The Green Ripple Effect: How Digital Transformation Reduces Carbon Emissions Across Industrial Chains
Abstract
This study examines the impact of digital transformation in focal enterprises on the carbon emission intensity of upstream and downstream firms within industrial chains, particularly in the context of global green and low-carbon development. The findings reveal that digital transformation significantly reduces carbon emission intensity by 9.97% in upstream enterprises and 11.9% in downstream enterprises, highlighting the substantial spillover effects across the industrial chain. These reductions are driven by three mechanisms: innovation integration, information spillover, and resource allocation. The study also finds that these spillover effects are more pronounced in regions with lower economic growth targets and stricter environmental regulations, particularly in central-eastern China. Additionally, the research identifies significant industry heterogeneity, with varying spillover effects across different industrial sectors. This research offers valuable policy insights for leveraging digital transformation to promote green and low-carbon industrial transformations, especially in developing countries.
1. Introduction
The global commitment to tackling climate change and fostering low-carbon development has gained significant momentum in recent years. Leading economies, including the United States, Europe, Japan, and South Korea, have pledged to achieve carbon neutrality by 2050. Similarly, China has set strategic goals to peak carbon emissions by 2030 and achieve carbon neutrality by 2060. According to the 2024 Global Carbon Neutrality Annual Progress Report, 151 countries have set carbon neutrality targets as of May 2024, with 120 of them establishing legal standing for these targets and 86 outlining detailed carbon neutrality roadmaps. This global push toward a green and low-carbon development strategy underscores the importance of identifying effective methods for reducing carbon emissions, a challenge that is relevant for both developed and developing nations.
As the digital economy accelerates the global shift toward digitalization, networking, and intelligence, the digital transformation of enterprises has emerged as a crucial process in this transition. Digital transformation facilitates cost reductions, enhances operational efficiency, optimizes resource allocation, and has the potential to significantly impact carbon emissions across various stages of production [1, 2]. Existing studies have documented the positive environmental outcomes associated with digitalization, highlighting its influence on reducing carbon emissions at both the macro- and microlevels [3–5]. At the macrolevel, regional digitalization has been shown to reduce carbon emission intensity (CEI), with the flow of innovative elements being a key factor in promoting low-carbon development [6, 7]. At the microlevel, the digital transformation of enterprises plays a pivotal role in achieving low-carbon technological innovations, which are crucial for meeting the requirements of low-carbon development [8–10].
The industrial chain concept, first introduced by Porter in competitive strategy, has been extensively discussed in the context of enterprise transformation. The industrial chain is defined as a network of enterprises, organizations, and individuals through which raw materials, components, products, and services flow [11]. Huang and Ni argue that the industrial chain encompasses processes such as production, service provision, and value realization, while Wang, Chen, and Li propose that it consists of dimensions including enterprise chain, technology chain, value chain, product chain, and spatial chain [12, 13]. The digital transformation of focal enterprises, which creates a linkage effect across the industrial chain, has become an increasingly important research area. Studies suggest that the behaviors of enterprises within the industrial chain influence one another, particularly when considering digital transformation initiatives. For example, digital technologies introduced by customer enterprises often have technological spillover effects on upstream suppliers, fostering vertical linkages that reduce transaction costs and enhance operational efficiencies [14, 15].
The relationship between digitization and the industrial chain has also attracted attention from scholars. Digital transformation not only is a strategic decision of individual enterprises but also influences their upstream and downstream partners. Fan, Wu, and He argue that the linkage effect within the industrial chain is crucial to understanding enterprise digital transformation, as digital collaboration enhances the transformation process of both focal and surrounding enterprises [14]. Similarly, Zhang, Gao, and Han highlight how digital transformation facilitates industrial chain integration by reducing internal production costs and transaction costs, thus strengthening the linkage among enterprises [9]. Moreover, digital transformation creates spillover effects that foster greener supply chains, as evidenced by research indicating positive information spillovers from customer enterprises to suppliers, reducing information search and verification costs [16].
In the context of low-carbon development, digital transformation offers significant potential for reducing carbon emissions throughout the industrial chain. Scholars like Peng et al. have explored how supply chain coordination, through carbon emission quotas and low-carbon practices, can reduce carbon footprints within the industrial chain [17, 18]. These studies suggest that low-carbon supply chain management involves not only direct carbon reduction strategies but also collaboration and information exchange among firms to maximize the environmental benefits [19].
The existing literature clearly demonstrates that digital transformation plays a vital role in facilitating low-carbon development. However, despite the growing body of research, the mechanisms through which digital transformation impacts carbon emissions across the industrial chain remain underexplored. This paper aims to fill this gap by analyzing the impact and mechanisms through which digital transformation influences the carbon emissions of upstream and downstream enterprises. Specifically, it investigates three potential mechanisms: innovation integration, information spillover, and resource allocation. Through this approach, the paper provides insights into how digital transformation can be leveraged to reduce carbon emissions in industrial chains, contributing to the broader goal of sustainable and low-carbon economic development.
This paper makes three key contributions. First, it expands the research on digital transformation by focusing on its externalities, particularly the spillover effects of carbon emissions within the industrial chain, including upstream and downstream enterprises. While previous studies have explored digitalization and low-carbon development, the mechanisms through which digital transformation affects carbon emissions across industries and regions remain underexplored. Second, the paper introduces a unique micropanel dataset to examine how digital transformation reduces CEI in upstream and downstream enterprises through mechanisms such as innovation integration, information spillover, and resource allocation. The results show significant reductions in carbon emissions, with spillover effects varying by region, industry, and economic growth targets. For example, carbon emission reductions are more pronounced in regions with stricter environmental policies and lower economic growth targets. The study also reveals substantial industry heterogeneity in the impact of digital transformation on carbon emissions, offering detailed insights for policy formulation. Lastly, from a policy perspective, the findings provide empirical support for government efforts to promote the green and low-carbon transformation of industrial chains, with targeted policies for specific industries and regions contributing to the sustainable development of the circular economy.
The remainder of the paper is structured as follows: Section 2 establishes the theoretical framework and proposes hypotheses to be tested; Section 3 presents the empirical design of the study; Section 4 includes the baseline regression and robustness analysis; Section 5 examines the three potential mechanisms; Section 6 conducts heterogeneity analysis; and finally, Section 7 concludes with policy recommendations.
2. Theoretical Analysis and Hypothesis
2.1. Basic Hypothesis
The digital transformation of focal enterprises is closely related to the industrial chain. Digital transformation is conducive to information fluency, business process optimization, and industrial chain collaboration [14, 20], which plays a crucial role in reducing carbon emissions of upstream and downstream enterprises in the industrial chain. First, digital transformation provides enterprises with advanced data management tools to monitor energy consumption in real time, thereby effectively identifying energy waste and inefficiencies. Through the optimization of energy management, the energy utilization efficiency is improved and the carbon emissions of the industrial chain are reduced. Second, the digitization process is accompanied by the automation of enterprise production. Automated systems precisely control the production process, and smart sensors help enterprises optimize the energy consumption pattern [21], which is conducive to reducing the carbon emissions of enterprises. Third, digital transformation facilitates remote collaboration and virtualized work arrangements. Through online meetings, collaboration platforms, and cloud storage, firms effectively reduce employee commuting and business travel, thereby reducing carbon emissions associated with transportation. Virtualized work arrangements also contribute to lower energy consumption in office facilities, further reducing carbon emissions throughout the industrial chain. Based on the aforementioned analysis, the paper proposes the following theoretical proposition.
