Digitalization Drives the Green Transformation of Agriculture-Related Enterprises: A Case Study of A-Share Agriculture-Related Listed Companies

Abstract

The advent of new digital technologies has catalyzed a disruptive technological revolution, fostering significant industrial changes and advancing the green transformation of the economy and society. This paper investigates the influence of digitization on the green transformation of agribusiness firms, focusing on agriculture-related companies listed on the Shanghai and Shenzhen A-share markets from 2013 to 2021. Employing fixed-effect and mediated-effect models, the study examines the mechanisms through which digitization impacts these enterprises. The findings indicate that the relationship between digitization and green transformation in agribusiness is non-linear; a certain threshold of digitization must be achieved before it positively affects green transformation. The effect of digitization on green transformation varies according to the nature of business ownership, company size, supply chain flexibility, and regional environmental regulations. The study reveals that digitization influences green transformation through several mechanisms that promote economies of scale, technological innovation, and structural adjustments. While economies of scale derived from digitization do not directly support green transformation, they facilitate technological innovation and structural adjustments that enhance green initiatives in agribusiness.

1. Introduction

In today’s modern society, technological progress has greatly increased the accessibility of information to new knowledge, social science can benefit greatly from the findings of natural science, and the environmental governance of enterprises has undergone drastic changes due to technological development [1]. The advent of new digital technologies, including artificial intelligence and blockchain, has initiated significant industrial transformations. These technologies have swiftly penetrated various sectors, driving the digitization of industries and consequently expanding the scale of the digital economy on a macro level. According to the China ICT Institute’s “China Digital Economy Development Research Report (2023)”, China’s digital economy was valued at 50.2 trillion yuan in 2022. Digital industrialization and industrial digitization represent the two pivotal components of this economy, with industrial digitization alone accounting for 80.7% or 41 trillion yuan in 2022. However, the distribution of digital penetration across different sectors varies considerably: 10.5% in primary, 24.0% in secondary, and 44.7% in tertiary industries. This disparity highlights that the digitization of agriculture remains a significant weak point in the broader framework of industrial digitization and digital economic development [2]. Moreover, the 20th Party Congress report emphasizes the need for comprehensive rural revitalization through enhanced integration of urban and rural dynamics. The digitization of agriculture could significantly streamline the agricultural supply chain and reduce the transaction costs associated with the flow of factors, thereby serving as a crucial catalyst for rural revitalization.

Advancing the green transformation of agribusiness presents a viable strategy for the modernization of agriculture [3,4]. As the foundational hue of the sector, green underscores the importance of developing ecological, low-carbon agricultural practices. These practices are critical for bolstering the construction of a robust agricultural nation and furthering the modernization of both agriculture and rural areas. However, green development in China’s agriculture is still nascent, and it encounters numerous challenges. These include the nascent overall level of agricultural green development, marked regional disparities [5,6], a superficial grasp of green development principles [7], rudimentary agricultural production methods [8], inadequate supply and branding of high-quality green agricultural products [9], and an underdeveloped policy incentive mechanism aligned with green development goals [10].

Protecting biodiversity and restoring the resilience of ecosystems requires the joint efforts of market players such as individuals, businesses, and governments [11]. In response, the National Agricultural Green Development Plan for the 14th Five-Year Plan, issued in 2021 by the Ministry of Agriculture and Rural Affairs, along with six other departments, emphasizes a collaborative governance approach. This approach calls for government leadership, market orientation, and active engagement from various stakeholders, including farmers, enterprises, and other social forces, to promote agricultural green development. Agriculture-related enterprises, defined as industrial entities directly involved in the production, processing, or utilization of agricultural, forestry, animal husbandry, and fishery products or those dependent on these resources, play a pivotal role. These enterprises are not only productivity drivers within the agricultural sector but also key players in its green transformation. Therefore, it is essential to examine agricultural green development from the perspective of these enterprises [12].

Numerous studies have affirmed that the digital economy represents a new economic paradigm, offering a viable pathway to foster green agricultural development [13,14,15,16]. From a technical standpoint, the adoption of digital technologies enhances smart agriculture and facilitates the green transformation of agricultural production methods. Structurally, the growth of the digital economy promotes technological innovation in agriculture and the development of productive agricultural services, which, in turn, supports the structural upgrading of the agricultural industry and alters traditional developmental pathways. Economically, the rise of e-commerce platforms and other business models addresses information asymmetry and reduces various costs associated with agricultural products, including inventory and transportation. This reduction in costs accelerates product sales, minimizes resource wastage from production to sales, and boosts the green total factor productivity of agriculture [17].

Despite considerable interest in the digital economy’s effects on green agricultural development, the scope of current research is primarily concentrated on the macro- and meso-level impacts, with less attention given to the micro-level dynamics involving agriculture-related business entities. Methodologically, most empirical studies focus on industry-wide, regional, or single-industry analyses, often neglecting detailed examinations of agriculture-related enterprises [18,19,20]. These enterprises play crucial roles in the digital and green transformation processes of agriculture. Thus, an important question arises: in the dual context of agricultural digitization and green development, what impact does the digital transformation of agriculture-related enterprises have on their green transformation?

Based on the above analysis, the overall question of this paper is proposed, which is to discuss the impact of digital transformation of agriculture-related enterprises on green transformation and the possible channels of impact in the dual context of agricultural digitalization and agricultural green development, using listed agriculture-related companies as the research sample. The specific questions are as follows: first, does the degree of digital transformation of agriculture-related enterprises affect green transformation? If so, what is the geometry of the impact effect, and what are the specific channels of action? Secondly, due to the wide range of listed agriculture-related companies, there are heterogeneous characteristics in terms of enterprise size, enterprise nature, main business characteristics, and supply chain length. Under the influence of these heterogeneous characteristics, is there any heterogeneity in the impact of digital transformation on the green transformation of agriculture-related enterprises represented by listed agriculture-related companies?

2. Theoretical Background

2.1. The Direct Impact of Digital Transformation on the Green Transformation of Agribusinesses

Digital technology, characterized by its disruptive, pervasive, and integrative nature, has rapidly merged with the real economy. This integration significantly alters the mode and efficiency of factor allocation, profoundly impacting the green transformation of the economy and society. The digital economy, emerging alongside the development of digital technologies, sets a new direction for the transformation of traditional enterprises.

