The Impact of Digital Innovation Cooperation on Green Total Factor Productivity of Chinese High-Energy-Consuming Enterprises

Abstract

In this study, a sample of 702 high-energy-consuming A-share listed companies in China from 2007 to 2024 was used to construct digital innovation cooperation (DIC) based on enterprises’ digital technology patent cooperation data. The impact of DIC on firms’ green total factor productivity (GTFP) is empirically investigated. The findings indicate that DIC can considerably enhance the GTFP of high-energy-consuming enterprises (HECEs), and this conclusion is validated using traditional econometric methods and machine learning techniques. Mechanism analysis indicates that DIC promotes GTFP through two main channels: improving firms’ innovation quality and facilitating the green cognitive transformation of top executives. Heterogeneity analysis further highlights that the effect of DIC varies with institutional environment and resource endowment. When local governments place greater emphasis on ecological and environmental issues, the positive effect of DIC becomes more pronounced. Moreover, the positive impact of DIC on GTFP is stronger among energy-type firms. This paper clarifies the internal mechanisms and contextual conditions through which DIC affects GTFP, providing policy implications for promoting the green transformation of HECEs.

1. Introduction

A global consensus has solidified around the need for low-carbon development to combat climate change. However, despite the various measures taken by countries worldwide, the total global carbon emissions continue to rise. A major source of anthropogenic CO₂ emissions is attributed to the combustion of coal, oil, and other fossil fuels [1]. According to the IEA, in 2024, the global energy-related CO₂ emissions reached 37.6 billion tons, setting a historical record. Among these, CO₂ emissions from emerging markets and developing economies rose by 1.5%, exceeding the total reduction in emissions in developed economies [2]. China, as the largest carbon emitter in the world, accounts for approximately 30% of global emissions [3]. Against this backdrop, HECEs (high-energy-consuming enterprises are found in industries such as petroleum and coal processing, chemical manufacturing, non-metallic mineral products, ferrous and non-ferrous metal smelting, and power generation) have attracted increasing attention as major sources of industrial carbon emissions. On the one hand, HECEs are crucial pillars of China’s national economy, historically accounting for around 30% of the total industrial output. Their internal linkages provide essential raw materials for infrastructure and manufacturing, including petrochemical products, steel, and non-ferrous metals, thus significantly supporting industrial development, employment, and economic growth. On the other hand, most HECEs in China are positioned in the mid-to-lower segments of the industrial chain, with limited technological innovation capabilities and low energy efficiency, which not only constrains high-quality enterprise development but also poses challenges to economic sustainability. Therefore, promoting technological innovation and green transformation in HECEs while ensuring economic benefits is critical not only for achieving China’s green transition but also for providing important references for other developing economies in reducing pollution and CO₂ emissions [4].

GTFP represents a fundamental metric in assessing the advancement of corporate green practices [5,6]. Traditional total factor productivity (TFP) does not incorporate pollution emissions, meaning that it does not fully reflect the quality of economic development. In contrast, GTFP addresses this limitation, offering a more accurate assessment of the efficiency between economic outputs and environmental inputs [7], aligning more closely with the practical requirements of green development. For traditional high-energy industries, digital technologies streamline production factors and boost efficiency, significantly enhancing enterprise GTFP. However, due to the complexity and cross-disciplinary nature of digital technologies, individual enterprises often struggle to achieve technological breakthroughs with internal resources alone. Additionally, many HECEs still rely on traditional innovation models, resulting in scattered innovation resources, low cooperation efficiency, and redundant R&D activities, which limit the overall effectiveness of the innovation system. In this context, to accelerate technological innovation and achieve green development, HECEs need to break away from closed innovation models and actively integrate into open and shared DIC, collaborating with individuals, universities, research institutes, complementary enterprises, and governments.

Digital technologies not only promote the convergence of technologies but also reshape inter-organizational innovation, leading to networked innovation activities. DIC refers to the process whereby firms leverage digital technologies as a medium to achieve innovation goals through cross-boundary collaboration, resource sharing, and knowledge co-creation [8]. Compared with traditional cooperation models, DIC effectively mitigates issues such as information asymmetry and resource fragmentation in the innovation process. Data-, algorithm-, and platform-based digital innovation cooperations are reshaping knowledge acquisition and production organization within enterprises [9], driving profound changes in innovation models and production systems. Previous studies indicate that DIC increases the complexity of inter-organizational relationships, making the factors arising from interactions and multi-party comparisons more prominent [10].

As such, this study focuses on the following core questions: at the micro level, how does digitalization affect enterprise greening? Can digital collaboration between HECEs and other innovation entities effectively enhance GTFP? What are the mechanisms at play, and which contextual factors influence the effectiveness of “digital-to-green” transformation? Clarifying these questions can help guide policy and provide a new pathway for the green development of HECEs. This study examines 702 Chinese A-share listed HECEs from 2007 to 2024, using enterprise digital technology patent collaboration data to systematically explore the effect of DIC on enterprise GTFP and its mechanisms. The findings demonstrate that DIC markedly enhances GTFP, mainly through two channels: enhancing enterprise innovation quality and promoting the green cognition transformation of executives. Heterogeneity analysis shows that the impact of DIC on GTFP is closely related to the institutional environment and resource endowment.

The marginal contributions of this research are threefold: (1) in terms of research perspective, it enriches the study of the connection between business innovation organization models and green development, based on open innovation collaboration. This study innovatively incorporates DIC into the GTFP research framework, providing micro-level empirical evidence of DIC enabling green development and supplementing the existing research on factors affecting GTFP. (2) Regarding mechanism analysis, although DIC has attracted wide attention in academia and industry, the process by which it promotes GTFP remains a “black box.” This research reveals the key factors through which DIC affects GTFP via “enhancing innovation quality” and “executive green cognition transformation”, enriching the theoretical understanding of digitalization empowering green development. (3) In terms of methodology, this study supplements traditional empirical analysis with the DDML method. This method helps alleviate endogeneity issues arising from estimation bias, thereby enhancing the reliability of empirical results.