Proposition 1. The digital transformation of focal enterprises reduces the carbon emissions of upstream and downstream enterprises in the industrial chain.
2.2. Mechanism Hypothesis
The digital transformation of focal enterprises facilitates the reduction of carbon emissions within the industrial chain. So, by what means does digital transformation contribute to carbon emission reduction? Subsequently, we delve into a further analysis of the mechanisms through which digital transformation influences the carbon emissions of upstream and downstream enterprises within the industrial chain.
2.2.1. The Innovation Integration Effect (IIE)
The digital transformation of focal enterprises is able to promote the innovative integrated development of the industrial chain, thereby reducing carbon emissions of upstream and downstream enterprises. First, the digital transformation of focal enterprises plays an important role in optimizing and innovating the structure of the industrial chain. Digital technologies and platforms provide convenience for enterprises to carry out cross-organizational collaboration, which improves the efficiency and flexibility of the industrial chain, thus boosting the replanning and layout of the industrial chain [22]. The optimization and innovation of the industrial chain structure promotes the decomposition and specialization of the production process. Therefore, enterprises can outsource highly polluting production stages to other specialized emission reduction entities within the industrial chain, thereby enabling precise control and management of carbon emissions at each stage.
Second, the digital transformation of focal enterprises contributes to green technology innovation within the industrial chain. The accessibility of information serves as a determining factor for enterprise innovation [23]. The digital transformation of enterprises implies a broader application of digital technologies, which offer cost advantages in cross-temporal information transmission, data acquisition, and processing [3]. Consequently, it positively promotes green technology innovation [24, 25]. In addition, the innovation integration resulting from digital transformation within the industrial chain integrates investment and financing resources across various chain links [26, 27], providing financial support for basic research in green technologies and expediting their research and application. These not only reduce the carbon emissions of individual enterprises but also reduce the carbon emissions throughout the industrial chain.
Proposition 2. The digital transformation of focal enterprises reduces carbon emissions of upstream and downstream enterprises within the industrial chain through the effect of innovation integration.
2.2.2. The Information Spillover Effect (ISE)
As industrial collaboration deepens, the relationships among enterprises in the supply chain are transforming from traditional customer–supplier partnerships to integrated strategic alliances [16]. The digital transformation of focus enterprises provides the possibility for upstream and downstream enterprises to realize information spillover with the help of digital technology, which promotes the sharing and transmission of information and reduces the carbon emissions of the industrial chain. First, tools such as online meetings and shared documents facilitate enterprises to communicate and collaborate in real time, improving the efficiency of information exchange and organizational coordination [28]. Strengthening cooperation and understanding between upstream and downstream enterprises enhances production efficiency [4] and responsiveness, avoiding unnecessary energy consumption and lowering carbon emissions in operational activities. Second, upstream and downstream enterprises can also use digital technology to mine and analyze data, so that enterprises can obtain more detailed information on production and business activities [20]. This helps to discover hidden correlations and trends in the industry chain and identify potential optimization opportunities, thereby reducing carbon emissions of enterprises in the industrial chain. Third, data-sharing platforms serve as an effective means for enterprises to access information about market trends, consumer preferences, raw materials, and product supply–demand dynamics Shaharudin et al., [18, 29]. These platforms effectively reduce information search costs for enterprises [3], providing room for comparison and selection among enterprises, thereby reducing the matching difficulties between trading partners [30, 31]. This enables enterprises in the industrial chain to make more accurate decisions and plans based on timely market trends and demand matching information, streamlining unnecessary processes and steps, and ultimately reducing carbon emissions. Based on the above, the paper proposes the third theoretical proposition.
Proposition 3. The digital transformation of focal enterprises reduces carbon emissions of upstream and downstream enterprises within the industrial chain through the effect of information spillover.
2.2.3. The Resource Allocation Effect (RAE)
Optimal resource allocation plays a crucial role in the digital transformation and carbon emission reduction of enterprises. On the one hand, the digital transformation of focal enterprises facilitates the promotion of optimized resource allocation within the industrial chain. The improper allocation or abuse of existing resources is an important factor restricting the economic development of a region [32]. Specific barriers to rational resource allocation in production include adjustment costs, incomplete information, financial frictions, and distortions arising from economic policies or other institutional characteristics [33]. However, enterprise digital transformation helps remove the barriers that distort the allocation of production factors. For instance, the study by Wu et al. demonstrates that the rise of the digital economy and the interrelation between new technologies and the real economy have a positive impact on the way resources are allocated [34].
On the other hand, optimizing the allocation of resources within the industrial chain contributes to carbon emission reduction. Studies indicate that the extent of resource misallocation in impoverished countries is significant enough to explain a substantial portion of the total factor productivity (TFP) gap between rich and poor nations. Improper allocation of input factors such as labor and capital among enterprises has adverse effects on overall productivity and output [35, 36]. In the context of negative impact on fundamental economic development, it becomes even more challenging to balance the requirements for high-quality development, including green and low-carbon initiatives. However, the optimization of industrial chain resource allocation driven by digitization can effectively improve this situation and provide an opportunity to achieve structural optimization and sustainable development goals. Wang et al. provide evidence that improving industrial resource allocation can reduce carbon emissions [37]. Li and Wang found that digital economy has an inverted U-shaped relationship with carbon emissions, and there is room for digital economy to reduce carbon emissions [38]. Chen confirms from the perspective of microenterprise that digital economy is conducive to reducing resource mismatch and thus reducing the CEI of enterprises [5]. Based on the aforementioned analysis, the fourth theoretical proposition is proposed.
Proposition 4. The digital transformation of focal enterprises reduces carbon emissions of upstream and downstream enterprises within the industrial chain through the effect of resource allocation.
3. Empirical Design
3.1. Data Source
3.1.1. The CSMAR Database
The CSMAR database not only contains comprehensive financial information of listed companies but also furnishes this paper with the list of the top five suppliers and customers for A-share listed companies. This list includes information such as the names of domestic suppliers and customers, purchase or sales amounts, their respective proportions, affiliations, and stock codes. Based on statistics, there are 131,485 valid observations for upstream enterprises and an average of 161,391 observations for downstream enterprises. With these data, the “upstream enterprises–focal enterprise–downstream enterprises” relationship is constructed, providing empirical support for the research on industrial chain transmission in this paper.
