Can Supply Chain Digitalization Reduce Corporate Carbon Emission Intensity? Evidence from the Annual Reports of Chinese Listed Companies

Abstract

In the context of a rapidly evolving data-driven economy and increasingly stringent carbon reduction policies, the impact of supply chain digitalization (SCD) on corporate carbon emission intensity (CEI) has become an important research topic. Using panel data on Chinese A-share listed firms from the Shanghai and Shenzhen stock exchanges over the period 2013–2023, this study employs Python-based text analysis of corporate annual reports to explore the effect of SCD on corporate CEI. The results show that SCD significantly reduces corporate CEI. Mechanism analysis further indicates that this effect operates through three channels: alleviating financing constraints, promoting green innovation, and reducing supply chain disruption risk. Heterogeneity analysis reveals that the mitigating effect of SCD on corporate CEI is more pronounced among non-state-owned firms, large-scale firms, firms in non-high-tech industries, firms in highly environmentally sensitive industries, and firms located in regions with more developed digital infrastructure. Further analysis shows that SCD contributes to improvements in both firms’ sustainability and financial performance. Overall, this study provides important policy implications for both governments and firms.

1. Introduction

Global climate change, characterized by extreme phenomena such as sea-level rise, droughts, and torrential rainfall [1,2], poses a significant global challenge. Carbon dioxide emissions are widely acknowledged as a primary driver of climate change [3], thereby necessitating coordinated global action. China, as one of the largest contributors to global greenhouse gas emissions, has actively participated in international efforts to mitigate carbon emissions. Six years ago, China proposed the “Dual Carbon Goals” to address carbon emissions matters, with supply chain decarbonization identified as a critical pathway toward achieving these targets. A supply chain constitutes an interconnected network linking upstream and downstream firms, within which carbon emissions and energy consumption occur at multiple stages [4], including raw material sourcing, energy use, and production processes [5]. Emissions from production and supply chain activities account for over 42% of global carbon emissions, while indirect supply chain emissions are typically 5.5 times greater than direct emissions, reaching as high as 10.7 times in the retail sector [6]. Effective supply chain management enhances resource allocation efficiency, improves operational performance, and reduces environmental burdens. Under increasingly stringent carbon regulatory frameworks, firms are required to strengthen supply chain management practices to ensure compliance. Accordingly, supply chain optimization has become an essential strategic approach for promoting carbon reduction and facilitating green transformation [7].

Existing studies primarily focus on the determinants of carbon emissions from the perspective of aggregate emission levels, emphasizing both internal corporate characteristics [1] and external political and economic environments [8,9]. However, relatively few studies employ carbon emission intensity (CEI), defined as carbon emissions per unit of output. Achieving the “Dual Carbon Goals” requires profound economic and social transformation, particularly in improving energy utilization efficiency and reducing corporate CEI [10]. Prior research has analyzed the drivers and governance mechanisms of corporate CEI. Government intervention contributes substantially to the reduction in corporate CEI. Some studies argue that fiscal and tax incentives promote the adoption of clean technologies, thereby generating positive environmental externalities [11]. Other studies find that increased government fiscal expenditure contributes to reductions in corporate CEI [12]. Furthermore, favorable financing conditions, particularly green credit policies, encourage heavily polluting firms to engage in technological innovation and low-carbon transformation [13]. Technological innovation is also identified as a key driver of corporate CEI reduction. Evidence shows that technological progress in China’s steel industry improved energy efficiency by 60% between 1994 and 2003 [14], while digital inclusive finance has been shown to reduce urban CEI through innovation channels [15]. However, existing studies exhibit several limitations. First, prior research predominantly focuses on macro-level factors and corporate governance mechanisms, with limited attention to the supply chain perspective. Second, much of the existing literature relies on qualitative analysis, with insufficient quantitative empirical evidence.

Supply chain digitalization (SCD) represents a critical management transformation in the digital economy. It is characterized by the use of technologies, including the IoT and artificial intelligence, to reconfigure the collaborative efficiency of supply chain components [16,17]. In practice, firms and their partners utilize digital technologies to restructure organizational workflows, coordination mechanisms, and value creation across both internal and external supply chain networks, including logistics, production, and information systems [18]. Existing literature primarily emphasizes the economic benefits of SCD. Empirical evidence shows that digitalization improves supply chain efficiency [19] and enhances operational stability [20]. Evidence from Malaysian manufacturing firms shows that digitalization significantly improves supply chain processes [13]. Other studies document that SCD enhances organizational resilience [21] and strengthens competitive advantage [22]. However, research on environmental impacts remains limited, with existing studies mainly focusing on corporate ESG performance [23], green innovation [24,25], and carbon emissions [26]. Recent studies have begun to explore how supply chains can reduce carbon emissions through digital solutions, including optimization algorithms [27] and artificial intelligence technologies [28]. Nevertheless, an integrated understanding of the impact of SCD on corporate CEI and its underlying mechanisms remains lacking. Furthermore, existing research primarily relies on dynamic capability theory or the resource-based view [29], often employing difference-in-differences methods or survey-based approaches, resulting in limited empirical measurement and insufficient empirical evidence regarding the effects of SCD.

Therefore, using a panel dataset of A-share listed firms from the Shanghai and Shenzhen stock exchanges spanning 2013 to 2023, this study employs machine learning-based text analysis of corporate annual reports to measure SCD. The results indicate that SCD significantly reduces corporate CEI. This effect operates through three primary channels: alleviating financing constraints, promoting green innovation, and lowering supply chain disruption risk. Heterogeneity analysis further shows that the suppressive effect is more pronounced in non-state-owned enterprises, large-scale enterprises, non-high-tech industries, highly environmentally sensitive industries, and firms located in regions with more developed digital infrastructure.

This study makes several contributions. First, it proposes a novel analytical perspective by examining corporate CEI reduction at the firm level through the lens of SCD, thereby adding to the growing body of literature on the non-economic consequences of digitalization from the perspectives of Dynamic Capability Theory, the Resource-Based View, and Institutional Theory. Second, it develops a quantitative measurement framework for SCD using machine learning techniques implemented in Python 3.14.4, contributing to the advancement of empirical measurement and indicator construction in this field. Third, it identifies multiple mechanism pathways—including financing constraints, green innovation, and supply chain disruption risk—thereby enriching the understanding of how SCD affects corporate CEI. Finally, it investigates heterogeneity across firm-level, industry-level, and regional dimensions, thereby yielding a more comprehensive understanding of the conditional effects of SCD.

The rest of the study is organized below. Section 2 outlines the theoretical framework and develops hypotheses. Section 3 introduces the variable definitions and model specification. Section 4 reports descriptive statistics, multicollinearity tests, baseline regression results, and robustness tests; Section 5 introduces mechanism effects, heterogeneity analysis, and economic consequence analysis; Section 6 summarizes conclusions, theoretical implications, practical implications, and limitations.

2. Hypothesis Development

2.1. Policy Background

To facilitate supply chain transformation and upgrading, the Chinese government issued the “Notice on Carrying out Pilot Projects for Supply Chain Innovation and Application,” aimed at promoting supply chain innovation in response to evolving economic demands. The policy targets both pilot cities and pilot firms. Pilot cities are required to implement supporting policies, improve operational conditions, and promote governance innovation, while pilot firms are required to adopt advanced digital technologies to upgrade their supply chains. In October 2018, a total of 55 cities and 266 firms were selected as pilot units, providing a representative sample for policy implementation. This policy initiative provides a quasi-natural experimental setting to observe variations in the degree of SCD at the firm level, thereby enabling an empirical examination of its impact on corporate CEI and the underlying mechanisms.

