Taking the largest carbon emission trading scheme in the world, EU ETS, as an example, the scheme has a positive inhibitory effect on the overall carbon dioxide emissions of the region (Abrell et al., 2011), but this effect has been uneven in different countries and different industries. For example, Borghesi et al. (2015) found that the implementation of European carbon trading policy effectively reduced the carbon emissions of the Italian manufacturing sector. For the German manufacturing industry, compared with companies that have not joined the carbon emission trading scheme, the companies that have joined the carbon emission trading scheme have reduced the consumption of natural gas and petroleum products, and reduced carbon dioxide emissions by 20% between 2007 and 2010 (Petrick and Wagner, 2014). An empirical study of more than 5,000 Lithuanian companies from 2003 to 2010 by Jaraite-Kažukauske and Di Maria. (2016) shows that EU ETS did not achieve any emission reduction results during the first year of implementation, and was only effective from 2006 to 2007. the EU ETS did not represent a drag on the profitability of participating firms. Wang et al. (2021) showed that the implementation of China’s ETS pilot policies can promote the upgrading of industrial structure. However, differences in economic development and technological levels mean that different regions have different policy effects. Empirical analysis supports Porter’s hypothesis that technological innovation has an intermediary effect on the impact of emissions trading on the upgrading of industrial structure. Zhang S. et al. (2021) verified the positive effects of China’s ETS and found that China’s ETS significantly improved the efficiency of green development and regional carbon equality. Mechanism analysis further shows that China’s ETS promotes regional carbon equality by increasing green total factor productivity and reducing investment in carbon-intensive industries in pilot provinces, confirming the Porter hypothesis and the effect of investment transfer.

Other studies have shown that the carbon emission trading scheme has not yet reached a significant reduction in emissions. There is little evidence the implementation of the NZ ETS has led to any reduction in greenhouse gas emissions (Evison, 2017). Streimikiene and Roos (2009) studied the changes in carbon emissions in the Baltic countries and concluded that the EU ETS has not yet delivered its real potential to reduce emissions in a cost-effective way because of the surplus of greenhouse gas (GHG) emission allowances in first trading period. Cao et al. (2021) conducted a retrospective analysis on the effectiveness of the China’s ETS pilot policy in reducing carbon emissions in the power industry, and found that the China’s ETS pilot policy did not significantly improve the carbon efficiency of coal-fired power plants, because the production activities of power plants are subject to the instructions of the competent authorities, and the power industry in China is highly regulated. Chen et al. (2021) analyzed green patent data of listed companies in 31 provinces (municipalities or autonomous regions) from 1990 to 2018 and found that the China’s ETS pilot policy significantly reduced the proportion of green patents. In addition, companies mainly achieve emission reduction targets by reducing production. However, due to the decrease in cash flow and expected income, companies have reduced their investments in research and development, which is not conducive to green innovation.

There are many determinants of carbon emission intensity. Researches have been deepened step by step from the national level, the regional level to the firm level. Researches are relatively mature on the national level, have been in full swing in the past 2 years on the regional level, while are relatively rare on the firm level.

On the national level, many scholars have found that political factors [such as political polarization (Aller et al., 2021)], economic factors [such as economic growth (Dong et al., 2020b), urbanization (Wang et al., 2021), industrialization (Zhang et al., 2014), and foreign direct investment (Zhang et al., 2020)], population factor [such as population growth (Dong et al., 2020b), and per capita income (Balado-Naves et al., 2018)], technological factors [such as ICT (Xuan et al., 2020), and technological innovation (Zhang et al., 2016; Lin and Zhu, 2019)], legal factors [such as carbon emissions trading scheme (Zhou et al., 2019)], and other factors [such as the expansion of natural gas infrastructure (Dong et al., 2020a), and force majeure (Zaman et al., 2017; Wang and Wang, 2020; Zhang X. et al., 2021)] will directly or indirectly affect the carbon emission intensity and efficiency of each country. Grossman and Krueger (1991) studied Mexico’s further trade liberalization and found that the reduction of trade barriers would affect the environmental quality by expanding the scale of economic activities, changing the composition of economic activities, and changing production technologies, and also described the inverted U-shaped relationship between economic growth and pollutant emissions (sulfur dioxide and “smog”), which is called the Environmental Kuznets Curve (EKC) hypothesis. This hypothesis has been verified by many scholars (Al-Mulali et al., 2015; (Balado-Naveset et al., 2018; Hao et al., 2018; Leal and Marques, 2020). However, Munir et al. (2020) considered the Cross-sectional Dependence and re-examined the relationship between CO₂ emissions, energy consumption (EC) and economic growth (GDP) in the five countries of the Association of Southeast Asian Nations (ASEAN) between 1980–2016, and found that traditional research has misleading conclusions about the Granger causality between EKC and CO2, EC and GDP. The relationship between CO2, EC and GDP among the five ASEAN countries is heterogeneous. Cheng et al. (2019) used the panel quantile regression method to study the impact of economic growth, renewable energy, and increased patents on carbon emissions. The results did not support the EKC hypothesis, and the impact of increased patents on carbon emissions was not significant. It can be seen that the “black box” of the relationship between economic growth and pollutant emissions needs to be further “opened”. In terms of technical factors, Lahouel et al. (2021) confirmed that the popularization of information and communication technology (ICT) in Tunisia can promote economic growth, thereby increasing the carbon emission efficiency. Zhang X. et al. (2021)compared the effects of SARS and COVID-19 on CO₂emissions, and found that COVID-19 had a much greater impact.