3.1.2. The Chinese National Taxation Survey Database (CNTSD)
The database is jointly implemented by the Ministry of Finance and the State Administration of Taxation, and local tax authorities are responsible for data reporting, collection, and verification. It boasts a comprehensive and diverse sampling framework, ensuring the authority, scientific rigor, and rationality of the data [39]. This database covers a sample of over 600,000 enterprises annually, providing rich information on enterprise characteristics and financials, including the required indicator of the corporate income tax rate, with minimal missing values. Given the availability of the data, this study ultimately obtained nationwide panel data from the CNTSD between 2007 and 2016. The carbon emission data for upstream and downstream enterprises are derived from the CNTSD. As only a limited number of upstream and downstream enterprises are available in the CSMAR database, and their reported carbon emission data are essentially nonexistent, the CNTSD is utilized as a supplementary source to maximize data availability. For some companies where subsidiary information is not provided, manual searches are conducted using databases such as the Chinese Corporate Registration Database, Tianyancha official website, and company websites. The obtained information is then matched and completed with the CNTSD. In cases where industry information is not identified for certain companies, a comparison is made based on their primary products to determine the industry classification. Any unidentified data are subsequently excluded. Additionally, the control variables for upstream and downstream enterprises in this study primarily originate from the CNTSD.
In addition, this paper also involves other data sources, such as the National Bureau of Statistics and provincial statistical yearbooks over the years, which mainly provide macrolevel control variable information of upstream and downstream enterprises. The patent data used in the mechanism analysis is mainly from the State Intellectual Property Office.
After merging all the databases, the unified sample period is set as 2007–2016, as it represents the most recent data period in the CNTSD. Following the practices of Meng et al., the paper applies the following principles to screen the sample: (1) exclude listed companies in the financial sector; (2) exclude samples classified as ST, PT, or those with insolvency issues; and (3) exclude samples with missing values for relevant variables [40–42]. To mitigate the influence of outliers, a winsorization technique is employed, truncating all continuous variables at the 1st and 99th percentiles. After implementing the aforementioned procedures, this study ultimately obtains a distinctly unbalanced panel dataset comprising 14,235 observations.
3.2. The Empirical Model
Within the model, the variable notation is as follows: i represents the focal enterprise, up represents upstream enterprises, down represents downstream enterprises, and t represents the year. The terms CEIup,t and CEIdown,t, respectively, signify the CEI of upstream and downstream enterprises. Digitali,t represents the digital transformation of the focal enterprise. and refer to a series of control variables for upstream and downstream enterprises, including the corresponding micro- and macrolevel control variables. σup and σdown represent the fixed effects of upstream and downstream enterprises, respectively, which control for unobservable heterogeneity in these entities. τt represents the time fixed effects, accounting for unobserved shocks across different years. εup,t and εdown,t are the error terms for upstream and downstream enterprises, respectively. The core coefficients of interest in this study are denoted as βup and βdown, capturing the impact of the focal enterprise’s digital transformation on the CEI of upstream and downstream enterprises. The expected coefficients are anticipated to be significantly negative.
3.3. The Definition of Variables
The dependent variable in question pertains to the CEI of upstream and downstream enterprises. Following the approach of He et al, it is defined as the “ratio of carbon emissions to enterprise value added,” effectively mitigating the influence of varying enterprise scales [43, 44]. Due to the limited availability of carbon emission disclosure data in China, as highlighted by Pan and Wang, we reference the research of Yu et al. to calculate the carbon emissions of enterprises. This involves multiplying the “total industry carbon emissions” by the “ratio of enterprise operating costs to industry operating costs” [45, 46]. The total industry carbon emissions are obtained from the Carbon Emission Accounts and Datasets (CEADS), which are computed based on the consumption of each energy source and their corresponding carbon emission coefficients, following the IPCC National Greenhouse Gas Inventory Guidelines from 2006. Industry operating costs are derived from aggregating enterprise-level operating costs. Now, the key challenge lies in determining the carbon emissions of upstream and downstream enterprises. The data structure of this study is characterized by “enterprise-self–year–upstream enterprises–downstream enterprises.” This implies that a specific enterprise in a given year may correspond to multiple upstream and downstream enterprises. To address this issue of dimensionality inconsistency arising from the shared digital transformation among multiple upstream and downstream enterprises, we adopt the approach proposed by Cai, Tang, and Han. By weighting the CEI of multiple upstream and downstream enterprises based on their respective proportions of primary business revenue, we establish a one-to-one correspondence between the dependent variable and the core explanatory variable [47]. The primary business revenue data are sourced from the CNTSD and CSMAR. Lastly, for the sake of result interpretation, we take the logarithm of CEI.
The core explanatory variable pertains to the focal enterprise’s digital transformation. Digital transformation in enterprises is a comprehensive process. Previous studies have primarily constructed regional or industry-level digital economic indicators as proxies for measuring the degree of enterprise digitalization from a macro perspective. However, due to the pervasive nature of digital applications across various fields and industries, accurately capturing its extent at the macrolevel presents challenges. Therefore, this study attempts to measure the degree of digitalization at the microlevel of enterprises. Drawing inspiration from the research of Zhao et al,this study begins by constructing a dictionary of digitalization terms for enterprises [41, 42, 48]. Using Python, we process and tokenize relevant policy documents related to the digital economy at the national level, sourced from the official websites of the Central People’s Government and the Ministry of Industry and Information Technology in China. Through this process, we extract 99 high-frequency vocabulary items as the foundation for the enterprise digitalization term dictionary. Subsequently, we expand these 99 terms from the digitalization dictionary into the “jieba” Chinese word segmentation library within the Python software package. Employing machine learning techniques, we conduct text analysis on the “Management’s Discussion and Analysis” (MD&A) section of annual reports of A-share listed companies. By tallying the frequency of appearance for each of the 99 keywords within these reports, we obtain statistical data. Next, we standardize the frequency data and employ the entropy weight method to determine the weights of each indicator. Through weighted calculations, we derive a comprehensive index representing the degree of enterprise digitalization. Lastly, for the purpose of result interpretation, we apply a logarithmic transformation to the variable Digital.
Control variables encompass both microlevel and macrolevel variables. The microlevel variables consist of enterprise-level variables pertaining to upstream and downstream enterprises. These variables are as follows: (1) Return on Assets (Lnroa), calculated by dividing the net profit by total assets and taking the logarithm; (2) Capital Structure (Lev), calculated by dividing total liabilities by total assets and taking the logarithm; (3) Capital Output Ratio (Lnppe), obtained by dividing net fixed assets by total revenue and taking the logarithm; (4) Capital Intensity (Lncap), derived from the ratio of total assets to the total revenue and taking the logarithm; (5) Cash Flow from Operating Activities (Lncfo), computed by dividing the net cash flow from operating activities by total assets and taking the logarithm; and (6) Enterprise Age (Lnage), determined by subtracting the year of establishment from the current year, adding one, and taking the logarithm. Macrolevel control variables include (1) Economic Development Level (Agdp), measured by the logarithm of per capita GDP at the provincial level; (2) Industrial Structure (Struc), assessed by the proportion of the secondary sector within the province; (3) Population Growth (Popg), evaluated by the rate of population growth within the province; (4) Foreign Direct Investment (Fdi), gauged by the proportion of foreign investment to GDP; and (5) Government Fiscal Pressure (Press), quantified by the ratio of local government expenditure to revenue. It is worth noting that when multiple upstream or downstream enterprises are located in different provinces, this study considers the province where the largest enterprise operates. Furthermore, in subsequent analysis, we will eliminate these observations for robustness testing.
