Effects of Supply Chain Digitization on Different Types of Corporate Green Innovation: Empirical Evidence from Double Machine Learning (DML)

Abstract

Amid global resource shortage and severe climate problems, green innovation has become the key for enterprises to achieve sustainable development, and supply chain digitization brings a new opportunity to enhance the green innovation capability of enterprises. Therefore, this paper empirically investigates the differential effects of supply chain digitization (SCD) on two different green innovation strategies, namely substantive green innovation (SGI) and tactical green innovation (TGI), with 38,548 observations of Chinese listed companies in the 17-year period from 2007 to 2023 using an innovative double machine learning model. It is found that SCD can significantly enhance the substantive and tactical green innovation capabilities of enterprises, and the promotion effect on the former is more obvious. Mechanism analysis shows that SCD promotes substantive green innovation by improving the ESG (Environmental, Social, and Governance) performance of enterprises, and promotes tactical green innovation by improving the management efficiency of supply chain nodes. Heterogeneity analysis shows that SCD promotes green innovation more significantly for high-tech firms, firms with high degree of internal control and low financing constraints. Our paper can be informative in addressing this differential impact of supply chain digitization on different types of corporate green innovation.

1. Introduction

In the face of the impact of new information technology, the digitization of industry and the industry of digitization have become the main trends of the global new round of industrial transformation. Supply chain development has also entered a new stage of integration with digital technologies such as the Internet of Things. To better guide and promote the innovation and development of supply chains, the State Council of China issued the “Guidance on Promoting the Innovation and Application of Supply Chains” in October 2017. Under this guidance, the Ministry of Commerce of China and seven other departments jointly issued the “Notice on Pilot Projects for the Innovation and Application of Supply Chains” on 10 April 2018.

The supply chain innovation and application pilot project has seen the policy evolve and the system become more innovative [1,2,3]. Since the pilot project started in 2018, the policy has been developed to include “subject cooperation process control” governance framework, forming a supply chain reform paradigm with Chinese characteristics. In terms of policy implementation, a complete set of rules and regulations has been established based on the collaboration between government and enterprises, and the regional implementation strategy. This includes the application process, multi-dimensional evaluation, capacity building and performance verification. The cross-departmental expert review system focuses on the maturity of the regional supply chain, the potential for technological innovation and the ability to work together. Ultimately, 55 demonstration cities and 266 benchmark enterprises were selected as the experimental carriers. At the policy execution level, the digital regulatory platform monitors the efficiency of resource utilization, the level of decarbonization and the intensity of collaborative research and development. At the same time, the incentive mechanism is being improved with the help of special support funds. Empirical data shows that the average total logistics cost in the demonstration areas has decreased by 12%, with a 23.5% increase in enterprise inventory turnover rates and the emergence of innovative models such as the management of the entire life cycle of new energy batteries. This pilot experience has promoted the transformation of supply chain innovation from regional piloting to institutional arrangements, particularly in the form of supply chain governance networks formed across regional economic belts, providing enterprises with a scalable green transformation practice.

In the context of global climate change and resource depletion, green innovation is seen as a key way to achieve sustainable development [4] and is particularly important for businesses. By implementing green products and processes, businesses can not only improve their competitive advantage, but also respond to social and environmental responsibility. Green innovation in businesses can be divided into two categories: substantive green innovation and tactical green innovation. Substantive green innovation involves using technological innovation and management techniques to improve environmental performance and achieve sustainability. Tactical green innovation involves businesses taking green innovation measures to comply with government policies, improve their image or respond to regulatory pressure. Business green innovation is not just an internal activity, but a process involving multiple stakeholders. This is because green innovation requires efficient integration of resources and real-time information sharing at every stage, from green procurement of materials to green production, transportation and sales. Any disconnection in this process can affect the achievement of the overall green goal.

Digital transformation not only affects the operational efficiency of enterprises, but also makes an important contribution to green innovation. Digital economy has a direct effect, an indirect effect, a spatial effect, a nonlinear relationship and a policy effect on green innovation, and can indirectly increase the level of green innovation. Supply chain digitization opens up new paths to enhance the green innovation capabilities of enterprises. Digital transformation can increase information transparency and enable real-time data sharing, significantly reducing the coordination and execution costs in the supply chain [5]. Digital technology accelerates the flow of information between suppliers, distributors and customers, reducing external transaction costs. Furthermore, the construction of digital logistics platforms enables supply chain node enterprises to share information in real time, thereby increasing their inventory turnover rate. However, environmental issues are becoming increasingly apparent in the pursuit of efficient operations. In this context, supply chain digitization and environmental governance, which enhance the green performance of supply chain enterprises, have become a key driver of innovation [6].

However, most studies have only verified the direct promotion of green innovation by digital transformation, and there has been no in-depth research into the impact of digitization on the two types of green innovation. Given that China is the largest developing country and the second largest economy in the world, studying this issue in China can provide effective case studies for other developing countries and offer valuable insights to the rest of the world. Therefore, we have chosen Chinese listed companies as our research subjects and compiled relevant data from 2007 to 2023. Through a quasi-natural experimental design and the application of a dual machine learning model, we conducted an empirical analysis to investigate the differential impact of supply chain digitization on the two distinct strategies of substantive and tactical green innovation. Furthermore, we employed a mediation model to examine the channels through which these effects are transmitted, providing valuable insights for the further advancement of enterprise digitization and green innovation.

The marginal contributions of this paper are threefold: (1) we have effectively supplemented and added to the existing literature on supply chain digitization and green innovation by systematically analyzing the heterogeneous impact of supply chain digitization on different types of green innovation, thereby enriching the existing body of research in this area; (2) we have used dual machine learning models to process large amounts of complex enterprise data and accurately identify the differentiated impact of supply chain digitization on the green innovation strategies of enterprises in different regions and of different sizes; and (3) when examining the impact of supply chain digitization on substantial and tactical green innovation, we have considered enterprise ESG and the internal management efficiency of supply chain nodes, thereby breaking down the black box of the impact of supply chain digitization on tactical and substantial green innovation and greatly expanding the theoretical framework of green innovation strategies.

