent tables unless otherwise stated. Table 2Benchmark resultsVariableDependent variable: LnInequality(1)(2)(3)(4)(5)(6)(7) Treat*Post −0.043***−0.037**−0.035**−0.035**−0.041***−0.038***−0.036***(0.015)(0.015)(0.014)(0.015)(0.014)(0.013)(0.013) LnGDPPC −0.102***−0.091***−0.088***−0.072***−0.098***−0.091***(0.017)(0.018)(0.019)(0.018)(0.020)(0.020) LnCity −0.038***−0.038***−0.034**−0.036***−0.037***(0.015)(0.015)(0.014)(0.014)(0.014) Industry −0.0001−0.0003−0.001−0.001(0.001)(0.001)(0.001)(0.001) LnInternet −0.038***−0.038***−0.039***(0.007)(0.008)(0.008) Fiscal −0.261***−0.325***(0.087)(0.078) Finance 0.029***(0.011) Constant 0.971***2.015***1.945***1.928***1.898***2.226***2.139***(0.005)(0.176)(0.182)(0.184)(0.179)(0.211)(0.198)City FE√√√√√√√Year FE√√√√√√√Observations3,0913,0913,0913,0913,0913,0913,091R20.5550.5820.5870.5870.5980.6100.616***, **, * indicate the statistical significance at the 1%, 5%, and 10% levels, respectively. Robust standard errors are shown in parentheses. This note applies to all subsequent tables unless otherwise stated. Benchmark results ***, **, * indicate the statistical significance at the 1%, 5%, and 10% levels, respectively. Robust standard errors are shown in parentheses. This note applies to all subsequent tables unless otherwise stated. Robustness checks Parallel trend test A key requirement when utilizing the DID approach is that cities exposed to CET and those not exposed exhibit comparable outcome trajectories before policy implementation. Hence, we conduct a parallel trends test based on the following event study method.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&LnInequalit{y_{it}}=\alpha +\sum\limits_{{k=2}}^{4} {{\beta _{ - k}}Befor{e_{ - k,it}}} \\&+{\beta _0}Curren{t_{it}}+\sum\limits_{{b=1}}^{6} {{\beta _{+b}}Afte{r_{+b,it}}} \\&+\gamma Contro{l_{it}}+{\mu _i}+{\lambda _t}+{\varepsilon _{it}}\end{aligned}$$\end{document} where Before− k, it, Currentit, and After+ b, it, are dummy variables. For treated cities, Before− k, it (k = 2, 3, 4) equals 1 if k = 2014 − t, and 0 otherwise. Similarly, After+ b, it (b = 1, 2, 3, 4, 5, 6) equals 1 if b = 2014 − t, and 0 otherwise. Currentit equals 1 for treated cities in 2014, and 0 otherwise. Insignificant coefficients on Before− k, it, indicate parallel pre-policy trends between the treatment and control groups. Given the baseline findings that the CET significantly reduces income inequality, the coefficients on After+ b, it are expected to be significantly negative. The year 2013, one year prior to policy implementation, serves as the reference year. The estimated coefficients for the pre-policy periods are statistically indistinguishable from zero (Fig. 3), indicating the absence of systematic differences in trends between the treatment and control cities prior to the introduction of the CET policy. Moreover, except for the year 2020, the coefficients after policy implementation are significantly negative, indicating that CET significantly reduces income inequality. The insignificant effect in 2020 may be attributed to two factors. First, the outbreak of COVID-19 severely disrupted socioeconomic activities. Second, China’s carbon prices did not increase substantially after the policy was implemented, and firms may have already adapted to the carbon cost, resulting in a gradual weakening of the policy’s effect. Fig. 3Result of parallel trend test Result of parallel trend test Placebo test To exclude the possibility that observed differences between cities exposed to CET and those not exposed are driven by other policies or factors, we perform a placebo test as a robustness check. Specifically, cities are randomly assigned as CET cities to construct a pseudo-treatment variable, which is then used to re-estimate the baseline model and record the resulting coefficients. This randomization procedure is performed 500 times. The coefficients estimated for the pseudo-treatment variable are tightly centered around zero (Fig. 4), confirming that the generated pseudo-treatment variable does not significantly influence the income inequality. Moreover, the actual estimated coefficient in the benchmark regression result (−0.036) lies outside the entire simulated