2. Literature Review and Research Hypotheses

2.1. Literature Review

This article is grounded in the practical landscape of DIC and deeply explores the mechanism by which DIC affects the GTFP of HECE. The relevant literature mainly focuses on the following three aspects.

2.1.1. Assessment and Determinants of GTFP

The existing literature on GTFP primarily revolves around measurement approaches and drivers. Regarding the measurement of GTFP, scholars mainly use parametric and non-parametric approaches. Parametric methods are limited in application due to high requirements for model specification; non-parametric methods, particularly DEA, are mainstream, with derived models (CCR, BCC, SBM, EBM) flexibly addressing different analytical needs [11]. Regarding productivity fluctuations over time, Chung et al. (1997) constructed a directional distance function (DDF) that considers both desired outputs and undesirable outputs, as well as the ML index [12]. Building on this, scholars further optimized measurement methods by proposing improved forms such as the SBM-DDF model [13] and the GML index [14], which effectively enhanced the robustness and comparability of the measurement results. Regarding influencing factors, the relevant research emphasizes the examination of internal driving elements and external environmental variables. From an internal perspective, researchers have explored aspects such as production technology [15], financial constraints [16], digital transformation [17], and transaction costs [18]. From an external perspective, factors such as policy support [19], regional economic structure [20], trade openness [21], green credit [22], and environmental regulations [23] have all been confirmed to exert considerable influence on GTFP.

2.1.2. Digital Innovation, Collaborative Innovation, and DIC

Digital innovation refers to the process and outcomes of creating entirely new products, processes, organizations, and business models based on the implementation of digital technologies and their combinations [8]. Such innovative activities typically feature high investment and high risk, making traditional independent innovation models insufficient to support ongoing innovation in enterprises. In this context, collaborative innovation has emerged as a crucial method for companies to attain technological breakthroughs and share risks [24]. Early collaborative innovation was constrained by spatial distance and communication costs [25]; however, the development of digital technologies has effectively overcome these limitations, enabling knowledge spillover and transfer to occur at near-zero marginal costs, significantly reducing the costs and uncertainties associated with innovation. Meanwhile, the advancement of digital technologies has also created novel avenues for inter-organizational innovation collaboration, with enterprise digital innovation continuously evolving towards an open ecological cooperation model [26]. This suggests that studying digital technology innovation from the perspective of enterprise collaboration aligns with the practical needs of business innovation. In fact, digital innovation cooperation empowers enterprises to reshape both their innovation processes and organizational structures [27]. For example, JD.com, as a leading enterprise in China’s e-commerce sector, has optimized its warehouse network planning and inventory management strategies by closely collaborating with Volvo and utilizing big data and machine learning technologies to analyze shared operational data, enhancing the electronic logistics level of the enterprise.

2.1.3. Digital Technology Innovation and Green Development of Firms

Innovations in digital technology play a vital role in fostering eco-friendly growth [28]. In the context of economic transformation, digital technology innovation is crucial for driving green development and shaping the sustainable competitiveness of enterprises. Internally, digital technology integrates directly into production processes, enabling the efficient allocation of production factors [29] and enhancing enterprise competitiveness [30]. Furthermore, the utilization of digital technology facilitates the exchange and dissemination of knowledge and information [31], helping enterprises to monitor and manage their environmental impact more effectively [32], thereby fostering advancements in ESG performance metrics [33]. Externally, through collaboration with supply chain partners, firms can leverage digital technologies to extract deep values hidden in data [34], thereby stimulating innovation in green technologies and business models. Additionally, the increased transparency brought about by digitalization lays the foundation for enterprises to implement comprehensive environmental practices throughout the supply chain [35]. Overall, these studies reveal the strategic significance of digital innovation for enterprises’ green development, providing a solid theoretical foundation for exploring how digital innovation collaboration can promote GTFP.

In summary, the existing literature has several shortcomings: (1) most studies examine the influencing factors of GTFP from a single dimension, lacking an integrated perspective on internal and external factors, particularly a comprehensive assessment of the impact of digital technologies on GTFP from the standpoint of collaborative innovation. (2) Regarding digital innovation cooperation, the existing body of research primarily concentrates on inter-organizational innovative behaviors and network characteristics [36], with relatively limited investigation of how DIC affects the GTFP of HECEs, leaving the relevant transmission paths unclear. (3) Methodologically, most studies remain at the traditional econometric model stage, with few integrating emerging technological approaches such as machine learning and data mining, leading to limited identification capabilities for complex nonlinear relationships and high-dimensional data characteristics, and there is still room for improvement in the robustness of causal inference testing.

2.2. Research Hypotheses

2.2.1. DIC and GTFP

GTFP is an important indicator for measuring the quality of economic growth and the degree of environmental coordination, which can comprehensively reflect enterprises’ efficiency in resource input, technological progress, and pollution reduction [10,37]. Enterprise innovation is the core driving force for GTFP improvement, and the rise in DIC provides a new pathway for this process. Based on Schumpeter’s theory that innovation is the reorganization of factors, DIC can be defined as a new-type innovation mechanism for enterprises to enhance GTFP. By expanding the sources of organizational knowledge, integrating internal and external resources, and relying on the dynamic process of “multi-source input—collaborative transformation—value output”, this mechanism systematically optimizes resource input, effectively improves innovation efficiency, reasonably mitigates environmental externalities, and ultimately drives the sustained growth of enterprises’ GTFP.

In terms of multi-source input, DIC, with its open and interconnected architecture, greatly expands the sources and types of green innovation elements for enterprises. By DIC, enterprises can widely access patent technologies, energy-saving processes, and environmental management experiences from different fields and entities [38]. This connection not only enhances the accessibility of external knowledge resources but also structurally reshapes the way enterprises input resources, shifting from a closed, linear internal accumulation to an open, dynamic network integration, thereby laying a solid foundation for optimizing resource allocation from the source. In the manufacturing industry, engineers can collaborate on virtual platforms using DIC technology, which minimizes the need for physical prototypes during the design process. This approach not only significantly enhances R&D efficiency but also reduces capital investment [39].