3.4. Descriptive Statistics
The descriptive statistics for the main variables are presented in Table 1. The mean value of CEI in upstream enterprises is 2.055, slightly higher than that of downstream enterprises at 1.786. The mean value of Digital is 1.134, which remains the same for both upstream and downstream enterprises since it is the focal enterprises’ variable. There is little difference in the other control variables between upstream and downstream enterprises, and further elaboration is unnecessary.
| Variables | Observations | Upstream enterprises | Downstream enterprises | ||
|---|---|---|---|---|---|
| Mean | Std. dev. | Mean | Std. dev. | ||
| CEI | 14,235 | 2.055 | 1.998 | 1.786 | 1.765 |
| Digital | 14,235 | 1.134 | 1.378 | 1.134 | 1.378 |
| Lnroa | 14,235 | 0.043 | 0.049 | 0.049 | 0.055 |
| Lev | 14,235 | 0.342 | 0.161 | 0.278 | 0.131 |
| Lnppe | 14,235 | 0.913 | 1.367 | 0.762 | 1.14 |
| lncap | 14,235 | 1.39 | 1.19 | 1.772 | 1.516 |
| Lncfo | 14,235 | 0.064 | 0.05 | 0.073 | 0.056 |
| Lnage | 14,235 | 2.302 | 0.391 | 2.621 | 0.445 |
| Agdp | 14,235 | 8.293 | 3.31 | 11.19 | 4.466 |
| Struc | 14,235 | 0.886 | 0.303 | 1.291 | 0.442 |
| Popg | 14,235 | 0.038 | 0.022 | 0.048 | 0.027 |
| Fdi | 14,235 | 0.031 | 0.018 | 0.039 | 0.023 |
| Press | 14,235 | 1.848 | 0.606 | 2.532 | 0.831 |
- Note: Data are calculated by the authors.
4. Regression Analysis
In this section, we will test Proposition 1 proposed earlier based on (1) and (2). The testing process consists of two parts: the baseline regression and robustness tests.
4.1. Baseline Regression
The baseline regression results are presented in Table 2: In column (1), it is shown that the coefficient of Digital on CEI is −0.0997, which is statistically significant at the 1% level. This indicates that for every 1% increase in the digital transformation of the focal enterprise, the carbon intensity of upstream enterprises decreases by 0.0997%. In column (2), the coefficient of Digital on CEI is −0.119, also statistically significant at the 1% level. This suggests that for every 1% increase in the digital transformation of the focal enterprise, the carbon intensity of downstream enterprises decreases by 0.119%. For comparison purposes, in column (3), we report the impact of the focal enterprise on its own carbon emissions. The results reveal that the digital transformation of the focal enterprise leads to a reduction of 0.155% in its own carbon intensity.
| Variables | Carbon emission intensity (CEI) | ||
|---|---|---|---|
| Upstream enterprises | Downstream enterprises | Focal enterprises | |
| (1) | (2) | (3) | |
| Digital | −0.0997 ∗∗∗ | −0.119 ∗∗∗ | −0.155 ∗∗∗ |
| (0.0262) | (0.0233) | (0.0282) | |
| Lnroa | 0.144 | 1.168 ∗∗∗ | −0.384 |
| (0.437) | (0.333) | (0.392) | |
| Lev | 0.601 ∗∗ | 0.575 ∗∗ | 0.793 ∗∗∗ |
| (0.253) | (0.284) | (0.279) | |
| Lnppe | 0.0505 | 0.0363 | −0.0350 |
| (0.0555) | (0.0612) | (0.0729) | |
| Lncap | −0.0614 | −0.0240 | 0.0104 |
| (0.0651) | (0.0464) | (0.0587) | |
| Lncfo | −1.055 ∗∗ | −0.942 ∗∗ | −0.911 ∗∗ |
| (0.525) | (0.405) | (0.464) | |
| Lnage | −0.0117 | −0.136 | −0.420 ∗∗ |
| (0.195) | (0.147) | (0.189) | |
| Agdp | 0.00751 | 0.00168 | 0.00705 |
| (0.00570) | (0.00371) | (0.00545) | |
| Struc | 0.135 ∗∗ | 0.137 ∗∗∗ | 0.122 ∗∗ |
| (0.0634) | (0.0382) | (0.0525) | |
| Popg | −0.0953 | 0.409 | −0.116 |
| (0.833) | (0.600) | (0.837) | |
| Fdi | −1.046 | −0.00871 | −0.675 |
| (1.050) | (0.712) | (0.917) | |
| Press | 0.0577 ∗ | 0.0601 ∗∗∗ | 0.0459 ∗ |
| (0.0310) | (0.0201) | (0.0271) | |
| Firm FE | Y | Y | Y |
| Year FE | Y | Y | Y |
| Controls | Y | Y | Y |
| Robust | Y | Y | Y |
| Observations | 14,235 | 14,235 | 14,235 |
| R2 | 0.004 | 0.006 | 0.006 |
- Note: Firm FE represents firm fixed effects to control for firm heterogeneity. Year FE stands for year fixed effect to control for unobservable shock effects in different years. Controls represent a series of macro- and microlevel control variables, which have been described in detail above, and will not be repeated. Robust represents the use of robust standard error to control the effects of heteroscedasticity of the error term.
- ∗∗∗, ∗∗, ∗Significant levels at 1%, 5%, and 10%, respectively.
The aforementioned results provide us with three enlightening insights: First, the digital transformation of the focal enterprise significantly reduces the carbon intensity of both upstream and downstream enterprises, thus confirming the existence of industrial chain spillover effects and validating Proposition 1. Second, the spillover effects in downstream enterprises are more pronounced than those in upstream enterprises. One plausible explanation for this disparity lies in the stronger monopolistic capabilities of upstream enterprises compared to the relatively weaker monopolistic capabilities of downstream enterprises [49]. When the digital transformation of the focal enterprise propagates along the industrial chain, it is more likely to impact the production and operation conditions of downstream enterprises. Lastly, the digital transformation of the focal enterprise has the ability to reduce its own carbon emissions, aligning with the findings of Sheng et al., thus corroborating their research direction [6, 7].
Subsequently, let us examine the impact of control variables on the dependent variable: Lev exhibits a significant increase in CEI, possibly attributed to a decrease in environmental awareness when companies face high debt pressure. Lncfo displays a significant decrease in CEI, for similar reasons, as high cash liquidity and reduced debt pressure may prompt proactive adoption of low-carbon strategies. Stru significantly enhances CEI, given that the secondary sector contributes substantially to carbon emissions. Press significantly amplifies CEI, possibly due to the motivation to relax environmental regulations in times of significant government fiscal pressure, aiming to generate additional tax revenue. The remaining variables, for the most part, do not exhibit significant effects, potentially stemming from the nonsignificant average impact coefficients resulting from divergent influences.