The paradigm of the technological economy holds that technological change causes changes in key factors and further causes changes in economic structure and economic and social aspects [21]. Therefore, as a new low-cost factor, data appear in the relative cost structure of production input, attracting enterprises to adopt it and becoming the best choice for innovation and profit, which is conducive to promoting the green transformation of enterprises in many ways. In terms of technical application, digital technology’s permeability plays a crucial role in the digitalization of agriculture, impacting both technological innovation within the field and the application of digital tools in agricultural production and management. Agribusinesses are pivotal in integrating digital technology into agriculture. Chen et al. [22] identified three core characteristics of green agriculture: resource conservation, a people-centered approach, and integrated management. Furthermore, digital transformation enables the precise monitoring of all phases of agricultural cultivation and harvesting, promoting the standardized production of agricultural products. Standardization not only enhances production efficiency and product quality but also minimizes resource waste, contributing significantly to green agriculture [23]. Digital technology supports both preventive measures at the source and end-of-pipe management. Machines, for instance, can replace manual labor in handling high-risk waste, improving both the efficiency and quality of waste treatment and furthering the green transformation agenda. In terms of factor changes, digitalization promotes the “virtualization” and “dematerialization” of production and operational activities in agribusiness. It transforms traditional methods of production and operation, relocates agribusiness operations to the digital realm, and reduces inefficient and redundant physical processes in the value chain. This lessens the reliance on resources within the production chain, decreases resource consumption across production stages, and enhances operational efficiency, all of which are conducive to green transition [24,25]. Finally, in terms of structural changes, the advancement of digital technology has spurred the development of environmentally friendly digital business models, such as the sharing and platform economies. These models not only improve the efficiency of resource allocation and conserve resources but also foster the development of clean industries, which can replace high-energy-consuming and high-emission sectors [26,27]. The growth of these clean and green industries generates greater value-creation opportunities for agriculture-related enterprises and can even propel the green transformation of agriculture through green innovation.

Although digital transformation theoretically facilitates the green transformation of agriculture-related enterprises, the overall digitization level among Chinese enterprises remains low at this stage [28]. This is particularly true for enterprises whose primary business involves agricultural products or those that use agricultural products as raw materials. These businesses also face the dual challenges of agricultural market risks and natural agricultural vulnerabilities. Consequently, the digitization level within the agriculture-related industry is generally insufficient, with many enterprises either not yet initiating digital transformation or only in the preliminary stages of planning. Current research indicates that the initial phases of digital transformation, involving the establishment of substantial computational infrastructure, can lead to increased electricity usage and a consequent negative environmental impact [29]. Additionally, some studies suggest that green transformation can elevate production costs, potentially reducing profit margins and creating a “crowding-out effect” or “compliance cost effect” on investments in green technological innovation [30]. Given their smaller scale, agribusiness enterprises face inherent disadvantages compared to larger industrial firms, confronting double the costs due to both digital and green transformations. In addition, Cirillo et al. [31] used samples from Italian enterprises to prove that the adoption of digital technology is closely related to factors such as enterprise scale and industry attributes. According to the traditional decision-making paradigm, enterprises are profit-seeking. When the adoption of digital technology faces double cost obstacles [31], the adoption level of digital technology will be at a low level. Scuderi et al.’s research on Italian agricultural enterprises also supports this point. The adoption rate of digital technology in agriculture is still very low, because there are cultural and technical constraints [32]. Based on this analysis, this paper proposes the following research hypothesis:

Hypothesis 1 (H1): 

The relationship between digital transformation and green transformation in agriculture-related enterprises is non-linear. A facilitating effect on green transformation is observed only when the digitization level of these enterprises meets or exceeds a certain threshold.

2.2. Channels of Digital Transformation Impacting Green Transformation in Agribusinesses

Building on the environmental Kuznets curve, Grossman and Krueger [33] identified that economic growth influences environmental quality through three distinct pathways: scale, technological, and structural effects. These pathways form a foundational theoretical basis utilized by numerous studies to examine how industrial digitalization contributes to green transformation. Industrial digitalization results from the evolution and interplay of digital transformations across multiple enterprises within an industry. Following the analytical framework of published literature [34], this paper explores how digital transformation in agriculture-related enterprises influences green transformation through the scale effect, technological effect, and structural effect.

2.2.1. Scale Effect

The adoption of digital technology significantly lowers the production cost per unit of product, thereby fostering economies of scale. According to the theory of economies of scale, digitalization’s role in lowering unit production costs enables agribusinesses to achieve economies of scale and elevate output levels. As profit-maximizing entities, these businesses are likely to escalate their investments in production operations following gains from economies of scale. However, while the expansion of output scale and increased investments lead to greater consumption and utilization of energy and other production resources, this can adversely impact the green transformation of agribusinesses [35].

On the one hand, for agriculture-related enterprises that interact directly with consumers or those within downstream industrial chains, transitioning to cloud-based platforms can reduce transaction costs, enabling more targeted consumer engagement and facilitating large-scale personalized customization [36,37]. This technological adoption can expand consumer demand, leading to a “rebound effect” where the production scale is further increased, potentially undermining green transformation efforts. For instance, Xu et al. [38] found that while industrial intelligence can diminish the intensity of pollution emissions among industrial firms, it does not necessarily reduce overall pollution levels due to the concurrent expansion of output. Similarly, Yang et al. [39] observed that the use of industrial robots enhances firms’ productivity and energy efficiency. However, this also leads to an increase in the scale of production and consumption, exacerbating rather than alleviating air pollution. Based on these insights, this paper proposes the following hypothesis:

Hypothesis 2 (H2): 

The digital transformation of agribusiness firms negatively impacts their green transformation due to scale effects.

2.2.2. Technology Effects

Green technology innovation is typically characterized by higher input costs, more complex innovation mechanisms, and greater associated risks and uncertainties compared to other types of innovation. Agriculture-related enterprises, often limited by their scale, tend to have a weak foundation in green technology innovation. Digital transformation can significantly enhance the allocation efficiency of innovation resources and promote green technology development within these enterprises. Digital technology plays a crucial role in reducing transaction costs and alleviating information asymmetry. This enables agriculture-related enterprises to bridge the gap in their knowledge by learning from the green technological innovations of others, effectively compensating for their own knowledge deficiencies. Gaining knowledge spillovers from other enterprises facilitates significant improvements in their green technology innovation levels [40]. Digital technologies also enhance the research and development (R&D) capabilities of agriculture-related enterprises. For instance, certain artificial intelligence technologies can perform high-risk, precision-required experiments, extending the application of digital technologies in fields like green agriculture, energy management, intelligent environmental monitoring, the circular economy, and efforts in energy conservation and emission reduction. This not only stimulates diversified sub-innovations but also promotes the diffusion of green innovations. Moreover, digital transformation addresses the human capital challenges faced by agribusinesses, which vary in degree. Digitalization facilitates overcoming these challenges by leveraging platforms to expand recruitment opportunities and reduce information asymmetry in talent acquisition, as well as enhancing human resource management practices. Research by Agyemang et al. [41] underscores that human capital is a crucial precursor to the level of green technology innovation. Based on this analysis, this paper proposes the following hypothesis:

Hypothesis 3 (H3): 

Digital transformation of agribusinesses facilitates green transformation of agribusinesses through the channel of technological effects.