In terms of collaborative transformation, the interactive cooperation mechanism constructed by DIC not only strengthens the foundation of trust among enterprises but also directly boosts the productivity of collaborative R&D initiatives and outcome transformation. Specifically, collaborative innovation carried out by enterprises relying on DIC can effectively reduce redundant experiments and path dependence, thereby accelerating the iteration speed of products, services, and processes [40]. Furthermore, the innovation model shaped by DIC is an extension of the dynamic capabilities of enterprises. This capability grants enterprises high flexibility, enabling them to keenly sense changes in the external environment and timely adjust R&D directions, thus continuously optimizing production efficiency and environmental performance.

In terms of value output, DIC reshapes the value creation and realization model of enterprises. At the level of enhancing expected outputs, DIC empowers value chain collaboration, guiding enterprises and partners to innovate around products, services, and business models, thus expanding the value space and redistributing revenues [41]. In terms of mitigating undesirable outputs, the external resource channels constructed by DIC expedite the widespread adoption and implementation of sustainable technologies, helping enterprises to optimize emission management at the production end, thereby improving their environmental performance.

Based on the above analysis, in this study, we propose the following:

H1. 

DIC can significantly enhance the GTFP of HECEs.

2.2.2. DIC, Enterprise Innovation Quality, and GTFP

Adopting a collaborative innovation model helps enterprises improve their innovation absorption capacity and proprietary innovation capabilities [42]. DIC provides a more efficient pathway for this collaboration, significantly enhancing the quality of enterprise innovation. Firstly, DIC built on complementary advantages can integrate innovation resources far exceeding those of a single enterprise, making it easier to generate highly original, integrated, and valuable outcomes [43]. Secondly, DIC not only helps enterprises share R&D costs and reduce R&D risks but its endogenous intellectual property collaboration mechanism can also effectively alleviate the uncertainty of returns caused by knowledge spillovers, thus enhancing enterprises’ motivation to continuously participate in collaboration. The improvement in innovation quality is a core manifestation of the advancement of enterprise innovation capabilities and lays a solid foundation for enterprises to promote green transformation. Research shows that high-quality innovation can drive profound changes in enterprises with respect to energy savings, emission control, and clean production, directly enhancing their environmental performance. Furthermore, the improvement in innovation quality also sends constructive messages to the broader community, attracting the attention of governments, markets, and media organizations. This attention not only brings potential support to enterprises but also creates positive external awareness, urging them to fulfill more environmental responsibilities, thereby promoting the continuous improvement of their environmental performance. Thus, DIC can promote the growth of enterprises’ GTFP by enhancing their innovation quality.

Therefore, in this study, we propose the following:

H2. 

DIC enhances the GTFP of HECEs by improving their innovation quality.

2.2.3. DIC, Executives’ Green Cognition, and GTFP

Executives’ green cognition refers to their perceptions and judgments regarding resource and environmental issues, which are shaped by their personal characteristics such as values, educational background, and career experience [44]. In practice, although environmental management systems (EMSs) such as ISO 14001 [45] have institutionalized the identification and assessment of environmental impacts, executives’ strategic decisions are still significantly influenced by their own cognitive frameworks and values [46]. At the same time, the level of executives’ green cognition is also profoundly affected by the external environment. The existing studies indicate that green transformation can enhance executives’ awareness of green development and reshape their expectations of long-term returns [47]. Relying on DIC, enterprise managers can absorb diverse knowledge, broaden their cognitive horizons, and deepen their understanding and judgment of green innovation trends. Additionally, the value consensus and institutional norms formed within the collaborative network continuously shape the social recognition environment of enterprise managers, prompting them to reinforce their awareness of green responsibilities at the strategic level. Strategic cognition theory posits that enterprises’ behaviors are not merely passive responses to external environments but are driven by managers’ subjective cognitions. Executives with a high level of green cognition can not only strategically endorse sustainable development concepts but also exhibit sufficient patience to overcome short-term uncertainties, actively promoting the application of green technologies and process optimizations [48]. They are also more sensitive to changes in government environmental policies, proactively seeking opportunities through technological innovation and management optimization to enhance resource utilization efficiency comprehensively. These behaviors ultimately manifest directly as GTFP growth.

Accordingly, this research presents the subsequent hypothesis:

H3. 

DIC enhances the GTFP of HECEs by promoting the transformation of executives’ green cognition.

The structure of this research, illustrated by the proposed hypotheses and the subsequent research design, is presented in Figure 1.

3. Methods and Data

3.1. Model Specification

3.1.1. Two-Way Fixed Effects Panel Model

To examine the relationship between the integration of DIC and the GTFP of HECEs, in this study, the following model is constructed:

$${ GTFP}_{itz}={\alpha }_0+{\alpha }_1{DIC}_{itz}+{{\alpha }_{2}X}_{itz}+{\mu }_i+{\nu }_t+{\theta }_z+{\epsilon }_{itz}$$

(1)

where GTFP_itz is the dependent variable, DIC_itz is the independent variable, X_itz is a series of control variables, and ${\epsilon }_{itz}$ is the error term. The parameters ${\mu }_i, { \nu }_t$, and ${\theta }_z$ denote the controls for individual, time, and regional fixed effects, respectively, where the subscripts i, t, and z represent different individuals, time periods, and regions.

3.1.2. Mechanism Analysis Model

To empirically test the two mechanisms proposed above—improvement in enterprise innovation quality and transformation of executives’ green cognition—this research employs the two-step method suggested by Chen et al. (2020) [49], rather than the traditional stepwise regression method for mediation analysis. This approach emphasizes first establishing the effect of DIC on GTFP and then examining its influence on key intermediate variables. Drawing on theoretical arguments and existing empirical evidence regarding the relationship between these intermediate variables and GTFP, the following analysis focuses on testing whether DIC significantly affects enterprise innovation quality and executives’ green cognition. The corresponding empirical model is specified in Equation (2):

$$M_{itz}={\beta }_0+{\beta }_1{DIC}_{itz}+{{\beta }_{2}X}_{itz}+{\mu }_i+{\nu }_t+{\theta }_z+{\epsilon }_{itz}$$

(2)

where $M_{itz}$ denotes the mechanism variable, while the definitions of the rest of the variables are consistent with those in Equation (1).