2.2. Direct Impact

This study integrates insights from Dynamic Capability Theory, the Resource-Based View, and Institutional Theory, and demonstrates that the impact of SCD on corporate CEI can be understood within an analytical framework characterized by “external institutional drivers–internal resource foundations–dynamic capability transformation.”

First, Institutional Theory suggests that, under increasingly stringent environmental regulations [2] and the strengthening of digital economy policies [30], firms face mounting pressure to improve environmental performance and advance digital transformation, thereby acting as external institutional drivers that promote both the adoption of SCD and the reduction in corporate CEI. Second, the Resource-Based View posits that digital technologies and data elements have become critical strategic resources. These resources enable firms to overcome information barriers across production, sales, and transportation processes, thereby improving overall supply chain efficiency [31]. Through digital collaboration with suppliers in transportation scheduling and inventory allocation, firms are able to alter production modes more flexibly, reduce redundant production and inefficient energy consumption, and achieve low-carbon objectives [32]. Moreover, the implementation of SCD promotes information sharing and optimized management across the supply chain, as reflected in enhanced inventory management and the monitoring of supply chain operations and supplier capacity utilization, thereby improving resource utilization efficiency and reducing corporate CEI without compromising production efficiency [33]. Finally, in terms of Dynamic Capability Theory, the framework of sensing, seizing, and transforming capabilities proposed by Teece [34] provides a theoretical explanation for the effects of digitalization. Digitalization enhances firms’ flexibility in responding to environmental and market changes through real-time data monitoring, thereby promoting sustainable development and reducing unnecessary carbon emissions. Specifically, firms transform digital resources into operational optimization and improved environmental performance, while SCD, supported by real-time monitoring and intelligent decision-making, strengthens firms’ ability to respond to market fluctuations and environmental changes, thereby reducing additional carbon emissions caused by production volatility or resource misallocation [4].

Accordingly, driven by external institutional pressures, firms rely on digital resources and achieve resource reconfiguration through dynamic capability adjustments, thereby effectively reducing corporate CEI. Accordingly, this study proposes:

Hypothesis 1.

SCD reduces corporate CEI.

2.3. Indirect Impact

From a theoretical perspective, the financing constraint channel can be explained through Institutional Theory and the Resource-Based View. Institutional Theory suggests that, with the continuous improvement of financial regulatory environments and the advancement of digitalization policies, corporate information disclosure is enhanced, thereby enabling market participants and regulators to better assess firms’ financial conditions and improve the overall financing environment [35]. The Resource-Based View further emphasizes that information and data, as key strategic resources, can effectively alleviate information asymmetry in the financing process, thereby contributing to carbon reduction. Within supply chain relationships, information asymmetry often gives rise to opportunistic behavior, where firms may exploit contractual loopholes by deliberately reducing input or output, thereby harming other supply chain participants [36]. SCD reduces search costs and provides both customers and suppliers with more options, facilitating more efficient matching and mitigating information asymmetry [37]. In addition, SCD promotes the development of sharing platforms that enable previously inaccessible information, including price and demand information, to be widely shared [19], thereby improving supply chain efficiency.

Alleviating financing constraints plays a critical role in reducing CEI. Reduced cash flow constraints enable firms to increase investment in carbon reduction activities [38]. From a practical perspective, SCD improves financing conditions, expands financing channels, and incentivizes firms to engage in emission reduction [39]. Accordingly, this study proposes:

Hypothesis 2.

SCD reduces corporate CEI by alleviating financing constraints.

Institutional Theory posits that external institutional environments shape corporate behavior and strategic choices [40]. Digitalization policies enhance corporate management capabilities and efficiency by facilitating the application of digital technologies and data resources [4]. Guided by such policies and supported by fiscal and tax incentives, firms increase investment in green innovation and expand the development and application of green technologies [30], including low-carbon production equipment and intelligent monitoring systems. These measures improve information transparency and provide a foundation for green innovation.

The Resource-Based View suggests that competitive advantage arises from firm-specific resources and capabilities [41]. Green innovation, by leveraging unique resources and technologies, enables firms to reduce CEI. Green innovation allows firms to access technological resources that facilitate more efficient manufacturing processes, greater energy efficiency, and reduced waste generation [42]. These improvements do not undermine firms’ competitive advantage; rather, they contribute to improvements in operational efficiency and reduce corporate CEI. The development of difficult-to-imitate green technologies further strengthens sustainable competitive advantage. Accordingly, this study proposes:

Hypothesis 3.

SCD reduces corporate CEI by promoting green innovation.

As global economic uncertainty intensifies, supply chains are increasingly exposed to external shocks. Supply chain disruption risk may arise from excessive concentration of costs and risks, as well as from demand fluctuations driven and amplified by information asymmetry. By leveraging emerging digital technologies and related infrastructure, SCD enables real-time connectivity and dynamic sharing of information, logistics, capital, and knowledge flows [43]. From a theoretical perspective, this mechanism can be explained by Dynamic Capability Theory, which emphasizes that firms can respond more effectively to external shocks by strengthening their capacities for environmental sensing and adaptive response [44]. SCD enables firms to monitor supply chain operations and potential risks in advance—for example, through algorithmic analysis [27]—and dynamically adjust resource allocation, thereby reducing the likelihood of supply chain disruptions. Accordingly, firms can more promptly identify potential risk factors, assess disruption probabilities, and respond rapidly to unexpected events.

Reducing supply chain disruption risk helps ensure production continuity, improves resource allocation efficiency, and lowers additional energy consumption. A stable supply chain allows firms to conduct procurement and production activities as planned, avoiding energy-intensive outcomes such as inventory backlogs caused by material shortages or logistical disruptions [4]. Therefore, mitigating disruption risk reduces additional carbon emissions arising from production volatility, thereby lowering corporate CEI. Accordingly, this study proposes:

Hypothesis 4.

SCD reduces corporate CEI by lowering supply chain disruption risk.

Figure 1 presents the theoretical framework in detail.

3. Research Design

3.1. Sample and Data

This study uses a panel dataset of Chinese A-share listed companies from the Shanghai and Shenzhen stock exchanges from 2013 to 2023. The sample is screened based on the following criteria. First, firms identified as ST or *ST are excluded from the sample. Second, firms in the real estate and financial industries are excluded. Third, observations with considerable missing data in key variables are removed from the sample. Fourth, key continuous variables are adjusted using winsorization at the 1st and 99th percentiles. Data on SCD and supply chain disruption risk are obtained from text analysis of corporate annual reports. Green innovation data are sourced from the CNRDS database, while the remaining variables, including corporate CEI and financial variables, are obtained from the CSMAR database.

3.2. Measures of Variable

3.2.1. Independent Variable

The independent variable was Supply Chain Digitalization (SCD). Drawing on prior research [25], this study applies Python-based textual analysis to the Management Discussion and Analysis (MD&A) sections of corporate annual reports. Based on the Guidelines for Digital Supply Chain Management issued by the State Administration for Market Regulation and the Standardization Administration of China in 2022, which classify corporate SCD into five dimensions, a keyword dictionary is constructed, and keyword frequencies are calculated for A-share listed firms. To account for variation in document length, the digitalization measure is normalized by dividing the total keyword frequency by the length of the MD&A text. For presentation purposes, the index is further rescaled by a factor of 10, where higher values indicate a higher level of SCD.