On the regional level, A large number of academic articles in the field of regional carbon emissions have emerged in China in recent years. Chen et al. (2018)found that the main factors contributing to the increase in carbon emissions in Macau are economic scale effect and energy structure change. Based on the research at the city level in China, using the propensity score matching-DID method (PSM-DID), Zhou and Liu (2020)found that the China’s ETS pilot policy has a significant restrictive effect on the carbon emission intensity of pilot cities. As the policy continues to be implemented, the restrictive effect becomes more obvious. Adding intermediary variables, it is found that adjustment of industrial structure and energy conservation in pilot cities are beneficial to reducing carbon emission intensity. Zhang et al. (2020) conducted a study on panel data at the city level in China from 2004 to 2015 and found that pilot areas that adopted carbon emission trading policies reduced carbon emissions by about 16.2%. This impact is particularly significant in the economically developed areas of eastern China. Based on the DID model, Wen et al. (2020)selected 10 cities in Hubei Province and Hunan Province as the treatment group and control group respectively. They found that the implementation of ETS had little significant impact on industrial carbon emission, gross domestic product (GDP), energy consumption intensity and carbon emission intensity in Hubei Province.

On the firm level, there is relatively little research on related aspects. Liu et al. (2021)found that Chinese companies with higher profit pressure will also have higher sulfur dioxide emission intensity, which reflects the conflict between financial goals and environmental goals to a certain extent. Using the panel data of Chinese A-share listed companies from 2008 to 2016, Hu et al. (2020)constructed a quasi-natural experiment based on the DDD model and found that China’s ETS has a significant positive impact on the quantity and quality of innovation. However, the impact of China’s ETS on the innovation quantity, low-quality innovations, and high-quality innovations is reduced in turn. China’s ETS only promotes the innovation quantity and high-quality innovations in State-owned enterprises, large-scale enterprises and eastern enterprises.

Considering factors such as research objects and research conditions, Chinese scholars chose different quantitative research methods when studying effects of China’s ETS pilot policy. There are more representative of the quantitative research methods: Difference-in-Difference (DID) model, Propensity score matching and difference-in-difference (PSM-DID) model, Difference-in-Difference-in- Difference (DDD) model, and so on. DID is a classic method in the field of policy effect evaluation, and is widely used in researches on policy effects (Feng et al., 2017; Dong et al., 2019; Wen et al., 2020). When using DID we should realize that there are significant differences in the characteristics of the treatment and control groups. Non-overlapping bias and the density weighting bias could result if such differences are correlated to the dynamics of the dependent variables (Heckman et al., 1998). Compared with DID, PSM-DID eliminates the systematic differences in individual characteristics between the two groups (Zhang et al., 2019; Zhou et al., 2019; Huang et al., 2021). The parallel trend assumption of DID model is sometimes difficult to meet in actual research. In order to solve this problem, DDD model can be adopted (Gao et al., 2020; Hu et al., 2020). There are also some other methods used, such as the synthetic control approach (Cheng et al., 2019), the two-step panel quantile regression approach (Dong et al., 2020a), and the data envelopment analysis (DEA) model (Dong et al., 2019).