3.2. Variable Description

3.2.1. Explained Variables

GTFP. Based on the theoretical framework of the environmental production function, the GTFP of enterprises incorporates environmental factors into the accounting system of the traditional production function. This indicator not only covers traditional production factors such as labor and capital but also regards energy consumption as an input factor in the production process, while defining environmental pollution emissions as undesirable outputs. It can comprehensively measure the input–output efficiency between enterprises’ economic development and the environmental system, thereby accurately evaluating the effectiveness of enterprises’ green transformation [50]. Following the methodologies employed in the relevant literature, in this study, enterprise GTFP is measured using the super-efficiency SBM-GML model [51,52]. The measurement system of this model comprises three dimensions: input factors, expected outputs, and undesirable outputs. Specifically, the input side includes three elements, namely labor, capital, and energy; the output side is divided into expected outputs and undesirable outputs. The former is measured by the enterprise’s operating revenue, and the latter is evaluated through the emissions of key pollutants, as detailed in

Table 1. GTFP input–output variables.

Indicator Category	Names	Indicator Specification
Input Indicators	Labor input	Total workforce of the organization
	Capital input	Net fixed assets held by the company
	Energy consumption	Calculated from the industrial electricity consumption of the city where the company is situated, using the proportion of the company’s workforce to the total employment in the urban region.
Output Indicators	Expected output	Operating revenue of the enterprise
	Undesirable output	Converted from the discharge of industrial wastewater (containing chemical oxygen demand and ammonia nitrogen), sulfur dioxide, and dust, based on the proportion of the company’s employees relative to the overall employment in the urban region.

.

3.2.2. Explanatory Variables

DIC. Patent cooperation data has been widely used in innovation collaboration research [53]. Compared to patent citations and inventor information, the cooperation information of patent applicants can more clearly demonstrate the collaboration between enterprises and external entities in the context of DIC. Therefore, this study constructs the DIC of listed companies using patent application data. Specifically, first, this study collects patent application data from publicly listed HECEs from 2007 to 2024 and identifies patents related to digital technology innovation using patent classification numbers and text analysis based on the relevant classification table published by the NIPA. Second, based on the industry codes of HECEs, patents with two or more applicants are filtered. Finally, considering the relative stability of innovation networks, this study follows the approach of Yang and Yuan (2025) [54] by selecting a 5-year time window (t − 4 to t) to construct the enterprise DIC. Within this 5-year window, if a company collaborates on digital patents with other entities, the value is set to 1; otherwise, it is set to 0, thus obtaining the DIC data for HECEs.

3.2.3. Mechanism Variables

Enterprise innovation quality. Drawing from the study of Li et al. (2025) [55], this research constructs an indicator of knowledge breadth to evaluate the innovation quality of companies. A larger value of enterprise innovation quality indicates a greater breadth of knowledge in the patents held by the enterprise, which correlates with higher innovation quality. However, this study does not consider design patents in the calculation of this indicator for the following important reason: design patents cannot reflect the innovation capability of enterprises, and thus do not adequately represent the complexity of knowledge applied in the enterprise’s DIC process.

Executives’ green cognition. The text analysis method has demonstrated its effectiveness in measuring executive cognition and can be utilized for longitudinal data research [56]. Accordingly, this research adopts a textual analysis approach grounded in three key dimensions: green competitive advantage awareness, corporate social responsibility awareness, and external environmental pressure perception. Using a set of manually curated keywords, we quantify the level of green cognition among executives by calculating their occurrence frequency in annual reports spanning 2007 to 2024.

3.2.4. Control Variables

A company’s financial performance characteristics, to some extent, influence its production and operational decisions. Enterprises with higher profitability possess greater financial capacity to invest in long-term technological research and development planning and the green upgrading of production processes. In contrast, companies with a higher cost-to-expense ratio often lack the motivation to enhance long-term productivity and drive green transformation. Therefore, this study draws on the existing research practices [57,58] by selecting several control variables to mitigate the impact of omitted variables, primarily including the enterprise size (Size), enterprise age (Age), debt-to-asset ratio (Lev), return on assets (Roa), proportion of shares held by the top ten shareholders (Top10), enterprise growth capability (Growth), cash flow ratio (Cashflow), market value (TobinQ), and capital intensity (CI). At the macro level, we also control for economic development level (GDP) and industrial structure (IS), as these external contextual factors may systematically influence both corporate financial performance and the companies’ willingness or capacity to engage in green transformation initiatives.

3.3. Sample Selection and Data Sources

Referencing the work by Chen et al. (2025) [58], this study selects a sample of 702 HECEs from the A-share market (industry codes B08, C25, C26, C30, C31, C32, D44, and D45) between 2007 and 2024. Companies designated as ST or *ST were excluded, and all continuous variables underwent a 1% and 99% winsorization process. To reduce the impact of heteroscedasticity, the logarithm of all major continuous variables was taken. The data on energy consumption by industry is sourced from the China Energy Statistical Yearbook, while the data on urban “three wastes” emissions is sourced from the China City Statistical Yearbook. The financial and governance data of the listed companies are derived from the CSMAR database, the CNRDS database, annual reports, corporate social responsibility reports, and company website information. Patent data are mainly derived from the NIPA patent database, while other data are sourced from the China Statistical Yearbook. Further details on the data sources are presented in

Table 2. Definition of main variables.

