An examination of the impact of carbon emissions trading on corporate green technology innovation from the perspective of supply chain integration

Abstract

Based on the “double carbon target” environment, from the perspective of supply chain integration, the effect of carbon trading policy on enterprise green technology innovation is worthy of in-depth study. This paper selects the data of listed enterprises from 2011 to 2020 as samples, uses multi-time-point triple difference model and intermediary effect model to study the impact of carbon trading policy on green technology innovation of listed enterprises in China, discusses the intermediary role of supply chain integration, and analyzes the heterogeneity of enterprises. The results show that carbon trading policy has a significant promoting effect on enterprises’ green technology innovation. In terms of heterogeneity, the heterogeneity of enterprise nature, enterprise profitability and enterprise debt is discussed. The research shows that carbon trading policy of state-owned enterprises, enterprises with high profitability and enterprises with high debt has a significant promoting effect on green technology innovation. It has practical guiding significance for promoting the environmental governance efficiency of carbon trading policy and the innovation and development of green technology of enterprises.

Introduction

With the advancement of industrialization, global energy consumption and pollution emission issues have become increasingly prominent, posing severe challenges to the ecological environment. Against this backdrop, sustainable development has become a global consensus, and balancing economic, environmental, and social benefits has emerged as a core issue. As a market-incentive environmental regulation tool, carbon trading policy aims to promote cost-effective carbon emission reduction by setting emission allowances and permitting inter-enterprise trading. However, existing studies mostly focus on the direct impact of carbon trading policy on enterprises’ emission reduction behaviors, lacking systematic exploration of the transmission mechanism from this policy to green technology innovation behaviors in the context of supply chain integration. In particular, when facing carbon trading policy, enterprises may exhibit significant differences in their green technology innovation response paths due to heterogeneous characteristics such as ownership nature, profitability, and debt structure—this constitutes the starting point of this study.

Green technology innovation, featuring dual externalities of synergizing resource conservation, pollution reduction, and economic benefits, is regarded as the core path to achieve green transformation. Compared with end-of-pipe treatment and technology introduction, green technology innovation can address resource and environmental constraints at the source and promote industrial green upgrading. Meanwhile, as a core carrier of enterprise operations, the degree of supply chain integration directly affects the coordination and allocation efficiency of enterprises’ internal and external resources. Under carbon constraints, whether enterprises can transmit policy pressure and stimulate green technology innovation motivation through upstream and downstream supply chain collaboration and integration is not only a key link for the effective implementation of the policy but also an under-explored mechanism in existing studies.

Based on this, this paper takes listed enterprises as the research object and explores the impact of China’s carbon trading policy implementation on green technology innovation from the perspective of supply chain integration. Through literature research and empirical analysis, this paper discusses the policy effect (positive incentive or negative inhibition), clarifies the functional mechanism of supply chain integration, and conducts an exploratory analysis on enterprise heterogeneity. The structure of this paper is arranged as follows: Sect. 1 sorts out the research background of carbon trading policy, green technology innovation, and supply chain integration, puts forward research questions, clarifies research content and methods, and illustrates the innovation of the research topic. Section 2 defines core concepts such as the “dual carbon” goal, carbon trading market, green technology innovation, and supply chain integration, and reviews relevant domestic and foreign research literature and research status. Section 3 conducts theoretical analysis on core research questions, proposes research hypotheses around key issues including the relationship between carbon trading policy and green technology innovation, the mediating effect of supply chain integration, and the impact of enterprise heterogeneity, and sets up appropriate research models and variables with systematic explanations. Section 4 carries out empirical analysis and discussion on research questions, supplemented by robustness tests. Section 5 summarizes conclusions based on empirical results and puts forward policy implications.

Literature review

The impact of carbon trading policies and the Porter hypothesis

A substantial strand of research validates the impact of carbon trading policies on emissions and economic outcomes, often resonating with the Porter Hypothesis. Studies by Zhang et al.¹ and Yang et al.² employed CGE modeling to conclude that CET policies positively impact energy conservation and emission reduction. Empirical analyses using Difference-in-Differences (DID) models, such as those by Lu et al.³ and Zhou et al.⁴, have consistently found that China’s pilot CET schemes lead to significant emission reductions at regional and sectoral levels. Furthermore, evidence from Yu et al.⁵ and Hu et al.⁶ suggests these policies can promote both green performance and economic benefits, enhancing carbon emission efficiency. While this body of work robustly establishes a direct causal link, a prevailing limitation, noted by Zhao et al.⁷, is the insufficient exploration of the internal transmission channels. Most studies treat the firm as a “black box,” focusing on the existence of the policy effect rather than the specific operational pathways through which firms translate regulatory pressure into innovative outcomes.

Drivers of green technology innovation and the unexplored supply chain link

Concurrently, research on the drivers of Green Technology Innovation (GTI) has flourished. Following the conceptual groundwork by Ernest Braun et al.⁸, scholars have investigated a range of factors. Studies have examined the roles of government subsidies and executive incentives (Wang et al.⁹), price-based regulations (Wang et al.¹⁰), and technical efficiency (Luo et al.¹¹). The relationship between GTI and socioeconomic development has been explored using novel data sources like night-time light data (Yu et al.¹²). Furthermore, the influences of industrial convergence (He et al.¹³), environmental regulation (Wan et al.¹⁴), foreign direct investment (Yang et al.¹⁵), and high-tech industrial agglomeration (Yang et al.¹⁶) have been highlighted. However, this literature, while comprehensive, frequently overlooks the firm’s embeddedness within its supply chain network. The strategic interactions with key upstream suppliers and downstream customers—relationships that dictate resource flow, cost structures, and market access—constitute a critical yet under-theorized set of determinants for GTI in the context of environmental policy. Joston et al.¹⁷ investigates the impact of China’s Dual-Credit Policy on innovation capability in the new energy vehicle (NEV) industry using a difference-in-differences approach. Our findings reveal that the policy has a positive and significant effect on NEV innovation, partially mediated by R&D investment.

The economic consequences of supply chain integration

The finance and operations management literature provides a robust foundation for understanding the implications of supply chain integration (SCI), often proxied by supplier and customer concentration (Wang et al.¹⁸). Research by Zhao et al.¹⁹ shows that high supplier concentration influences a firm’s cost structure, while Chiu et al.²⁰ highlights that demand volatility from major customers can guide corporate investment decisions. The presence of large customers can reduce information asymmetry and affect the cost of capital, as shown by Chen et al.²¹, and signal positive firm prospects to investors (Peng et al.²²). Conversely, Chu et al.²³ argued that the supply chain integration effect can mitigate information concealment and reduce stock price crash risk. The fundamental insight from this stream is that the structure of a firm’s supply chain has profound implications for its financial performance, risk profile, and operational strategies.

Synthesis and positioning of this study

In summary, three robust streams of literature—on CET policy effects, GTI drivers, and the economic consequences of SCI—have developed in parallel but have seldom intersected. Research on CET has largely bypassed the question of how the policy reconfigures a firm’s core operational relationships. Conversely, the GTI literature has underappreciated the supply chain as a strategic conduit for innovation. This disconnect leaves a pivotal research question unanswered: Through what specific supply chain mechanisms does the carbon trading policy transmit its effects to ultimately influence green technology innovation?

This study aims to bridge this gap by constructing an integrated framework. We introduce supply chain integration—delineated into supplier and customer concentration—as a critical mediating mechanism that transmits the dual pressures of the CET policy. Furthermore, we employ a robust multi-period Difference-in-Differences framework to address the staggered rollout of the pilots, providing a more precise identification of the causal effects. By doing so, we move beyond the question of “if” the policy works to “how” it works through the reconfiguration of the supply chain, thereby contributing to a more granular and actionable understanding of the micro-foundations of environmental regulation.