2.2.3. Structural Effects

As the level of digital transformation of agriculture-related enterprises continues to improve, it will produce structural effects and promote enterprise upgrading. The structural effect of digitalization of agriculture-related enterprises is reflected in two aspects. From within the enterprise, as the level of digitization increases, the stronger its role in promoting green technological innovation; when the level of green technological innovation in agriculture-related enterprises increases, it will improve the internal green total factor productivity, change the traditional way of value creation, achieve green business model innovation, and promote enterprise upgrading. From the outside of the enterprise, as agriculture-related enterprises are associated with one, two, and three industries, they have the industry chain correlation effect. Therefore, the scope of the digital transformation of agriculture-related enterprises is not only in a single agriculture-related enterprise but also has a spillover effect. This improvement in the industry chain’s green transformation contributes to the optimization and upgrading of the industrial structure at a macro level, resulting in profound and transformative changes in the economic structure. These changes, in turn, support the green development of agriculture-related enterprises.

In addition, the analysis also extends to the level of social development, because digital technologies cover all levels of the economy and society, and digitalization is not only about change in the technology-driven economic growth model. Instead, it drives dynamic changes in social culture through interaction with consumer behavior, fostering the formation of a digital civilization [42]. When constructing the analytical framework of the culture tower, Khuc [43] believes that social norms, traditions, and values will affect the decision-making process of individuals and groups; moreover, different cultures are still interrelated. Therefore, digital civilization will affect the digital technology adoption of enterprises and the digital consumption behavior of consumers [42,44]. Digital civilization, through changing consumer behavior, facilitates agribusinesses in enhancing productivity and promoting more sustainable innovation through information sharing, innovation, collaboration, data-driven decision-making, sustainability, and customer orientation. This shapes and changes the development direction of agribusinesses and the entire agricultural industry, enhancing the level of green transformation in the economy and society. Based on this analysis, this paper proposes the following hypothesis:

Hypothesis 4 (H4): 

As digitization levels in agriculture-related enterprises increase, digital transformation will promote their green transformation through structural effects.

3. Model, Method and Data

3.1. Variable Setting

3.1.1. Explained Variables

According to the United Nations Environment Programme (UNEP), a green economy is an economic model that “one that results in improved human well-being and social equity, while significantly reducing environmental risks and ecological scarcities “. In its simplest expression, a green economy can be thought of as one that is low-carbon, resource-efficient, and socially inclusive. Other definitions have been proposed by various stakeholders, including some governments and coalition groups, but they generally describe the same core idea [45]. Green transformation refers to a development model guided by ecological civilization construction, based on a green economy, and guaranteed by green management. From the perspective of industrial development, the Institute of Industrial Economics of China Academy of Social Sciences put forward in 2011 that “green transformation” refers to taking the intensive use of resources and environmental friendliness as the guide, taking green innovation as the core, adhering to the new road of industrialization, and realizing the green and sustainable development of the whole process of industrial production [46]. This concept has been recognized by many documents.

Combined with the definitions of UNEP, the Institute of Industrial Economics CASS, this paper defines the green transformation of agriculture-related enterprises as a sustainable development process in which agriculture-related enterprises adopt more environmentally friendly agricultural production technologies and production methods, thereby reducing the use of chemical pesticides and fertilizers or promoting the effective utilization and recycling of resources so as to reduce pollution emissions and greenhouse gas emissions and achieve a balance between economic benefits, social benefits, and environmental benefits. In this analysis, the green transformation of enterprises is assessed based on two key dimensions: the reduction of pollution emissions and the reduction of carbon emissions. Although historically considered distinct issues, recent studies, including Feng et al. [47], have demonstrated a close relationship between pollution emissions and carbon emissions. Moreover, these studies suggest that promoting the synergy between pollution and carbon reduction can enhance the consistency of environmental policies, decrease the cost per unit of policy implementation, and produce a synergistic effect. The “Implementation Programme for Synergy in Pollution Reduction and Carbon Emission Reduction” initiated by the Ministry of Ecology and Environment in 2002 also supports this approach, indicating that the synergistic promotion of pollution and carbon reduction is a necessary strategy for the comprehensive green transformation of economic and social development in China’s new development stage. Given this backdrop, the reduction of pollution and carbon emissions constitutes the core focus of green transformation for agriculture-related enterprises. The subsequent sections will detail the specific measures employed to evaluate these two variables.

This paper adopts the chemical oxygen demand (COD) as a proxy for pollution emissions, informed by published literature [48,49]. In the agricultural sector, COD represents a significant proportion of pollution sources. For instance, results from the Second National Census of Pollution Sources in 2020 indicated that agricultural sources emitted 10,671,300 tonnes of COD, making it the predominant contributor to agricultural pollution. In this study, COD measurements are expressed in tonnes and are logarithmically transformed for regression analysis. To enhance the robustness of our findings, this paper also incorporates an alternative measurement approach from Qi et al. [50]. This approach calculates pollution emission equivalents by combining data on both water and air pollution emissions, serving as an additional proxy indicator for corporate pollution emissions.

Drawing on Li et al. [51], carbon emissions data were meticulously gathered from the social responsibility, sustainability, and environmental reports published annually by various enterprises. These reports provided the basis for calculating Scope 1 and Scope 2 emissions, in line with the “Guidelines on Methods of Accounting and Reporting of Greenhouse Gas Emissions of Enterprises” issued by the National Development and Reform Commission (NDRC). Specifically, Scope 1 emissions were derived from the direct combustion activities of agro-related enterprises, while Scope 2 emissions pertained to greenhouse gases from energy purchased by these enterprises. For enterprises reporting both Scope 1 and Scope 2 emissions, these figures were aggregated to determine the total carbon emissions at the enterprise level.

3.1.2. Explanatory Variable

The core explanatory variable of this study is the digital transformation of agriculture-related enterprises. Theoretically, Gong and Ribiere [52] and Vial [53] all systematically reviewed the existing literature and defined and interpreted the definition of digital transformation. However, their research consistently reflects that there is no unified definition of the digital transformation of enterprises. As we pointed out earlier, agriculture-related enterprises are different from other enterprises. Because of the wide range of industries and the small scale of agriculture-related enterprises, agriculture-related enterprises are not the focus of digital technology innovation, but their focus is more on the application of digital technology that promotes the development of new business models. Digital transformation, in this context, refers to employing advanced information technology and digital tools to enhance agricultural production, operations, and management. This process aims to boost efficiency, reduce costs, increase profitability, and foster modernization and sustainable development in agriculture. How to measure digital transformation is also a difficult problem. Yang et al. [54] offer a framework for understanding enterprise digital transformation, distinguishing between “underlying technology” and “digital technology application” based on how digital technologies are utilized in various enterprise scenarios. They analyzed the annual reports of listed companies using a web crawler to extract keywords associated with digitization. The frequency of these keywords serves as a metric of digitalization.