distribution. These results confirm that the mitigating effect of CET on income inequality is not driven by other concurrent policies or omitted variables. Fig. 4Result of placebo test Result of placebo test PSM-DID Significant socioeconomic differences exist between treatment and control cities. To mitigate the potential bias arising from these differences, we use the PSM-DID method for robustness testing [19, 54]. A logit regression model is employed to analyze the determinants of a city’s participation in the CET pilot and to calculate the propensity scores. Initially, one-to-one nearest-neighbor matching is adopted for estimation. Figure 5 presents the covariate bias plot after PSM matching. The bias of each variable is significantly reduced and remains within 10%, indicating that the matching method used in this study is effective. We also used radius matching and nearest neighbor matching to match the sets of covariates. The standardized differences after matching were also below 10%. Fig. 5Balance test after PSM matching Balance test after PSM matching Using matched samples generated through three matching approaches, we re-estimate the effect of CET on income inequality. Table 3 shows that Treat*Post consistently exhibits significantly negative effects across all three matching methods. The findings corroborate the baseline results, supporting the conclusion that the CET policy can help mitigate income inequality in China. Table 3PSM-DID resultsVariableDependent variable: LnInequality(1) Nearest-Neighbor Matching(2) Radius Matching(3) Kernel Matching Treat*Post −0.049***−0.171***−0.171***(0.019)(0.061)(0.061)Controls√√√City FE√√√Year FE√√√Observations764396396R20.7120.8010.801 Table 3PSM-DID resultsVariableDependent variable: LnInequality(1) Nearest-Neighbor Matching(2) Radius Matching(3) Kernel Matching Treat*Post −0.049***−0.171***−0.171***(0.019)(0.061)(0.061)Controls√√√City FE√√√Year FE√√√Observations764396396R20.7120.8010.801 PSM-DID results Synthetic difference-in-differences The Synthetic Difference-in-Differences (SDID) method is also an important method for evaluating policy effects. The weighting mechanism of SDID does not rely entirely on the parallel trends test. By combining the strengths of both Synthetic Control Method and DID, SDID offers advantages in handling systematic differences between the experimental and control groups. The coefficient of the SDID model is estimated as below:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\left( {{{\hat {\tau }}^{sdid}},\hat {\mu },\hat {\alpha },\hat {\beta }} \right)\\&=\underbrace {{\arg \hbox{min} }}_{{\tau ,\mu ,\alpha ,\beta }}\left\{ {\sum\limits_{{i=1}}^{N} {\sum\limits_{{t=1}}^{T} {{{\left( {{Y_{it}} - \mu - {\alpha _i} - {\beta _i} - {W_{it}}\tau } \right)}^2}\hat {w}_{i}^{{sdid}}\hat {\lambda }_{t}^{{sdid}}} } } \right\}\end{aligned}$$\end{document} where τ is the parameter to be estimated. After reweighting the control group to enhance comparability, the estimated coefficient of Treat*Post is −0.034 and significant, supporting that the benchmark findings are robust. Alternative measure of the dependent variable In the existing literature, the urban-rural income gap has been one of the most commonly used proxies for income inequality. However, relying on the urban-rural income ratio essentially substitutes a partial structural disparity for the overall distributional condition. In addition, official statistics may contain statistical errors during their collection and aggregation. To address these limitations and further test the robustness of our baseline results, we employ township-level nighttime light data to construct alternative measures of income inequality, namely the Gini coefficient and the Theil index. Compared with conventional statistical indicators, township-level nighttime light data help mitigate potential measurement errors arising from official statistics [53, 56, 57]. Moreover, by moving beyond the urban-rural dichotomy, these data provide a more comprehensive reflection of within-region income distribution disparities. Specifically, following Gao et al., [17], this study uses calibrated NPP-VIIRS-like nighttime light data from 2010 to 2020 to