4.2. Robustness Test
The above regression results show that the digital transformation of focal enterprises can significantly reduce the carbon intensity of upstream and downstream enterprises, which proves the existence of the industrial chain spillover effect. However, whether the baseline results are robust or not, a series of tests will be carried out in this section. The specific results are shown in Table 3.
| Variables | Carbon emission intensity (CEI) | |||
|---|---|---|---|---|
| Upstream enterprises | Downstream enterprises | Upstream enterprises | Downstream enterprises | |
| (1) | (2) | (3) | (4) | |
| Panel A: instrumental variables | ||||
| Digital | −0.131 ∗∗ | −0.161 ∗∗ | −0.133 ∗∗∗ | −0.162 ∗∗∗ |
| (0.0642) | (0.0645) | (0.0449) | (0.0396) | |
| K-P LM | 199.090 | 199.019 | 1459.329 | 1468.028 |
| K-P Wald | 43.742 | 43.735 | 4456.675 | 4494.141 |
| Observations | 13,305 | 13,305 | 13,780 | 13,780 |
| R2 | 0.001 | 0.001 | 0.004 | 0.006 |
| Panel B: exogenous policy | ||||
| Digital | −0.157 ∗∗ | −0.171 ∗∗ | −0.145 ∗ | −0.154 ∗∗ |
| (0.0785) | (0.0676) | (0.0756) | (0.0681) | |
| Observations | 13,496 | 13,496 | 14,235 | 14,235 |
| R2 | 0.003 | 0.002 | 0.003 | 0.004 |
| Panel C: change the core explanatory variable | ||||
| Digital | −0.120 ∗∗∗ | −0.121 ∗∗∗ | −0.104 ∗∗∗ | −0.135 ∗∗∗ |
| (0.0294) | (0.0265) | (0.0253) | (0.0230) | |
| Observations | 14,235 | 14,235 | 14,235 | 14,235 |
| R2 | 0.004 | 0.006 | 0.004 | 0.007 |
| Panel D: partial sample | ||||
| Digital | −0.0486 ∗∗ | −0.112 ∗∗∗ | −0.0942 ∗∗∗ | −0.123 ∗∗∗ |
| (0.0195) | (0.0286) | (0.0290) | (0.0193) | |
| Observations | 9431 | 10,331 | 12,227 | 13,272 |
| R2 | 0.004 | 0.007 | 0.004 | 0.008 |
| Firm FE | Y | Y | Y | Y |
| Year FE | Y | Y | Y | Y |
| Controls | Y | Y | Y | Y |
| Robust | Y | Y | Y | Y |
- Note: Firm FE represents firm fixed effects to control for firm heterogeneity. Year FE stands for year fixed effect to control for unobservable shock effects in different years. Controls represent a series of macro- and microlevel control variables, which have been described in detail above, and will not be repeated. Robust represents the use of robust standard error to control the effects of heteroscedasticity of the error term. The K-P LM refers to the Kleibergen–Paap rk LM statistic, which reports the results of the underidentification test and clearly rejects the null hypothesis, indicating a significant correlation between the core explanatory variable and the instrumental variable. The K–P Wald represents the Kleibergen–Paap rk Wald F statistic, reporting the results of the weak identification test. The results significantly exceed the critical value of 16.38 corresponding to the 10% statistical significance level in the Stock–Yogo weak identification test, indicating a strong correlation between the core explanatory variable and the instrumental variable.
- ∗∗∗, ∗∗, ∗Significant levels at 1%, 5%, and 10%, respectively.
4.2.1. Instrumental Variables
The fundamental regression results may raise concerns regarding potential endogeneity issues. The digital transformation of focal enterprises can influence the carbon intensity of both upstream and downstream enterprises. Conversely, excessive carbon emissions from these enterprises may trigger negative regulatory interventions by the government through the spillover effects of the industrial chain [50]. This, in turn, can impact the implementation of the digital transformation strategies of focal enterprises, potentially biasing the fundamental regression results. To overcome these potential endogeneity concerns, the paper opts for an instrumental variable approach. Following the practices of Yuan et al, the number of post offices per 100 people in prefecture-level cities in China in 1984, multiplied by the national internet users in the previous year, is employed as an instrumental variable to capture the level of internet development in each city [41, 51, 52]. This choice is motivated by the idea that if a city has a higher number of post offices, it implies a greater demand for information communication, meeting the relevance requirement of the instrumental variable. Moreover, the historical count of post offices and fixed telephone lines has had minimal influence on subsequent years’ corporate CEI, satisfying the exogeneity requirement of the instrumental variable. Specific regression results are presented in Panel A, columns (1)–(2), revealing that the digital transformation of focal enterprises leads to a respective decrease of 0.131% and 0.161% in the carbon intensity of upstream and downstream enterprises, providing robust evidence.
The present study also draws upon references such as Xiao et al, where the average level of digital transformation in peer firms within the same industry in the local area is selected as another instrumental variable [42, 53]. This choice is justified by the fact that the digital development level of the industry in which an enterprise operates influences its own degree of digitalization, satisfying the condition of relevance. Furthermore, the digital development level of peer firms in the same industry within the local area does not directly impact the CEI of the focal enterprise, fulfilling the exogeneity requirement. Specific regression results can be found in Panel A, columns (3)–(4), revealing that the digital transformation of focal enterprises leads to a respective decrease of 0.133% and 0.162% in the carbon intensity of upstream and downstream enterprises, confirming the robustness of the fundamental regression results.
4.2.2. Exogenous Policy
To alleviate potential endogeneity concerns, this paper also leverages exogenous policy shocks related to the digital transformation of focal enterprises. In 2012, the Ministry of Housing and Urban-Rural Development of China initiated a comprehensive evaluation for the national smart city pilot program, subsequently announcing three batches of “smart city” pilot lists in the following years. The construction of smart cities serves as a crucial pathway to promote intensive, intelligent, green, and low-carbon urban development, driving industrial transformation and upgrading. It facilitates the establishment of urban public information platforms, enabling comprehensive cross-industry and cross-departmental applications as well as data sharing, through the full integration of existing information resources and application systems at the local level. Additionally, broadband network serves as strategic public infrastructure for socioeconomic development in the new era. To seize the commanding heights of international economic, technological, and industrial competition, China launched the “Broadband China” strategy in 2013, selecting a total of 120 cities or city clusters as demonstration areas in three batches in 2014, 2015, and 2016. The “smart city” and “Broadband China” pilot policies can be regarded as excellent exogenous shocks that enhance the digital transformation of enterprises in the pilot regions.
Therefore, the paper treats the “smart city” and “Broadband China” urban pilot policies as quasi-natural experiments, employing the method of difference-in-differences for causal identification. Specific results can be found in Panel B: columns (1)–(2) showcase the impact of the smart city pilot policy, revealing a respective decrease of 0.157% and 0.171% in the carbon intensity of upstream and downstream enterprises. Columns (3)–(4) present the effects of the Broadband China pilot policy, indicating a respective reduction of 0.145% and 0.154% in the carbon intensity of upstream and downstream enterprises. The fundamental regression results remain robust.