2. Literature Review and Theoretical Analysis

2.1. Literature Review

In the context of global climate change and resource depletion, green innovation is considered an important means of achieving sustainable development. Among these, the role of enterprises in green innovation is particularly important. By implementing green products and process innovations, enterprises can not only improve their competitive advantage, but also actively respond to social and environmental responsibility [4,7]. The antecedents of enterprise green innovation are multifaceted, including internal capabilities [8,9], external institutional pressures, consumer demand and digital empowerment [10]. Clearly, green innovation requires the efficient integration of resources and the real-time sharing of information at every stage, from the green procurement of raw materials to the green production, transportation and sale of products. Any disruption to this process can affect the overall green objective. Therefore, green innovation is not just an internal activity for companies, but a process involving multiple stakeholders. The supply chain network is key [11]. However, existing research has paid little attention to the role of supply chain networks in helping companies to develop green innovation strategies, especially in terms of digitization.

The advent of Industry 4.0 has brought unprecedented opportunities and challenges to supply chain management [12]. As the digital transformation wave continues to penetrate the supply chain sector, supply chain operating models are undergoing profound changes. Digital transformation can enhance information transparency and facilitate real-time data sharing, significantly reducing the coordination and execution costs in supply chains [12]. Empirical analysis of listed companies indicates that digital technology accelerates the flow of information between suppliers, distributors and customers, reducing external transaction costs by 12–15% [13]. Further research has shown that building digital logistics platforms can enable supply chain nodes to share information in real time, increasing their inventory turnover rate by an average of 22% and reducing order delivery cycles by 18%. However, environmental issues are becoming increasingly apparent as companies pursue efficient operations. In this context, supply chain digitization and environmental governance have become key drivers of innovation, enhancing the green performance of supply chain enterprises [6]. Digital transformation not only affects operational efficiency, but also makes a significant contribution to green innovation. In recent years, an increasing number of scholars have been paying close attention to the relationship between digitization and green innovation. For example, some scholars have found that digitization promotes green economic growth through “local effects” and “neighborhood effects”, and in a spatial perspective it exhibits “synchronous resonance” characteristics [14].

Digital economy has a direct effect, an indirect effect, a spatial effect, a nonlinear relationship and a policy effect on green innovation, and can indirectly raise the level of green innovation. From the perspective of financing constraints, scholars have used endogenous growth models to study the impact of the digital economy on corporate innovation and provided empirical evidence [4].

However, although existing research exists on the impact of enterprise digitization on green innovation, there is relatively little literature from the perspective of supply chain digitization, while the identification of causality may have some shortcomings in methodological design. Secondly, fewer studies on green innovation nowadays classify it into two categories—substantive and tactical—while this paper innovatively classifies it into two categories—substantive and tactical—and establishes a study on the relationship between supply chain digitization and the impact of corporate green innovation based on this, which can further enrich the impact of supply chain digitization on the different green innovation strategies of corporations in a differentiated way. Therefore, we selected Chinese listed companies as the research object and organized the relevant data from 2007 to 2023. Through a quasi-natural experimental design and a dual machine learning model, we conducted an empirical analysis to investigate the different impacts of supply chain digitization on two different types of green innovations, substantive and tactical. In addition, we investigate the propagation channels of these impacts using mediation models, which provide valuable insights for further advancing enterprise digitization and green innovation.

2.2. Theoretical Analysis

In the current study, green invention patents were selected as an alternative indicator of SGI and green utility model patents as an alternative indicator of TGI. Numerous studies have shown that green invention patents are generally considered to have a higher technical content and innovativeness and can represent a company’s substantive green innovation [15,16,17]. For example, green invention patents are more likely to be used to represent substantial green technological innovation due to their high technical content and innovative nature in solving environmental problems or improving environmental benefits. In contrast, green utility patents are more likely to be used to represent tactical green technological innovation due to their focus on improving existing technologies to meet environmental standards or regulatory requirements [15,17,18]. These studies provide theoretical support for using green invention patents as a measure of substantive innovation and green utility patents as a measure of tactical green innovation, thereby helping to improve the scientific and accurate nature of the research and providing more targeted policy recommendations for businesses and governments [19].

2.2.1. Heterogeneous Effects of SCD on Tactical and Substantive Green Innovation

Substantial green innovation usually refers to fundamental innovation in products, technologies or processes, with the aim of achieving significant improvements in environmental performance through technological innovation. It involves making fundamental improvements to products, processes or technologies to achieve significant environmental benefits. Tactical green innovation focuses more on how companies can enhance their market competitiveness and sustainability through green strategies and management practices. It is more about meeting external regulatory requirements or improving corporate image than intrinsic environmental motivation.

Supply chain digitization (SCD) is a core driver of Industry 4.0 and is reshaping the underlying logic and practices of enterprise green innovation [20,21]. Digitization not only reconfigures the flow of information and resource allocation in supply chains, but also drives enterprises to transition from end-of-pipe management to source innovation, achieving the substantive integration of environmental and economic benefits. Research indicates that supply chain digitization, by enhancing transparency, agility and traceability throughout the supply chain, can embed environmental risk management in operational decision-making systems, while also catalyzing breakthroughs in green technology development and clean production models [20,21,22]. Furthermore, supply chain digitization optimizes resource allocation efficiency through real-time data collection and analysis, essentially driving the practical implementation of green technology iteration. When enterprises discover the cost savings and environmental benefits of digitization, they are more likely to invest in clean energy equipment or circular economy technologies. Empirical studies have shown that for every 10% increase in supply chain visibility, there is a corresponding 6.8% increase in the number of green patents granted to enterprises, and this correlation is particularly evident in industries with high pollution levels [23]. It is worth noting that the organizational changes brought about by digitization provide a framework for tactical green innovation. Digital transformation requires companies to establish cross-functional digital governance systems, which align with the cross-departmental collaboration required for green innovation [24,25].

Manufacturing enterprises can use digital platforms to access supplier carbon emission data, enabling them to accurately identify high-pollution processes and develop targeted low-carbon replacement technologies. Unlike strategy-based green innovation, which focuses on compliance improvement, substantial innovation requires cross-organizational collaboration. Improved supply chain collaboration enables the formation of technology research alliances and the efficient sharing of green patent pools. This technology collaboration, based on digital trust mechanisms, shortens the development cycle of breakthrough environmental technologies. However, strategy-based innovation, with its short-term focus, struggles to leverage these synergies. Digitalization’s impact is often limited to process automation, failing to transform the core of innovation.

Therefore, the following hypotheses are proposed:

H1a. 

SCD has a significant positive effect on substantive green innovation;

H1b. 

SCD has a significant positive effect on tactical green innovation;

H1c. 

The impact of SCD on substantive green innovation is more pronounced than tactical green innovation.