It is worth noting that, on the one hand, if firms engage in exaggerated disclosures regarding supply chain digitalization in their annual reports, the SCD indicator constructed in this study may overestimate the actual level of digitalization. From an econometric perspective, such measurement error can be characterized as classical measurement error, which typically biases estimated coefficients toward zero. Therefore, in the presence of disclosure bias, the estimated impact of SCD on carbon emission intensity is more likely to be underestimated rather than overestimated, suggesting that the results reported in this study are conservative. On the other hand, in terms of signaling theory, frequent disclosures regarding digital transformation in annual reports carry a certain degree of credibility, as inaccurate disclosures may entail regulatory penalties and reputational risks. Therefore, this indicator not only reflects disclosure behavior but also, to some extent, captures firms’ actual digital investment and strategic priorities. Second, with respect to information quality, although external stakeholders may obtain information on SCD through informal channels, such information is often less reliable. In contrast, audited and publicly disclosed annual reports enhance the credibility of such information [45]. Even when the MD&A section does not provide entirely new information, it can still serve to validate previously acquired information from informal channels [46], thereby improving the overall reliability of SCD-related information.

3.2.2. Dependent Variable

The dependent variable was Carbon Emission Intensity (CEI). There is no regulatory requirement in China for enterprises to disclose carbon-related data; therefore, it is difficult to directly obtain such information [47]. Furthermore, some listed companies may engage in greenwashing, leading to misreporting or underreporting, which may introduce selection bias. Consequently, estimating carbon emissions based on well-established industry-level data improves the representativeness of the sample. Following prior studies [48,49,50], corporate CEI is measured as carbon dioxide emissions per unit of operating revenue (in 10,000 yuan). The specific formulas are presented in Equations (1) and (2). According to the CO₂ accounting guidelines of the Xiamen Energy Conservation Center, the carbon conversion coefficient θ is set at 2.493 per ton of standard coal.

$$\begin{gathered} Corporate carbon dioxide emissions= \\ {\displaystyle \frac{Business Operating Costs}{Operating costs in the company\mathit{’}s industry}}\times \\ Total energy consumption in the industry\times \theta \end{gathered}$$

(1)

$$CEI={\displaystyle \frac{Corporate carbon dioxide emissions}{Company Revenue}}\times {10}^5$$

(2)

3.2.3. Control Variables

To ensure model rigor and following prior studies, control variables are included to capture general firm characteristics, including firm size (SIZE), firm age (AGE), leverage ratio (LEV), operating cash flow (CFO), and profitability (ROA), as well as corporate governance and innovation-related characteristics, including state ownership (SOE), ownership concentration (TEG), independent director ratio (IDP), and R&D intensity (RES).

Table 1. Definitions of variables.

Variable	Definition	Measurement
CEI	Carbon emission intensity	See Section 3.2.2
SCD	Supply chain digitalization	See Section 3.2.1
SIZE	Size of firms	ln (total assets for the year + 1)
AGE	Age of firms	ln (current year—the year of establishment + 1)
LEV	The ratio of leverage	Total liabilities/total assets
CFO	Cash flow from operations	Net cash flows from operating activities/total assets
ROA	Profitability	Net profit/average shareholders’ equity
SOE	State-owned enterprise	State-controlled enterprise = 1, otherwise = 0
TEG	Concentration of ownership	Shareholding percentage of the largest shareholder
IDP	Percentage of Independent Directors	The number of independent directors divided by the total number of board members
RES	R&D intensity	R&D expenditure as a percentage of revenue

provides detailed information.

3.3. Model

A fixed-effects model is employed:

$${CEI}_{i,t}=\alpha +\beta {SCD}_{i,t}+\gamma {Controls}_{i,t}+Ind+Year+\epsilon$$

(3)

In model (3), ${CEI}_{i,t}$ represents the CEI of firm $i$ in period $t$, ${SCD}_{i,t}$ is the SCD of firm $i$ in period $t$, ${Controls}_{i,t}$ is the group of control variables, $Ind$ represents industry fixed effects, $Year$ represents time fixed effects, and $\epsilon$ represents the random disturbance term. Standard errors are clustered at the firm level to account for potential within-firm correlation.

4. Results and Discussion

4.1. Descriptive Statistical Analysis

As reported in

Table 2. Descriptive statistic.

Variable	N	Mean	S.D.	Min	Max
CEI	22,119	0.449	0.567	0.040	2.279
SCD	22,119	0.006	0.012	0.000	0.236
AGE	22,119	2.989	0.289	2.197	3.611
LEV	22,119	0.400	0.189	0.060	0.859
SIZE	22,119	22.227	1.213	20.102	26.000
SOE	22,119	0.277	0.447	0.000	1.000
TEG	22,119	0.334	0.141	0.094	0.727
ROA	22,119	0.037	0.063	−0.226	0.201
CFO	22,119	0.053	0.065	−0.129	0.241
IDP	22,119	0.378	0.054	0.333	0.571
RES	22,119	0.050	0.044	0.000	0.266

, CEI has a mean of 0.449, with values spanning from 0.040 to 2.279, indicating substantial variation in CEI across firms. The SCD variable has a minimum value of 0 and a mean of 0.006, which is higher than the median (0.002), suggesting that although some firms exhibit relatively high levels of digitalization, the overall level of SCD remains low, with considerable dispersion across firms.

4.2. Multicollinearity Test

The variance inflation factor (VIF) test is conducted to detect multicollinearity.

Table 3. VIF test.

Variable	VIF	1/VIF
LEV	1.73	0.58
SIZE	1.60	0.62
ROA	1.60	0.63
CFO	1.35	0.74
SOE	1.26	0.79
RES	1.15	0.87
TEG	1.11	0.90
AGE	1.09	0.91
IDP	1.01	0.99
SCD	1.01	0.99
Mean	1.29

shows that the VIF values of all variables are well below the commonly accepted threshold of 10. The highest VIF value is 1.73 for leverage (LEV), and the mean VIF is 1.29. These results indicate no serious multicollinearity concerns in the regression model.

4.3. Baseline Regression Results

First, the Hausman test results (χ² (10) = 1301.70) justify the use of a fixed-effects model.

Table 4. Baseline regression.

Variables	CEI	CEI	CEI	CEI	CEI
	(1)	(2)	(3)	(4)	(5)
SCD	−0.417 ***	−0.413 ***	−0.403 ***	−0.421 ***	−0.845 ***
	(0.104)	(0.099)	(0.097)	(0.101)	(0.232)
SIZE		0.002	−0.000	−0.000	−0.000
		(0.002)	(0.002)	(0.002)	(0.002)
AGE		0.025 ***	0.021 ***	0.011	0.011
		(0.007)	(0.007)	(0.007)	(0.007)
LEV		0.060 ***	0.059 ***	0.028 **	0.028 **
		(0.012)	(0.012)	(0.012)	(0.012)
CFO		−0.177 ***	−0.176 ***	−0.196 ***	−0.194 ***
		(0.023)	(0.023)	(0.023)	(0.023)
ROA		−0.420 ***	−0.427 ***	−0.483 ***	−0.484 ***
		(0.032)	(0.031)	(0.032)	(0.032)
SOE			0.022 ***	0.020 ***	0.020 ***
			(0.005)	(0.005)	(0.005)
TEG			0.052 ***	0.044 ***	0.045 ***
			(0.014)	(0.014)	(0.014)
IDP			−0.018	−0.009	−0.010
			(0.032)	(0.031)	(0.031)
RES				−0.584 ***	−0.588 ***
				(0.046)	(0.046)
SCD22					5.835 ***
					(2.256)
Constant	0.451 ***	0.322 ***	0.383 ***	0.452 ***	0.450 ***
	(0.002)	(0.047)	(0.052)	(0.051)	(0.051)
Ind	Yes	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes	Yes
R2	0.914	0.919	0.919	0.921	0.921
N	22,119	22,119	22,119	22,119	22,119

Note: *** p < 0.01; ** p < 0.05. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

reports the baseline regression results. Column (1) excludes control variables and controls only for fixed effects, while Columns (2)–(4) progressively incorporate general firm characteristics, governance structure, and innovation-related variables. The findings indicate that across all specifications, the effect of SCD on corporate CEI is significantly negative at the 1% level, and the coefficient variation is less than 5%, indicating strong robustness. This indicates that SCD significantly reduces corporate CEI, thereby supporting Hypothesis H1.