Theoretical model

Research hypotheses

Carbon trading policy and enterprise green technology innovation

The carbon trading policy imposes a dual pressure on firms. On one hand, to ensure raw materials are environmentally compliant and to reduce monitoring costs, firms are forced to strengthen cooperation with a few core suppliers, leading to increased supplier concentration (Wen et al.²⁴). However, this increased dependency enhances the suppliers’ bargaining power, squeezing profits that could have been used for R&D, thereby inhibiting green innovation (the negative pathway). On the other hand, the policy pressure also drives firms to seek cooperation with large customers who have a higher willingness to pay for green products or have stricter environmental requirements, leading to increased customer concentration. These core customers provide stable market demand and stronger external monitoring, which in turn incentivizes corporate green innovation.

Therefore, based on the Porter effect, this paper argues that carbon trading policy brings more pressure on enterprises in heavily polluting industries in terms of environmental legitimacy, and better stimulates them to carry out green upgrading and reform.

Therefore, the following hypothesis is proposed:

Hypothesis 1 (H1)

The implementation of carbon trading policy can effectively promote enterprises’ green technology innovation.

Intermediation effect of supply chain integration

The implementation of the carbon emissions trading (CET) policy imposes compliance costs and transition pressures that firms must address. Within this context, the supply chain, as the core operational network of a firm, becomes a critical channel for transmitting policy pressure and influencing corporate green innovation decisions. This study focuses on two key dimensions of supply chain integration: supplier integration and customer integration.

It is noteworthy that the CET policy may induce firms to adopt seemingly contradictory yet economically rational integration strategies across their upstream and downstream supply chains, consequently exerting opposing influences on green innovation. On one hand, to meet environmental compliance requirements, ensure the green attributes of raw materials, and reduce monitoring costs, firms may be compelled to strengthen cooperation with a few core suppliers possessing the necessary technical and environmental certifications, leading to an increase in supplier concentration (SI). However, this enhanced reliance on upstream core suppliers strengthens their bargaining power, potentially squeezing the profit margins that firms could otherwise allocate to R&D and innovation, thereby exerting an inhibitory effect on green technology innovation.

On the other hand, facing policy pressure and market demand for green products, firms are motivated to seek stable relationships with major customers who have a green preference, are willing to pay a premium for environmentally friendly products, or impose stringent environmental standards themselves, leading to an increase in customer concentration (CI). Such core customers not only provide stable market demand but their high standards also constitute an effective form of external monitoring and incentive, thereby promoting green technology innovation as firms strive to maintain partnerships and enhance market competitiveness.

The “dual pressure” mechanism proposed in this study refers to the fact that the transmission effects of carbon trading policies through the upstream supply chain (SI) and downstream supply chain (CI) are in opposite directions: ① Upstream path: Policy reduces SI→ weakens the positive support of SI for GTI → negative transmission effect; ② Downstream path: Policy increases CI→ strengthens the positive promotion of CI on GTI → positive transmission effect. The transmission directions of the two paths are opposite, ultimately forming a dynamic balance of “dual pressure” of policy influence on GTI, rather than the direct effects of SI and CI on GTI being in opposite directions.

Based on the theoretical analysis of this “dual-pressure” transmission mechanism, the following mediating effect hypotheses are proposed:

Hypothesis 2a (H2a)

The carbon trading policy reduce the concentration of suppliers and weaken their positive support for green technological innovation, thereby generating negative indirect effects.

Hypothesis 2b (H2b)

The carbon trading policy exerts a positive indirect effect on enterprise green innovation by increasing customer concentration.

Green technology innovation under firm heterogeneity: a moderated mediation perspective

The strength of the aforementioned mediating paths (H2a and H2b) is likely to vary depending on firms’ inherent characteristics. To delve deeper into the boundary conditions of how the CET policy influences green technology innovation through supply chain integration, this study further examines how firm heterogeneity—specifically in terms of ownership, profitability, and debt structure—moderates the above mediating mechanisms. The focus is not merely on testing for which firms the policy’s total effect is more significant, but on analyzing for which firms the process of “how the policy works through the supply chain” is stronger or weaker.

The moderating role of firm ownership

State-Owned Enterprises (SOEs) typically bear more policy burdens and maintain closer ties with the government and other SOEs within the supply chain. This implies that under the pressure of the CET policy, the profit-squeezing effect resulting from increased bargaining power of core suppliers is likely weaker for SOEs, due to their long-term collaborative relationships and political connections. Simultaneously, their motivation and capacity to respond to policies and meet the green demands of large customers (especially government clients) are stronger.

Therefore, this paper proposes:

Hypothesis 3 (H3)

Firm ownership moderates the mediating role of supply chain integration in the relationship between the policy and green innovation. Specifically:

H3a

The negative indirect effect of the policy via increased supplier concentration (H2a) is weaker for State-Owned Enterprises (SOEs).

H3b

The positive indirect effect of the policy via increased customer concentration (H2b) is stronger for State-Owned Enterprises (SOEs).

The moderating role of firm profitability

A firm’s financial performance is a key resource for coping with change and making long-term investments. High profitability provides firms with ample internal funds. This can both buffer against the profit pressure arising from enhanced supplier bargaining power and enable a stronger capacity for prospective R&D investment targeted at the green demands of core customers.

Hypothesis 4 (H4)

Firm profitability moderates the mediating role of supply chain integration in the relationship between the policy and green innovation. Specifically:

H4a

Compared to low-profitability firms, the negative indirect effect of the policy via increased supplier concentration (H2a) is weaker for high-profitability firms.

H4b

Compared to low-profitability firms, the positive indirect effect of the policy via increased customer concentration (H2b) is stronger for high-profitability firms.

The moderating role of firm debt structure

A firm’s debt structure, rather than its total debt level, better reflects its financial strategy and risk. Long-term debt is generally considered a stable source of capital, potentially used to support long-term asset investments. In contrast, short-term debt often implies greater liquidity pressure. Therefore, this study focuses on long-term debt, positing that firms with more long-term debt have a financial structure more inclined to support long-cycle, high-risk investments like green innovation. This enables them to better weather the financial shocks during supply chain integration and seize market opportunities.

Hypothesis 5 (H5)

The level of firm long-term debt moderates the mediating role of supply chain integration in the relationship between the policy and green innovation. Specifically:

H5a

Compared to firms with low long-term debt, the negative indirect effect of the policy via increased supplier concentration (H2a) is weaker for firms with high long-term debt.

H5b

Compared to firms with low long-term debt, the positive indirect effect of the policy via increased customer concentration (H2b) is stronger for firms with high long-term debt.

Model construction

It has been more than 10 years since China’s carbon trading pilot policy was announced in 2011, and relevant policies and data have been published. Therefore, triple difference method has the conditions to evaluate the effectiveness of carbon trading policy.

To identify the net effect of carbon trading policies, this paper constructs a multi-period triple difference model. Although the estimation methods for interlaced DIDs have developed in recent years, in order to facilitate comparison with classic literature and visually present the policy effects, our benchmark regression mainly reports the estimated coefficients of the traditional triple-difference interaction terms. As a robustness test, we simultaneously adopted the method of Callaway & Sant ‘Anna (2021) for estimation, and the results were consistent with the baseline conclusion. This method effectively resolves biases caused by treatment effect heterogeneity and negative weighting issues by using never-treated firms as the counterfactual benchmark, thereby providing a more reliable foundation for identifying the policy’s causal effect.

The treatment variable treat_it is defined as firms located in pilot cities and belonging to pilot industries. We estimate the Average Treatment Effect (ATT) of the policy based on the CS2021 method.

Prior to reporting the ATT, we first test the parallel trends assumption—the “gold standard” in DID design.

To examine the mediating effect of supply chain integration, we follow the steps outlined by Wen et al. (2014) and construct the following system of equations for mediation effect testing within the CSDID framework:

First, a basic carbon trading policy implementation model is adopted, which uses a regression criterion based on triple difference.