2.2.2. Mediating Effect of ESG Performance

ESG performance has surfaced as a pivotal metric for assessing corporate sustainability, providing a comprehensive evaluation of corporations’ practices spanning environmental stewardship, social equity and corporate governance through a multi-dimensional indicator system [26]. The enhancement of ESG performance is not merely an isolated corporate initiative but is intrinsically linked to the digital transformation of supply chains. This technological innovation exerts a systemic influence on environmental efficiency optimization, social responsibility fulfillment and governance capacity enhancement by reconfiguring the information and financial flows within the value chain.

SCD is transforming enterprise ESG performance through tech innovations and data-fueled strategies. This transformation not only alters the operational logic of traditional supply chains but also facilitates the systematic achievement of ESG goals through multi-dimensional synergies [27,28]. The deployment of artificial intelligence and other cutting-edge technologies has significantly enhanced the accuracy and transparency of carbon footprint tracking. This technology-driven transparency not only strengthens trust between enterprises and regulators but also steers market preferences toward low-carbon products through consumer behavior, thereby creating a virtuous cycle of positive incentives [28].

ESG performance is crucial in driving significant environmental advancements across three vital sectors. First, SCD enables corporations to enhance their environmental performance by reducing resource wastage and carbon emissions through real-time data monitoring and resource optimization [29]. Second, digital technologies augment supply chain transparency and traceability, thereby empowering firms to better fulfill their social responsibilities, which include ensuring labor rights within the supply chain and mitigating environmental pollution [30]. The heightened sense of social responsibility incentivizes firms to invest in substantive green innovations, leading to the advancement of technologies that integrate social and environmental sustainability. Third, digital tools enhance corporate governance frameworks, streamlining the advancement and implementation of sustainable innovation strategies via data-informed decision-making systems [31]. This enhanced governance capacity enables firms to prioritize long-term environmental goals, thereby fostering substantive green innovations with enduring impact.

Drawing on these insights, we formulate the following hypothesis:

H2. 

SCD fosters significant eco-friendly advancements within corporations through enhancing their ESG scores.

2.2.3. Mediating Effect of the Efficiency of Supply Chain Management

As a cross-industry theoretical framework, the supply chain operation reference model (SCORM) dissects node activities into five core processes: planning, purchasing, production, delivery and returns. SCORM provides enterprises with a standardized tool for process reconstruction and benchmarking [32]. The bullwhip effect highlights the efficiency losses resulting from information delays between nodes and deviations in demand forecasts. This phenomenon, which amplifies demand fluctuations upstream in the supply chain, has spurred SCD to reshape the information flow and collaboration model between traditional supply chain nodes. SCD enhances management effectiveness through the incorporation of technologies like the Internet of Things [33].

SCD has fundamentally altered the information flow and collaboration patterns among traditional supply chain nodes, leading to substantial improvements in management efficiency. The utilization of digital platforms has intensified the degree of supply chain collaboration. For instance, enterprises utilizing cloud-based collaboration systems have experienced a 28% increase in inventory turnover, underscoring the role of digital tools in optimizing resource scheduling [34]. Additionally, the development of supply chain analytics capabilities is contingent upon the digital upgrading of three key element—management, technology, and talent—which directly impacts the market responsiveness and operational cost control of node firms by enhancing supply chain agility.

The optimization of key processes based on supply chain operation reference models can systematically reduce resource consumption and pollution emissions, thereby creating both technical feasibility and economic incentives for innovative practices [34]. Moreover, information dissemination and collaborative decision-making mechanisms between supply chain nodes can accelerate the diffusion of green technologies [34]. Moreover, the dynamic responsiveness of supply chain nodes can alleviate the high cost pressures associated with the initial stages of green innovation. By shortening the innovation cycle through risk-sharing mechanisms, supply chain agility enhances firms’ willingness to make tactical investments in green technologies [34].

Therefore, we propose the following hypothesis:

H3. 

SCD promotes corporate tactical green innovation by improving the efficiency of supply chain management.

Supply chain digitization can significantly promote green innovation in enterprises by accelerating supply chain integration. Specifically, supply chain digitization enhances the responsiveness of enterprises to market demand and promotes the development and promotion of green products by improving inventory turnover and information transparency. This mechanism not only helps enterprises allocate resources more efficiently, but also reduces resource waste and carbon emissions by optimizing supply chain processes, achieving a substantial breakthrough in green innovation. Supply chain digitization enables enterprises to achieve more efficient resource allocation and decision support in the supply chain by improving inventory turnover and information transparency. At the same time, improved information transparency helps enterprises share data with upstream and downstream partners, enhancing synergies and reducing communication costs and information asymmetry. Thus, supply chain digitization can accelerate supply chain integration to promote the development of substantive green technology innovation in enterprises [35].

Therefore, we propose the following hypothesis:

H4. 

SCD can accelerate supply chain integration to promote the development of substantive green technology innovation in enterprises.

The theoretical framework of this study is depicted in Figure 1.

3. Methodology

3.1. Identification Strategies

We harness a supply chain innovation and application pilot initiative as a quasi-natural experiment to gauge the cumulative impact of SCD on corporate green innovation, and employ a double machine learning (DML) method following [35,36]. The model advantages of DML are as follows: first, for high-dimensional control variable problems, DML uses orthogonalization techniques to separate the core variables from the interfering variables and eliminates the interfering effects of the high-dimensional control variables by fitting and combining them with a machine learning model (e.g., Lasso, random forests, or neural networks) to retain the net effect of the target variables, thus avoiding dimensionality catastrophe and improving estimation efficiency. In contrast, traditional linear regression is prone to estimation bias due to covariance under high-dimensional data, while simple variable screening methods (e.g., stepwise regression) may miss important confounders and affect the accuracy of causal identification. Second, in model selection, DML does not rely on a single model, but compares the predictive performance of different algorithms through cross-validation (e.g., based on R² or mean square error), prioritizes models with high predictive accuracy for both the interfering variables and the core variables, and ensures that the models satisfy Naïve Orthogonality to minimize bias. In contrast, traditional structural equation modeling (SEM) or instrumental variable methods (IV) usually rely on strong parametric assumptions, which can lead to serious estimation bias if the model is set up incorrectly. Finally, hyper-parameter tuning is performed independently during the training phase through grid search or Bayesian optimization, with the goal of minimizing the prediction error of the auxiliary model rather than directly intervening in causal estimation to avoid overfitting; key steps include sample partitioning (data chunking for orthogonalization and prediction), iterative validation and final parameter tuning.