4.2.3. Change the Core Explanatory Variable
The present study undertakes a robustness examination by reconfiguring the focal enterprise’s digital transformation indicator. First, following the approach of Wu et al. [1], the logarithm of the key cumulative term frequency in the enterprise digitalization terminology lexicon is employed as the measure of digital transformation. Specific regression results can be found in Panel C, columns (1)–(2): the digital transformation of focal enterprises leads to a respective decrease of 0.12% and 0.121% in the carbon intensity of upstream and downstream companies. Second, apart from the entropy weighting method, principal component analysis (PCA) also serves as an objective weighting method widely adopted in the academic community. This approach eliminates the influence of interrelated indicators and ensures the objectivity of weight coefficients. In this study, the digital transformation indicator derived from PCA is employed for robustness examination. Specific regression results can be found in Panel C, columns (3)–(4): the digital transformation of focal enterprises results in a respective reduction of 0.104% and 0.135% in the carbon intensity of upstream and downstream enterprises. The fundamental regression results remain robust.
4.2.4. Delete Partial Sample
Initially, in the previous discussion regarding multiple upstream and downstream enterprises located in different provinces, this paper focuses on the provinces where the companies with the largest main business income are located. However, such treatment may lack precision. Therefore, we exclude these samples and conduct a new regression analysis. Specific results can be found in Panel D, columns (1)–(2): the digital transformation of focal enterprises resulted in a respective decrease of 0.0486% and 0.112% in the carbon intensity of upstream and downstream enterprises. Additionally, certain upstream and downstream enterprises are affiliated with the focal enterprise as associated entities, exhibiting various degrees of equity control, and in some cases, even absolute controlling relationships. Given the close interconnectedness among these samples, it is likely that the estimation of industry spillover effects could be overstated. Consequently, we exclude these samples and perform a new regression analysis. Specific results can be found in Panel D, columns (3)–(4): the digital transformation of focal enterprises leads to a respective reduction of 0.0942% and 0.123% in the carbon intensity of upstream and downstream enterprises. The fundamental regression results remain robust.
5. Mechanism Analysis
The preceding discourse adequately demonstrates that the digital transformation of focal enterprises indeed engenders industry spillover effects. Now, what are the underlying mechanisms at play? The theoretical framework has anticipated the existence of three potential mechanisms, and this section aims to empirically examine them. Detailed results can be found in Table 4.
| Variables | Carbon emission intensity (CEI) | |||||
|---|---|---|---|---|---|---|
| IIE | ISE | RAE | ||||
| Upstream enterprises | Downstream enterprises | Upstream enterprises | Downstream enterprises | Upstream enterprises | Downstream enterprises | |
| (1) | (2) | (3) | (4) | (5) | (6) | |
| Digital | 0.0108 ∗∗∗ | 0.0542 ∗∗∗ | 0.0333 ∗∗∗ | 0.0489 ∗∗∗ | 0.0362 ∗∗∗ | 0.0294 ∗∗ |
| (0.00131) | (0.00653) | (0.00623) | (0.0156) | (0.0107) | (0.0114) | |
| Firm FE | Y | Y | Y | Y | Y | Y |
| Year FE | Y | Y | Y | Y | Y | Y |
| Controls | Y | Y | Y | Y | Y | Y |
| Robust | Y | Y | Y | Y | Y | Y |
| Observations | 14,235 | 14,235 | 14,235 | 14,235 | 14,235 | 14,235 |
| R2 | 0.871 | 0.834 | 0.858 | 0.898 | 0.899 | 0.879 |
- Note: Firm FE represents firm fixed effects to control for firm heterogeneity. Year FE stands for year fixed effect to control for unobservable shock effects in different years. Controls represent a series of macro- and microlevel control variables, which have been described in detail above, and will not be repeated. Robust represents the use of robust standard error to control the effects of heteroscedasticity of the error term.
- ∗∗∗, ∗∗Significant levels at 1% and 5%, respectively.
5.1. The IIE
Existing research has provided a measurement method for the IIE, assessing the degree of integration within the innovation chain of the industry from five perspectives: technological integration capability, market integration capability, clustering, research and development stage capability, and outcome transformation stage capability. However, this method is limited to macrolevel data. In this study, we have chosen to measure IIE by taking the weighted average of patent applications and patent citations (with main business revenue as the weight) of upstream and downstream enterprises. This is because patents represent the outcomes of enterprise innovation, where the number of patent applications can accurately reflect the level of innovation, and the number of patent citations can effectively reflect the quality of innovation [54]. Thus, the weighted average of both can reflect IIE. The specific regression results are as follows: Columns (1)–(2) demonstrate a significant enhancement of IIE in upstream and downstream enterprises due to the digital transformation of focal enterprises, with respective coefficients of 0.0108 and 0.0542, indicating the presence of an IIE mechanism, thus confirming Proposition 2.
5.2. The ISE
Drawing upon the practices of Shan et al., we adopt the approach of using the main business costs of upstream and downstream enterprises as a proxy for enterprise demand. Additionally, we employ the difference between the main business costs and the net value of year-end and year-beginning survival as a proxy for enterprise production, following the notion that the ISE can be measured by the ratio of production variance to demand variance [55, 56]. When information cannot be shared, it leads to distortions and amplifications, resulting in increased demand fluctuations and ultimately a reduction in ISE. For ease of interpretation, we take the reciprocal of ISE. The regression results are as follows: Columns (3)–(4) reveal a significant enhancement of ISE in upstream and downstream enterprises due to the digital transformation of focal enterprises, with respective coefficients of 0.0333 and 0.0489, indicating the presence of an ISE mechanism, thus confirming Proposition 3.
5.3. The RAE
Drawing upon the approach of Chen, we employed the variance of TFP in upstream and downstream enterprises as a measure of the RAE [5]. A higher value of this indicator signifies lower efficiency in resource allocation between upstream and downstream enterprises. Increased volatility indicates suboptimal resource allocation and management during the production process, leading to excessive input and frequent adjustments of certain production factors, thereby amplifying the fluctuation of TFP. For ease of interpretation, we take the reciprocal of RAE. The regression results are as follows: Columns (5)–(6) reveal a significant enhancement of RAE in upstream and downstream enterprises due to the digital transformation of focal enterprises, with respective coefficients of 0.0362 and 0.0294, indicating the presence of a RAE mechanism, thus confirming Proposition 4.
6. Heterogeneity Analysis
In order to enrich the connotation of the spillover effect of the industrial chain, this paper also carries out a heterogeneity analysis, and the specific analysis results are shown in Table 5.
| Variables | Carbon emission intensity (CEI) | |||
|---|---|---|---|---|
| Upstream enterprises | Downstream enterprises | |||
| (1) | (2) | (3) | (4) | |
| Panel A: economic growth target | High | Low | High | Low |
| Digital | −0.105 ∗∗ | −0.108 ∗∗∗ | −0.0985 ∗∗∗ | −0.130 ∗∗∗ |
| (0.0416) | (0.0413) | (0.0380) | (0.0356) | |
| Observations | 7006 | 7229 | 7006 | 7229 |
| R2 | 0.009 | 0.006 | 0.011 | 0.009 |
| Panel B: environmental regulation | High | Low | High | Low |
| Digital | −0.0897 ∗∗ | −0.0860 ∗∗ | −0.116 ∗∗∗ | −0.101 ∗∗ |
| (0.0429) | (0.0399) | (0.0345) | (0.0392) | |
| Observations | 6813 | 7422 | 7422 | 6813 |
| R2 | 0.008 | 0.007 | 0.009 | 0.006 |
| Panel C: regional difference | Western | Central-eastern | Western | Central-eastern |
| Digital | −0.0707 | −0.112 ∗∗∗ | −0.0430 | −0.130 ∗∗∗ |
| (0.0781) | (0.0278) | (0.0578) | (0.0254) | |
| Observations | 1719 | 12,516 | 1719 | 12,516 |
| R2 | 0.011 | 0.004 | 0.022 | 0.007 |
| Firm FE | Y | Y | Y | Y |
| Year FE | Y | Y | Y | Y |
| Controls | Y | Y | Y | Y |
| Robust | Y | Y | Y | Y |
- Note: Firm FE represents firm fixed effects to control for firm heterogeneity. Year FE stands for year fixed effect to control for unobservable shock effects in different years. Controls represent a series of macro- and microlevel control variables, which have been described in detail above, and will not be repeated. Robust represents the use of robust standard error to control the effects of heteroscedasticity of the error term.