Digital transformation must rely on the development of a new generation of digital technologies; that is, digital technology is the core of transformation [52,53]. Based on the basic framework of the existing literature [54], this paper used the word frequency related to digital technology and digital transformation in the annual reports of companies to measure it. The idea behind this method is as follows: firstly, determine the keywords of enterprise digital transformation through the existing academic literature, important policy documents, and research reports and form a characteristic thesaurus. Among them, the keywords of enterprise digital transformation are divided into two dimensions: “underlying technology” and “digital technology application”(Note: the words of artificial intelligence technology include: artificial intelligence, business intelligence, image understanding, investment decision support system, intelligent data analysis, intelligent robot, machine learning, deep learning, semantic search, biometric technology, face recognition, voice recognition, identity verification, automatic driving, natural language processing and other characteristic words; Blockchain technologies include: digital currency, smart contract, distributed computing, decentralization, bitcoin, alliance chain, differential privacy technology, and consensus mechanism; Cloud computing technologies include: memory computing, cloud computing, stream computing, graph computing, Internet of Things, multi-party secure computing, brain-like computing, green computing, cognitive computing, converged architecture, billion-level concurrency, EB storage, and cyber-physical systems; Big data analysis technologies include: big data, data mining, text mining, data visualization, heterogeneous data, credit reporting, augmented reality, mixed reality, and virtual reality; Digital technology applications include: mobile Internet, industrial Internet, mobile internet, Internet medical care, e-commerce, mobile payment, third-party payment, NFC payment, B2B, B2C, C2B, C2C, O2O, networking, smart wear, smart agriculture, intelligent transportation, intelligent medical care, intelligent customer service, smart home, intelligent investment, intelligent travel, intelligent environmental protection, smart grid, and intelligent travel.); the underlying technologies are further subdivided into four categories: artificial intelligence, blockchain, cloud computing, and big data. Then, we used Python to extract the annual report text of agricultural-related companies in the Shanghai and Shenzhen A stock markets. We matched the annual report text with the keywords of digital transformation to get the measurement index of digital transformation. The data mainly come from the CSMAR database. Because the data have obvious right deviation characteristics, the original data are added with 1 for natural logarithm processing.

3.1.3. Mediating Variables

As hypothesized in the previous section, this paper contends that the digitization of agribusinesses significantly influences green transformation through three primary mechanisms: scale effect, technology effect, and structural effect. Accordingly, mediating variables are chosen to represent each of these dimensions.

(1)

Scale Effect. This analysis employs the operating income of agriculture-related enterprises (expressed as the natural logarithm) as a proxy for scale economy, thus illustrating the scale effect.

(2)

Technology Effect. Informed by the research of published literature [55,56], this study selects the number of green invention patent authorizations as a proxy for green technological innovation, demonstrating the technology effect.

(3)

Structural Effect. Based on the findings of Miao et al. [57], this paper used labor productivity, calculated as the ratio of enterprise operating income to the number of employees, as a proxy for enterprise upgrading to depict the structural effect.

3.1.4. Control Variable

In addition to digital transformation, the core explanatory variable in this paper, there are other enterprise-level factors that also affect the green transformation of agriculture-related enterprises. Referring to the existing literature [58,59], this paper selects a series of variables such as enterprise age, scale, ownership, asset–liability ratio, dual occupation, Tobin Q value, cash flow ratio, operating income growth rate, SA index, ROA, and management shareholding as control variables.

3.2. Empirical Model

In order to validate the research hypotheses related to digitization and green transformation of enterprises, combined with the published literature [54,58,59], this study selected the following econometric models:

$$y_{it}={\beta }_0+{\beta }_{1}D{T}_{it}+{\beta }_{2}D{T}_{it}^2+\mu Control{s}_{it}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(1)

In Equation (1), $y_{it}$ denotes the green transformation of agribusiness firms measured by two indicators: pollution emission and carbon emission. DT is the digital transformation of the firms, and ${DT}_{it}^2$ is the squared term denoting the digital transformation of the firms. The squared term captures non-linear correlations between variables. The variable ‘controls’ encompasses a variety of control variables previously identified, including firm age, size, ownership type, balance sheet ratio, dual occupation, Tobin’s Q value, cash flow ratio, operating income growth rate, SA index, return on assets (ROA), and management shareholding, among others. The model employs robust standard errors. Our analysis primarily concentrates on the magnitude and statistical significance of the coefficients in the model. The subscripts i and t represent individual companies and time, respectively. ${\gamma }_i$ is the individual fixed effect, ${\delta }_t$ is time fixed effect, ${\epsilon }_{it}$ is the random disturbance term subject to the white noise process, ${\beta }_0$ is a constant term, and ${\beta }_1$, ${\beta }_2$, and $\mathsf{\mu }$ are regression coefficients.

In order to verify the channel mechanism of digital transformation to green transformation of agricultural-related enterprises, combined with the empirical route of published literature [60,61,62], this study used a three-stage intermediary effect model for verification.

First, since the effect of digitization on the scale effect channel is linear, the following model is designed:

$$Scal{e}_{it}={\alpha }_0+{\alpha }_{1}D{T}_{it}+\mu Control{s}_{it}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(2)

$$y_{it}={\phi }_0+{\phi }_{1}D{T}_{it}+{\phi }_{2}Scal{e}_{it}+{\phi }_{3}Scal{e}_{it}^2+\mu Control{s}_{it}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(3)

In Equations (2) and (3), Scale denotes the size effect, which is measured using firms’ operating revenues, as mentioned earlier. The other variables are set up in line with Equation (1). The prerequisite for the establishment of the intermediary effect is the statistical significance of the digitized regression coefficient ${\beta }_1$. On this basis, it is also necessary for regression coefficients ${\alpha }_1$, ${\phi }_1$, and ${\phi }_2$ to be statistically significant.

Secondly, since the channel of influence of digitalization on the technology effect is also linear, the following model is designed:

$$GT{I}_{it}={\alpha }_0+{\alpha }_{1}D{T}_{it}+\mu Control{s}_{it}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(4)

$$y_{it}={\phi }_0+{\phi }_{1}D{T}_{it}+{\phi }_{2}GT{I}_{it}+{\phi }_{3}GT{I}_{it}^2+\mu Control{s}_{it}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(5)

In Equations (4) and (5), GTI is the technology effect, measured using the number of green invention patents granted by firms, and the settings of other variables are consistent with Equation (1).

Finally, since the process of digitization to structural effect is non-linear, in order to verify the structural effect channel, this paper constructs the following measurement model:

$$S{J}_{it}={\alpha }_0+{\alpha }_{1}D{T}_{it}+{\alpha }_{2}D{T}_{it}^2+\mu Control{s}_{it}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(6)

$$y_{it}={\phi }_0+{\phi }_{1}S{J}_{it}+{\phi }_{2}D{T}_{it}+{\phi }_{3}D{T}_{it}^2++\mu Control{s}_{it}+{\gamma }_i+{\delta }_t+{\epsilon }_{it}$$

(7)

In Equations (6) and (7), ${\alpha }_0$ and ${\phi }_0$ are constant terms, SJ is structural effect, and the meaning of the remaining symbols is consistent with Equation (1).