Variable Type	Variable Name	Variable Definition
Explained variables	Green total factor productivity	Super-SBM-GML
Explanatory variables	Digital innovation cooperation	If an enterprise has a digital innovation cooperation with other entities, the value is coded as 1; otherwise, it is coded as 0
Mechanism variables	Enterprise innovation quality	Assessed by the breadth of knowledge
	Executives’ green cognition	The logarithm of word frequency +1
Control variables	Enterprise size	Natural logarithm of total assets for the year
	Enterprise age	Natural logarithm of company age
	Debt-to-asset ratio	Total liabilities/total assets
	Net return on assets	Net profit over average total assets balance
	Shareholders’ shareholding ratio	Shares held by the largest 10 shareholders over total shares
	Enterprise growth capacity	Operating income for current year over previous year, minus 1
	Cash flow ratio	Net cash flow from operating activities/total assets
	Market value	Market capitalization/total assets
	Capital intensity	Total assets/operating income
	Economic development level	The logarithm of the GDP of each province
	Industrial structure	Value added by the tertiary sector as a percentage of GDP

.

4. Results

4.1. Descriptive Statistics and Correlation Analysis

Figure 2 shows the total number of enterprises and the proportion of DIC in the sample between 2007 and 2024 (the chart was compiled by the authors based on patent data that were manually collected from the official website of the National Intellectual Property Administration of China (https://www.cnipa.gov.cn)). The figure indicates that the proportion of DIC is generally low, with significant fluctuations across different years. Furthermore,

Table 3. Descriptive statistics.

Variable	N	Mean	SD	Max	Min
GTFP	7492	1.0013	0.1285	1.2188	0.7311
DIC	7492	0.1468	0.3540	1.0000	0.0000
Size	7492	22.5229	1.3785	26.4523	19.4058
Age	7492	2.9374	0.3529	3.6636	1.0986
Lev	7492	0.4678	0.2015	0.9347	0.0274
ROA	7492	0.0352	0.0629	0.2552	−0.3750
Top10	7492	0.5725	0.1529	0.9097	0.2040
Growth	7492	0.1526	0.3745	3.8082	−0.6132
Cashflow	7492	0.0573	0.0655	0.2825	−0.2262
TobinQ	7492	1.7843	1.1188	16.6472	0.7888
CI	7492	2.2265	1.8540	19.4809	0.3292
GDP	7492	10.3695	0.8930	11.8278	7.7223
Ris	7492	3.9096	0.1838	4.4279	3.5235

displays the descriptive statistics for the primary variables. To examine the linear correlation and collinearity risk among the variables, this research calculated the Pearson correlation coefficients. The findings suggest that the variables exhibit no significant linear correlation. Additionally, the VIFs are below 2, implying that there are no multicollinearity issues, which can be used for subsequent regression analysis.

4.2. Basic Regression Analysis

Table 4. Baseline regression results.

Variable	(1)	(2)	(3)	(4)
DIC	0.0036 **	0.0042 ***	0.0036 **	0.0042 ***
	(0.0014)	(0.0014)	(0.0014)	(0.0014)
Constant	1.0009 ***	1.0482 ***	1.0009 ***	1.0482 ***
	(0.0002)	(0.0464)	(0.0002)	(0.0465)
Controls	NO	YES	NO	YES
Region FE	NO	NO	YES	YES
Firm/year FE	YES	YES	YES	YES
N	7484	7484	7484	7484
adj. R2	0.981	0.981	0.981	0.981

Note: *** and ** indicate significance levels of 1% and 5%, respectively. The values in parentheses represent standard errors clustered at the firm level, and the same applies hereafter.

presents the outcomes of the main effect test after controlling for time effects and individual fixed effects. In column (1), without adding control variables and not controlling for regional fixed effects, the regression coefficient for DIC is 0.0036 (p < 0.05); in columns (2) and (3), control variables or regional fixed effects are added based on column (1), and the results are comparable to those in column (1). In column (4), both control variables and regional fixed effects are included based on column (1), and the findings suggest that DIC markedly improves the GTFP of HECEs, thereby validating Hypothesis 1. This finding is consistent with previous research [59,60,61] and lays a robust groundwork for subsequent analyses, implying that digital transformation can function as a key driver for a firm’s sustainability transition.

4.3. Robustness Tests

4.3.1. Traditional Methodology

To ensure the robustness of the findings, this study employs the following methods for robustness checks: ① replacement of explanatory variables. The window period for DIC is adjusted from 5 years to 3 years, and the indicators for corporate DIC are re-measured, followed by regression analysis. ② Lagged control variables. Considering that publicly listed companies typically release annual reports a year later, in order to eliminate estimation biases that may arise from data lags, this study adopts a one-period lag for the control variables. ③ Exclusion of special samples. The baseline regression sample spans from 2007 to 2024, during which it may have been affected by market fluctuations, such as the 2008 financial crisis. To diminish the impact of uncontrollable factors, this study re-selects the period from 2009 to 2024 as the regression sample. ④ Adjustment of standard error clustering levels. This study adjusts the clustering level of standard errors to industry for re-examination.

Table 5. Robustness test (I)—traditional method.

Variable	(1)	(2)	(3)	(4)
	Replacement of Explanatory Variables	Lagged Control Variables	Exclusion of Special Samples	Clustering to Industry
DIC		0.0040 ***	0.0047 ***	0.0042 ***
		(0.0015)	(0.0015)	(0.0008)
Window	0.0028 **
	(0.0014)
Constant	1.0494 ***	1.0759 ***	1.0630 ***	1.0482 ***
	(0.0467)	(0.0550)	(0.0548)	(0.0168)
Controls	YES	YES	YES	YES
FE	YES	YES	YES	YES
N	7484	6579	6983	7484
adj. R2	0.981	0.978	0.974	0.981

Note: *** and ** indicate significance levels of 1% and 5%, respectively. The values in parentheses represent standard errors clustered at the firm level.

exhibits the results of the aforementioned series of robustness checks, which indicate that the central finding of this study still holds, namely that the DIC contributes positively to the GTFP of HECEs.