construct two prefecture-level inequality indicators: the nighttime light Gini coefficient (LightGini) and the nighttime light Theil index (LightTheil). These indicators are then used as alternative dependent variables for the robustness analysis. The calculation for LightGini and LightTheil are as follows:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$LightGin{i_{it}}=\frac{{\sum\limits_{{k=1}}^{{{n_i}}} {\sum\limits_{{r=1}}^{{{n_i}}} {\left| {L{C_{ikt}} - L{C_{irt}}} \right|} } }}{{2{n_i}^{2}{{\overline {{LC}} }_{it}}}}$$\end{document}6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$LightThei{l_{it}}=\sum\limits_{{k=1}}^{n} {\frac{{L{C_{ikt}}}}{{L{C_{it}}}}} Ln\left( {\frac{{{{L{C_{ikt}}} \mathord{\left/ {\vphantom {{L{C_{ikt}}} {L{C_{it}}}}} \right. \kern-0pt} {L{C_{it}}}}}}{{{n_i}}}} \right)$$\end{document} where LightGiniiₜ denotes the nighttime light Gini coefficient. LightTheiliiₜ denotes the nighttime light Theil. LCiₖₜ represents the nighttime light value of township-level unit k within city i in year t. LCirₜ represents the nighttime light value of township-level unit r within city i in year t; LC is the total nighttime light value. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline {{LC}}$$\end{document}denotes the average nighttime light value across all township-level units. and ni is the number of township-level units within city i. Larger values of LightGini and LightTheil indicate higher levels of income inequality, while lower values imply a narrower income gap. The regression results obtained by replacing the core income inequality measure with the nighttime light Gini coefficient and the nighttime light Theil index are reported in Table 4. The coefficient on Treat*Post remains significantly negative across specifications, consistent with the baseline findings. This suggests that our main conclusion—that the CET policy alleviates income inequality—remains robust after addressing the potential limitations of the urban-rural income gap measure. Table 4Results of replacing the measurement of the dependent variableVariableDependent variable(1) LnLightGini(2) LnLightGini(3) LnLightTheil(4) LnLightTheil Treat*Post −0.039**−0.034**−0.086**−0.076**(0.017)(0.015)(0.036)(0.032)Controls×√×√City FE√√√√Year FE√√√√Observations3,0693,0693,0693,069R20.6400.6740.6570.697 Table 4Results of replacing the measurement of the dependent variableVariableDependent variable(1) LnLightGini(2) LnLightGini(3) LnLightTheil(4) LnLightTheil Treat*Post −0.039**−0.034**−0.086**−0.076**(0.017)(0.015)(0.036)(0.032)Controls×√×√City FE√√√√Year FE√√√√Observations3,0693,0693,0693,069R20.6400.6740.6570.697 Results of replacing the measurement of the dependent variable Eliminating the interference of other policies In September 2016, China’s government mandated the implementation of the System for the Paid Use and Trading of Energy Consumption Rights (SPTEC) in Zhejiang, Fujian, Henan, and Sichuan. The energy-use quota trading is similar to CET [41]. The introduction of this pilot policy may influence income and employment, thereby potentially interfering with the baseline results. Therefore, we exclude the samples from SPTEC provinces and re-estimate the DID model. The coefficients of CET are consistent with Table 1 after removing the potential influence of SPTEC (Table 5). Table 5Eliminating the interference of other policiesDependent variable: LnInequalityEliminating SPTECEliminating PPGPAVariable(1)(2)(3)(4) Treat*Post −0.052***−0.041***−0.055***−0.041***(0.015)(0.014)(0.014)(0.013)Controls×√×√City FE√√√√Year FE√√√√Observations2,4862,4861,9671,967R20.5000.5720.4700.531 Table 5Eliminating the interference of other policiesDependent variable: LnInequalityEliminating SPTECEliminating PPGPAVariable(1)(2)(3)(4) Treat*Post −0.052***−0.041***−0.055***−0.041***(0.015)(0.014)(0.014)(0.013)Controls×√×√City FE√√√√Year FE√√√√Observations2,4862,4861,9671,967R20.5000.5720.4700.531 Eliminating the interference of other policies In March 2016, the Chinese government issued the Opinions on Implementing Photovoltaic Power Generation Poverty Alleviation Work (PPGPA). Through measures such as rural power grid upgrades and photovoltaic poverty alleviation, the income