1

CSDID_{it} is the treatment effect variable calculated based on the CS2021 method, i represents the industry, t represents the time, and Xit represents a set of control variables.

Y is the dependent variable, that is, the representation of green technology innovation of different enterprises at different times, treat is the virtual variable of fixed effect of city (1 in pilot area and 0 in non-pilot area), period is the virtual variable of fixed effect of year (1 in pilot area and 0 in non-pilot area), and group is the variable of fixed effect of industry (1 in pilot industry, 0 in non-pilot area). And 0 for non-pilot industries). In the above specification, regional city fixed effects control is used for time-invariant differences across regions, year fixed effects control is used for common macroeconomic shocks across time, and industry fixed effects control for cross-industry shocks. Other control variables M of the firm itself are included in the model to account for the factors of change in each firm. Finally, β is the policy effect of the implementation of the carbon trading policy, and is the difference estimate of the impact of the carbon trading policy entries on the relevant variables. When a particular enterprise implements a carbon trading policy in a certain year, the enterprise is considered to be classified as the “treatment” group. Due to the adoption of multi-time DDD, the implementation time of policies in different regions is different, and the absolute period variable cannot be separated. When the period binary virtual variable is set, the factor of treat region will be taken into account. Therefore, the period variable and period* group variable are not separately added into the control variable. Its meaning is repeated with other variables, and the following model is the same.

Secondly, in order to study whether supply chain integration plays a mediating role, we use the mediation effect analysis method (Hasan et al.²⁵). summarized by Wen Zhonglin et al., and set the specific model as follows.

2

3

4

5

In Eq. (2), SI is the first intermediate variable supply chain concentration. Similarly in Eq. (3), CI is the second customer integration. The other variables are consistent with Eq. (1).

Data sources and variable settings

This paper uses the panel data of listed companies from 2011 to 2020. Since the carbon trading policy was announced in 2011, it has attracted wide attention in China, and pilot carbon trading policies have been implemented successively in several regions since 2013. The information shows the development of carbon trading markets in different provinces and cities in the past ten years. In 2013, five provinces and cities, including Shenzhen and Beijing, started pilot operations successively. In 2014, Chongqing and Hubei joined in, and in 2017, Fujian emissions trading started.

The region of the treatment group is different in all dimensions, including scale, industry and other factors are quite different, which makes it an ideal choice for analysis.

Second, the regions of the sample implemented carbon trading policies at different points in time, which rules out the possibility of results driven by shocks in a particular year. From the first pilot in 2013 to 2017, there were three added time nodes in the five years: 2013, 2014 and 2017. Different time nodes provided us with the possibility of adopting multi-time point DDD model, and also improved the robustness to a certain extent.

It should be noted that the establishment time of the pilot in Beijing, Shanghai, Guangdong and Tianjin was from November 26, 2013 to December 27, 2014, and all were treated as 2014. This paper obtains relevant data from CSMAR (Guotai ‘an) database, CNRDS database (China Research Data Service Platform) and corporate financial reports as samples. For the pilot projects launched in the second half of 2013 (usually in the fourth quarter) in Beijing, Shanghai, Guangdong and Tianjin, we define the starting year for the policy to take effect as 2014. For the Shenzhen pilot program launched in June 2013, the starting year is defined as 2013. Chongqing, Hubei (launched in 2014) and Fujian (launched in 2017) are all coded according to the actual year of launch. In addition, in order to ensure the reliability of the results, the samples of financial industry, ST, *ST or PT are excluded to reduce the error, and the tail processing is done. Due to the absence of some data, the samples with missing values are deleted in this paper.

Dependent variables

The measurement of green technology innovation focuses on two methods. One is to use indicators related to green technology innovation, such as the number of green patents: the number of patent applications finally authorized in a year, which can capture the number of innovation output. The second is to measure the efficiency of green innovation to represent green technology innovation, which is measured by shuju envelope analysis and other methods. Based on the accuracy of data in the actual analysis process and to avoid the deviation of regression analysis caused by intermediate calculation results, this paper adopts the first method to measure green technology innovation. In general, in order to improve the core competitiveness, organizations or individuals often apply for patents to maintain their own technological advantages, so as to obtain greater profits. The World Intellectual Property Organization classifies all patents into 35 technical areas, one of which is “environmental technology”, which this article uses as a measure of green innovation. The number of green patent applications is used as the measurement standard for green technological innovation, and the natural logarithm of the number of green patent applications of the enterprise in the current year plus 1, that is, log(1 + GP), is adopted as the main dependent variable. The identification of green patents is based on the China Research Data Services (CNRDS) platform classification. The CNRDS green patent taxonomy is constructed with direct reference to the World Intellectual Property Organization (WIPO) “International Patent Classification (IPC) Green Inventory.” This ensures a consistent and internationally comparable definition of environmental technologies. The classification encompasses patents across all key environmental domains outlined by WIPO, including but not limited to Climate change mitigation technologies (e.g., renewable energy, energy efficiency), Pollution abatement (e.g., waste management, water purification), Sustainable materials and resource conservation. Since patent application and maintenance have a certain cost of time and money, it is usually carried out for commercial purposes, which shows the determination of the applicant to carry out green technology innovation to a certain extent (Shi et al.²⁶). Therefore, the number of patent applications can be regarded as the most direct reflection of innovative activities.

Dummy variables and control variables

Dummy variable is also called binary variable. In this paper, dummy variable DDD is used as the dependent variable to measure whether the carbon trading policy is implemented, which represents whether the enterprise implements the carbon trading policy at a certain time. 1 represents the implementation, that is, the processing group, and 0 represents no implementation, that is, the control group. In the model, the existence of dummy variables can be divided into samples to test the impact of the dependent variable, namely the implementation of carbon trading. Secondly, the classification of 0 and 1 also means that each type of sample integrates other samples of various attribute types, which greatly increases the sample size and improves the accuracy of modeling. In addition, it can also separate the influence of abnormal factors.

Following the existing literature, this paper controls the enterprise feature vectors that may affect the output of regional green innovation. Control variables include asset size LnAssets, net profit margin on total assets ROA, Tobin’s Q value Q, Leverage of asset-liability ratio, operating Growth rate Growth, Cash flow ratio cash, return on equity ROE, total asset turnover ratio ATO, book-to-market ratio BM, years of establishment LnAge, For the detailed description of variables, please refer to Table 1 variables and definitions.-.

Table 1.

Variables and definitions.

Types of variables	Variable name	Symbol	Variable definition
Explained variable	Number of green patent applications	GP	Number of green applications filed that year
	Number of green invention applications filed	GI	Number of green inventions filed that year
	Number of green utility models	GUMI	Number of green utility models applied for that year
	Proportion of green inventions	PGI	Green inventions as a percentage of total patents granted annually in the district
	Proportion of green utility models	PGUMI	Green utility models accounted for the total number of utility models obtained in the region
Explanatory variables	Whether to process effects	DDD	Whether to implement carbon trading policy, the interaction item of individual time point industry
Individual variables	Whether the individual is processed	treat	Yes, take 1; No, take 0
	Ticker symbol	stkcd	Ticker symbol
Time point variable	Whether to process time	period	Yes, take 1; No, take 0
	A given year	year	Year
Industry variables	Whether to deal with industry	group	Yes, take 1; No, take 0
	Industry	Industry	Industry code
Control variables	Asset size	LnAssets	Natural log of total annual assets
	Net profit rate on total assets	ROA	Net profit/average balance of total assets
	Tobin Q value	Q	(Market value of outstanding shares + number of non-tradable shares x net assets per share + book value of liabilities)/total assets
	Asset-liability ratio	Leverage	Total year-end liabilities divided by total year-end assets
	Operating growth rate	Growth	Current Year operating income/Previous year operating income − 1
	Cash flow ratio	Cash	Net cash flow from operating activities divided by total assets
	Return on equity	ROE	Net profit/average balance of shareholders’ equity
	Turnover of total assets	ATO	Operating income/average total assets
	Book-to-market ratio	BM	Book value/total market value
	Number of years established	LnAge	ln(year of the year - year of incorporation + 1)
	Corporate ownership	SOE	The value is 1 for state-controlled enterprises and 0 for others
Mediating variables	Vendor concentration	SI	Top five supplier purchases as a percentage of total annual purchases
	Customer concentration	CI	Ratio of top five customer sales to total annual sales

Intermediary variables

As a mediator variable, supply chain integration is mainly represented from two aspects: supplier integration and customer integration. In terms of measurement, two variables are used: CI customer concentration, i.e., top five customer sales/total sales. And SI supplier concentration, which uses top five supplier purchases/total purchases.