4.3.2. Double/Debiased Machine Learning

Compared to traditional econometric methods, double/debiased machine learning (DDML) presents notable strengths in the processes of model estimation and the identification of relevant variables. On the one hand, DDML improves the estimation of the non-parametric part of the partially linear regression model, thereby relaxing the assumption of linear correlation among variables and allowing for the existence of nonlinear and interactive relationships between variables, effectively avoiding model misspecification issues [62]. On the other hand, DDML controls for high-dimensional variables through various machine learning algorithms and regularization methods, effectively alleviating the curse of dimensionality and improving the estimation accuracy [63].

Table 6. Robustness test (II)—DDML method.

Variable	(1) RF	(2) Lassocv	(3) Gradboost
DIC	0.0044 **	0.0044 ***	0.0050 **
	(0.0020)	(0.0014)	(0.0024)
Constant	0.0003	0.0000	0.0000
	(0.0004)	(0.0001)	(0.0008)
Controls	YES	YES	YES
FE	YES	YES	YES
N	7492	7492	7492

Note: *** and ** indicate significance levels of 1% and 5%, respectively. The values in parentheses represent standard errors clustered at the firm level.

illustrates the outcomes of the robustness verification conducted using three double/debiasing machine learning algorithms: random forest (RF), Lasso regression (lassocv), and gradient boosting (gradboost). As indicated in the table, compared to the baseline regression findings, DIC continues to significantly augment the GTFP of HECE, with a slight increase in the coefficients, further validating the reliability of the benchmark regression.

4.4. Handling Endogeneity

4.4.1. Heckman Two-Step Method

Since the decision of HECEs to engage in digital innovation collaboration is not random but rather a result of the enterprises’ active choices based on their characteristics (such as size, profitability, governance level, etc.), this non-random selection can lead to sample bias, thereby interfering with the research conclusions. Therefore, this study utilizes the Heckman two-step method for correction. Applying the empirical approach of Lee et al. (2023) [64], this analysis employs the lagged values of DIC from two periods prior as instrumental variables for DIC. This selection is informed by two reasons: first, there is a certain correlation between adjacent DIC values, and second, the previous period’s DIC does not directly affect the current GTFP. Specifically, in the first stage, L2.DIC is used as the selection variable for regression, and the IMR is calculated; in the second stage, the IMR is used to correct for sample selection bias. The selection equation in column (2) of

Table 7. Endogeneity test.

Variable	(1)	(2)	(3)	(4)
	DIC	GTFP	DIC	GTFP
DIC		0.0059 ***		0.0045 ***
		(0.0020)		(0.0016)
MIR		−0.0020 **
		(0.0009)
L2.DIC	3.8176 ***
	(0.1299)
Lewbel IV			1.5255 ***
			(0.0546)
Constant	0.0746	1.0927 ***	1.2822 ***	1.0476 ***
	(2.3580)	(0.0623)	(0.3758)	(0.0468)
Controls	YES	YES	YES	YES
FE	YES	YES	YES	YES
N	5883	5855	7484	7484
F test of excluded IV			779.98 [0.0000]
Kleibergen–Paap rk LM			55.98 [0.0000]
Cragg–Donald Wald F			24762.76 {16.38}

Note: *** and ** indicate significance levels of 1% and 5%, respectively. The values in parentheses represent standard errors clustered at the firm level. The values in [ ] represent the p-values of the Kleibergen–Paap rk LM statistic, while the values in { } indicate the critical values of the Stock–Yogo test at the 10% significance level.

indicates that the lagged values of DIC from two periods prior significantly promote the current DIC. The data displayed in column (1) of Table 7 reveal that the IMR is negative at the 5% significance level, indicating the presence of sample selection bias in the benchmark regression results. Nevertheless, DIC still notably promotes GTFP for HECE.

4.4.2. Instrumental Variable Method

When exploring the relationship between DIC and GTFP, the analysis may be affected by endogeneity issues such as reverse causality and omitted variable bias. On the one hand, firms with higher technological capabilities may establish more digital innovation collaboration relationships, thereby increasing the level of DIC, which can lead to reverse causality. On the other hand, there are various determinants of a company’s GTFP. Although this study has accounted for control variables, issues may still remain related to the omitted variables. To mitigate these endogeneity problems, this study constructs an instrumental variable based on the higher-order moments of the endogenous explanatory variable. Specifically, the cube of the deviation between a firm’s DIC and the average DIC of firms in the same industry and year is employed as the instrumental variable (Lewbel IV), and a two-stage least squares (2SLS) estimation is conducted. The rationale for this approach is twofold. First, as the instrument is generated from the higher-order moments of the explanatory variable, it is correlated with DIC, thereby satisfying the relevance condition. Second, Lewbel (1997) [65] demonstrates that, in the presence of heteroskedasticity in the error term, the third-order moment of the explanatory variable is uncorrelated with the error term, which ensures the exogeneity of the instrument. The first-stage regression results reported in Column (3) of Table 7 show that the estimated coefficient of the instrumental variable is significantly positive, indicating a strong correlation between the instrument and digital innovation collaboration. The second-stage results presented in column (4) of Table 7 reveal that the estimated coefficient of DIC remains significantly positive, suggesting that the positive impact of DIC on HECEs’ GTFP persists after accounting for endogeneity issues. According to the Kleibergen–Paap Wald F statistic and the Kleibergen–Paap rk LM test results, the estimations do not suffer from weak instrument or under-identification problems, confirming the validity of the instrumental variable.

5. Further Analysis

5.1. Mechanism Analysis

5.1.1. Enterprise Innovation Quality

DIC enables a smoother flow and greater sharing of innovative elements across firms, which allows HECEs to activate innovation resources and enhance the quality of innovation. As shown in column (1) of

Table 8. Mechanism test.