3.3. Data Sources

Integrating the research questions and variable design outlined in Section 4.1, this study analyzes agriculture-related companies listed on the main boards of the Shanghai and Shenzhen stock exchanges from 2013 to 2021. The selection of agriculture-related enterprises follows criteria established in prior research. Specifically, Huang et al. [63] included enterprises within several subsectors, such as agriculture, food processing, beverage manufacturing, and textile industries. In contrast, Xie et al. [64] focused on enterprises where the main business income related to agriculture represents a significant proportion of their total business income. Due to limitations in the availability of detailed operating income data, this study adopts the approach of Fan et al. [65], further including companies primarily engaged in agricultural machinery. This study also excludes companies that have been delisted or suspended, those with missing data for key variables, and those listed for less than one year. Following these criteria, a final sample of 1049 observations was obtained for analysis. The descriptive statistical analysis of each variable is shown in

Table 1. Descriptive statistics of variables.

Variable	Obs	Mean	Standard Error	Min	Max
Pollution emission	1049	5.7336	1.0514	1.8245	6.8993
Carbon emission	1049	3.1642	1.2101	0.4813	6.4279
Digital transformation	1049	1.2948	1.0702	0.0000	3.7377
Scale effect	1049	12.5397	1.2101	9.4276	16.9345
Technology effect	1049	0.0433	0.2065	0.0000	1.3863
Structural effect	1049	13.6863	0.6661	12.3566	15.5477
Ownership type	1049	0.3375	0.4731	0.0000	1.0000
Dual occupation	1049	0.2984	0.4578	0.0000	1.0000
ROA	1049	0.0561	0.0639	−0.1584	0.2164
Tobinq	1049	2.3557	1.5188	0.9049	9.6433
SA index	1049	3.8811	0.2064	3.4261	4.3980
Age	1049	2.9207	0.2760	2.0794	3.4340
Balance sheet ratio	1049	0.3535	0.1687	0.0438	0.7583
Firm size	1049	8.0591	1.0662	5.8777	10.8976
Management shareholding	1049	0.1223	0.1949	0.0000	0.6909
Revenue growth rate	1049	0.2389	1.1093	−0.9471	8.4453
Cash flow ratio	1049	0.1177	0.1325	−0.3419	0.5102

.

From Table 1, we can observe that the average digital transformation level of the sampled agricultural-related enterprises is 1.2948, with a standard deviation of 1.2101. This indicates a significant disparity in the digitalization levels among the sampled enterprises. This disparity might be due to the fact that some of the agricultural-related enterprises are purely agricultural, while others are involved in manufacturing agricultural production materials, such as agricultural machinery and pesticides, which tend to have higher levels of digitalization and automation. Regarding carbon emissions and pollution emissions, the average pollution emissions of agricultural-related enterprises are 5.7336, and the average carbon emissions are 3.1642. This implies that although industrial enterprises are the main contributors to pollution and carbon emissions, agricultural-related enterprises also generate considerable pollution and carbon emissions during their production processes. Additionally, from the characteristics of control variables, we can see that the proportion of state-owned enterprises is approximately 33.75%, indicating that private enterprises are predominant among agricultural-related enterprises. The average firm age is 2.9207, and the average firm size is 8.0591, suggesting that most of the sampled enterprises are relatively young and small in size.

Upon identifying the research subject, this study delineates its data sources comprehensively. The primary data, including financial statistics and corporate information, were sourced from the CSMAR database, with a preference for data from consolidated financial statements. Data concerning corporate pollution and carbon emissions were primarily collected from the annual and social responsibility reports of listed companies, as well as their official websites. Information on corporate green patents was retrieved from the China Research Data Service Platform (CNRDS). For the robustness tests, Baidu search indices were obtained from Baidu’s website, while city-level control variables were sourced from the EPS database.

4. Results and Discussion

4.1. Analysis of Benchmark Regression Results

The benchmark regression results of this study are presented in

Table 2. Benchmark regression results.

Variable		(1)	(2)	(3)	(4)
		Pollution Emissions	Carbon Emissions	Pollution Emissions	Carbon Emissions
Digital Transformation		−0.034	−0.011	0.337 **	0.090 *
		(0.058)	(0.020)	(0.138)	(0.048)
Digital Transformation Squared				−0.133 ***	−0.036 **
				(0.045)	(0.015)
Age of Business		−1.120	−0.078	−1.146	−0.085
		(1.256)	(0.370)	(1.250)	(0.368)
Ballpark		0.052	0.603 ***	0.106	0.617 ***
		(0.155)	(0.067)	(0.158)	(0.067)
System of Ownership		0.002	0.290 ***	0.054	0.304 ***
		(0.301)	(0.089)	(0.307)	(0.088)
Gearing		0.380	0.948 ***	0.473	0.974 ***
		(0.528)	(0.189)	(0.517)	(0.188)
Merging of Two Functions		−0.029	−0.027	−0.026	−0.026
		(0.125)	(0.043)	(0.125)	(0.044)
Tobin Q-value		−0.009	0.037 ***	−0.009	0.037 ***
		(0.039)	(0.013)	(0.038)	(0.012)
Cash flow ratio		−0.267	0.384 **	−0.238	0.392 **
		(0.394)	(0.157)	(0.394)	(0.157)
Growth		0.846	−0.542	0.853	−0.540
		(1.119)	(0.355)	(1.087)	(0.358)
SA index		0.263	2.252 ***	0.196	2.234 ***
		(1.005)	(0.416)	(1.007)	(0.412)
ROA		−0.331	−0.078	−0.590	−0.148
		(0.755)	(0.230)	(0.779)	(0.230)
Percentage of Management’s Shareholding		0.005	0.050 ***	0.009	0.051 ***
		(0.032)	(0.013)	(0.033)	(0.013)
Constant		5.300	−0.028	4.784	−0.168
		(5.628)	(1.723)	(5.507)	(1.721)
Observed Values		1028	1028	1028	1028
Firm Fixed Effects		YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES
R2		0.178	0.938	0.186	0.938
U-test	Extreme Point			1.272	1.244
	Interval			[0, 3.738]	[0, 3.738]
	Slope			[0.337, −0.654]	[0.090, −0.180]
	p > |t|			0.007	0.030

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. To enhance the reliability of the findings, the paper also provides estimates for the scenario of a linear relationship, detailed in Columns (1) and (2) of Table 2. Analysis from these columns indicates the absence of a linear relationship between digital transformation in agribusinesses and both pollution and carbon emissions. Upon incorporating the squared term of digital transformation into the model, the revised estimates are displayed in Columns (3) and (4) of Table 2. These results reveal significant coefficients for both the digital transformation and its squared term. Specifically, the coefficients for digital transformation are significantly positive, while those for the squared term are significantly negative, suggesting an inverted U-shaped relationship between digital transformation and green transformation in agribusiness. To verify the robustness of these findings, a U-test was conducted on the inverted U-shaped relationship observed in Columns (3) and (4), following the methodology of published literature [66,67,68]. Consistent with the conclusions of the published literature, this study reveals a U-shaped relationship between digitization and carbon and pollution reduction [69,70]. As observed in the existing literature [71,72,73], enterprises are in the initial stage of digital transformation because digital equipment and digital technology in the transformation of product production and daily business ability are not outstanding, and digital equipment with energy consumption will aggravate the carbon emissions and pollution emissions of enterprises. However, with product transformation, production line restructuring, and green innovation resources, digital technology is fully integrated into the strategic management and daily office of enterprises. Under the influence of digital transformation, the total factor productivity and production mode of the company will become greener, and the carbon emission and pollution emission associated with the production of the final product will gradually decrease [74]. As the researchers observed in their analysis of digital transformation and enterprise ESG, there is a significant time lag between digital and green transformation [75]. Our conclusions are significantly different from the existing literature, showing that there is no non-linear relationship between digital and green transformation [76].