levels in rural areas were effectively improved [47]. We exclude the relevant samples for the years 2017–2020 and re-estimate the benchmark regression model. We can observe that, after removing the potential influence of PPGPA, the coefficients of Treat*Post also exhibit significantly negative (Table 5). Mechanism analysis In the Research hypotheses, we propose that if carbon emission allowances are treated as an input factor, introducing carbon costs may encourage firms to shift part of their input structure away from carbon-intensive production activities toward greater reliance on labor input, thereby increasing LIS. Since the mitigating effect of LIS on income inequality has been widely confirmed in the existing literature from both theoretical and empirical perspectives [8, 11, 42], the empirical analysis in this subsection focuses primarily on examining the impact of the CET policy on LIS. Table 6 reports the regression results with LIS as the dependent variable. Column (1) presents the estimation without control variables, while Column (2) reports the results after including the full set of controls. As shown in Table 6, the coefficient on Treat*Post is significantly positive, indicating that the CET policy has a significant positive effect on increasing the labor income share. Accordingly, H2 is confirmed. In Column (2), the coefficient on Treat*Post is 0.119, implying that the CET policy increases the LIS in pilot regions by approximately 12.64% on average (e0.119−1 ≈ 0.1264). This result suggests that the CET policy assigns market value to carbon emission allowances and increases the demand for labor through factor substitution effects, thereby raising LIS. It should be noted that, given the well-established relationship between LIS and income inequality in the literature, this study focuses mainly on examining the effect of CET on LIS in the mechanism analysis. We do not implement a formal causal mediation analysis to strictly identify the mediating role of LIS. Nevertheless, the results in this subsection provide empirical support for understanding the potential mechanism through which CET influences income inequality. Table 6The effect of CET on LISVariableDependent variable: LnLIS(1)(2) Treat*Post 0.090*0.119**(0.053)(0.048)Controls×√City FE√√Year FE√√Observations3,0913,091R20.3820.456LnLIS = Ln[LIS/(1 − LIS)] Table 6The effect of CET on LISVariableDependent variable: LnLIS(1)(2) Treat*Post 0.090*0.119**(0.053)(0.048)Controls×√City FE√√Year FE√√Observations3,0913,091R20.3820.456LnLIS = Ln[LIS/(1 − LIS)] The effect of CET on LIS LnLIS = Ln[LIS/(1 − LIS)] Heterogeneity analysis Initial income inequality In this subsection, the average income inequality level of each prefecture-level city during 2010–2013 is used as the initial income inequality measure. We divide the research sample into two parts based on the overall sample mean. CET significantly reduces income inequality in both groups (Table 7). In cities with lower initial income inequality, the coefficient remains negative but is insignificant. These results support that CET has a stronger inequality-reducing effect in cities with higher initial income inequality. Table 7Heterogeneity analysis from the perspective of initial income inequality and marketizationDependent variable: LnInequalityInitial income inequalityMarketizationVariable(1) High(2) Low(3) High(4) Low Treat*Post −0.070***−0.008−0.040*−0.030*(0.015)(0.036)(0.022)(0.016)Controls√√√√City FE√√√√Year FE√√√√Observations1,5401,5511,3641,727R20.6000.6390.7580.538 Table 7Heterogeneity analysis from the perspective of initial income inequality and marketizationDependent variable: LnInequalityInitial income inequalityMarketizationVariable(1) High(2) Low(3) High(4) Low Treat*Post −0.070***−0.008−0.040*−0.030*(0.015)(0.036)(0.022)(0.016)Controls√√√√City FE√√√√Year FE√√√√Observations1,5401,5511,3641,727R20.6000.6390.7580.538 Heterogeneity analysis from the perspective of initial income inequality and marketization Marketization Marketization facilitates the transmission and acquisition of supply and demand and price information in the carbon market. This enables carbon market participants to make more rational decisions, thereby