The definitions of all variables are reported in Table 1 Variables and Definitions.

Descriptive statistical analysis

All variables are defined as shown in Table 1.

Before the empirical analysis, descriptive statistical analysis of the samples was conducted first, as shown in Table 2.

Table 2.

Descriptive statistical analysis.

Variable	Mean	Std.Dev	Min	Max
GP	0.972	1.25	0	7.386
GI	0.662	1.042	0	7.231
GUMI	0.663	1.007	0	6.443
PGP	0.103	0.179	0	1
PGI	0.105	0.203	0	1
PGUMI	0.084	0.167	0	1
DDD	0.066	0.249	0	1
DD	0.347	0.476	0	1
Treat	0.078	0.269	0	1
Periods	0.066	0.249	0	1
Group	0.257	0.437	0	1
Treat	0.411	0.492	0	1
Period	0.347	0.476	0	1
LnAssets	22.189	1.278	15.577	28.253
Leverage	0.423	0.208	0.008	1.758
ROA	0.038	0.076	−1.859	0.675
ROE	0.05	0.365	−29.144	1.751
ATO	0.656	0.545	0.003	12.373
Cash	0.042	0.077	−1.938	0.661
Growth	9.205	1093.218	− 0.985	134607.06
BM	1.017	1.208	0.003	18.49
Q	2.184	3.204	0.153	259.146
SOE	0.332	0.471	0	1
LnAge	2.841	0.346	1.099	3.951
CI	30.105	22.171	0	100
SI	34.71	20.351	0	100

Empirical analysis and research for example analysis

The impact of carbon trading policy on enterprises’ green technology innovation

In this section, empirical analysis will be conducted on the effect of carbon trading policies on enterprises’ green technology innovation, including basic regression, substitution regression, heterogeneity analysis, crowding out analysis and robustness test.

Basic regression

The data analysis results of the constructed DID model are shown in Table 3 basic regression. To address potential concerns about pre-existing differences and to validate our identification strategy, we conducted a formal event-study analysis following the methodology proposed by Callaway & Sant’Anna (2021). The results, presented in Fig. 1, demonstrate that the estimated coefficients for all pre-treatment periods are statistically insignificant and cluster around zero. This pattern provides strong empirical support for the parallel trends assumption, confirming that treatment and control groups followed similar trajectories in green technology innovation prior to policy implementation.

Table 3.

Basic regression.

LnAssets		0.474***
		[0.0089]
Leverage		0.056
		[0.0410]
ROA		−0.146
		[0.0981]
	(1)	(2)
	preGP	GP
DDD	0.177 * *	0.233***
	[0.0803]	[0.0636]
DD	0.395***	0.232***
	[0.0296]	[0.0292]
Treat	0.164 * *	0.0274
	[0.0720]	[0.0579]
Treat	0.196***	0.103***
	[0.0277]	[0.0273]
Group	0.103***	−0.027
	[0.0200]	[0.0185]
ROE		0.0312
		[0.0191]
Cash		0.379***
		[0.0908]
Growth		−0.00000411** -
		[0.0000]
BM		0.102***
		[0.0085]
Q		0.0114***
		[0.0016]
LnAge		0.207***
		[0.0206]
SOE		0.133***
		[0.0161]
_cons	0.831***	8.920***
	[0.0112]	[0.1925]

Fig. 1.

Distribution of estimated coefficients and corresponding P-values.

Columns (1)–(2) report baseline DDD estimates using GP as the dependent variable, without and with control variables, respectively.

PreGP and GP are regression results without and with control variables respectively under the condition of GP (the number of green patents applied in the current year) as explanatory variables. The empirical analysis results of the controls with fixed time, individual and industry effects are added with the implementation of carbon trading policy as the dependent variable. Where DDD stands for, DD stands for Using the various controls outlined above, you can see that the DDD coefficient statistics show significant results. The results show that there is a positive correlation between the implementation of carbon trading policy and enterprises’ green technology innovation, indicating that enterprises begin to invest in green technology innovation under carbon trading policy and are driven to carry out low-carbon transformation.

The statistical data in the table comes from all the samples, including all the treated and controlled enterprises, which provides strong support for the confirmation of the hypothesis of the effect of carbon trading policy.

Robustness test

Partial robustness has been demonstrated by the previous partial results. Fixed effects were added to the model to reduce endogeneity and the original result remained unchanged, confirming that the result was not attributable to potential spillovers from other control groups within the same area. In addition, the launch of carbon trading markets occurred at different time nodes in different regions, which to some extent avoids the possibility of shocks in special years.

However, green technology innovation may also be driven by other factors. In this section, substitution-dependent regression and two sets of falsification experiments are conducted to prove that the basic carbon trading policy actually positively affects green technology innovation. In addition, the endogenicity is analyzed, and the above methods are used to test whether the findings of basic regression are reliable for these alternative measures.

Alternative dependent variable regression

In the baseline results, this paper uses GP, the number of green patents filed in the current year, as the primary variable of concern. In this section, two alternative explanatory variables for green technology innovation are constructed, and the fundamental regression findings are checked to see if they are robust to these alternative measures.

The first alternative dependent variable is the number of subdivided patents: the number of green inventions filed in the year GI and the number of green utility models filed in the year GUMI. Among them, invention patents are of higher quality. The second alternative dependent variable is the proportion of green patents. The explanatory variable is PGP, PGI and PGUMI respectively, the proportion of total patent number. Compared with the number, the proportion of green patent can better reflect the importance of enterprises.

According to the alternative dependent variables, the regression results were obtained, as shown in Table 4.

Table 4.

Regression of alternative dependent variables.

	(1)	(2)	(3)	(5)	(4)
	GI	GUMI	PGI	PGUMI	PGP
DDD	0.140 * *	0.239***	0.0224 *	0.0209 *	0.0147 *
	[0.0540]	[0.0532]	[0.0120]	[0.0111]	[0.0114]
DD	0.196***	0.118***	0.0313***	0.00582	0.0237***
	[0.0236]	[0.0232]	[0.0054]	[0.0049]	[0.0050]
treat	0.0298	0.0696	0.0248 * *	−0.0127	−0.0165
	[0.0485]	[0.0474]	[0.0111]	[0.0104]	[0.0108]
treat	0.0696***	0.0659***	0.0104 * *	0.00481	−0.00477
	[0.0218]	[0.0217]	[0.0050]	[0.0047]	[0.0047]
group	0.0584***	0.0543***	0.0270***	0.0224***	0.0280***
	[0.0157]	[0.0148]	[0.0038]	[0.0031]	[0.0034]
LnAssets	0.407***	0.326***	0.0244***	0.0143***	0.0183***
	[0.0083]	[0.0077]	[0.0013]	[0.0011]	[0.0012]
Leverage	0.0790 * *	0.201***	0.0172 * *	0.0344***	0.0256***
	[0.0344]	[0.0328]	[0.0077]	[0.0065]	[0.0071]
ROA	0.172 * *	−0.0915	0.0377 *	−0.0156	0.0344 *
	[0.0777]	[0.0841]	[0.0218]	[0.0164]	[0.0207]
ROE	0.0037	0.0439 * *	0.00889 * *	0.0108 * *	0.0105***
	[0.0111]	[0.0219]	[0.0037]	[0.0042]	[0.0041]
Cash	0.306***	0.283***	0.0514***	0.0534***	0.0571***
	[0.0752]	[0.0720]	[0.0174]	[0.0137]	[0.0161]
Growth	−0.00000234	0.00000274 *	0.000000633***	000000426***	0.000000603***
	[0.0000]	[0.0000]	[0.0000]	[0.0000]	[0.0000]
BM	0.0908***	0.0480***	0.00596***	0.00299***	0.00386***
	[0.0076]	[0.0074]	[0.0012]	[0.0011]	[0.0011]
Q	0.0148***	0.00656***	0.000968 * *	0.00121***	0.00121***
	[0.0026]	[0.0010]	[0.0004]	[0.0004]	[0.0005]
LnAge	0.145***	0.208***	0.00673 *	0.0102***	−0.00379
	[0.0174]	[0.0171]	[0.0037]	[0.0031]	[0.0033]
SOE	0.0784***	0.106***	0.0164***	0.0177***	0.0169***
	[0.0137]	[0.0129]	[0.0029]	[0.0024]	[0.0026]
_cons	7.867***	6.052***	0.417***	0.215***	0.297***
	[0.1779]	[0.1632]	[0.0293]	[0.0241]	[0.0262]