Variable	(1)	(2)
	Enterprise Innovation Quality	Executives’ Green Cognition
DIC	0.0924 ***	0.1280 *
	(0.0219)	(0.0772)
Constant	2.0476 **	−0.8931
	(0.9897)	(2.4242)
Controls	YES	YES
FE	YES	YES
N	7484	7468
adj. R2	0.639	0.592

Note: ***, **, and * indicate significance levels of 1%, 5%, and 10%, respectively. The values in parentheses represent standard errors clustered at the firm level, and the same applies hereafter.

, DIC improves the quality of enterprise innovation—with particular significance at the 1% level—suggesting that the quality of enterprise innovation is one of the important mechanisms through which DIC drives GTFP, thereby validating Hypothesis 2. Practical evidence shows that cross-organizational collaboration mechanisms also facilitate the division of labor and professional complementarity during the innovation process, helping enterprises to more quickly identify technological bottlenecks and improvement paths, effectively promoting the enhancement of innovation quality. For instance, Huawei Technologies Co., Ltd. has established a DIC network with complementary enterprises and individual developers through business collaborations and open source communities, significantly advancing breakthroughs in key technological areas such as 5G communication and operating systems.

5.1.2. Executives’ Green Perception

In addition to enhancing innovation quality, DIC can also help enterprise managers to transcend geographical and industry boundaries to engage in communication and collaboration with different types of entities. This cross-boundary knowledge exchange and sharing mechanism has, to some extent, facilitated the refinement of executives’ green cognition levels. As shown in column (2) of Table 8, DIC can enhance executives’ green cognition levels, thereby validating Hypothesis 3. The level of executives’ green cognition integrates the concept of green growth into the management process, which not only directly drives enterprises to allocate resources to areas conducive to GTFP enhancement but also assists managers in improving production processes by implementing intelligent and digital production methods, and thus promoting the green transformation of companies.

5.2. Heterogeneity Analysis

5.2.1. Heterogeneity Analysis of Government Environmental Attention

HECEs, due to their characteristics of high pollution, high emissions, and high energy consumption, often face enormous pollution control costs. In the absence of effective external regulation, they generally lack the intrinsic motivation to proactively engage in green transformation. In emerging economies, government attention plays a critical role in shaping the effectiveness of corporate strategies. The existing studies indicate that the degree of local government attention to ecological and environmental issues can significantly influence firms’ GTFP [66]. It should be emphasized that such governmental attention does not constitute specific regulatory instruments or technical standards; rather, it represents a political signal embedded within the environmental governance system. This signal reflects governmental strategic priorities and policy preferences and conveys governance emphases through policy orientation, thereby influencing firms’ strategic decision-making and behavioral choices. To validate this context, this study draws on a comprehensive compilation of Chinese provincial government work reports (The Provincial Government Work Report is a comprehensive document issued annually by provincial-level administrative units in China. Its core content includes a review of the government’s work over the past year, an analysis of the current economic and social development situation, and work objectives and policy arrangements for the upcoming year) covering the period from 2007 to 2024, extracting 53 keywords related to the ecological environment to construct a government environmental attention index. Using the annual sample median, the index is divided into high-attention and low-attention groups for testing. The results in columns (1) and (2) of

Table 9. Heterogeneity analysis.

Variable	(1) High GEA	(2) Low GEA	(3) Energy	(4) Non-Energy
DIC	0.0058 ***	0.0032	0.0050 ***	0.0028
	(0.0019)	(0.0020)	(0.0019)	(0.0021)
Constant	0.9944 ***	1.0177 ***	1.0317 ***	1.0516 ***
	(0.0761)	(0.0580)	(0.0801)	(0.0479)
Controls	YES	YES	YES	YES
FE	YES	YES	YES	YES
N	3630	3558	4518	2964
adj. R2	0.984	0.977	0.981	0.981
Groupwise coefficient differences	−0.007 *		−0.013 **

Note: ***, **, and * indicate significance levels of 1%, 5%, and 10%, respectively. The values in parentheses represent standard errors clustered at the firm level, and the same applies hereafter. The tests for coefficient differences across groups are based on 1000 bootstrap replications.

reveal that when government attention to digital technology is higher, DIC significantly promotes the GTFP of enterprises; however, when government attention to digital technology is lower, there is no significant positive effect. Therefore, local government attention to the environment is an important institutional condition for the effective promotion of firms’ GTFP by DIC.

5.2.2. Heterogeneity Analysis of Enterprise Resource Types

Companies with different types of resource dependencies face varying regulatory pressures and green transformation goals, which affect the enhancement effect of DIC on GTFP. To this end, in this study, the sample is stratified by the Energy Classification and Codes (GB/T 29870-2013) [67], being divided into energy-type enterprises (Energy) and non-energy-type enterprises (Non-Energy) (Industry codes C25, C26, and D44 belong to energy-type enterprises, while codes B08, C30, C31, C32, and D45 belong to non-energy-type enterprises). As shown in columns (3) and (4) of Table 9, DIC dramatically elevates the GTFP of energy-type enterprises, contrasting with its insignificant effect on non-energy-type firms. The distinction potentially originates in two aspects: first, regarding regulatory pressure, energy-type enterprises are core sectors for national energy security and carbon reduction, facing stricter government regulation; thus, they have a stronger willingness to utilize DIC for green transformation. Second, in terms of business attributes, the core business of energy-type enterprises is closely related to energy development, processing, and sales, allowing them to more directly apply the results of digital innovation collaboration to energy-saving and emission-reduction processes; therefore, they achieve significant GTFP enhancement. In contrast, non-energy-type enterprises are less closely associated with energy consumption, requiring more transformation steps for DIC to have an effect, resulting in a less significant enhancement of GTFP.