Analyzing column (3), the interval for digital transformation values ranges from 0 to 0.378. Using the extreme point calculation method, the identified extreme value is 1.272, which falls within the observed data range. Consequently, the initial hypothesis is statistically rejected at the 1% level. Additionally, the negative sign of the slope within this interval indicates an inverted U-shaped relationship between digital transformation and pollution emissions in agribusiness. Similarly, in column (4), the extreme value is determined to be 1.243, also lying within the data range for digital transformation. Here, the original hypothesis is dismissed at the 5% statistical level. The slope’s negative sign aligns with the results, further corroborating the inverted U-shaped relationship between digital transformation and carbon emissions in the sector.

The baseline regression results presented in Table 2 reveal that the initial stages of digital transformation in agribusinesses do not decrease pollution or carbon emissions. This paper identifies two primary reasons for this phenomenon. Firstly, many agribusiness enterprises have a weak informational foundation and must begin their digital transformation from scratch. This process involves significant investment in digital infrastructure, such as servers and storage, and in the case of direct agricultural product producers, the establishment of smart agriculture systems. These infrastructures initially increase energy and electricity consumption, thus exacerbating pollution and carbon emissions, as corroborated by existing studies [77]. However, as digital transformation matures, it significantly reduces the marginal costs associated with the adoption of digital technologies, enhancing the input-output efficiency of agribusinesses and facilitating reductions in pollution and carbon emissions. Secondly, both digital and green transformations entail substantial costs. Agribusinesses aiming to maximize profits and faced with significant market and natural disaster risks are initially deterred from investing in green transformation due to the high costs of beginning digital transformation. As digital transformation progresses, the costs can be offset by reductions in management expenses, improvements in efficiency, and gains from economies of scale, which foster innovation. At this advanced stage, enterprises are better equipped and more motivated to invest in measures that reduce pollution and support green transformation.

4.2. Robustness Test

This study acknowledges potential biases due to variable selection errors, model setup, and strategic firm behavior, which may influence estimation results. To enhance the robustness of our findings, this section focuses on three main areas of robustness testing: substituting explanatory and interpretative variables and excluding samples without digital investments.

4.2.1. Replacement of Explanatory Variables

Initially, we replace the variable representing pollution discharge. Since Chemical Oxygen Demand (COD) captures only a segment of aquatic pollution and the operational scope of agribusinesses is diverse, other pollutants, such as ammonia nitrogen, total nitrogen, and total phosphorus, are also considered. Following the methodology of Du et al. [78] and the “Regulations on the Administration of the Collection and Use of Sewage Charges” (2002) by the State Council, we sum and log-transform these measures into an integrated pollution equivalent for a comprehensive reflection of water body pollution by enterprises. The results in column (1) of

Table 3. Regression results with replacement of core variables.

		(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variable		Logarithm of Combined Pollution Equivalent of Water Bodies	Production Process Emissions	Green Investment	Pollution Emissions	Carbon Emissions	Pollution Emissions	Carbon Emissions
Digital Transformation		0.001 **	0.094 *	0.188 *	0.306 **	0.064	0.273	0.100
		(0.000)	(0.052)	(0.101)	(0.132)	(0.047)	(0.192)	(0.068)
Digital Transformation Squared		−0.0003 ***	−0.038 **		−0.124 ***	−0.031 **	−0.101 *	−0.051 **
		(0.000)	(0.017)		(0.046)	(0.016)	(0.060)	(0.021)
Constant		0.128 ***	8.845 ***	−17.770	5.203	−0.064	8.147	2.562
		(0.012)	(1.996)	(11.920)	(5.534)	(1.711)	(8.320)	(2.361)
Control Variable		YES	YES	YES	YES	YES	YES	YES
Observed Values		1028	1028	1049	1028	1028	637	637
Firm Fixed Effects		YES	YES	YES	YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES	YES	YES	YES
R2		0.671	0.912	0.622	0.184	0.938	0.224	0.942
U-test	Extreme Point	1.049	1.254		1.231	1.021	1.358	0.978
	Interval	[0, 3.738]	[0, 3.738]		[0, 3.738]	[0, 3.738]	[0, 3.738]	[0, 3.738]
	Slope	[0.001, −0.002]	[0.094, −0.187]		[0.306, −0.622]	[0.064, −0.170]	[0.273, −0.478]	[0.100, −0.284]
	p > |t|	0.009	0.035		0.010	0.08	0.078	0.070

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

confirm the persistence of an inverted U-shaped relationship between digital transformation and pollution emissions using this revised explanatory variable. Next, we substitute the carbon emission indicator. Whereas the baseline regression utilized total carbon emissions, this phase employs emissions from the production process as an alternative metric. The findings, detailed in column (2) of Table 3, affirm an inverted U-shaped relationship between digital transformation and carbon emissions when considering production process emissions, thus underscoring the robustness of the baseline regression conclusions.

We also used the amount of investment in green and environmental protection by enterprises as a proxy variable for green transformation. Environmental protection investment is a direct reflection of a company’s green transformation [79]. If digitalization promotes investment related to green transformation, it indicates that it is beneficial for the company’s green transformation. We screened out investment projects related to green and environmental protection from the investment details of agricultural-related enterprises and aggregated the amounts by year. The results are shown in column (3) of Table 3. From Table 3, it can be observed that digital transformation has a direct promoting effect on green investment rather than an inverted U-shaped relationship. However, this result still supports our conclusion in the baseline regression, as with the increase in the level of digitalization, after reaching a certain threshold, the impact of digital transformation on green transformation will also be positive.

4.2.2. Replacement of Core Explanatory Variables

Given the unique production and operational characteristics of agribusiness enterprises, not all are engaged in deploying digital technology. Particularly for smaller agribusinesses, the entry costs and usage barriers associated with digital technology are substantial. Building on the observations by Zhou et al. [80], who emphasize the significance of digital technology application in agribusinesses, this section adopts the degree of digital technology application as a proxy indicator. The regression results, presented in Columns (4) and (5) of Table 3, demonstrate that the robustness of our research findings is maintained when utilizing this proxy indicator for digital technology applications.