(1) GI: GI (number of green inventions filed that year) regression result as explanatory variable.

(2) GUMI: GUMI (number of green utility models filed that year) as regression result of explanatory variable.

(3) PGI: PGI (the ratio of the number of green inventions to the total number of inventions in that year) as the regression result of the explanatory variable.

(4) PGUMI: PGUMI (the ratio of the number of green utility models to the number of total utility models in that year) as the regression result of the explanatory variable.

(5) PGP: PGP (the ratio of the number of green patents filed in the year to the total number of patents) as the regression result of the explanatory variable.

It can be clearly seen that the interaction term coefficients are significantly positive. Both the quantity and the proportion of green patents have been promoted, which shows that enterprises have a certain willingness to carry out green technology innovation, not only pay attention to quantity, but also pursue quality. Green technology innovation and industrial upgrading run through all aspects. There is a positive correlation between the implementation of carbon trading policy and enterprises’ green technology innovation. Enterprises’ green technology innovation has a certain output, which is consistent with the result of baseline regression, and enhances its robustness.

Placebo analysis

In addition to the control variables mentioned in this article, green technology innovation may also be driven by other factors, but we did not take these factors into account when setting up the model, resulting in the lack of corresponding controls in the model. For this situation, the corresponding placebo test will be performed in this section.

The placebo test is conducted by constructing fictitious treatment groups or fictitious policy implementation times. The evaluation based on this approach requires focusing on the coefficient of the “pseudo-policy dummy variable.” If this coefficient is statistically insignificant, it indicates that no significant “treatment effect” exists under the fictional policy scenario, providing strong support for the robustness of the baseline regression results and demonstrating that the observed effect is unlikely to be driven by other potential factors. Conversely, if the coefficient is significant, it implies that the baseline results likely reflect a spurious correlation rather than a genuine policy effect.

(1) Construct a treatment group.

Randomly fabricate the treatment group, randomly select a certain number of individuals in the overall sample environment, assume that these samples are the pilot carbon trading policy, take them as the treatment group, and other samples are the control group. Repeat this process for 500 or 1000 times, and then generate “pseudo-policy dummy variables” (interaction terms) for regression, to see whether the coefficient of “pseudo-policy dummy variables” is significant. This placebo test mainly monitors the effect of unobservable features.

A random sample of 300 firms was selected to form a new treatment sample for regression analysis, and the above operation was repeated 1000 times. If the coefficient is still significant, it indicates that the reliability of the results of basic regression needs to be improved. On the contrary, if it is not significant, it proves robustness.

Using this method, we conducted a statistical test, and the relevant results were obtained as shown in Fig. 1, the distribution of the estimated coefficients and the corresponding P-values.

As shown in Fig. 1, the distribution of estimated coefficients and corresponding P-values shows that the coefficients are distributed near zero and follow normal distribution, and the test passes. This result shows that in the case of random sampling, it is less likely that the results of basic regression estimation are affected by unobservable factors.

(2) The time distribution is disturbed.

There remains the potential problem that missing variables consistent with the implementation of carbon trading policies may be the real root cause of the change in green technology innovation by firms. If this is the case, then our attribution of changes in firms undertaking green technology innovation to carbon trading policy implementation merely reflects an association, not causation. Our baseline identification strategy takes into account shocks that affect different firms at different times, so omitted variables unrelated to carbon trading policy implementation are unlikely to fluctuate at the same frequency every time (or even most of the time). To further rule out this possibility, a forgery test was performed.

The time span of the data in this paper is ten years, the enterprises are 2000, and the data presents the type of “big N small T”, so the processing method of randomly fictionalized policy entry time is of little significance. Therefore, the policy year is unified in this paper as 2012, that is, the entry time point is 2 or 3 years earlier, and based on this, the basic regression is performed again. To see if the coefficient of policy dummy variables in the newly constituted sample data is still significant. The results are shown in the regression of disrupted time distribution in Table 5.

Table 5.

Scrambled time distribution regression.

	(1)	(2)
	preGP	GP
DDD	0.0149	0.0944
	[0.1286]	[0.1006]
DD	0.372***	0.218***
	[0.0432]	[0.0420]
treat	0.316 *	0.142
	[0.1233]	[0.0967]
treat	0.222***	0.118 * *
	[0.0421]	[0.0408]
group	0.0755***	−0.0375
	[0.0221]	[0.0208]
LnAssets		0.484***
		[0.0105]
Leverage		−0.0237
		[0.0465]
ROA		−0.194
		[0.1136]
ROE		0.0148
		[0.0147]
Cash		0.402***
		[0.1020]
Growth	-	0.00000364
		[0.0000]
BM		0.126***
		[0.0109]
Q		0.0135***
		[0.0024]
LnAge		0.180***
		[0.0229]
SOE		0.157***
		[0.0181]
_cons	0.806***	9.168***
	[0.0128]	[0.2237]

(1) PreGP: the regression result of GP as the dependent variable without control variable after the time distribution is disrupted.

(2) GP: After the time distribution is disrupted, GP is added to the regression result of the control variable as the dependent variable.

The coefficient estimates of DDD are all statistically insignificant, and the results of these falsification tests confirm that the main results of our paper are not driven by the omitted variable problem.

Other problems

The underlying problem in the data is the possible endogeneity of carbon trading policies. It should be noted that the selection of pilot regions was determined by the central government based on administrative and industrial considerations rather than being random. However, the policy was assigned centrally at the macro level, making it plausibly exogenous to individual firm-level outcomes, particularly to specific green innovation activities. Our empirical strategy helps address potential concerns: the staggered adoption design accounts for time-varying shocks, and our inclusion of city, industry, and year fixed effects controls for time-invariant regional characteristics and common temporal trends.

Furthermore, the stability of our main coefficients across different model specifications, while standard errors vary with the inclusion of additional controls, demonstrates the robustness of our core findings rather than serving as direct evidence of exogeneity. This pattern suggests that our results are not driven by omitted variable bias in a systematic way. Therefore, we maintain that the carbon trading policy implementation can be reasonably treated as exogenous to firm-level green innovation decisions in our empirical setting.”

In addition to the falsification test above, in fact, the size of the coefficient of DDD does not change much when we use different sets of control variables. However, the standard error does change when we increase or decrease the number of control variables, which further suggests that carbon trading policies may be exogenous (Roberts et al.²⁷).

Mediating effects of supply chain integration

This section determines whether SI (supplier integration) and CI (customer integration) have intermediary effect tests respectively. The regression results are shown in Table 6.

Table 6.

Mediating effects.