6. Research Conclusions and Policy Recommendations

6.1. Discussion

Against the backdrop of the increasing dominance of the digital economy, DIC not only drives profound changes in enterprise innovation cooperation models but also opens new pathways for corporate green development. In this study, from the perspective of collaborative innovation, the internal and external resources of enterprises were integrated, and DIC was innovatively incorporated into the GTFP research framework, overcoming the limitations of traditional studies that focus on a single factor. Building on traditional econometric regression, this research further introduces the DDML model to empirically test the impact of DIC on GTFP, providing a deeper understanding of the pathways and contextual factors through which DIC empowers GTFP. This methodological innovation not only broadens the research approach but also deepens our understanding of how DIC influences GTFP, advancing the existing literature. The findings of this study are consistent with those of Jiao et al. (2025) [68], who also showed that DIC can effectively enhance GTFP. However, unlike their research at the city level, this study provides micro-level empirical evidence at the enterprise level, offering a clearer theoretical framework for understanding the relationship between digitalization and green development.

Although this study provides valuable insights, it still has some limitations. First, there are limitations regarding the research sample and data. This research primarily focuses on listed HECEs and constructs DIC based on digital patent cooperation application data. However, due to constraints in data availability and statistical criteria, non-listed companies are not included in the study, which may lead to insufficient sample representativeness. Second, digital patent cooperation cannot fully capture implicit cooperation or non-patent innovation activities (e.g., survey data on informal alliances, joint project announcements analysis), which may introduce certain biases in measuring DIC. Third, while this research focuses on the Chinese context, which has significant policy and practical implications, DIC as a global phenomenon presents further opportunities for exploration regarding cross-national cooperation, knowledge spillovers, and institutional adaptation. To conclude, further research should endeavor to achieve a more comprehensive sampling across diverse enterprise types while introducing multidimensional data sources, such as project cooperation, industry alliances, and data platforms, to achieve a more refined characterization of DIC. Furthermore, expanding the research focus based on international patent cooperation data could construct a theoretical framework for digital-driven green transformation by synthesizing insights from global value chains and international innovation networks, thereby providing more in-depth evidence to elucidate the interactive correlation between DIC and corporate green development.

6.2. Conclusions

In this study, a DIC was constructed for 702 HECEs in China’s A-share market from 2007 to 2024 based on digital patent cooperation application data; the impact of DIC on the GTFP of HECEs was empirically tested, and its transmission mechanisms and contextual factors were examined. Our findings show that DIC markedly enhances the GTFP of HECEs. We employed a combination of traditional econometric methods and machine learning techniques for robustness checks, additionally incorporating the Heckman two-step method and instrumental variable method for endogeneity analysis, with the results indicating that the research conclusions remain valid. Mechanism analysis reveals that DIC reconstructs the innovation model of enterprises, enhancing the GTFP of HECEs through two dimensions: strengthening the quality of innovation and fostering the green cognitive transformation of executives. These results are broadly consistent with the literature on the dynamic capabilities theory of companies. DIC helps firms access external digital knowledge and complementary resources, thereby enhancing their capacity for green development and improving their GTFP. From this perspective, DIC serves as an effective institutional arrangement that supports firms’ dynamic capabilities to reconfigure resources and innovation processes in response to environmental changes.

From the results of the heterogeneity analysis, the effects of DIC vary according to the institution and resource characteristics. Local government environmental attention strengthens the beneficial role of DIC, with the promotion of GTFP being stronger in energy-type and high-tech enterprises. The disparities resulting from governmental environmental attention highlight the critical role of government focus in driving corporate green transformation. Among energy-intensive enterprises, this effect is particularly pronounced, indicating that firms facing greater regulatory pressure possess stronger motivation for green development and are more inclined to pursue green transition through various means, including DIC. Overall, this study not only provides robust empirical evidence on the GTFP of DIC but also offers new theoretical insights and practical guidance for promoting sustainable development in heavy-carbon-emitting industries.

6.3. Policy Recommendations

First, enterprises should proactively roll out DIC to accelerate the formation of technological advantages for GTFP enhancement. According to the data compiled for HECEs, the overall level of DIC in enterprises is low. On the one hand, some enterprises are still inclined to maintain a closed innovation model due to concerns about technology leakage and data security in R&D cooperation. On the other hand, due to the high threshold of digital technology innovation, enterprises also find it difficult to locate ideal cooperation partners in the market. To overcome these issues, enterprises should adjust their open innovation strategies in a timely manner to adapt to emerging innovation cooperation paradigms. First, they should increase investment in digital technology innovation research, actively construct cross-sector cooperation mechanisms, break down industry barriers, and promote the efficient flow of knowledge, technology, and data to build a DIC framework centered on enhancing GTFP. Second, they should enhance the breadth and depth of DIC. This involves not only expanding the scope of cooperation and actively engaging in digital innovation collaboration with diverse entities such as individuals, firms, universities, and government departments but also focusing on collaborative innovation around key issues to improve DIC efficiency.

Second, enterprises should fully leverage the synergistic effects of DIC to enhance innovation quality and elevate executives’ green awareness. This study’s conclusions further indicate that DIC can enhance GTFP by improving the EIQ and EGC. For HECEs, leveraging the open innovation model constructed by DIC can promote resource integration and knowledge sharing, thereby solidifying the foundation for green transformation. Specifically, HECEs should use DIC as a vehicle and an opportunity for strengthening their capacity to absorb external knowledge, optimizing innovation processes, enhancing R&D efficiency, and shifting innovation activities from quantity to quality, thereby increasing the originality and application value of innovation outcomes. Additionally, utilizing the cooperative platform established by DIC, enterprises should focus on building cooperation mechanisms in low-carbon industries, enhancing executives’ understanding of green technologies and sustainable strategies, and accelerating the digital and green transformation of enterprises.

Third, governments should focus on resource supply and institutional design to optimize the external environment for enterprise DIC. The results of this study show that under certain conditions, although some enterprises have performed DIC, their GTFP has not significantly improved. For instance, non-energy-type enterprises face less regulatory pressure, resulting in insufficient motivation to enhance GTFP. In light of this, the government should adopt multiple measures to actively improve support policies and remove obstacles for enterprises to engage in DIC. First, financial support should be increased for enterprises’ green transformation to stimulate participation in DIC and achieve breakthroughs in green technology. Second, platforms should be established for enterprise communication and cooperation to enhance digital innovation capabilities and create opportunities for collaborative innovation.