This finding can be interpreted through the theoretical framework. From the perspective of Institutional Theory, increasingly stringent environmental regulations and digital economy policies incentivize firms to promote SCD and improve environmental performance. From the perspective of the Resource-Based View, digital technologies and data resources reduce information asymmetry and enhance resource allocation efficiency, thereby lowering unnecessary energy consumption. In addition, Dynamic Capability Theory suggests that firms leverage digital tools to strengthen sensing and reconfiguration capabilities, improving energy efficiency and ultimately reducing corporate CEI.

Considering the potential nonlinear impact of digitalization on carbon emissions [51], the quadratic term of SCD is further included. The results in the last column of Table 4 show a negative coefficient for SCD and a positive coefficient for its squared term, indicating a U-shaped relationship. The validity of this relationship is further examined by calculating the inflection point, which is estimated at 0.072. Given that the observed range of SCD is [0.000, 0.236], the inflection point lies within the sample range. This implies that at lower levels of SCD, digitalization reduces corporate CEI; however, beyond the threshold level (0.072), further increases in SCD may lead to a rise in corporate CEI. However, the distributional characteristics of the sample suggest that this nonlinear effect does not alter the overall conclusion. Specifically, the mean and median values of SCD are 0.006 and 0.002, respectively, and the distribution is highly right-skewed (skewness = 4.917). Moreover, the 99th percentile (0.061) remains below the estimated threshold. Therefore, since the majority of observations fall within the decreasing segment of the curve, the overall effect of SCD on corporate CEI remains negative. Accordingly, a linear specification is retained in subsequent analyses to ensure consistency and comparability.

4.4. Robustness Tests

4.4.1. Lag Testing

Given that the potential lagged effects of SCD and the possible bidirectional causality between SCD and corporate CEI, SCD lagged by one and two periods are employed as explanatory variables, denoted as L_SCD and LL_SCD, respectively [52]. The estimates presented in Columns (1) and (2) of

Table 5. The Results of Lag Testing and the IV Method.

Variables	Lag Testing		IV
	(1)	(2)	(3)SCD	(4)CEI
	CEI	CEI	First Stage	Second Stage
L_SCD	−0.536 ***
	(0.110)
LL_SCD		−0.395 ***
		(0.147)
IV			0.000 ***
			(0.000)
SCD				−0.787 ***
				(0.289)
Control variables	Yes	Yes	Yes	Yes
Constant	0.445 ***	0.541 ***	0.001
	(0.052)	(0.061)	(0.003)
Ind	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes
R2	0.919	0.923	0.454	0.068
N	19,714	13,019	12,170	12,170
Kleibergen–Paap rk Wald F statistic			37.941
Kleibergen–Paap rk LM statistic			36.235

Note: *** p < 0.01. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

remain statistically significantly negative, with values of −0.536 and −0.395, respectively, indicating that the baseline results are robust and providing further support for H1.

4.4.2. The IV Method

First, the number of post offices in each city in 1984 is employed as an instrumental variable, representing historical communication infrastructure at the city level. Cities with a higher number of post offices provide more favorable conditions for digital transformation, thereby satisfying the relevance condition. As a communication service infrastructure, post offices are unlikely to directly affect firms’ production efficiency, suggesting that the exogeneity condition is plausibly satisfied. Following prior studies [53], the instrumental variable is constructed as the interaction between the 1984 post office count and the firm’s SCD lagged by two periods.

The first-stage results in Column (3) confirm the relevance of the instrumental variable, as the coefficient is significantly positive. The Kleibergen–Paap rk LM statistic (36.235) rejects the null hypothesis of under-identification at the 1% level, while the rk Wald F statistic (37.941) exceeds the Stock–Yogo critical value (16.38 at the 10% level). The Hansen J statistic fails to reject the null hypothesis of valid overidentifying restrictions. The second-stage regression in Column (4) shows a significantly negative coefficient on SCD, indicating that the baseline results are robust to potential endogeneity concerns.

4.4.3. Placebo Test Analysis

To further rule out potential bias arising from model misspecification or unobservable factors, a placebo test is conducted to examine potential endogeneity concerns [54]. The underlying rationale is as follows: if the significant effect of SCD observed in the baseline regression is driven by omitted variables or other unobservable factors, similar significant results may still emerge after randomly permuting the key explanatory variable.

The results presented in Figure 2 indicate that the coefficients from the random regressions are centered at zero and display an approximately normal distribution, whereas the original coefficient (−0.4) from the baseline regression lies well outside the main distribution. The findings provide evidence that the inverse association between SCD and CEI is unlikely to be explained by model misspecification or omitted variable bias.

4.4.4. Alternative Measures of SCD

Following prior studies [25],

Table 6. The Results of Altering Measures of Variables and Adjusting the Sample Selection.

Variables	CEI	CEI	CEI1	CEI2	CEI	CEI
	(1)	(2)	(3)	(4)	(5)	(6)
SCD			−0.292 ***	−0.134 **	−0.377 ***	−0.336 ***
			(0.076)	(0.059)	(0.108)	(0.101)
SCD1	−0.296 ***
	(0.098)
SCD2		−0.051 ***
		(0.017)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Constant	0.450 ***	0.445 ***	0.592 ***	0.611 ***	0.457 ***	0.461 ***
	(0.051)	(0.051)	(0.036)	(0.103)	(0.052)	(0.053)
Ind	Yes	Yes	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.921	0.921	0.949	0.973	0.951	0.921
N	22,119	22,119	22,119	22,119	11,270	18,329

Note: *** p < 0.01; ** p < 0.05. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

reports robustness tests using alternative measures of SCD. To account for industry heterogeneity, SCD1 is constructed by measuring the deviation of firm-level SCD from the annual industry mean. In addition, SCD2 is defined as the natural logarithm of the total SCD-related word count disclosed in firms’ annual reports plus one. The results in Columns (1) and (2) of Table 6 show that the coefficients remain significantly negative across alternative specifications, indicating that the baseline results are robust and providing further support for H1.

4.4.5. Alternative Measures of CEI

Furthermore, alternative measures of CEI are employed. CEI1 is operationalized as the ratio of carbon emissions to firms’ total operating costs. CEI2 is constructed as a dummy variable [55], taking the value of 1 if a firm operates in high-emission industries (e.g., power generation, petrochemicals, chemicals, building materials, steel, non-ferrous metals, paper, and civil aviation), and 0 otherwise. As shown in Columns (3) and (4), the coefficients on SCD remain significantly negative, providing additional support for H1.

4.4.6. Adjust the Sample Selection

First, alternative sample specifications are considered. Following prior studies [56], observations from 2020 to 2023 are excluded to mitigate potential bias arising from pandemic-related exogenous shocks (Column (5) of Table 6). In addition, given that observations from 2013 to 2015 account for less than 15% of the full sample, the regression is re-estimated using the 2016–2023 subsample [57] (Column (6) of Table 6). The coefficients in these specifications remain statistically significant, providing further support for the robustness of the baseline results and H1.