4.2.3. Remove Samples That Have Not Invested in Digital Intangible Assets

Due to the fact that the word frequency method is based on annual reports of enterprises, the information disclosed in these reports can be influenced by corporate behavior. Some companies might exaggerate their digital transformation in their disclosures, leading to an overestimation of their digitalization level; conversely, some companies might hide their actual level of digitalization in their annual report texts out of fear that competitors might imitate them, resulting in an underestimation of their digitalization level. Compared to the textual portions of annual reports, financial statements are less influenced by corporate disclosure behavior due to government regulations. We identified intangible assets related to digitalization from the intangible assets section of the financial statements, including software, platforms, and patents related to digital technology. In our regression analysis, we excluded samples with zero intangible assets related to digital technology, as shown in columns (6) and (7) of Table 3. After excluding samples with zero intangible assets, the regression results remain robust.

4.3. Endogeneity Test

This study is subject to endogeneity concerns. Firstly, despite controlling for firm-level fixed effects and a series of control variables, our regressions might still be influenced by omitted variables. Secondly, the research may be affected by sample selection and self-selection biases. Specifically, sample selection bias arises because the study includes only agriculture-related listed companies, excluding non-agriculture-related listed companies, which could skew the results. To address these issues, this paper employs several strategies to mitigate endogeneity.

4.3.1. Adding Control Variables

There are a lot of internal and external factors that affect the green transformation of agriculture. The benchmark regression model mainly considers the internal variables of corporate governance, while the external factors of urban environment, economic form, and climate change are not considered. Potential missing variables will make the regression results unbiased. To lessen the impact of omitted variables, this study incorporates additional control variables at both provincial and city levels. At the provincial level, we introduce GDP per capita (expressed in yuan and log-transformed for regression purposes), the proportion of agricultural value added to GDP, and the total power of agricultural machinery (measured in 10,000 kilowatts). At the city level, the degree of digital technological innovation (logarithm of the total number of patents related to digital technology in the city) and carbon emissions from industrial processes are included as control variables. The regression outcomes presented in Columns (1) and (2) of

Table 4. Endogeneity test results.

Variable		(1)	(2)	(3)	(4)
		Pollution Emissions	Carbon Emissions	Pollution Emissions	Carbon Emissions
Digital Transformation		0.367 ***	0.085 *	0.4293 *	−0.1196
		(0.137)	(0.048)	(1.8360)	(−0.8678)
Digital Transformation Squared		−0.141 ***	−0.034 **	−0.1164 **	0.0389
		(0.045)	(0.015)	(−1.9756)	(1.1182)
Constant		−4.129	−4.660	6.4194 ***	−1.8792 **
		(10.252)	(3.384)	(4.1687)	(−2.1681)
Inverse Mills Ratio				−1.5649	0.4433
				(−0.9115)	(0.5214)
Control Variable		YES	YES	YES	YES
Observed Values		991	991	1049	1049
Firm Fixed Effects		YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES
R2		0.197	0.939
U-test.	Extreme Point	1.299	1.233	1.844	1.538
	Interval	[0, 3.738]	[0, 3.738]	[0, 3.738]	[0, 3.738]
	Slope	[0.367, −0.689]	[0.085, −1.172]	[0.429, −0.441]	[−0.120, 0.171]
	p > |t|	0.004	0.040	0.030	0.193

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

affirm that the inverted U-shaped relationship between digital transformation and green transformation in agribusiness persists even after these controls are applied.

4.3.2. Heckman Two-Stage Model

Digital transformation may have certain industry attributes; that is, some specific industries will be more inclined to improve the degree of digitalization, and these industries just happen to have a high level of risk [81]. In the era of the digital economy, some leading enterprises have taken the initiative to adopt the strategy of digital transformation in order to meet technological updates and customer needs. Some market players themselves have the ability to choose whether to carry out digital reform, so they are likely to take some actions that affect the sampling process so that the randomness of the research sample is lost, resulting in self-selection bias. In addition, this study mainly used the sample data of listed companies in China, while the agriculture-related micro, small, and private enterprises were ignored, and the samples of farmers using digital technology to carry out green production were also omitted. Therefore, the research samples had the defects of selection bias and non-randomness. The Heckman two-step method mainly solves the problem of sample selection. Due to the survey design, some samples cannot be observed, and the samples we analyze are selective, which leads to biased estimation results [82,83].

To mitigate the impact of sample selection bias and further reduce endogeneity, this paper employs a Heckman two-stage model. In the first stage, a full-sample probit regression is conducted using the binary dummy variable that indicates whether digital transformation word frequency is disclosed. The control variables include the Baidu search index related to environmental concerns, the Herfindahl index, and the enterprises’ investments in green and environmental protection. These variables were selected because the Baidu search index reflects societal environmental concerns and acts as an informal environmental regulation; the Herfindahl index gauges industry competition, influencing digital transformation decisions; and investments in green and environmental protection are related to the costs of enterprise transformation.

Following this, the inverse Mills ratio calculated from the probit model is incorporated into the second-stage regression. Regression results from Heckman’s two-stage model are presented in Columns (3) and (4) of Table 4. As detailed in Column (3), an inverted U-shaped relationship between digital transformation and pollution emissions in agribusiness remains evident in post-bias reduction. Column (4) shows that while the coefficients of digital transformation and its squared term are not significant, the coefficients of the inverse Mills ratios are also insignificant, suggesting minimal endogeneity from sample selection bias. This indicates that listed agribusiness companies, as proxies for agricultural sector productivity, adequately reflect the degree of digital transformation within the sector.

We also used Propensity Score Matching (PSM) as a method to mitigate bias. PSM is an effective way to address self-selection bias [84]. Drawing on the design of the PSM method from Shipman et al. [84], we used the median digitalization level of agricultural-related enterprises as the standard, classifying enterprises above the median as the treatment group and those below as the control group. Based on this, we used control variables as covariates and applied both nearest neighbor matching and kernel matching. In nearest neighbor matching, we matched each treatment group enterprise with three control group samples. The ATT value after nearest neighbor matching was 3.413 with a t-statistic of 2.63, and only four observations were excluded from the estimation, indicating that the choice of matching method was appropriate. The results after nearest neighbor matching are shown in Columns (1) and (2) of

Table 5. Results of PSM method.