The model is structured as follows.

$${SGI}_{it}={\theta }_0{SCD}_{it}+g\left(X_{it}\right)+U_{it} E\left(U_{it}/{SCD}_{it},X_{it}\right)=0$$

(1)

$${TGI}_{it}={\alpha }_0{SCD}_{it}+g\left(X_{it}\right)+V_{it} E\left(V_{it}/{SCD}_{it},X_{it}\right)=0$$

(2)

where SGI and TGI are the dependent variables, that is, substantive and tactical green innovation; i is the enterprise; t is the year; SCD is the independent variable, which takes the value of 1 if firm i joins the demonstration enterprise of supply chain innovation and application in year t, and 0 otherwise; the coefficients θ₀ and α₀ denote the effects of SCD on substantive and tactical green innovation, respectively; $X_{it}$ represents the set of high-dimensional control variables that influence the dependent variables via function $G\left(X_{it}\right)$, the specific form of which remains unknown, and the estimation $G ^ \left(X_{it}\right)$ will be derived through machine learning methods; $U_{it}$ and $V_{it}$ are error terms that comply with the zero-mean assumption.

Considering the limited size of the data [35], the auxiliary regression model is developed to ensure the unbiasedness of the disposal coefficient estimates:

$${SCD}_{it}=m\left(X_{it}\right)+{\epsilon }_{it} E\left({\epsilon }_{it}/X_{it}\right)=0$$

(3)

In this context, $m\left(X_{it}\right)$ denotes a function reliant on dispositional and high-dimensional variables. The estimator, denoted as $m ^ \left(X_{it}\right)$, is achieved through the use of machine learning techniques. The error term, ${\epsilon }_{it}$, adheres to the zero-mean assumption. Utilizing Equations (1)–(3), the unbiased estimates for the coefficients ${\theta ^ }_0$ are derived using a three-step procedure, inspired by the work of [37].

The distinct influence of SCD policies on businesses varies significantly among regions, and we rely on [36] to construct more general interaction models that take into account the effect of heterogeneity in the empirical estimation:

$${SGI}_{it}=g\left({SCD}_{it},X_{it}\right)+U_{it}, E\left(U_{it}/X_{it,}{SCD}_{it}\right)=0$$

(4)

$${TGI}_{it}=g\left({SCD}_{it},X_{it}\right)+V_{it}, E\left(V_{it}/X_{it,}{SCD}_{it}\right)=0$$

(5)

$${SCD}_{it}=m\left(X_{it}\right)+{\epsilon }_{it} E\left({\epsilon }_{it}/X_{it}\right)=0$$

(6)

The specific estimation of the relevant parameters aligns with the partial linear model.

3.2. Variables

3.2.1. Dependent Variable

In this study, the dependent variable is corporate green innovations. Drawing upon the previous research [4,7], we gauge the extent of corporate substantive green innovation (SGI) by tallying the patents awarded for environmentally friendly creations that demonstrated a notable advancement in technology. And we use the count of green utility model patents owned by enterprises to assess the tactical green innovation (TGI).

3.2.2. Independent Variable

In this study, the independent variable is supply chain digitization (SCD). In the present research, SCD serves as the main explanatory factor, and it is defined as a dummy variable [38,39]. The variable is assigned a value of 1 if the enterprise engages in the supply chain innovation and application trial initiative, while it is set to 0 if they do not participate in.

3.2.3. Mediating Variables

Following [40,41], we select corporate ESG performance (ESG) and the efficiency of supply chain management (SCE) as the mediating variables. These variables are measured using the median ESG score and the median of SCE, respectively. ESG scores are typically measured through qualitative metrics, quantitative data and a combination of assessments by Wind [42]. The efficiency of supply chain management is often measured through a combination of operational metrics, sustainability metrics and resilience metrics [43].

3.2.4. Control Variables

Drawing on the latest research on the drivers of corporate green innovation, our study incorporates extra variables that might impact green innovation. These additional elements encompass the debt-to-equity ratio, company scale, asset return, Tobin’s Q, ownership distribution, the proportion of independent directors and the margin of corporate sales profitability. The definitions and metrics for each of these variables are detailed in the subsequent section in

Table 1. Variable definitions and measures.

Variable Type	Variable Name	Variable Symbol	Variable Measurement
Dependent variables	Substantive green innovation	SGI	Number of patents for green inventions filed independently during the year
	Tactical green innovation	TGI	Number of green utility models filed independently during the year
Independent variables	Supply chain digitization	SCD	SCD policy dummy variable
Mediating variables	Corporate ESG performance	ESG	ESG score
	The efficiency of supply chain management	SCE	Ln (365/inventory turnover)
Control variables	The debt-to-equity Ratio	DER	Total liabilities/total assets
	Company scale	Size	Total assets in logarithms
	Asset return	ROA	Net profit/total assets
	Tobin’s Q	Tobinq	Market capitalization/total assets
	Centralization of ownership	TOP10sh	Sum of shareholdings of the top ten shareholders’ shareholders
	Independent director	Ind.Dir	Proportion of independent directors
	Corporate sales Margin	NP	Net sales margin
	Time fixed effect	Year	Year dummy variable
	Firm fixed effect	ID	Firm dummy variable

. From the existing literature, the selection of these control variables is usually based on the potential impacts of corporate finance, governance structure, and market environment on corporate innovation capabilities, and thus has the theoretical basis and empirical support [44,45,46,47,48,49].

3.3. Data and Samples

Our study encompasses data gleaned from Chinese corporations listed between 2007 and 2023. We secure information on green innovations through the National Intellectual Property Office’s (CNIPA) repository. To obtain detailed financial and operational stats at the company level, we tap into the China Securities Market and Accounting Research (CSMAR) and China National Research Data Service (CNRDS) databases. These resources furnish a wealth of financial details about Chinese publicly traded entities. We employ stock ticker identifiers to cross-reference and align firms within different data sets.

To guarantee data integrity and uniformity, we omit companies lacking financial data in any year of the sample. Additional selection requirements for the sample are

(1)

Exclusion of financial companies (e.g., banks, insurance companies) due to their distinct regulatory frameworks;

(2)

Removal of *ST/ST-classified firms to mitigate biases in innovation activity reporting;

(3)

Adjustment for mergers, acquisitions, or name changes to maintain longitudinal continuity;

(4)

Continuous financial variables are winsorized at the 1st and 99th percentiles to address extreme values. After these procedures, the final sample comprises 38,548 firm–year observations.

Table 2. Results of descriptive statistics of variables.