4.4.7. Other Robustness Tests

First, model specification is further refined. To account for unobserved firm-level characteristics and provincial-level policy shocks, additional interactive fixed effects at the firm–year and province–year levels are included (Columns (1) and (2) of

Table 7. The Results of Other Robustness Tests.

Variables	CEI
	(1)	(2)	(3)	(4)
SCD	−0.495 ***	−6.379 ***	−0.421 ***	−0.421 ***
	(0.189)	(0.534)	(0.123)	(0.101)
Control variables	Yes	Yes	Yes	Yes
Constant	0.660 **	0.101	0.452 ***	0.452 ***
	(0.256)	(0.243)	(0.048)	(0.047)
Ind	No	No	Yes	Yes
Year	Yes	Yes	Yes	Yes
Code	Yes	No	No	No
Province	No	Yes	No	No
R2	0.941	0.156	0.921	0.921
N	21,817	22,119	22,119	22,119

Note: *** p < 0.01; ** p < 0.05. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

, respectively). Second, given that firms may exhibit systematic differences across multiple dimensions, standard errors are further clustered at the city and province levels to better capture intra-group correlation and improve the precision of the estimates (Columns (3) and (4) of Table 7, respectively). These results show that the estimates are robust to alternative model specifications.

Finally, to alleviate concerns regarding potential sample selection bias, 500 bootstrap replications are performed. The results show that the coefficient on SCD is −0.421, with a Z-statistic of −4.12, and the 95% confidence interval [−0.621, −0.221] does not include zero, indicating that the baseline results are highly robust.

5. Additional Analyses

5.1. Mechanism Analysis

To address potential endogeneity concerns associated with traditional three-step mediation models, this study employs a two-step mediation approach to examine Hypotheses H2–H4 [58]:

$$M_{i,t}=\alpha +\beta {SCD}_{i,t}+\gamma {Controls}_{i,t}+Ind+Year+\epsilon$$

(4)

In model (4), $M_{i,t}$ represents mechanism variables: financing constraints, green innovation, and supply chain disruption risk; other variables align with Equation (3).

5.1.1. Alleviating Financing Constraints (KZ)

Drawing on prior studies [59], this study adopts the KZ index to measure corporate financing constraints. SCD has a significantly negative effect on corporate CEI in Column (1) of

Table 8. The Results of the Mechanism Test.

Variables	CEI	KZ	GI	SCDRISK
	(1)	(2)	(3)	(4)
SCD	−0.421 ***	−0.305 ***	0.192 ***	−0.247 ***
	(0.101)	(0.113)	(0.096)	(0.053)
Control variables	Yes	Yes	Yes	Yes
Constant	0.452 ***	0.417 ***	−0.425 ***	0.200 ***
	(0.051)	(0.038)	(0.047)	(0.020)
Ind	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes
R2	0.921	0.727	0.162	0.243
N	22,119	22,119	22,119	22,119
Sobel test z-value		−5.600 ***	−3.667 ***	−6.291 ***

Note: *** p < 0.01. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

. Column (2) further shows that the coefficient of SCD is significantly negative, implying that SCD significantly alleviates corporate financing constraints. Furthermore, as corporate financing constraints decrease, firms exhibit a greater ability to access and effectively utilize financial resources, which ultimately contributes to reducing corporate CEI.

Specifically, in terms of Institutional Theory, improvements in the regulatory environment and digital transformation enhance information disclosure, thereby reducing financing frictions and constraints. In addition, in terms of the Resource-Based View, SCD enhances information transparency and provides key informational resources to stakeholders. This, in turn, leads to a reduction in information asymmetry and a decline in risk evaluation costs, alleviating financing constraints and enabling firms to allocate more funds to carbon reduction activities, thereby ultimately reducing corporate CEI.

The Sobel test further supports the robustness of the results, with a significant coefficient of −0.129. In summary, the financing constraint mechanism linking SCD and corporate CEI is confirmed, and Hypothesis H2 is supported.

5.1.2. Promote Green Innovation (GI)

Building on prior studies [60], green innovation is proxied in this study by firms’ total number of green patent applications filed during the current period, with the natural logarithm taken after adding 1. The total number of green patent applications is calculated as the sum of green invention patent applications and green utility model patent applications. Table 8 shows that SCD has a significantly positive effect on corporate green innovation, with a coefficient of 0.192 in Column (3). In addition, greater corporate green innovation enhances firms’ technological capabilities and resource integration capacity, thereby promoting reductions in corporate CEI.

Specifically, in terms of Institutional Theory, green innovation can be stimulated and strengthened through external regulatory support, including the development and application of R&D technologies and equipment. Through green innovation, firms are able to access emerging technological resources, which helps improve production efficiency and optimize supply chain operations, making them more digitalized, intelligent, and advanced. In terms of the Resource-Based View, this process improves energy efficiency, reduces resource waste, and ultimately promotes carbon reduction.

The Sobel test further supports the robustness of the results, with a significant coefficient of −0.062. In summary, the green innovation mechanism linking SCD and corporate CEI is confirmed, and Hypothesis H3 is supported.

5.1.3. Lower Supply Chain Disruption Risk (SCDRISK)

This study constructs a supply chain disruption risk (SCDRISK) indicator using a word embedding model based on the MD&A section of annual reports of listed companies in China [61,62]. First, Chinese word segmentation is performed using Python’s module, followed by the removal of stop words. Second, based on the characteristics of supply chain disruption risk, the MD&A texts are manually reviewed to extract features and identify 20 seed words related to supply chain disruption risk. Third, the preprocessed corpus is input into a Word2Vec CBOW model, and words semantically similar to the seed terms are identified using cosine similarity. Then, the seed terms are used to generate 200 candidate keywords. Using a cosine similarity threshold of 0.6, 70 core keywords are selected to construct the vocabulary. The SCDRISK indicator is defined as the proportion of keyword occurrences relative to the total number of words. This approach reduces the subjectivity associated with manually constructed dictionaries while ensuring that no important terms are omitted.

As reported in Table 8, SCD has a significantly negative effect on supply chain disruption risk in Column (4), with a coefficient of −0.247, suggesting that greater SCD is linked to a reduced risk of supply chain disruption. Furthermore, the reduction in supply chain disruption risk enhances supply chain stability, improves production continuity, and reduces resource waste caused by unexpected disruptions, thereby lowering corporate CEI per unit of output.

Specifically, in terms of Dynamic Capability Theory, SCD strengthens firms’ sensing and response capabilities through real-time monitoring, enabling them to anticipate potential disruptions and adjust resource allocation in a timely manner. This process enhances supply chain stability, avoids production interruptions and excessive resource consumption, and ultimately reduces corporate CEI.

The Sobel test further supports the robustness of the results, with a significant coefficient of −0.503. In summary, the supply chain disruption risk mechanism linking SCD and corporate CEI is confirmed, and Hypothesis H4 is supported.

5.2. Heterogeneous Analysis

To further investigate the origins of heterogeneity, a unified explanation for the cross-firm, cross-industry, and cross-regional heterogeneity in SCD effects is developed in this study through the integration of Institutional Theory and the Resource-Based View. Specifically, Institutional Theory explains how differences in the external market environment influence firms’ SCD transformation and carbon emission reduction outcomes. Meanwhile, the Resource-Based View emphasizes that differences in firms’ resource endowments determine their capacity to effectively utilize digital technologies. Therefore, the heterogeneous effects observed across firms can be interpreted as the joint outcome of institutional pressures and differences in resource endowments.