	(1)	(2)	(3)
	GP	GP_SI	GP_CI
DDD	0.2256 * *	0.1881 * *	0.2293 * *
	(0.0915)	(0.0898)	(0.0925)
SI		0.0077***
		(0.0004)
CI			0.0039***
			(0.0004)
DD	0.2184***	0.2209***	0.2056***
	(0.0430)	(0.0429)	(0.0427)
treat	−0.0082	−0.0170	−0.0314
	(0.0853)	(0.0836)	(0.0864)
treat	0.0930 * *	0.0785 *	0.0810 * *
	(0.0408)	(0.0408)	(0.0405)
group	0.0692***	−0.0002	0.0847***
	(0.0247)	(0.0248)	(0.0248)
LnAssets	0.4858***	0.4562***	0.4954***
	(0.0123)	(0.0125)	(0.0123)
Leverage	0.1425***	0.1008 *	0.1427***
	(0.0551)	(0.0549)	(0.0548)
ROA	−0.0684	−0.1551	0.0130
	(0.1137)	(0.1147)	(0.1134)
ROE	0.0000***	0.0000***	0.0000***
	(0.0000)	(0.0000)	(0.0000)
Cash	0.1221***	0.1160***	0.1180***
	(0.0126)	(0.0124)	(0.0125)
Growth	0.0114***	0.0147***	0.0099***
	(0.0018)	(0.0026)	(0.0016)
BM	0.1702***	0.1467***	0.1666***
	(0.0276)	(0.0273)	(0.0275)
Q	0.1218***	0.1128***	0.1342***
	(0.0222)	(0.0220)	(0.0221)
LnAge	9.2754***	8.4337***	9.6107***
	(0.2681)	(0.2742)	(0.2701)

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

(1) GP: Basic regression with GP as the dependent variable.

(2) GP_SI: regression with GP as the dependent variable and SI (supplier concentration) added.

(3) GP_CI: regression with GP as dependent variable and CI (customer concentration) added.

As can be seen from the mediating effect in Table 6, for supplier concentration, we can see, and all of them are significant. Then, it conforms to the mediating effect .For customer concentration and both are significant, and conform to the intermediary effect. Supply chain integration plays a mediating role in this causal effect. The impact coefficient of carbon trading policies on SI is negative (γ=−0.032, p < 0.01), and on CI is positive (ρ = 0.041, p < 0.01). Policies do indeed change the supply chain integration strategies of enterprises. The coefficient of SI to GTI is positive (0.0077, p < 0.001), so the indirect effect =(−0.032)×0.0077≈−0.000246, which is consistent with the direction of the indirect effect (−0.519, after standardization treatment) in the Sobel test report. It conforms to the negative path proposed by H2a that “policies suppress GTI by increasing the concentration of suppliers”.

For the customer concentration (CI) path, the indirect effect is significantly positive, while the Bootstrap confidence interval of its direct effect contains zero ([−5.224, 2.940]). This is consistent with the result of the Sobel test that the direct effect is not significant, jointly suggesting that the impact of carbon trading policies on green innovation may be achieved entirely through the intermediary channel of enhancing customer concentration, that is, there is evidence of a complete mediating effect.

Sobel test and bootstrap test were carried out to test the mediating effect, and both passed. These tests again verify hypothesis H2, indicating that supply chain integration plays a mediating role in the mechanism of carbon trading policy on enterprises’ green technology innovation, mainly through reducing supplier integration and improving customer concentration.

The results of Sobel test and bootstrap test are as follows (Tables 7, 8, 9 and 10).

Table 7.

Sobel test for SI.

	Coef	Std Err	Z	P>|Z|
Sobel	− 0.51936481	0.23357072	−2.224	0.02617624
Goodman-1 (Aroian)	− 0.51936481	0.23398852	−2.22	0.0264448
Goodman-2	− 0.51936481	0.23315217	−2.228	0.02590862
a coefficient	0.225582	0.100531	2.24391	0.024838
b coefficient	−2.30233	0.139028	−16.5602	0
Indirect effect	− 0.519365	0.233571	−2.22359	0.026176
Direct effect	−4.33491	1.72043	−2.51966	0.011747
Total effect	−4.85428	1.73559	−2.7969	0.00516
Proportion of total effect that is mediated: 0.10699116
Ratio of indirect to direct effect: 0.11980975

Ratio of total to direct effect: 1.1198097.

Table 8.

Sobel test for CI.

	Coef	Std Err	Z	P>|Z|
Sobel	0.31822197	0.14606435	2.179	0.02935825
Goodman-1 (Aroian)	0.31822197	0.14689346	2.166	0.0302848
Goodman-2	0.31822197	0.1452305	2.191	0.02844086
a coefficient	0.225582	0.100531	2.24391	0.024838
b coefficient	1.41067	0.155028	9.09946	0
Indirect effect	0.318222	0.146064	2.17864	0.029358
Direct effect	−1.29353	1.91843	−0.674268	0.500141
Total effect	−0.975311	1.92328	−0.507108	0.612079
Proportion of total effect that is mediated: − 0.32627741
Ratio of indirect to direct effect: − 0.24600993
Ratio of total to direct effect: 0.75399007

Table 9.

Bootstrap test for SI.

	Observed Coef.	Bias	Bootstrap Std. Err.	[95% Conf. Interval]
_bs_1	− 0.51936481	0.0030371	0.2260519	−1.013644-0.0836662 (P)
				− 1.004243–0.0766108 (BC)
_bs_2	−4.3349129	0.004098	1.7571895	−7.939486-1.080164 (P)
				−7.823617 −0.8965528 (BC)

Table 10.

CI’s bootstrap test.

	Observed Coef.	Bias	Bootstrap Std. Err.	[95% Conf. Interval]
_bs_1	0.31822197	0.0026019	0.13304437	0.0724188 0.5835074 (P)
				0.0879061 0.6122158 (BC)
_bs_2	−1.2935331	0.0623617	2.1970519	−5.224239 2.939756 (P)
				−5.224239 2.939756 (BC)

Firm heterogeneity analysis: testing for moderated mediation

To rigorously examine whether firm heterogeneity genuinely moderates the “policy-supply chain-innovation” mediation pathway, this study employs the interaction term approach for testing. This method, which introduces interaction terms into a full-sample model, allows for direct testing of the statistical significance of differences between groups, thereby avoiding the statistical fallacy of “comparing significance levels”.

We construct the following moderating effect model, using firm ownership as an example:

6

SOE_i is the firm ownership dummy variable (State-Owned Enterprise = 1, Non-State-Owned Enterprise = 0). The key coefficient to be tested is β₃, which captures the difference in the policy treatment effect between SOEs and non-SOEs. If β₃ is statistically significant, it indicates that firm ownership indeed exerts a significant moderating effect.

Heterogeneity of enterprise ownership

In order to explore the heterogeneity of ownership, the regression results of samples of state-owned enterprise group and non-state-owned enterprise group are shown in Table 11.

Table 11.

Heterogeneity of enterprise ownership.