Variables	Obs	Mean	SD	Min	Max.
SCD	38,548	0.0473453	0.2123943	0	1
SGI	38,548	0.9436929	3.544605	0	23
TGI	38,548	0.569327	1.989745	0	14
ESG	38,548	3.989	1.3773	0	8
SCE	38,548	4.465	1.379	−9.28402	12.63707
Size	38,548	22.63473	1.348441	19.9956	26.0627
DAR	38,548	0.4873997	0.19668	0.057563	0.873221
ROA	38,548	0.0473308	0.0729296	−0.272016	0.241733
Ind.Dir	38,548	36.91397	4.999	33.33	57.14
NP	38,548	0.0571155	0.1495529	−0.89866	0.494229
Tobinq	38,548	1.821334	1.139036	0	7.47242
TOP10sh	38,548	55.23507	15.24044	23.3735	90.272

provides an overview of the descriptive statistics for the primary variables. The average score for substantial green innovation (SGI) is 0.944, with a variance of 3.545. On the other hand, the mean for tactical green innovation (TGI) stands at 0.569, accompanied by a standard deviation of 1.990. These results suggest that both types of green innovation are relatively low overall, although substantive green innovation exhibits a slightly higher mean value compared to tactical green innovation. Additionally, the maximum values for SGI and TGI are 23 and 14, respectively, indicating that some firms have potentially invested significantly in green innovation activities. The mean value of the SCD is 0.047, with a standard deviation of 0.212. This suggests that the proportion of pilot firms within the sample is relatively small, suggesting that SCD, as measured by participation in the pilot program, is not widely adopted among the firms in our sample.

4. Results

4.1. Baseline Regression Results

We investigated the result of SCD on corporate green innovation. To achieve this, we employ a semi-parametric approach combining a partially linear model and a random forest algorithm. The partially linear model is selected for its flexibility in capturing nonlinear relationships, while the random forest algorithm is adopted due to its high predictive accuracy. To ensure robustness, an interaction term model is introduced, and both time- and firm-fixed effects are incorporated in the regression analysis. The sample is split into a 1:4 ratio (training vs. testing sets), and detailed results are reported in

Table 3. Results of the baseline regression.

Panel A: Impact of SCD on SGI
Variables	(1)	(2)	(3)	(4)
	Partial Linear Model		Interactive Model
	SGI	SGI	SGI	SGI
SCD	3.388 *** (0.747)	3.282 *** (0.714)	0.771 *** (0.226)	0.794 *** (0.229)
Control variable with one term in the hierarchy	Y	Y	Y	Y
Quadratic term of the control variable	N	Y	N	Y
Time fixed effect	Y	Y	Y	Y
Firm fixed effect	Y	Y	Y	Y
N	38,548	38,548	38,548	38,548
Panel B: Impact of SCD on TGI
Variables	(1)	(2)	(3)	(4)
	Partial Linear Model		Interactive Model
	TGI	TGI	TGI	TGI
SCD	0.967 *** (0.351)	1.007 *** (0.360)	0.285 ** (0.118)	0.297 ** (0.119)
Control variables with one term in the hierarchy	Y	Y	Y	Y
Quadratic term of the control variables	N	Y	N	Y
Time fixed effect	Y	Y	Y	Y
Firm fixed effect	Y	Y	Y	Y
N	38,548	38,548	38,548	38,548

Note: ** and *** represent significant at the 5% and 1% statistical level, respectively. Robust standard errors are in parentheses. Y represents “the factors are controlled”, and N denotes “the factors are not controlled”. Same below.

.

In Panel A and Panel B, Columns (1) and (2) present predictions from the partially linear model, and Columns (3) and (4) present results of the interaction model. Columns (1) and (3) account for primary covariate influences, while Columns (2) and (4) factor in both primary effects and interaction components.

The findings reveal that the coefficients of SCD on both TGI and SGI are significantly positive, which confirms that SCD robustly enhances corporate green innovation. Notably, the effect of SCD on SGI consistently exceeds its impact on TGI, and the difference between the two coefficients is verified by the test of coefficient difference between the groups with the method of SUR estimation. These findings strongly support hypotheses H1a, H1b and H1c, and they are also confirmed by Figure 2, supporting H1c.

In this paper, when exploring the impact of supply chain digitization on substantive green innovation and tactical green innovation, baseline regression shows that enterprise supply chain digitization promotes enterprise substantive green innovation more significantly. The above empirical study is mainly theoretical, and now an economic significance study is conducted to verify the heterogeneity between the two at the actual economic significance level. In this paper, by introducing four economic significance indicators—ES1–ES4 [50,51,52]—the results shown in

Table 4. Comparison of the results of economic significance.

Variables	(1)	(2)	(3)	(4)
	ES1	ES2	ES3	ES4
SGI	0.037	0.166	0.131	0.587
TGI	0.019	0.084	0.036	0.161

indicate that the economic significance measurements of SGI are greater than those of TG. Therefore, SGI is more economically significant in the process of supply chain digitization affecting green innovation of enterprises.

4.2. Robustness Checks

4.2.1. Resetting Machine Learning Models

To combat any possible biases stemming from how the model is set up, we ran a series of tests that were robust to change. We experimented with different data splits, changing the initial 1:4 ratio for 1:2 and then 1:7. We also replaced the random forest method with other machine learning techniques, such as support vector machines, gradient boosting, and neural networks. The results of these adjustments are detailed in Columns (1)–(5) of Table 4, spanning both Panel A and Panel B. The results indicate that altering the sample splitting ratio and resetting the machine learning model do not affect the effect of SCD on TGI and SGI, which validates the robust nature of our baseline regression results.

4.2.2. Tailing Treatment

To more rigorously mitigate the influence of extreme values within the sample regarding the estimation results, we apply a 2% and 98% quantile winsorization to all variables in the sample. We then conduct regression analysis based on this adjusted sample. The regression results are presented in Column (6) of Panel A and Panel B in

Table 5. Results of robustness checks.