5.2.1. State Ownership

Based on differences in corporate property rights, grouped regressions are conducted for non-state-owned enterprises and state-owned enterprises. SOE are assigned a value of 1, while other firms are assigned 0. The coefficient of SCD in Column (1) is significantly negative, whereas it is statistically insignificant in Column (2) of Table 9. This suggests that the effect of SCD on CEI is more pronounced for non-state-owned enterprises.

Table 9. Analysis of Heterogeneity at the Firm Level.

Variables	CEI
	(1)	(2)	(3)	(4)
	SOE = 1	SOE = 0	ES = 1	ES = 0
SCD	−0.152	−0.530 ***	−0.495 ***	−0.300 *
	(0.204)	(0.111)	(0.116)	(0.174)
Control variables	Yes	Yes	Yes	Yes
Constant	0.472 ***	0.460 ***	0.262 **	0.581 ***
	(0.094)	(0.062)	(0.107)	(0.081)
Ind	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes
R2	0.926	0.918	0.927	0.917
N	6123	15,996	11,057	11,056
Suest test p-value			0.002

Note: *** p < 0.01; ** p < 0.05; * p < 0.1. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables. Intergroup coefficient differences are tested using Fisher’s combined test based on 500 bootstrap replications, and this procedure is also applied in Table 10.

Table 9. Analysis of Heterogeneity at the Firm Level.

Variables	CEI
	(1)	(2)	(3)	(4)
	SOE = 1	SOE = 0	ES = 1	ES = 0
SCD	−0.152	−0.530 ***	−0.495 ***	−0.300 *
	(0.204)	(0.111)	(0.116)	(0.174)
Control variables	Yes	Yes	Yes	Yes
Constant	0.472 ***	0.460 ***	0.262 **	0.581 ***
	(0.094)	(0.062)	(0.107)	(0.081)
Ind	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes
R2	0.926	0.918	0.927	0.917
N	6123	15,996	11,057	11,056
Suest test p-value			0.002

Note: *** p < 0.01; ** p < 0.05; * p < 0.1. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables. Intergroup coefficient differences are tested using Fisher’s combined test based on 500 bootstrap replications, and this procedure is also applied in Table 10.

Table 10. Analysis of Heterogeneity at the Industry and Region Level.

Variables	CEI
	(1)	(2)	(3)	(4)	(5)	(6)
	Ht = 1	Ht = 0	HES = 1	HES = 0	DIL = 1	DIL = 0
SCD	−0.373 ***	−0.660 **	−2.423 ***	−0.089 *	−0.505 ***	−0.251 *
	(0.107)	(0.267)	(0.514)	(0.0459)	(0.149)	(0.132)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Constant	0.394 ***	0.616 ***	0.938 ***	0.101 ***	0.419 ***	0.492 ***
	(0.058)	(0.107)	(0.115)	(0.026)	(0.067)	(0.075)
Ind	Yes	Yes	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.933	0.857	0.891	0.819	0.916	0.925
N	16,729	5390	9271	12,848	11,037	11,079
Suest test p-value	0.004		0.000		0.000

Note: *** p < 0.01; ** p < 0.05; * p < 0.1. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

Table 10. Analysis of Heterogeneity at the Industry and Region Level.

Variables	CEI
	(1)	(2)	(3)	(4)	(5)	(6)
	Ht = 1	Ht = 0	HES = 1	HES = 0	DIL = 1	DIL = 0
SCD	−0.373 ***	−0.660 **	−2.423 ***	−0.089 *	−0.505 ***	−0.251 *
	(0.107)	(0.267)	(0.514)	(0.0459)	(0.149)	(0.132)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Constant	0.394 ***	0.616 ***	0.938 ***	0.101 ***	0.419 ***	0.492 ***
	(0.058)	(0.107)	(0.115)	(0.026)	(0.067)	(0.075)
Ind	Yes	Yes	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.933	0.857	0.891	0.819	0.916	0.925
N	16,729	5390	9271	12,848	11,037	11,079
Suest test p-value	0.004		0.000		0.000

Note: *** p < 0.01; ** p < 0.05; * p < 0.1. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

A possible explanation, from the perspective of Institutional Theory, is that state-owned firms benefit from strong governmental support and enjoy a range of policy advantages, including financial subsidies and tax incentives [63], which may weaken their incentives to actively respond to market competition. In addition, state-owned enterprises typically bear more social objectives and obligations [64]. Due to constraints imposed by industrial structure, some traditional state-owned enterprises, such as China National Petroleum Corporation, find it difficult to avoid high-carbon operations; even with the development of SCD, its effect on reducing corporate CEI remains limited. In contrast, in terms of the Resource-Based View, non-state-owned enterprises, with greater strategic flexibility and capability, can more effectively leverage digital resources. They also tend to prioritize supply chain and inventory optimization and exhibit higher responsiveness to market fluctuations, thereby enhancing their capacity to reduce corporate CEI. Meanwhile, non-state-owned enterprises tend to encounter more intense reputational constraints and therefore have stronger incentives to promote low-carbon transformation through SCD, which results in a stronger effect of SCD on reducing corporate CEI.

5.2.2. Enterprise Scale

Taking the mean of corporate asset scale as the threshold, firms are classified into large-scale and small-scale groups based on asset size, where firms with assets equal to or above the sample mean are classified as large-scale enterprises, while those below the mean are classified as small-scale enterprises. The ES variable is set to 1 for large-scale enterprises and 0 otherwise. The SCD coefficients in Columns (3) and (4) of Table 9 are consistently negative and statistically significant, with estimated values of −0.495 and −0.300, respectively. However, the effect of SCD on reducing corporate CEI is weaker for small-scale enterprises than for large-scale enterprises and is less statistically significant, and the two groups exhibit statistically significant differences in their estimated coefficients.

A possible explanation is that large-scale enterprises exhibit a more pronounced effect of SCD in reducing corporate CEI, primarily due to economies of scale. First, from the perspective of the Resource-Based View, large-scale enterprises tend to possess more substantial financial resources, technological capabilities, and organizational reserves [4], which can support the high costs of digital transformation and enable more precise data collection for supply chain optimization and upgrading, thereby reducing energy consumption. Second, as core firms within the supply chain, large-scale enterprises can generate network effects through digital platforms [65], driving upstream and downstream firms to jointly reduce carbon emissions. The magnitude and effectiveness of emission reductions generated through such cross-firm collaboration exceed those achievable by individual small-scale enterprises. Finally, in terms of Institutional Theory, large enterprises are more likely to be subject to regulatory and public scrutiny and thus face stronger regulatory pressure. At the same time, they possess greater capacity to influence industry standards, and digitalization is often accompanied by the coordinated implementation of green standards, including carbon emissions, across the industry, thereby reducing corporate CEI at a broader scale across firms.

5.2.3. Industry Heterogeneity Analysis: High-Tech Industry

Different levels of technological complexity across industries imply variations in production models and innovation capabilities, which may affect the extent to which SCD reduces corporate CEI. Following prior studies [66] and based on the 2012 industry classification guidelines for listed companies issued by the China Securities Regulatory Commission (CSRC), firms with classification codes C25–C29, C31–C32, C34–C41, I63–I65, and M73 are classified as high-tech enterprises, while the remaining firms are classified as non-high-tech enterprises. The HI variable is set to 1 for high-tech enterprises and 0 otherwise. Significantly negative coefficients for SCD are observed in the first two columns of Table 10; however, SCD has a smaller impact on reducing corporate CEI in high-tech industries, and the two groups exhibit statistically significant differences in their estimated coefficients.