	(1)	(2)	(3)	(4)	(5)	(6)
	M1	M2	M3	M11	M21	M31
DDD	0.5207***	0.4756***	0.5328***	−0.0217	−0.0547	−0.0242
	(0.1429)	(0.1401)	(0.1442)	(0.1177)	(0.1165)	(0.1185)
SI		0.0074***			0.0081***
		(0.0007)			(0.0005)
CI			0.0061***			0.0019***
			(0.0007)			(0.0005)
DD	0.1527 * *	0.1600 * *	0.1424 *	0.2276***	0.2237***	0.2220***
	(0.0773)	(0.0774)	(0.0754)	(0.0520)	(0.0517)	(0.0519)
treat	0.0790	0.0588	0.0290	−0.0303	−0.0386	−0.0330
	(0.1313)	(0.1284)	(0.1326)	(0.1120)	(0.1108)	(0.1127)
treat	0.2313***	0.2085***	0.2157***	−0.0315	−0.0162	−0.0273
	(0.0731)	(0.0733)	(0.0715)	(0.0495)	(0.0493)	(0.0495)
group	0.1955***	0.0983 * *	0.2547***	0.0031	0.0561 *	0.0015
	(0.0431)	(0.0442)	(0.0438)	(0.0304)	(0.0303)	(0.0304)
LnAssets	0.5716***	0.5529***	0.5743***	0.4464***	0.4077***	0.4539***
	(0.0207)	(0.0208)	(0.0206)	(0.0159)	(0.0161)	(0.0161)
Leverage	−0.1118	−0.1536	−0.1337	0.3606***	0.3288***	0.3529***
	(0.1023)	(0.1020)	(0.1013)	(0.0674)	(0.0667)	(0.0674)
ROA	0.8592 * *	0.9673***	0.7433 * *	0.1952	0.1816	0.2052
	(0.3438)	(0.3411)	(0.3399)	(0.1414)	(0.1415)	(0.1414)
ROE	0.0269 *	0.0240	0.0225 *	0.0568***	0.0467 * *	0.0566***
	(0.0143)	(0.0151)	(0.0117)	(0.0204)	(0.0223)	(0.0202)
ATO	0.1469***	0.1262***	0.1825***	0.0963***	0.1006***	0.0881***
	(0.0360)	(0.0357)	(0.0370)	(0.0172)	(0.0172)	(0.0172)
Cash	0.5136 * *	0.5084 * *	0.4723 *	0.4109***	0.4961***	0.3692 * *
	(0.2508)	(0.2450)	(0.2485)	(0.1445)	(0.1431)	(0.1441)
Growth	0.0000***	0.0000***	0.0000***	0.0070 * *	−0.0051	0.0073 * *
	(0.0000)	(0.0000)	(0.0000)	(0.0035)	(0.0034)	(0.0034)
BM	0.1020***	0.1027***	0.0904***	0.1841***	0.1706***	0.1814***
	(0.0166)	(0.0165)	(0.0165)	(0.0209)	(0.0205)	(0.0208)
Q	0.0564***	0.0646***	0.0480***	0.0053***	0.0081***	0.0048 * *
	(0.0122)	(0.0122)	(0.0123)	(0.0020)	(0.0015)	(0.0020)
LnAge	0.1780***	0.1338 * *	0.1710***	0.1289***	0.1095***	0.1292***
	(0.0634)	(0.0634)	(0.0622)	(0.0298)	(0.0295)	(0.0298)
_cons	11.2805***	10.7503***	11.5486***	8.5125***	7.4472***	8.7348***
	(0.4975)	(0.5017)	(0.4934)	(0.3413)	(0.3486)	(0.3488)

(1) M1: The effect of state-owned enterprise carbon trading policy on green technology innovation.

(2) M2: The mediating role of SI (Supplier concentration) in soes.

(3) M3: The mediating role of CI (customer concentration) in state-owned enterprises.

(4) M11: The effect of carbon trading policies on green technology innovation in non-state-owned enterprises.

(5) M21: The mediating role of SI (Supplier concentration) in non-state-owned enterprises.

(6) M31: Mediating role of CI (customer concentration) in non-state-owned enterprises.

From the heterogeneity of enterprise ownership in Table 11, it can be seen that the policy promotion effect of state-owned holding samples is significant, in which SI and CI play a certain intermediary role. In the sample of non-state-owned enterprises, the promoting effect of carbon trading policies on green technological innovation has not passed the statistical significance test, which indicates that the policy effect of carbon trading will be affected by enterprise ownership. Compared with non-state-owned enterprises, state-owned enterprises have strong financial strength and small financing constraints, which can resist the instability brought by green technology innovation. In addition, soes have a stronger sense of social responsibility and mission, and they are responsive to national policies. To sum up, under the influence of carbon trading policies, state-owned enterprises are more inclined to green technology innovation, and supply chain integration plays a certain intermediary effect, which is consistent with the expected hypothesis.

Heterogeneity of corporate profitability

Take the median of the net asset ratio variable as the critical value, and divide it into two groups (high-profit enterprises and low-profit enterprises respectively) for regression. The regression results are shown in Table 12 as the heterogeneity of corporate profitability.

Table 12.

Heterogeneity of corporate profitability.

	(1)	(2)	(3)	(4)	(5)	(6)
	M1	M2	M3	M11	M21	M31
DDD	0.274 *	0.243 *	0.305 * *	0.185	0.144	0.179
	[0.1415]	[0.1380]	[0.1443]	[0.1179]	[0.1168]	[0.1181]
SI		0.00868***			0.00689***
		[0.0007]			[0.0005]
CI			0.00540***			0.00193***
			[0.0006]			[0.0006]
DD	0.284***	0.282***	0.263***	0.126 * *	0.132 * *	0.122 * *
	[0.0597]	[0.0595]	[0.0589]	[0.0615]	[0.0615]	[0.0614]
treat	−0.0646	−0.0780	−0.115	0.0124	0.00773	0.00546
	[0.1328]	[0.1295]	[0.1359]	[0.1095]	[0.1082]	[0.1098]
treat	0.145 * *	0.126 * *	0.125 * *	−0.0194	−0.00853	−0.0146
	[0.0569]	[0.0568]	[0.0561]	[0.0583]	[0.0583]	[0.0582]
group	0.0570	0.126***	0.0227	0.171***	0.103***	0.175***
	[0.0385]	[0.0385]	[0.0389]	[0.0321]	[0.0322]	[0.0321]
LnAssets	0.504***	0.473***	0.521***	0.471***	0.445***	0.473***
Leverage	[0.0179]	[0.0182]	[0.0180]	[0.0169]	[0.0170]	[0.0169]
	0.204 *	0.149	0.188 *	0.00552	−0.0363	0.0127
	[0.1115]	[0.1104]	[0.1106]	[0.0683]	[0.0682]	[0.0683]
ROA	0.943 *	0.973 *	0.897 *	0.158	0.132	0.190
	[0.5089]	[0.5026]	[0.4991]	[0.1482]	[0.1488]	[0.1485]
ROE	−0.145	−0.0325	−0.173	−0.000495	−0.00683	−0.000251
	[0.2465]	[0.2437]	[0.2373]	[0.0111]	[0.0110]	[0.0108]
Cash	0.452 * *	0.518 * *	0.349 *	0.313 * *	0.370 * *	0.278 *
	[0.2118]	[0.2066]	[0.2085]	[0.1560]	[0.1550]	[0.1567]
Growth	−0.00000315***	−0.00000464***	−0.00000317***	0.00636***	0.00720***	0.00619 ***
	[0.0000]	[0.0000]	[0.0000]	[0.0019]	[0.0017]	[0.0020]
BM	0.188***	0.180***	0.181***	0.0628***	0.0575***	0.0608***
	[0.0196]	[0.0192]	[0.0194]	[0.0171]	[0.0168]	[0.0171]
Q	0.0134***	0.0188***	0.0115***	0.0117***	0.0140***	0.0108***
	[0.0039]	[0.0041]	[0.0039]	[0.0026]	[0.0033]	[0.0024]
LnAge	0.215***	0.200***	0.213***	0.107***	0.0787 * *	0.105***
	[0.0419]	[0.0414]	[0.0416]	[0.0361]	[0.0359]	[0.0361]
SOE	0.129***	0.111***	0.148***	0.114***	0.113***	0.120***
	[0.0347]	[0.0345]	[0.0342]	[0.0284]	[0.0282]	[0.0284]
_cons	9.418***	8.520***	9.959***	9.115***	8.374***	9.232***
	[0.3847]	[0.3935]	[0.3883]	[0.3674]	[0.3719]	[0.3697]

(1) M1: The effect of carbon trading policy on green technology innovation in high-margin enterprises.

(2) M2: The mediating role of SI (Supplier concentration) in high-margin firms.

(3) M3: Mediating role of CI (customer concentration) in high margin firms.