- ∗∗∗, ∗∗, ∗Significant levels at 1%, 5%, and 10%, respectively.
6.1. Economic Growth Target
Due to the excessive emphasis on short-term performance and political achievements in traditional promotion tournaments, local government officials often prioritize economic growth and political accomplishments over the long-term interests of the environment, society, and sustainable development goals. This tendency makes them prone to adopting irresponsible development models, leading to issues such as resource wastage, environmental pollution, and social conflicts. These problems hinder economic sustainability. Consequently, we anticipate that the industrial chain spillover effects of focal enterprise digital transformation would be weaker in the high-growth target sample. Leveraging the annual government work reports of various provinces, we collected panel data on economic growth targets and divided the sample into two categories: high-growth targets and low-growth targets based on the median value. The results of the grouped regression analysis are presented in Panel A: Columns (1)–(2) demonstrate the heterogeneity spillover effects in upstream enterprises, indicating that the carbon reduction effect of focal enterprise digital transformation is smaller in the high-growth target sample (0.105 < 0.108). Columns (3)–(4) illustrate the heterogeneity spillover effects in downstream enterprises, confirming that the carbon reduction effect of focal enterprise digital transformation is also smaller in the high-growth target sample (0.0985 < 0.13). This confirms our expectations.
6.2. Environmental Regulation
The government, in its pursuit of environmental protection and sustainable development, will establish a series of regulatory frameworks and institutions to guide and constrain enterprises and industries in their environmental conservation and low-carbon transformation efforts. It is expected that the industrial chain spillover effects of focal enterprise digital transformation would be stronger in the high environmental regulation sample. Drawing upon the work of Chen, Zhang, and Liu, we construct a measure of environmental regulation enforcement intensity at the prefecture level based on the frequency of “environmental protection” related terms in government work reports from various regions1 [57]. Using the median value, we categorized the sample into high environmental regulation and low environmental regulation groups. The results of the grouped regression analysis are presented in Panel B: Columns (1)–(2) demonstrate the heterogeneity spillover effects in upstream enterprises, indicating that the carbon reduction effect of focal enterprise digital transformation is greater in the high environmental regulation sample (0.0897 > 0.086). Columns (3)–(4) illustrate the heterogeneity spillover effects in downstream enterprises, confirming that the carbon reduction effect of focal enterprise digital transformation is also greater in the high environmental regulation sample (0.116 > 0.101). This validates our expectations.
6.3. Regional Difference
The impact of the digital economy on carbon emissions shows significant regional heterogeneity [38]. Compared to the central and eastern regions, the digital economy in the western region exhibits a predominantly positive effect on carbon emissions. Additionally, factors such as industrial structure, technological level, environmental regulations [58], and economic growth patterns in the western region may lead to a relatively weaker carbon emission reduction effect. Therefore, it is expected that in the western regions of China, the spillover effects of digital transformation in focal enterprises on the industrial chain are weaker. Based on the regional classification by the National Bureau of Statistics of China2, we divide the sample into two categories: the western region and the central-eastern region. The results of the grouped regression analysis are presented in Panel C: Columns (1)–(2) demonstrate the heterogeneous spillover effects in upstream enterprises, indicating that the carbon reduction effect of focal enterprise digital transformation is smaller in the western region (0.0707 < 0.112). Columns (3)–(4) illustrate the heterogeneity spillover effects in downstream enterprises, confirming that the carbon reduction effect of focal enterprise digital transformation is also smaller in the western region sample (0.043 < 0.13). This verifies our expectations.
6.4. Industrial Difference
The impact of the digital economy on carbon emissions exhibits significant industry heterogeneity [59]. Table 6 presents the results of the industry heterogeneity analysis: Column (1) illustrates that digital transformation in focal enterprises has varying effects on upstream firms across different sectors. In the manufacturing industry, the CEI decreased by 8.94%, suggesting a moderate impact of digitalization in optimizing production efficiency and resource management. In contrast, the wholesale and retail industry experienced a larger reduction of 23%, indicating the substantial role of digitalization in logistics and inventory management in significantly reducing carbon footprints. The information transmission, software, and information technology services sector saw a more pronounced decrease of 42.3%, pointing to the higher potential of digitalization in optimizing operational processes and reducing energy consumption. Industries such as real estate (37.7%) and scientific research and technical services (106.2%) exhibited even larger reductions, reflecting the role of digital transformation in fostering smarter building practices, energy-saving technologies, and low-carbon innovation. Additionally, the education sector (367.2%) and public administration, social security, and social organizations (48.6%) showed considerable reductions, highlighting the potential of digitalization in improving administrative efficiency and resource optimization. These differences underscore that the carbon emission reduction effects of digital transformation are shaped by industry-specific characteristics, including energy consumption patterns, technological readiness, and operational structures.
| Variables | Carbon emission intensity (CEI) | |
|---|---|---|
| Upstream enterprises | Downstream enterprises | |
| (1) | (2) | |
| Agriculture, forestry, animal husbandry, and fishery | 0.0326 | 0.385 |
| (0.283) | (0.249) | |
| Mining industry | −0.0907 | −0.0563 |
| (0.214) | (0.149) | |
| Manufacturing industry | −0.0894 ∗∗∗ | −0.144 ∗∗∗ |
| (0.0324) | (0.0295) | |
| Electricity, heat, gas, and water production and supply industry | −0.0470 | −0.0962 |
| (0.155) | (0.119) | |
| Construction industry | −0.0913 | −0.413 ∗∗ |
| (0.194) | (0.186) | |
| Wholesale and retail trade | −0.230 ∗∗ | −0.189 ∗ |
| (0.112) | (0.108) | |
| Transportation, storage, and postal services | 0.000104 | −0.0910 |
| (0.133) | (0.0993) | |
| Accommodation and catering services | 0.387 | 0.680 |
| (0.479) | (0.433) | |
| Information transmission, software, and information technology services | −0.423 ∗∗∗ | −0.329 ∗∗ |
| (0.156) | (0.136) | |
| Financial industry | −0.263 | 1.100 |
| (0.480) | (0.772) | |
| Real estate industry | −0.377 ∗ | −0.147 |
| (0.191) | (0.183) | |
| Leasing and business services | 0.413 | 0.185 |
| (0.317) | (0.366) | |
| Scientific research and technical services | −1.062 ∗ | 0.195 |
| (0.542) | (0.563) | |
| Water conservancy, environment, and public facilities management | −0.350 | −0.182 |
| (0.384) | (0.370) | |
| Resident services, repair, and other personal services | 3.062 | 5.759 |
| — | — | |
| Education | 3.672 ∗∗∗ | −1.735 ∗∗∗ |
| (1.02e − 06) | (1.02e − 06) | |
| Health and social work | −5.958 | −13.97 ∗∗∗ |
| (5.628) | (0.451) | |
| Culture, sports, and entertainment | 0.0696 | −0.190 |
| (0.445) | (0.194) | |
| Public administration, social security, and social organizations | 0.486 ∗∗ | −0.224 |
| (0.234) | (0.227) | |
| Firm FE | Y | Y |
| Year FE | Y | Y |
| Controls | Y | Y |
| Robust | Y | Y |
- Note: Firm FE represents firm fixed effects to control for firm heterogeneity. Year FE stands for year fixed effect to control for unobservable shock effects in different years. Controls represent a series of macro- and microlevel control variables, which have been described in detail above, and will not be repeated. Robust represents the use of robust standard error to control the effects of heteroscedasticity of the error term.