making the CET market more effective [32]. We use the average marketization level of each prefecture-level city during 2010–2013 as its marketization indicator, and divide the research sample into two parts based on the overall sample mean. The marketization index is constructed following [35, 43]. As shown in columns (3)-(4) of Table 7, CET significantly reduces income inequality in both groups. However, the mitigating effect of CET on income inequality is more pronounced when the cities have a higher degree of marketization. Environmentally targeted fiscal expenditure Government expenditures in environmental protection typically encompass investments in renewable energy, clean technologies, and environmental infrastructure. Such expenditures can enhance green innovation, expand the supply of renewable energy, and improve firms’ technical conditions for energy use and pollution control, thereby reducing their carbon abatement costs. Lowering emission reduction costs allows companies to trade more carbon allowances in the market, increasing market liquidity. In other words, government green fiscal expenditure can create synergistic effects with the CET market. Therefore, it is necessary to conduct a heterogeneity analysis based on the government’s environmentally targeted fiscal expenditure. We classify cities based on the average level of the government’s environmentally targeted fiscal expenditure during 2010–2013, before policy implementation. Cities with expenditures above the mean value are categorized as the high-expenditure part, and the other cities constitute the low-expenditure part. Columns (1)-(2) of Table 8 show that CET exhibits a significant narrowing of income inequality in both groups. However, the magnitude of the coefficients indicates that CET is more effective when the government has a higher level of spending on environmental protection. Such expenditures enable businesses to acquire more advanced green technologies and infrastructure, thereby allowing them to participate more actively in CET markets. Labor market mobility Labor mobility refers to the phenomenon in which workers relocate across different regions, industries, sectors, employment types, enterprises, or even job positions in pursuit of higher income. Labor market mobility refers to the ability of labor to move freely within the market [1]. A higher degree of labor market mobility implies fewer barriers for workers in seeking better job opportunities. Therefore, higher labor market mobility may have a synergistic effect with CET. CET can alter both the scale and structure of firms’ labor demand. Consequently, in regions with higher labor market mobility, CET may be more effective in mitigating income inequality. The research sample is divided into two parts based on the median value of labor market mobility: a high mobility group and a low mobility group. Regression analyses are then conducted for each group using Eq. (1). We can observe that higher labor market mobility strengthens the inequality-reducing effect of CET (Table 8). Therefore, the reform and improvement of the labor market are also crucial. In particular, the government should optimize the policy environment to facilitate cross-regional and cross-industry labor mobility and reduce the costs associated with labor migration. Table 8Heterogeneity analysis: Environmentally targeted fiscal expenditure and labor market mobilityDependent variable: LnInequalityEnvironmentally targeted fiscal expenditureLabor market mobilityVariable(1) High(2) Low(3) High(4) Low Treat*Post −0.053**−0.033**−0.059***−0.005(0.036)(0.015)(0.017)(0.020)Controls√√√√City FE√√√√Year FE√√√√Observations8032,2881,4901,320R20.5260.6510.6530.552 Table 8Heterogeneity analysis: Environmentally targeted fiscal expenditure and labor market mobilityDependent variable: LnInequalityEnvironmentally targeted fiscal expenditureLabor market mobilityVariable(1) High(2) Low(3) High(4) Low Treat*Post −0.053**−0.033**−0.059***−0.005(0.036)(0.015)(0.017)(0.020)Controls√√√√City FE√√√√Year FE√√√√Observations8032,2881,4901,320R20.5260.6510.6530.552 Heterogeneity analysis: Environmentally targeted fiscal expenditure and labor market mobility Analysis