Panel A: Impact of SCD on SGI
Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Changing the Split Ratio		Reinventing Machine Learning Models			Tailing Treatment	Substitution of Explanatory Variables
	1:2	1:7	Svm	Gradboost	Nnet	(2, 98)	Substitution Variables
	SGI	SGI	SGI	SGI	SGI	SGI	SGI
SCD	2.655 *** (0.769)	2.224 ** (0.999)	2.571 *** (0.430)	1.127 *** (0.330)	2.904 *** (0.095)	2.947 *** (0.657)	3.947 *** (0.717)
Control variables with one term in the hierarchy	Y	Y	Y	Y	Y	Y	Y
Quadratic term of the control variables	Y	Y	Y	Y	Y	Y	Y
Time fixed effect	Y	Y	Y	Y	Y	Y	Y
Firm fixed effect	Y	Y	Y	Y	Y	Y	Y
N	38,548	38,548	38,548	38,548	38,548	38,548	38,548
Panel B: Impact of SCD on TGI
Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Changing the Split Ratio		Reinventing Machine Learning Models			Tailing Treatment	Substitution of Explanatory Variables
	1:2	1:7	Svm	Gradboost	Nnet	(2, 98)	Substitution Variables
	TGI	TGI	TGI	TGI	TGI	TGI	TGI
SCD	0.428 *** (0.133)	0.433 ** (0.172)	0.768 *** (0.213)	0.455 ** (0.202)	0.685 *** (0.163)	0.989 *** (0.353)	0.826 ** (0.373)
Control variables with one term in the hierarchy	Y	Y	Y	Y	Y	Y	Y
Quadratic term of the control variables	Y	Y	Y	Y	Y	Y	Y
Time fixed effect	Y	Y	Y	Y	Y	Y	Y
Firm fixed effect	Y	Y	Y	Y	Y	Y	Y
N	38,548	38,548	38,548	38,548	38,548	38,548	38,548

Note: ** and *** represent significant at the 5% and 1% statistical level, respectively.

. The findings indicate that the exclusion of sample extremes does not alter the significance of the coefficients of SCD, thereby further confirming the robustness of our baseline regression results.

4.2.3. Substitution of Explanatory Variables

Supply chain digitization is a systematic technological change to supply chain processes, structures and management models. According to the Supply Chain Digitization Management Guide jointly issued by the State Administration of Market Supervision and Administration and the National Standardization Administration in 2022, enterprise supply chain digitization is divided into five dimensions—planning digitization, purchasing digitization, production digitization, sales digitization and logistics digitization—and with the help of text analytics and machine learning methodology, we aim to build a more objective and complete indicator reflecting the degree of supply chain digitization of Chinese enterprises. The purpose is to construct a more objective and complete indicator reflecting the degree of digitization of Chinese enterprises’ supply chain. With reference to the keywords provided, the number of keyword occurrences in the annual report is counted on the basis of word division processing; the final structure of the number of keyword occurrences/total number of words in the annual report is the degree of digital transformation of supply chain [53]. The larger the value, the higher the degree of digital transformation. The regression results are presented in Column (7) of Panel A and Panel B in Table 5. The findings indicate that the exclusion of sample extremes does not alter the significance of the coefficients of SCD, thereby further confirming the robustness of our baseline regression results.

4.2.4. PSM-DML Model

To address the endogeneity issue arising from sample self-selection, this study applies propensity score matching (PSM) to preprocess the sample data, followed by regression analysis on the matched sample. Specifically, we utilize a Logit model, with policy implementation behavior as the dependent variable and a set of control variables as covariates, to conduct 1:1 nearest-neighbor matching within a caliper range of 0.01. Figure 3 presents the kernel probability density plots of the propensity scores before and after matching. Post-matching, the heightened similarity in the curves between the treatment and control groups suggests the shared support condition holds.

By utilizing the matched cohort, we revisited the regression model. The results, as illustrated in

Table 6. Results of PSM-DML model.

Variables	PSM-DML
	1:1 Nearby 0.01 Caliper
	SGI	TGI
SCD	3.295 *** (0.981)	2.036 *** (0.574)
Control variables with one term in the hierarchy	Y	Y
Quadratic term of the control variables	N	Y
Time fixed effect	Y	Y
Firm fixed effect	Y	Y
N	38,548	38,548

Note: *** represents significant at the 1% statistical level.

, reveal that the significance of the coefficients of SCD held steady. This consistency across various models validates the robustness of our estimation findings.

4.2.5. Heckman Two-Stage Treatment Effect Model

In order to mitigate endogeneity concerns stemming from sample selection caused by companies’ non-random conduct, this research utilizes the two-step Heckman model to estimate treatment effects. We utilize the mean of industry dichotomies and years as instrumental variables. The overall situation of the dichotomous industry affected by policies can predict to some extent whether individual firms within the industry will be subject to policy shocks. Given that the overall industry situation cannot directly influence the investment and financing decisions of individual firms, this variable satisfies both the relevance and exclusion restrictions required for instrumental variables. In the first stage of the Heckman regression, industry dichotomies and yearly averages are used as instrumental variables. In the second stage, the inverse Mills ratios (IMRs) computed from the first stage are included as additional adjustment covariates in the baseline model, and the coefficients of the explanatory variables are re-estimated. The statistical estimates are presented in

Table 7. Results of Heckman two-stage treatment effect model.

Variables	(1)	(2)	(3)
	SCD	SGI	TGI
IV	6.847 *** (0.357)
SCD		2.570 *** (0.434)	1.142 *** (0.229)
Control variables with one term in the hierarchy	Y	Y	Y
Quadratic term of the control variables	N	N	Y
Time fixed effect	Y	Y	Y
Firm fixed effect	Y	Y	Y
N	38,548	38,548	38,548

Note: *** represents significant at the 1% statistical level.

.

Column (1) reports the results of the first stage, indicating that the coefficients of the instrumental variables are significantly positive at the 1% statistical level. This finding suggests that the instrumental variables have a statistical meaningful influence on the explanatory variables, confirming the existence of a correlation. At the second phase in Columns (2) and (3), the coefficients of SCD continue to demonstrate significant positivity, suggesting that the findings of baseline regression are unaffected by selection bias-related endogeneity.

4.2.6. Instrumental Variables

To mitigate the potential issue of two-way causality, this study controls for the quadratic terms of the control variables and incorporates as many factors influencing corporate green innovation as possible. However, due to data limitations, omitted variable bias may still be present in the regression model, which could introduce endogeneity concerns. To address this issue, we employ the instrumental variable (IV) method. Specifically [36], we employ the mean of industry dichotomies and years as instrumental variables to build a semi-linear IV model for double machine learning. The grouped mean design captures the heterogeneity of policy shocks through two dimensions: cross-industry and cross-temporal variations. On one hand, it incorporates industry-specific differences in sensitivity to digital policies (e.g., manufacturing vs. services) into the model, avoiding industry-specific biases; on the other hand, year-grouped means control for macroeconomic shocks (e.g., pandemic impacts) that may affect short-term SCD adoption, thereby isolating long-term policy effects. The model is specified as follows.

$${SGI}_{it}={\theta }_0{SCD}_{it}+g\left(X_{it}\right)+U_{it}$$

(7)

$${TGI}_{it}={\theta }_0{SCD}_{it}+g\left(X_{it}\right)+U_{it}$$

(8)

$${IV}_{it}=m\left(X_{it}\right)+{\epsilon }_{it}$$

(9)

where IV_it denotes the instrumental variable for SCD_it and alternative parameter estimation techniques align with the standard regression model.