From the previous literature review, it can be seen that scholars have done a lot of research on the policy effects of domestic and foreign carbon emission trading scheme and on the determinants of carbon emission intensity, using different research methods. However, there are still research shortcomings. First of all, on the effects of China’s ETS pilot policy, a large number of studies have focused on the regional level, from the macro perspective of provinces, regions, and cities. However, few scholars research on the firm level. After all, enterprises are cells of the economy, and macro data is also aggregated from micro data of enterprises. The agglomeration effect of corporate carbon emission intensity is reflected in the intensity of regional carbon emissions. The regional carbon emission effects have been studied by scholars from multiple perspectives, while research on the micro level is less. Secondly, the existing firm-level literatures focus on the impacts of China’s ETS pilot policy on corporate green innovation, or research on the impact of corporate carbon mitigation in a certain industry. Few literatures pay attention to the comprehensive reflection of the carbon emission performance of listed companies in all industries under the impact of policies. Thirdly, with the opening of China’s national carbon emission trading market on July 16, 2021, the government’s determination to control carbon emissions has become increasingly firm, and the measures to control emissions have become increasingly stringent. All walks of life should take precautions. So, what mechanism does the China’s ETS policy use to affect the carbon emission performance of listed companies? What measures should be taken by listed companies to improve carbon emission performance? Researches in these areas are very important, but there are very few existing literatures that need to be supplemented.

Our sample is composed of listed companies on the Shanghai Stock Exchange (SHSE) and Shenzhen Stock Exchange (SZSE) in 293 prefecture-level cities in China from the China Stock Market and Accounting Research (CSMAR) system over the period of 2008–2018. Since the China’s ETS pilot policy started from June 2013 to June 2014, the year 2014 will be taken as the time point for external policy shocks. Listed companies registered in the pilot provinces are set as the experimental group, and listed companies registered in the non-pilot provinces are set as the control group. In order to ensure the validity of the data processing results, this article screens the samples as follows before the analysis: 1) Exclude ST or *ST companies; 2) Exclude companies with incomplete data or abnormal financial data. Finally, 105 unbalanced panel data were constructed. The sample distribution by industry is presented in Table 1.

The Difference-in-Difference method obtains the net effect of policy impact by eliminating the interference of time effect and fixed effect on the explained variables (Ashenfelter and Card, 1985), which is a classic powerful tool in the evaluation method of policy effect. Because the policy is exogenous, the DID model can avoid the endogenous problem to a great extent. At the same time, the use of fixed effect estimation can also alleviate the problem of missing variables to a certain extent. This paper regards the impact of China’s ETS pilot policy in 2014 as a quasi-natural experiment, and uses the DID model to compare the carbon emission performance and the corporate innovation level of Listed Companies in the experimental group and the control group before and after the implementation of the policy, so as to present the policy effect more intuitively. Therefore, the benchmark panel DID model is constructed as follows: Y = α 0 + α 1 T r e a t × T + ∑ C o n t r o l + y e a r + i n d u s t r y + p r o v i n c e + ε (1)

Carbon emission performance and corporate innovation are used as the dependent variables, denoted by CO₂ and R&D (Ting et al., 2021), respectively. T r e a t is a regional dummy variable that equals 1 if the province is in the treatment group; otherwise, it is 0. T equals 1 if the emission trading scheme is implemented; otherwise, it is 0. The interaction term, the core explanatory variable, is denoted by T r e a t × T , and it tests whether the China’s ETS pilot policy can bring double dividends of carbon emission performance and corporate innovation. C o n t r o l represents the control variables. y e a r , i n d u s t r y , and p r o v i n c e represent the time fixed effect, the industry fixed effect and the province fixed effect, respectively. ε is the error term.

Carbon emission intensity is a traditional measure of corporate carbon emission performance (Ding et al., 2019). On the macro level, carbon emission intensity is equal to carbon dioxide emissions per unit of GDP (Zhu and Li, 2021). By analogy with this method and considering the availability of sample data, on the firm level, we use the carbon dioxide emissions per unit of assets or income to measure carbon emission intensity. R&D is measured by R&D investment/Total assets based on input of corporate innovation (Ting et al., 2021). To control any factors that may explain the change in carbon emission performance and corporate innovation, we select the following control variables (He and Tian, 2013; Hu et al., 2020; Wen et al., 2020; Lv and Bai, 2021; Ting et al., 2021): Company Size (Size), Leverage level (Lev), Company listing year (Age), Cash flow level (CFO), Proportion of fixed assets (Tangibility), Company growth ability (Growth), Board size (Board), Proportion of independent directors (Indd), and The development of Local economic in province level (GDP). Detailed variable definitions are presented in Table 2.