of spatial spillover effects CET can impact technological progress, industrial structure, and employment. These impacts may also spill over into surrounding regions. Income inequality tends to exhibit spatial clustering characteristics. Therefore, building on the benchmark regression results, this subsection incorporates spatial dependence into the empirical analysis to identify both the direct and spatial spillover effects of the CET policy. Spatial autocorrelation analysis We employ the Moran’s I index, originally developed by Moran [31] and extensively adopted in spatial econometric research, to examine whether income inequality exhibits spatial dependence. The formulation of the Moran’s I index is presented as follows.7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I=\frac{n}{{\sum\limits_{{i=1}}^{n} {\sum\limits_{{j=1}}^{n} {{W_{ij}}} } }}\frac{{\sum\limits_{{i=1}}^{n} {\sum\limits_{{j=1}}^{n} {{W_{ij}}({Y_i} - \bar {Y})({Y_j} - \bar {Y})} } }}{{\sum\limits_{{i=1}}^{n} {{{({Y_i} - \bar {Y})}^2}} }}$$\end{document} Where I denotes the Moran’s I value. Yi and Yj represent the income inequality of city i and j, respectively. W denotes the spatial weight matrix (SWM). To conduct a comprehensive examination, we construct four SWMs: (1) Geographical distance SWM (W1); (2) Economic distance SWM (W2); (3) Geo-economic distance SWM (W3); (4) Geo-economic nested spatial weight matrix (W4). The detailed construction methods of the SWMs are provided in the Appendix. Fig. 6Result of Moran’s I. The Moran’s I values are statistically significant at a 1% significance level. Result of Moran’s I. The Moran’s I values are statistically significant at a 1% significance level. Based on Eq. (7), we calculate the Moran’s I index, and the results are presented in Fig. 6. Under all four SWMs, the Moran’s I values are significantly positive at the 1% level, indicating that income inequality is not spatially independent but exhibits significant spatial clustering. This finding suggests that local income inequality is influenced not only by region-specific factors but also by spatial spillover effects from neighboring areas. Therefore, it is necessary to employ spatial econometric models to further examine the spatial impact of the CET on income inequality. Spatial durbin DID Model The Moran’s I index indicates that income inequality exhibits significant spatial clustering. Hence, we employ the Spatial Durbin DID Model (SDM-DID) to conduct further analysis. Based on Eq. (1), the SDM-DID is defined as follows:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&LnInequalit{y_{it}}\\&={\alpha _0}+{\alpha _1}Treat*Pos{t_{it}}\\&+\beta WTreat*Pos{t_{it}}+\rho WLnInequalit{y_{it}} \\&+\gamma Contro{l_{it}}\\&+\theta WContro{l_{it}}\\&+{\mu _i}+{\lambda _t}\\&+{\varepsilon _{it}} \hfill \\ \end{aligned}$$\end{document} Where W denotes SWM, the same as in Eq. (7). The coefficient β is to be estimated and captures the spillover effect of CET. WLnInequalityiₜ represents the spatial lag term of income inequality, and its estimated coefficient ρ reflects the effect of income inequality in neighboring regions on that in the local region. WControlit denotes the spatial lag of the control variables. Table 9Results of Spatial Durbin DID ModelsVariableDependent variable: LnInequality(1)W1(2)W2(3)W3(4)W4 Treat*Post 0.009−0.036***−0.024***−0.017***(0.008)(0.006)(0.006)(0.006) WTreat*Post −0.116**−0.041*−0.086***−0.204***(0.046)(0.021)(0.020)(0.037) ρ 0.856***0.093**0.324***0.523***(0.041)(0.036)(0.032)(0.061)Controls√√√√WControls√√√√City FE√√√√Year FE√√√√Observations3,0913,0913,0913,091R20.1060.2040.2000.181Standard errors are shown in parentheses Table 9Results of Spatial Durbin DID ModelsVariableDependent variable: LnInequality(1)W1(2)W2(3)W3(4)W4 Treat*Post 0.009−0.036***−0.024***−0.017***(0.008)(0.006)(0.006)(0.006) WTreat*Post −0.116**−0.041*−0.086***−0.204***(0.046)(0.021)(0.020)(0.037) ρ 0.856***0.093**0.324***0.523***(0.041)(0.036)(0.032)(0.061)Controls√√√√WControls√√√√City FE√√√√Year FE√√√√Observations3,0913,0913,0913,091R20.1060.2040.2000.181Standard errors are shown in parentheses Results of Spatial Durbin DID Models Standard errors are shown in parentheses We can