Table 8. Results of the instrumental variables.

Variables	(1)	(2)
	SGI	TGI
SCD	4.309 *** (0.691)	4.037 *** (1.592)
Control variables with one term in the hierarchy	Y	Y
Quadratic term of the control variables	Y	Y
Time fixed effect	Y	Y
Firm fixed effect	Y	Y
N	38,548	38,548

Note: *** represents significant at the 1% statistical level.

shows that the coefficients of SCD hold steadfastly positive. This suggests that the initial regression results remain stable even when accounting for potential endogeneity issues.

5. Further Studies

5.1. Mediating Mechanism Analysis

We select ESG performance, the efficiency of supply chain management and supply chain integration as mediating variables and design the mediating mechanism model. The main regression for constructing the partial linear model is as follows:

$${ESG}_{it}={\theta }_0{SCD}_{it}+g\left(X_{it}\right)+U_{it} E\left(U_{it}/X_{it,}{SCD}_{it}\right)=0$$

(10)

$${SCE}_{it}={\theta }_0{SCD}_{it}+g\left(X_{it}\right)+V_{it} E\left(V_{it}/X_{it,}{SCD}_{it}\right)=0$$

(11)

$${SCI}_{it}={\theta }_0{SCD}_{it}+g\left(X_{it}\right)+V_{it} E\left(V_{it}/X_{it,}{SCD}_{it}\right)=0$$

(12)

The results of employing partial linear models are presented in

Table 9. Results of the mediating mechanism analysis.

Variables	(1)	(2)	(3)
	ESG	SCE	SCI
SCD	0.362 ** (0.164)	0.668 *** (0.048)	0.258 *** (0.048)
Control variables with one term in the hierarchy	Y	Y	Y
Quadratic term of the control variables	Y	Y	Y
Time fixed effect	Y	Y	Y
Firm fixed effect	Y	Y	Y
N	38,548	38,548	38,548

Note: ** and *** represent significant at the 5% and 1% statistical level, respectively.

. The mediating mechanism models take into account both the key and ancillary control factors, and they apply dual fixed effects for both time and company dimensions. Furthermore, these models apply the random forest technique, splitting the data at a 1:4 ratio. As shown in Column (1) of Table 9, SCD has a substantial positive effect on ESG, which implies that SCD indirectly supports the growth of SGI by improving ESG, thereby confirming the hypothesis H2.

Meanwhile, as shown in Column (2) of Table 9, SCD has a significant promoting influence on the efficiency of supply chain management. Therefore, SCD indirectly improves the capability of TGI by enhancing the efficiency of supply chain management, thereby verifying the hypothesis H3.

Finally, as shown in Column (3) of Table 9, SCD significantly contributes to supply chain integration. Therefore, SCD indirectly enhances the capability of SGI by improving supply chain integration, thus testing hypothesis H4.

5.2. Heterogeneity Analysis

Given that the influence of SCD on corporate green innovation may be contingent upon firm-specific attributes, internal management practices and the extent of financing constraints encountered by firms, we examine the heterogeneity of baseline regression results across three dimensions: (a) the qualification of a business as a high-tech entity, (b) the effectiveness of corporate internal control, and (c) the extent of corporate financial constraints.

5.2.1. Firm Attributes

We divide the sample into high-tech and non-high-tech companies, and then perform individual regression analyses for each group.

Table 10. Results of heterogeneity analysis for firm attributes.

Variables	(1)	(2)	(3)	(4)
	High-Tech Companies	Non-High-Tech Companies	High-Tech Companies	Non-High-Tech Companies
	SGI	SGI	TGI	TGI
SCD	1.446 ** (0.717)	0.813 (0.547)	2.597 * (1.369)	0.435 (0.346)
Control variables with one term in the hierarchy	Y	Y	Y	Y
Quadratic term of the control variables	Y	Y	Y	Y
Time fixed effect	Y	Y	Y	Y
Firm fixed effect	Y	Y	Y	Y
N	13,877	24,671	13,877	24,671

Note: * and ** represent significant at the 10% and 5% statistical level, respectively.

displays the results. The coefficients of SCD in high-tech companies exhibit significantly positive correlations with SGI and TGI. Comparatively, the results for non-high-tech companies lack statistical significance. Therefore, the above findings indicate that there is a stronger impact of SCD on green innovation specifically within high-tech companies. The possible reason is that high-tech firms typically possess a robust technological foundation and strong R&D capabilities, enabling them to more effectively understand, absorb, and apply SCD technologies, and then integrate such technologies into their supply chain management more efficiently, thereby providing substantial support for green innovation [54].

5.2.2. Corporate Internal Control

We divide enterprises into two groups as per the midpoint of the internal control measure: those with high internal control and those with low internal control. We then conduct separate regressions for each group. The results are presented in

Table 11. Results of heterogeneity analysis for corporate internal control.

Variables	(1)	(2)	(3)	(4)
	High Internal Control	Low Internal Control	High Internal Control	Low Internal Control
	SGI	SGI	TGI	TGI
SCD	0.701 *** (0.407)	2.118 (0.84)	0.985 * (0.652)	1.127 (0.507)
Control variables with one term in the hierarchy	Y	Y	Y	Y
Quadratic term of the control variables	Y	Y	Y	Y
Time fixed effect	Y	Y	Y	Y
Firm fixed effect	Y	Y	Y	Y
N	21,838	16,710	21,838	16,710

Note: * and *** represent significant at the 10% and 1% statistical level, respectively.

. For enterprises with high internal control, the effect of SCD on green innovation is highly favorable. However, this effect is not observed for enterprises with low internal control. This divergence may be attributed to superior information communication systems and collaborative platforms within enterprises that have high internal control. With the support of SCD, they can better coordinate procurement, production, sales and other supply chain activities with the objective of green innovation [55]. Consequently, these enterprises can swiftly bring green innovative products to market by adjusting their production plans and supply chain strategies in response to market demand for green products.