In terms of the Resource-Based View, low- and medium-technology industries face more limited access to capabilities and data, and they typically exhibit lower resource efficiency and higher energy intensity of production, thereby leaving greater room for emission reduction. In this context, SCD reduces the risk of supply chain disruptions and more readily generates significant resource-saving effects in low- and medium-technology industries, thereby substantially lowering CEI. In contrast, firms in high-technology industries generally possess stronger innovation capabilities and relatively lower energy consumption, resulting in limited marginal improvement potential; accordingly, the capacity of SCD to reduce corporate CEI is constrained [67]. In addition, from the perspective of Institutional Theory, traditional industries with higher pollution intensity (primarily non-high-tech enterprises) tend to face more stringent environmental regulations, which incentivize firms to adopt digital technologies to reduce carbon emissions.

5.2.4. Industry Heterogeneity Analysis: Environmentally Sensitive Industries

Firms operating in industries with different levels of environmental sensitivity tend to exhibit heterogeneous environmental and social performance outcomes [68] and therefore may influence the effect of SCD on corporate CEI. Following relevant studies [69] and based on the “Guidelines for the Industry Classification of Listed Companies” revised by the China Securities Regulatory Commission and the “Classified Management Directory of Environmental Protection Verification for Listed Companies” issued by the Ministry of Environmental Protection, the following industries are classified as heavily polluting industries: B06, B07, B08, B09, C15, C17, C18, C19, C22, C25, C26, C27, C28, C29, C31, C32, D44, and D45. Accordingly, the sample firms are classified into high and low environmental sensitivity groups depending on whether they are classified as heavily polluting industries. The HES indicator is defined to equal 1 for firms in highly environmentally sensitive industries and 0 otherwise. The SCD coefficients in Columns (3) and (4) of Table 10 are significantly negative; however, the effect of SCD on reducing corporate CEI is significantly stronger for firms in highly environmentally sensitive industries than for those in less sensitive industries. The effect is also more statistically significant, and the two groups exhibit statistically significant differences in their estimated coefficients.

Environmentally sensitive industries are characterized by stricter regulation and higher emission reduction requirements; accordingly, SCD can promote improvements in production arrangements through supply chain optimization and enhanced monitoring capabilities, thereby generating stronger incentives and more pronounced effects in reducing carbon emissions within highly polluting industries [70,71], which is consistent with Institutional Theory. Furthermore, in terms of the Resource-Based View, these firms can obtain greater marginal returns from digitalization, as improvements in resource allocation and production efficiency can substantially reduce corporate CEI. In contrast, industries with low environmental sensitivity tend to exhibit lower corporate CEI and face weaker environmental pressure, and thus, the marginal effect of SCD on reducing corporate CEI is limited.

5.2.5. Regional Heterogeneity Analysis: Digital Infrastructure Level

The evaluation index for regional digital infrastructure in this paper is constructed based on two dimensions [72]: digital infrastructure construction and application. A total of 12 indicators are included: number of websites per 100 enterprises, optical cable density, per capita internet broadband access ports, number of domain names per 100 people, mobile base station density, mobile phone penetration level, computer usage level, proportion of broadband internet users, per capita total telecommunications service output, share of urban unit employees engaged in the information transmission, software, and information technology services industry, cable radio and television household penetration rate, and the average sales of e-commerce enterprises. The index is calculated using the entropy method. By considering the level of digital infrastructure in firms’ respective provinces, this study further conducts a heterogeneity analysis. Grouped regressions are performed based on whether provincial digital infrastructure is above or below its mean value. The DIL variable is set to 1 for firms located in provinces with more developed digital infrastructure and to 0 otherwise. The SCD coefficients in both Columns (5) and (6) of Table 10 are significantly negative; however, the effect of SCD on reducing corporate CEI is stronger for firms located in provinces with more developed digital infrastructure. The effect is also more statistically significant, and the two groups exhibit statistically significant differences in their estimated coefficients.

Digital infrastructure, a core enabler of the digital ecosystem, directly affects information transmission, data processing, and the level of interfirm collaboration. In terms of the Resource-Based View, digital infrastructure constitutes an important external resource that facilitates firms’ access to data and technological support, thereby enhancing the effectiveness of SCD in reducing CEI [73]. In regions with weak digital infrastructure, low data transmission efficiency, and limited digital application scenarios make it difficult for firms to achieve efficient resource integration, thereby weakening the inhibitory effect of SCD on corporate CEI. In contrast, regions with strong digital infrastructure, including well-developed data centers, provide solid support for firms to conduct digital collaboration [74]. Firms can not only obtain upstream and downstream information in real time, but also optimize supply chain configurations through data analysis, reduce energy waste, and thereby effectively lower corporate CEI per unit of output. Finally, in terms of Institutional Theory, regions with more advanced digital infrastructure tend to have more supportive policies, which further promote digital transformation and improvements in carbon-related performance. These findings indicate that environmental factors play an important role in SCD and corporate carbon reduction, while also highlighting the importance of regional coordinated development.

5.3. Analysis of Economic Consequences

As the primary creators of social wealth and major consumers of natural resources, enterprises serve as key agents in coordinating ecological protection and economic development [75]. Therefore, examining the economic implications of improving enterprises’ sustainable carbon emission capacity can support the effective realization of China’s “dual-carbon” objectives. Existing studies indicate that innovation compensation resulting from increased corporate green technology innovation can significantly enhance firm performance [76] and that supply chain optimization and upgrading can also generate efficiency gains within the supply chain [77]. Given the mediating mechanisms discussed above, how does the reduction in corporate CEI induced by SCD affect firm economic performance? First, following prior studies [78], a corporate sustainable development performance indicator (SUS) is constructed by integrating financial and environmental performance indicators. Additionally, return on equity (ROE) and main business revenue (MR) are used as proxies for corporate economic performance. Across Columns (1)–(3) in

Table 11. Analysis of Economic Consequences.

Variables	SUS	ROA	MR
	(1)	(2)	(3)
SCD	0.140 **	0.111 **	0.334 *
	(0.065)	(0.048)	(0.201)
Control variables	Yes	Yes	Yes
Constant	−0.128 ***	−0.160 ***	−20.39 ***
	(0.017)	(0.014)	(1.032)
Ind	Yes	Yes	Yes
Year	Yes	Yes	Yes
R2	0.253	0.397	0.533
N	21,840	21,840	22,105

Note: *** p < 0.01; ** p < 0.05; * p < 0.1. Robust standard errors clustered at the firm level are reported in parentheses throughout all tables.

, SCD coefficients are consistently positive and statistically significant, with estimated coefficients of 0.140, 0.111, and 0.334, respectively. Taken together, SCD facilitates low-carbon transformation by reducing corporate CEI, thereby supporting the stable implementation of long-term corporate strategies and promoting the achievement of sustainable development goals. Furthermore, SCD improves return on equity and main business revenue, effectively alleviates financial pressure, enhances firm performance, and promotes the coordinated development of economic and environmental outcomes.

6. Conclusions and Discussion

6.1. Conclusions

Based on a panel sample of Chinese A-share listed firms from the Shanghai and Shenzhen stock exchanges during 2013–2023, this study employs machine learning techniques to conduct text analysis of corporate annual reports to measure SCD and empirically examine its impact on corporate CEI.