Table 3 reports the descriptive statistics of the variables. The mean value of carbon emission intensity in our sample is 0.092 kg per yuan, measured by amount CO₂ emissions divided by asset, suggesting that unitary asset will generate 0.092 kg CO₂. However, there is a big difference of carbon emissions among companies. The minimum value of CO2 is 0.001, the maximum value is 1.596. We have the similar finding when using CO₂2, measured by amount CO₂ emissions divided by revenue, the mean (median) value is 0.174 (0.017). The results of other variables are consistent with prior literature.

Based on the benchmark regression analysis, we apply the DID method to investigate the average treatment effect of carbon emission performance of listed companies in the pilot provinces The results for carbon emission performance as the explained variable are reported in Table 4.

The basic regression results in the first column show that the coefficient of T r e a t × T is significantly negative at the 10% level ( α 1 = − 0.286 , p < 0.10 ), indicating that the China’s ETS pilot policy reduce the carbon emission intensity of listed companies in the pilot provinces and improve carbon emission performance, which is consistent with Hypothesis 1 .

Based on the basic regression analysis, a heterogeneity analysis is conducted. According to the marketization index, the samples are divided into high and low groups. The second and third columns of Table 4 show that the carbon emission performance of listed companies in the eastern coastal pilot provinces has improved significantly, but not in the central and western pilot provinces. It can be seen that the higher the degree of marketization, the better the company’s business environment and the more focused on carbon emission performance.

The test results of Eq. 1 for corporate innovation as the explained variable are reported in Table 5.

The results show that the coefficient of T r e a t × T is significantly positive at the 5% level ( α 1 = 0.006 , p < 0.05 ), indicating that the China’s ETS pilot policy can significantly improve innovation capabilities of listed companies, suggesting that innovation is a channel for the impact of the China’s ETS pilot policy on carbon emission performance of listed companies in the pilot provinces, which is consistent with Hypothesis 2 .

Through the basic regression analysis, it is known that the China’s ETS pilot policy can reduce the carbon emission intensity of listed companies in the pilot provinces and improve their carbon emission performance. In order to reduce the error of DID estimation and ensure the reliability of the conclusion, we use different methods to conduct robustness tests. The results are shown in Table 6.

Method 1: Replacing the indicator of carbon emission intensity. Carbon emission intensity is measured by CO₂ emissions per unit of income (unit: kg/yuan). The first column of Table 6 shows the regression results. It can be seen intuitively that the coefficient of T r e a t × T is significantly negative at the 10% level ( α 1 = − 0.516 , p < 0.10 ), indicating that the China’s ETS pilot policy can reduce the carbon emission intensity of listed companies in the pilot provinces and improve carbon emission performance.

Method 2: Eliminating the interference caused by the policy implementation years of 2013 and 2014. The second column of Table 6 shows the regression results. It can be seen intuitively that the coefficient of T r e a t × T is significantly negative at the 5% level ( α 1 = − 0.457 , p < 0.05 ), indicating that the China’s ETS pilot policy can reduce the carbon emission intensity of listed companies in the pilot provinces and improve carbon emission performance.

Method 3: Controlling the variable of area characteristic. The third column of Table 6 shows the regression results. It can be seen intuitively that the coefficient of T r e a t × T is significantly negative at the 10% level ( α 1 = − 0.297 , p < 0.10 ), indicating that the China’s ETS pilot policy can reduce the carbon emission intensity of listed companies in the pilot provinces and improve carbon emission performance.

Method 4: Eliminating Finance industry. The fourth column of Table 6 shows the regression results. It can be seen intuitively that the coefficient of T r e a t × T is significantly negative at the 10% level ( α 1 = − 0.295 , p < 0.10 ), indicating that the China’s ETS pilot policy can reduce the carbon emission intensity of listed companies in the pilot provinces and improve carbon emission performance.

In Table 6, we find that the estimated coefficients of T r e a t × T are still significant, which indicates that the conclusion is robust.