(4) M11: Effect of carbon trading policies on green technology innovation in low-margin firms.

(5) M21: The mediating role of SI (Supplier concentration) in low-margin firms.

(6) M31: Mediating role of CI (customer concentration) in low-margin firms.

As can be seen from the heterogeneity of profitability of enterprises in Table 12, in the sample of high-margin enterprises, the effect of policy on promoting green technology innovation is significant, and the intermediary effect is the same as that of the total sample. For low-profit margin firms, there is no correlation. It shows that high-profit margin enterprises are more inclined to carry out technological innovation to drive industrial upgrading, which is consistent with the expected hypothesis.

Heterogeneity of corporate debt

The median of the asset-liability ratio variable is taken as the critical value, and two groups (high-debt enterprises and low-debt enterprises respectively) are divided into regression. The regression results are shown in Table 13 as the heterogeneity of corporate debt.

Table 13.

Corporate debt heterogeneity.

	(1)	(2)	(3)	(4)	(5)	(6)
	M1	M2	M3	M11	M21	M31
DDD	0.341***	0.279 * *	0.374***	−0.0352	−0.0504	−0.0375
	[0.1262]	[0.1239]	[0.1284]	[0.1312]	[0.1292]	[0.1313]
SI		0.00902***			0.00612***
		[0.0006]			[0.0005]
CI			0.00645***			0.000542
			[0.0006]			[0.0005]
DD	0.312***	0.318***	0.284***	0.114 * *	0.120 * *	0.113 * *
	[0.0703]	[0.0700]	[0.0690]	[0.0517]	[0.0517]	[0.0517]
treat	0.0513	0.0690	−0.0177	−0.0474	−0.0747	−0.0475
	[0.1179]	[0.1157]	[0.1204]	[0.1235]	[0.1213]	[0.1236]
treat	0.244***	0.234***	0.210***	0.0462	0.0548	0.0474
	[0.0669]	[0.0667]	[0.0657]	[0.0490]	[0.0491]	[0.0490]
group	0.143***	0.0775 * *	0.176***	0.0227	0.0882***	0.0215
	[0.0374]	[0.0374]	[0.0376]	[0.0315]	[0.0317]	[0.0315]
LnAssets	0.564***	0.532***	0.577***	0.350***	0.327***	0.351***
	[0.0175]	[0.0177]	[0.0174]	[0.0165]	[0.0167]	[0.0166]
Leverage	0.730***	0.708***	0.735***	0.692***	0.539***	0.700***
	[0.1310]	[0.1304]	[0.1298]	[0.1258]	[0.1258]	[0.1261]
ROA	0.0741	0.0282	0.125	2.524***	2.567***	2.486 * *
	[0.1843]	[0.1854]	[0.1825]	[0.9660]	[0.9584]	[0.9705]
ROE	0.0418 * *	0.0460 * *	0.0420 * *	2.104***	2.094***	2.084***
	[0.0187]	[0.0193]	[0.0185]	[0.7248]	[0.7193]	[0.7271]
Cash	−0.169	−0.215	−0.118	0.743***	0.796***	0.727***
	[0.1811]	[0.1776]	[0.1785]	[0.1640]	[0.1628]	[0.1648]
Growth	0.00000252***	−0.00000426*** -	0.00000253***	0.0168***	0.0141***	0.0170***
	[0.0000]	[0.0000]	[0.0000]	[0.0035]	[0.0032]	[0.0036]
BM	0.129***	0.127***	0.120***	0.0754 *	−0.0656	0.0749 *
	[0.0145]	[0.0143]	[0.0144]	[0.0422]	[0.0419]	[0.0422]
Q	0.0271***	0.0318***	0.0225***	0.00413***	0.00620***	0.00402***
	[0.0052]	[0.0058]	[0.0047]	[0.0013]	[0.0013]	[0.0013]
LnAge	0.248***	0.209***	0.240***	0.0846***	0.0733 * *	0.0843***
	[0.0469]	[0.0466]	[0.0466]	[0.0321]	[0.0318]	[0.0321]
SOE	0.0988***	0.0827***	0.129***	0.125***	0.122***	0.126***
	[0.0321]	[0.0318]	[0.0318]	[0.0282]	[0.0282]	[0.0282]
_cons	10.27***	9.405***	10.78***	6.739***	6.036***	6.789***
	[0.4073]	[0.4121]	[0.4068]	[0.3483]	[0.3563]	[0.3538]

(1) M1: The effect of carbon trading policy for highly indebted enterprises on green technology innovation.

(2) M2: The mediating role of SI (Supplier concentration) in highly indebted enterprises.

(3) M3: The mediating role of CI (customer concentration) in highly indebted enterprises.

(4) M11: The effect of carbon trading policies on green technology innovation in low-debt enterprises.

(5) M21: The mediating role of SI (Supplier concentration) in low debt firms.

(6) M31: Mediating role of CI (customer concentration) in low debt enterprises.

As can be seen from the heterogeneity of corporate debt in Table 13, the sample policy of the high debt group has a significant role in promoting green technology innovation, and supply chain integration plays a mediating role. After the implementation of the policy, the green technology innovation level of enterprises with high debt has improved, while that of enterprises with low debt has not changed. Enterprises with high debt ensure that enterprises have long-term capital supply and effectively guarantee the progress of green technology innovation investment, which is consistent with the expected hypothesis H5a.

Discussion and conclusion

Key findings

This study investigates the impact of China’s carbon emissions trading (CET) policy on enterprise green technology innovation (GTI) through the lens of supply chain integration. Employing a robust multi-period DID estimator and mediating effect analysis on listed firm data from 2011 to 2020, we derive three key findings: (1) The CET policy exerts a net positive effect on GTI. (2) The policy weakens the positive support for green innovation by reducing the concentration of suppliers, thus creating a negative indirect effect. (3) These effects are strongly moderated by firm heterogeneity, being most pronounced in state-owned enterprises (reflecting their policy role and financing advantages), high-profitability firms (with greater internal financial resilience), and firms with higher long-term debt (providing stable capital for innovation).

Theoretical and practical implications

Our findings refine the Porter Hypothesis by revealing how environmental regulation’s effects are filtered through supply chain dynamics. The discovered “dual-pressure” mechanism shows that the same policy creates both inhibiting forces (upstream) and promoting forces (downstream), with their net effect depending on firm-specific characteristics. In China’s institutional context, SOEs’ unique position - bearing policy burdens while enjoying financing advantages - makes them particularly effective at navigating these dual pressures. Meanwhile, internal profitability and long-term debt provide the crucial financial resilience other firms need to transform regulatory pressure into innovation. Furthermore, supplementary analyses suggest that the observed increase in green innovation represents a net gain rather than a mere reallocation from non-green domains, addressing potential concerns about crowding-out effects.

Policy recommendations

Policymakers should: Develop complementary measures to mitigate upstream supply chain frictions, particularly for non-SOEs and less profitable firms. Leverage demand-side mechanisms like green procurement to strengthen the positive customer concentration effect. Design targeted financing programs to help financially constrained firms undertake green innovation.

In conclusion, this study demonstrates that the CET policy’s effectiveness depends critically on how firms reconfigure their supply chains and their capacity to manage the resulting dual pressures. Future research could benefit from qualitative measures of supply chain integration to complement the concentration ratios used in this study. Future research could also further explore potential crowding-out effects across different technology domains to fully understand the policy’s comprehensive impact on corporate innovation portfolios^28–31.

Author contributions

C. W. D : Writing – review & editing. Y. G.Y : Writing –Original draft, review & editing. Z. B. B : Writing – Methodology, Data curation, Analyzed the empirical data. Z. J. C : Writing – review & editing.

Funding

National Social Science Foundation of China (grant no. 2014B1-0130), National Natural Science Foundation of China (grant no.71373173), National Social Science Foundation of China (grant no. 19CGL006).

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.