Several conclusions are drawn. First, SCD significantly reduces corporate CEI, and this finding remains robust after robustness checks, including alternative variable specifications and placebo tests. Second, this reduction is achieved by alleviating financing constraints, promoting green innovation, and reducing supply chain disruption risk. Third, the negative effect is more pronounced in non-state-owned enterprises, large firms, non-high-tech industries, environmentally sensitive industries, and firms located in regions with more developed digital infrastructure. Finally, SCD enhances firms’ sustainable development and economic performance, thereby supporting their long-term stable development.

6.2. Theoretical Implications

This study integrates Institutional Theory, the Resource-Based View, and Dynamic Capability Theory into a unified analytical framework to explain how SCD affects corporate CEI, thereby enriching the literature on SCD and carbon reduction. Specifically, Institutional Theory is employed to capture external regulatory and competitive pressures that incentivize firms to advance digital technologies; the Resource-Based View posits that digital resources and data constitute strategic assets that enhance supply chain efficiency and competitive advantage; and Dynamic Capability Theory further elucidates how firms transform accumulated resources into environmental and carbon performance through enhanced capabilities in sensing, decision-making, and resource reconfiguration. This integrated framework has been found to align with previous research that highlights the complementary roles of external institutions and internal capabilities in shaping environmental outcomes.

More importantly, this study extends our existing understanding by demonstrating that the effect of SCD on corporate CEI is not a simple direct effect but rather a multi-layered transmission mechanism encompassing financial optimization, innovation transformation, and risk mitigation. In particular, it is found that digital transformation, including SCD, influences environmental outcomes through the joint improvement of resource allocation efficiency and organizational adaptability, thereby contributing to carbon reduction. Furthermore, the financing channel indicates that digitalization reduces information asymmetry and improves capital allocation efficiency, implying that environmental benefits partly stem from financial optimization rather than solely from technological upgrading. In addition, the green innovation channel suggests that digital resources provide a foundation for innovation-driven carbon reduction, supporting the resource reconfiguration emphasized by the Resource-Based View and Dynamic Capability Theory. Finally, the supply chain risk channel suggests that digitalization improves firms’ disruption management capabilities, thereby stabilizing production processes and reducing resource waste, which introduces a risk governance perspective into the analysis of digitalization and environmental performance. Overall, these findings reposition SCD from a tool for operational efficiency to a comprehensive governance mechanism that facilitates carbon reduction.

Heterogeneity analysis provides deeper theoretical insights into the conditional effects of SCD. Specifically, the effect of SCD on corporate CEI is contingent upon the alignment among institutional pressure, resource endowment, and firms’ dynamic capabilities. In terms of Institutional Theory, it is observed that stronger effects occur in non-state-owned enterprises and environmentally sensitive industries, indicating that firms facing greater market competition, regulatory, or reputational pressures have stronger incentives to adopt digitalization for environmental improvement. In terms of the Resource-Based View, it is evident that large firms and those located in regions with more advanced digital infrastructure possess more abundant digital and financial resources, enabling more effective utilization of SCD. In terms of Dynamic Capability Theory, it is further suggested that firms characterized by strong flexibility and adaptability can be better positioned to translate digital technologies into efficiency gains and carbon reduction outcomes. Taken together, these results suggest that the environmental outcomes associated with digitalization vary depending on the synergy among external institutional conditions, internal resource bases, and firms’ resource reconfiguration capabilities, thereby reframing SCD as a context-dependent mechanism for sustainable development rather than a universally effective tool.

6.3. Practical Implications

The results indicate that SCD should no longer be regarded solely as a tool for efficiency enhancement or cost reduction, but rather as a governance mechanism capable of systematically reducing corporate CEI. Baseline regression results demonstrate that SCD significantly reduces corporate CEI, while mechanism analyses further suggest that this effect is not driven by a single channel but is jointly realized through multiple pathways. This result aligns with prior studies highlighting the multifaceted environmental effects of digital transformation.

At the firm level, this study extends the existing literature by suggesting that low-carbon transformation depends not only on digital technological advancement but also on the coordinated improvement of resource efficiency and operational stability. Firms are therefore encouraged to incorporate SCD into their low-carbon development strategies and promote digital upgrading across key stages, including procurement, production, and logistics. Specifically, the deployment of cloud computing and IoT technologies enables enhanced real-time monitoring and risk identification, thereby reducing disruption risk and limiting additional energy consumption caused by production volatility. Moreover, empirical evidence indicates that green innovation serves as a mediating mechanism, suggesting that firms should strengthen the role of green R&D within SCD implementation, facilitate the integration of digital technologies with operational processes and ultimately reduce corporate CEI.

At the government level, first, sustained efforts are needed to enhance digital infrastructure development, with greater investment in emerging infrastructure, including industrial internet systems and computing facilities, to provide robust support for SCD. Such improvements are expected to facilitate the diffusion of digital technologies in SCD and production scheduling, thereby creating the necessary conditions for carbon reduction. Second, in terms of financing constraints, a favorable financing environment plays a critical role in linking digitalization to low-carbon transformation. Governments should reduce firms’ financing costs through fiscal subsidies or green financial instruments, while inducing firms to increase investment in green R&D activities, thereby achieving emission reduction through technological progress. Third, heterogeneity analysis reveals that the effect of SCD on reducing corporate CEI varies significantly across regions, industries, and firms, highlighting the uneven effectiveness of digitalization policies. Therefore, policy design should adopt a more targeted approach, with greater support for firms in less-developed regions, small enterprises, and firms in highly polluting or low- and medium-technology industries, to alleviate resource and financing constraints and fully realize the environmental benefits of SCD.

6.4. Limitations

First, the sample is composed of A-share listed firms from the Shanghai and Shenzhen exchanges, with those in the financial and real estate sectors excluded. Given that research conclusions may be affected by firm-specific internal factors, the generalizability of the findings to a broader set of firms may be limited. The sample size can be expanded to include a more diverse set of firms, thereby increasing the external validity of the study.

Second, following prior studies, SCD is decomposed into multiple dimensions, and a quantitative indicator is constructed using text analysis and machine learning methods. Although this indicator captures the dynamic evolution of SCD, it primarily reflects managerial disclosure intensity and may not fully represent the actual depth of digital technology integration in physical supply chain processes. Firstly, substantial heterogeneity may exist across firms in terms of information disclosure. Some firms may strategically disclose digitalization-related information for reputation management purposes, whereas others may adopt more conservative disclosure practices, leading to an uneven distribution of measurement errors across the sample. Secondly, such measurement biases may also interfere with mechanism identification. For example, if firms strategically disclose information related to green innovation or financing constraints, the corresponding mechanism pathways may partially reflect disclosure tendencies rather than actual economic behavior. Thirdly, the findings are more applicable to publicly firms with high disclosure standards and stringent regulation, and their applicability may be limited in environments with lower information transparency. Future studies may incorporate firm-level indicators of digital investment—such as IT-related capital expenditures and digital patent outputs—to more accurately capture the environmental impacts of SCD.

Furthermore, the underlying channels through which SCD affects corporate CEI are analyzed. However, additional mechanisms may also exist. Future research could explore mechanisms related to upstream and downstream supply chain partners, as well as information asymmetry [30].

Finally, while robustness tests account for the lagged effects of SCD, the sample period (2013–2023) includes major digital transformation policies and global economic shocks. Linear models may be insufficient to capture non-linear or threshold effects of carbon reduction across different stages of digitalization. Future research could employ panel threshold models to examine whether a threshold effect exists, whereby CEI initially increases before declining, thereby providing more precise guidance for firms and policymakers.