- ∗∗∗, ∗∗, ∗Significant levels at 1%, 5%, and 10%, respectively.
Column (2) reveals similar industry-specific heterogeneity in the impact of digital transformation on the CEI of downstream firms. In the manufacturing sector, CEI decreased by 14.4%, suggesting that digital transformation positively influences downstream firms by improving production efficiency and reducing resource waste. The construction industry demonstrated a significant reduction of 41.3%, reflecting the transformative potential of digitalization in optimizing construction material management and enhancing the efficiency of construction processes, leading to substantial carbon emission reductions. The wholesale and retail industry saw a 18.9% decrease, highlighting the role of digitalization in optimizing supply chain management and enhancing logistics efficiency, which reduces the carbon footprint of downstream enterprises. Similarly, the information transmission, software, and information technology service sector experienced a 32.9% reduction, demonstrating that digitalization drives higher efficiency in information flow and lower operational energy consumption. Notably, the education (173.5%) and health and social work (1397%) industries exhibited drastic reductions in carbon emissions, illustrating the transformative effect of digitalization in improving service efficiency, optimizing resource allocation, and reducing energy consumption in sectors with high social service demand. Overall, these findings illustrate that the carbon emission reduction effects of digital transformation are closely tied to the technological and operational characteristics of each industry, emphasizing the critical role of sector-specific factors in shaping the environmental impact of digitalization.
7. Conclusions and Policy Recommendations
This study demonstrates that the digital economy is driving a technological and industrial revolution, with digital transformation in enterprises having a significant impact on the environment across industrial chains. Unlike traditional informatization, digital transformation aligns internal processes with market demands, resulting in a notable reduction in carbon emissions. By constructing a digital transformation index using natural language processing, the study finds that digital transformation in focal enterprises significantly reduces CEI in both upstream and downstream firms through three key mechanisms: innovation integration, information spillover, and resource allocation. Industry heterogeneity analysis reveals significant variation in how different sectors respond to digital transformation, with particularly pronounced carbon reduction effects observed in the manufacturing, information technology services, real estate, and education sectors. Furthermore, the carbon reduction effects of digital transformation are more pronounced in regions with lower economic growth targets and stricter environmental regulations, particularly in central and eastern China.
Based on the findings of this study, we offer the following policy recommendations to help developing countries leverage digital transformation for green and low-carbon development while addressing potential rebound effects in order to meet the 2030 greenhouse gas reduction targets: (1) Promote sector-specific digital transformation. Governments should develop targeted policies based on the characteristics of each industry, especially in sectors with high potential for digital transformation (e.g., information technology services, real estate, and education), to promote high-efficiency applications and carbon emission reductions. In traditional industries (e.g., manufacturing and construction), efforts should focus on accelerating infrastructure development and fostering green technology innovation. (2) Strengthen digital infrastructure and market management reforms. Improving digital infrastructure, particularly in the western regions, is crucial to facilitate enterprise digital transformation. At the same time, market management reforms should promote green innovation in enterprises and ensure that the digital transformation does not lead to negative environmental outcomes. (3) Improve carbon pricing mechanisms and support green technologies. Carbon pricing mechanisms, such as carbon taxes and emission trading systems, should be implemented to internalize environmental costs and align digital transformation with low-carbon goals. Additionally, governments should increase support for the research, development, and application of green technologies to drive the adoption of low-carbon solutions. (4) Enhance cross-sector collaboration and policy coordination. Governments should foster collaboration between enterprises, local governments, and academia, coordinate the goals of different interest groups, and promote green and low-carbon collaborative development across regions. Successful experiences in regions with advanced technologies and infrastructure should be leveraged to drive transformation in other regions. (5) Flexibly adjust policies to address rebound effects. To avoid potential rebound effects from digital transformation, governments should set flexible carbon emission control targets and strengthen oversight and adjustments in industry transformation processes, ensuring that the transformation reduces carbon emissions while maintaining long-term sustainability.
While this study makes a significant contribution to understanding the environmental impacts of digital transformation across industrial chains, it has some limitations. Notably, despite considering industry heterogeneity, the analysis could be further refined, particularly with respect to the rebound effects that may counteract the carbon reduction benefits of digital transformation. Future research should explore industry-specific rebound effects and their overall impact on carbon emissions. Additionally, potential conflicts between enterprise objectives and government policy goals, particularly in high energy-consuming industries, warrant further investigation. Future studies could integrate more precise mathematical models that incorporate industry characteristics and rebound effects, while also considering regional and sectoral policy differences, thereby enhancing the comprehensiveness and applicability of the findings and providing more actionable policy recommendations for achieving sustainable development goals.
Conflicts of Interest
The authors declare no conflicts of interest.
Author Contributions
Lin Guo: writing – review and editing, writing – original draft, visualization, data curation, and conceptualization. Shanna Li: formal analysis, validation, writing – review and editing. Ying Liu: formal analysis, validation, writing – review and editing. Bin Sang: writing – review and editing, writing – original draft, visualization, data curation, methodology, and validation. Chunyuan Zhang: writing – review and editing, data curation, formal analysis, and investigation. All authors have read and agreed to the published version of the manuscript. Lin Guo, Shanna Li, and Ying Liu contributed equally to this paper and shared the first authorship.
Funding
This work was supported by the project of Weifang Science and Technology Bureau “Enterprise Digital Transformation Smooth Green low-carbon Big Cycle: Effect and Mechanism” (2023RKX133) and the Fundamental Research Funds for the Provincial Universities of Zhejiang (2024ZD25).
Endnotes
1Among them are 15 published works: environmental words, low carbon, environmental protection, air, green, PM2.5, chemical oxygen demand, carbon dioxide, PM10, ecology, pollutant discharge, emission reduction, pollution, environmental protection, sulfur dioxide, energy consumption, etc.
2For more details, see https://www.stats.gov.cn/hd/lyzx/zxgk/202107/t20210730_1820095.html.
Open Research
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