5.2.3. Corporate Financial Constraints

We segregate enterprises into two groups based on the median score of the SA index for financial constraints: those exhibiting high constraints and those facing low constraints. We then conduct separate regressions for each group. The results are presented in

Table 12. Results of heterogeneity analysis for corporate financial constraints.

Variables	(1)	(2)	(3)	(4)
	High Financing Constraints	Low Financing Constraints	High Financing Constraints	Low Financing Constraints
	SGI	SGI	TGI	TGI
SCD	2.737 (1.616)	1.191 ** (0.502)	0.865 (0.772)	0.882 * (0.496)
Control variables with one term in the hierarchy	Y	Y	Y	Y
Quadratic term of the control variables	Y	Y	Y	Y
Time fixed effect	Y	Y	Y	Y
Firm fixed effect	Y	Y	Y	Y
N	19,274	19,274	19,274	19,274

Note: * and ** represent significant at the 10% and 5% statistical level, respectively.

. For enterprises experiencing low financial constraints, there is a marked positive impact of SCD on green innovation. However, this effect is not observed for enterprises with high financial constraints. This divergence may be attributed to the differing abilities of enterprises to invest in green innovations. Enterprises with low financial constraints possess the financial capacity to continuously invest in green technological advancements. They can leverage the information advantages brought by SCD to quickly acquire new processes from the market, thereby accelerating the upgrading and application of green innovation technologies.

6. Conclusions and Discussion

6.1. Conclusions

In the context of global resource shortage, green innovation by enterprises can help them achieve sustainable development, and supply chain digitization (SCD), as an emerging tool, can enhance the green innovation capability of enterprises. Using data from Chinese listed companies from 2007 to 2023, this study innovatively applies a dual machine learning model to systematically explore the impact of SCD on corporate green innovation. We find that (1) supply chain digitization can significantly enhance firms’ substantive and tactical green innovation capabilities, which validates H1a and H1b, and supply chain digitization is found to have a stronger impact on substantive than tactical green innovation, which validates H1c; (2) SCD can promote substantive green innovation by improving firms’ ESG performances, which validates H2; (3) SCD can improve firms’ green innovation capabilities by improving the internal management efficiency of node firms, which indirectly promotes tactical green innovation, verifying H3; and (4) the optimization effect of SCD on firms shows significant differences in different contexts, specifically, the promotion effect of SCD on firms’ green innovation is more pronounced in high-tech firms, firms with a higher degree of internal control and firms with a lower degree of financing constraints.

6.2. Discussion

The deepening implications of this study at the theoretical level are reflected in the following: Firstly, by subdividing green innovation into two dimensions—substantive innovation and strategic innovation—the differentiated influence mechanism of SCD on the two is systematically investigated. This classification framework not only makes up for the lack of attention to the heterogeneity of green innovation in the existing literature, but also provides a more refined analytical perspective for subsequent research by revealing the asymmetric effect path of SCD on the two types of innovation. Secondly, ESG performance, the efficiency of supply chain management and supply chain efficiency are innovatively introduced as mediating variables, and a transmission mechanism model of “SCD→green innovation” is constructed. This model not only confirms the direct effect of SCD on green innovation, but also clarifies its indirect path to drive green innovation by improving ESG performance, the efficiency of supply chain management and supply chain efficiency from the perspective of resource allocation theory, breaking through the limitations of traditional single-mediation mechanism research. Furthermore, by revealing the moderating effect of enterprise size, industry attributes and regional characteristics on the enabling effect of SCD, this study identifies the boundary conditions of contextual factors in green innovation, and opens up a new path for the intersection of supply chain management and green innovation. Importantly, this study responds to the call for exploration of “how supply chain learning mechanisms affect green innovation”, and deepens the application of dynamic capability theory in the field of sustainable supply chain through the moderating effect analysis of environmental uncertainty. These findings jointly construct an analytical framework with theoretical depth, which not only fills the research gap of the mechanism of digital technology-driven green innovation, but also provides a theoretical basis for enterprises to differentiate the implementation of green innovation strategies.

The practical implications of this study are as follows: first, we should look for better ways to address tactical green innovation. Second, we should develop differentiated green innovation strategies for tactical green innovation and substantive green innovation. Firms should improve their ESG performance to further promote substantive green innovation and optimize SCE to promote the growth of tactical green innovation. Third, our findings suggest that firms with tight internal controls have a greater advantage in leveraging supply chain digitization for green innovation. Therefore, firms should focus on improving their internal control level and establishing a sound internal control system to better utilize the advantages of SCD and promote the growth of green innovation. Fourth, for high-tech enterprises, SCD is not only an important means to improve productivity and management level, but also an important way to promote green innovation and realize sustainable development. Therefore, high-tech enterprises should actively embrace SCD, strengthen technology R&D and application, and continuously improve their innovation ability and competitiveness. Fifth, China is the world’s largest developing country and the world’s second-largest economy, and choosing the case of China to study this issue can provide other developing countries with effective cases to learn from, and give other countries in the world some experience and inspiration. On this basis, the conclusions of this paper have particularly significant practical implications for relevant practitioners [56], such as business managers, policymakers and supply chain professionals, under carbon reduction plans: by implementing SCD and optimizing internal control systems, enterprises can integrate green innovation strategies more efficiently, thereby reducing carbon emissions while improving operational efficiency [57]. At the same time, differentiated strategies for tactical and substantive green innovation can help enterprises balance short-term costs with long-term sustainability under carbon emission reduction goals, avoid the risk of “greenwashing” and respond to global sustainable development goals [58,59].

The limitations of this study are as follows: first, the sample selection is limited. We used Chinese listed companies as the sample mainly due to the availability of data. Listed companies have larger sizes and more comprehensive management systems, making them more representative. However, our sample does not include small and medium-sized enterprises (SMEs), which is a limitation. In future studies, we will try to include SMEs in the sample to make our conclusions more representative. Second, the indicators selected are limited. We discussed the differentiated impact of supply chain digitization on different types of green innovation. The selection of indicators for different types of green innovation is limited and does not consider other indicators. In the future, we will consider more indicators to measure corporate green innovation to improve the accuracy of our study. Third, the model is limited. The limitations of the dual machine learning model are due to its complexity and sensitivity to model settings. If the orthogonality condition is not met, the estimated results may be biased. In the following research, we will adopt more scientific causal inference methods to accurately identify the causal impact of supply chain digitization on corporate green innovation.