observe that the spatial lag terms of CET are significantly negative (Table 9), supporting that the CET in neighboring cities also narrows the income inequality in the local region. In fact, when enterprises in neighboring cities adopt labor-substitution or cost-adjustment strategies in response to rising carbon compliance costs, these adjustments can reshape regional production networks. Through industrial chain linkages, upstream and downstream input-output relationships, and cross-regional labor mobility, such responses may influence local factor allocation and ultimately alter the income inequality in surrounding areas [6]. Consequently, the impact of the CET is not confined to pilot cities but may transmit across regions via interregional economic connections and factor flows, giving rise to spatial spillover effects. Concluding remarks As a market-based tool, CET is becoming an important tool for addressing the concern about carbon reduction. This paper aims to explore, from the perspective of factor substitution, whether China’s CET has a mitigating effect on income inequality for the period of 2010 to 2020. Our analysis indicates that: (1) CET significantly mitigates income inequality, and the conclusion passed the robustness tests. (2) When faced with low carbon prices and the early stages of CET markets, firms tend to substitute labor input for carbon emission allowances, which raises the LIS and ultimately reduces income inequality. (3) The equalizing effect is more pronounced in cities with higher initial income inequality and greater marketization. Increased government environmental expenditure and higher labor market mobility further enhance the mitigating effect of the CET on income inequality. (4) CET inevitably has an impact on surrounding regions. We have the policy recommendations as follows: (1) Building on its existing CET markets, China can gradually expand the scope of industries covered by the CET. China’s experience shows that developing countries, when introducing carbon markets, can initially adopt low carbon prices and small-scale pilot policies to avoid exacerbating income inequality. (2) Companies may adopt labor substitution strategies in response to carbon emission allowance constraints, which could affect labor demand and employment. Therefore, the government should further improve labor market policies by reducing barriers to labor mobility, strengthening vocational training and reskilling programs, and enhancing employment services, thereby facilitating labor reallocation and supporting workers during the low-carbon transition. (3) A differentiated regional strategy should be implemented. To fully leverage the dual benefits of emission reduction and promoting equity carbon market, reforms can be prioritized in the regions with well-developed market mechanisms. Due to the spillover effects of CET, the government can establish the regional CET markets in urban clusters with high levels of economic development and complementary industrial structures. (4) For other developing countries, China’s experience suggests that a gradual approach to carbon market development, such as introducing low initial carbon prices and implementing small-scale pilot programs before nationwide expansion, may help balance emission reduction objectives with social equity. Such a phased approach allows firms to gradually adapt to carbon pricing while mitigating adverse impacts on employment and income distribution. The theoretical and empirical analyses in this study primarily focus on the initial stage of China’s CET policy, when carbon prices are relatively low. As the CET system continues to improve, carbon prices gradually rise, and the international climate policy environment evolves (e.g., the implementation of the EU Carbon Border Adjustment Mechanism), China’s future carbon pricing level and the associated environmental regulatory intensity are likely to increase further. Under such circumstances, Firms’ responses to the CET policy may differ from those observed in the initial stage. Therefore, it is important for future research to continue examining the income distribution effects of China’s CET policy across its different stages.