Variable		(1)	(2)	(3)	(4)
		Nearest Neighbor Matching		Kernel Matching
		Pollution Emissions	Carbon Emissions	Pollution Emissions	Carbon Emissions
Digital Transformation		0.335 **	0.090 *	0.337 **	0.090 *
		(0.139)	(0.048)	(0.139)	(0.048)
Digital Transformation Squared		−0.131 ***	−0.036 **	−0.133 ***	−0.036 **
		(0.045)	(0.015)	(0.045)	(0.015)
Constant		5.424	−0.321	4.785	−0.172
		(5.551)	(1.762)	(5.563)	(1.738)
Control Variable		YES	YES	YES	YES
Observed Values		1044	1044	1049	1049
Firm Fixed Effects		YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES
R2		0.211	0.940	0.211	0.940
U-test	Extreme Point	1.274	1.261	1.272	1.244
	Interval	[0, 3.738]	[0, 3.738]	[0, 3.738]	[0, 3.738]
	Slope	[0.335, −0.646]	[0.090, −0.177]	[0.337, −0.654]	[0.090, −0.180]
	p > |t|	0.008	0.031	0.008	0.032

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. To ensure the robustness of the study results, Columns (3) and (4) of Table 5 provide the estimation results using kernel matching. Compared to nearest neighbor matching, kernel matching does not discard any samples, as it calculates a weighted average using all control group samples. Therefore, as long as control group samples are available, all treatment group samples can be matched without any loss of observations. The results in Table 5 show that whether using nearest neighbor matching or kernel matching, the estimation results remain robust after mitigating selection bias through PSM.

5. Further Discussion

5.1. Mechanism Test

Drawing on the theoretical framework discussed in the preceding section, this paper posits that the inverted U-shaped relationship between digital transformation and the green transformation of agribusinesses arises from scale effects, technological effects, and structural effects induced by digital transformation.

Table 6. Impact mechanism test results.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Variable	Economy of Scale	Pollution Emissions	Carbon Emissions	Green Technology Innovation	Pollution Emissions	Carbon Emissions	Corporate Upgrade	Pollution Emissions	Carbon Emissions
Digital Transformation	0.033 **	−0.036	−0.019	0.021 *	−0.022	0.018	0.061 *	0.334 **	0.029
	(0.013)	(0.058)	(0.015)	(0.012)	(0.057)	(0.020)	(0.034)	(0.138)	(0.048)
Digital Transformation Squared							−0.026 **	−0.130 ***	−0.002
							(0.012)	(0.044)	(0.015)
Economy of Scale		−0.285	0.032
		(1.945)	(0.452)
Economies of Scale Squared		0.016	0.039 **
		(0.077)	(0.018)
Green Technology Innovation					0.684	−0.088
					(0.497)	(0.228)
Green Technology Innovation Squared					−0.853 **	−0.018
					(0.397)	(0.189)
Corporate Upgrade								−0.020	0.537 ***
								(0.165)	(0.073)
Constant	12.040 ***	5.477	−6.919 ***	0.266	5.441	2.019	17.063 ***	5.496	−7.206 ***
	(1.710)	(9.021)	(2.276)	(1.926)	(5.636)	(2.096)	(1.373)	(6.032)	(2.323)
Observed Values	1028	1028	1028	1028	1028	1028	1028	1028	1028
R2	0.968	0.178	0.959	0.486	0.181	0.927	0.917	0.186	0.934
Firm Fixed Effects	YES	YES	YES	YES	YES	YES	YES	YES	YES
Year Fixed Effects	YES	YES	YES	YES	YES	YES	YES	YES	YES

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

presents the results of testing these influence mechanisms. The regression analysis reveals that the coefficient of digital transformation on economies of scale is significantly positive at the 5% statistical level, suggesting that digital transformation facilitates the formation of economies of scale. Column (2) of Table 6 indicates that while the coefficient of economies of scale is positive, it is not statistically significant when the dependent variable is pollution emissions. In contrast, the results in Column (3) demonstrate that the coefficient of economies of scale is significantly positive, confirming its role as a key mechanism in the inverted U-shaped relationship between digitalization and carbon emissions. Employing the mediation effect test methodology developed by Lin and Feng [85] for non-linear models, the product of the squared term of digital transformation and economies of scale is calculated at 0.0012. This result suggests that economies of scale intensify the inverted U-shaped relationship between digitalization and carbon emissions, indicating that while economies of scale developed in the early stages of digital transformation may initially exacerbate carbon emissions, they contribute to reductions in emissions as digital transformation progresses.

Column (3) of Table 6 reveals that the regression coefficient of digital transformation on green technology innovation is significantly positive at the 10% statistical level, suggesting that digital transformation fosters green technology innovation in agribusiness. Column (5) demonstrates that the coefficient of green technology innovation, serving as a mediating variable, is negative and significant at the 5% statistical level. According to the same methodology, the product of the squared term of digitalization and green technology innovation is −0.018, indicating that green technology innovation helps mitigate the non-linear relationship between digitalization and pollution emissions. This result underscores that digital transformation in agribusinesses promotes green technological innovation, thus reducing pollution emissions and fostering green transformation.

Column (7) of Table 6 showed an inverted U-shaped relationship between digitization and the upgrading of agribusinesses. The U-test reveals that when regression of business upgrading is modeled with digitization and its squared term, the extreme point is 1.156 within an interval of [0, 3.738], and the slope interval ranges from 0.061 to −0.136, significantly rejecting the original hypothesis at the 5% statistical level. This suggests a non-linear relationship between digitization and business upgrading. The findings in Column (8) show that the coefficient for enterprise upgrading is not significant; however, results in Column (9) show that enterprise upgrading is significant at the 1% statistical level, where the product of the squared term of digital transformation and enterprise upgrading is −0.014, indicating that enterprise upgrading contributes to weakening the non-linear relationship between digitalization and carbon emissions. Thus, digital transformation in agriculture-related enterprises is conducive to business upgrading, which, in turn, curbs carbon emissions and promotes green transformation through structural effects.

As carbon and pollution emissions share common root characteristics, and based on the empirical test results, this paper posits that digital transformation exerts both additive and subtractive superposition effects on the green transformation of enterprises, which underpins the formation of the inverted U-shaped relationship. Initially, digital transformation generates economies of scale that increase carbon emissions and impede green transformation. However, as digitization progresses and surpasses a certain threshold, it simultaneously fosters green technological innovation and triggers enterprise upgrading. Through technological and structural effects, this enhanced digitization level enables enterprises to reduce both pollution and carbon emissions, thereby facilitating their green transformation. In summary, scale effects, technological effects, and structural effects collectively form the mechanisms underlying the non-linear relationship between digital transformation and the green transformation of agribusinesses.

5.2. Heterogeneity Test

5.2.1. Heterogeneity in Firm Size

While this study’s sample comprises agriculture-related listed companies, these entities typically enjoy a size advantage over unlisted counterparts. Nonetheless, the presence of agriculture-related enterprises across agriculture and manufacturing industries introduces significant size variations among listed enterprises, potentially influencing the relationship between digitalization and green transformation. An analysis of enterprise size variability by using the xtsum command in STATA 16 reveals a within-group standard deviation of 0.218 and a between-group standard deviation of 1.027, highlighting substantial size disparities among these enterprises. Such discrepancies could cause variations in the degree of digitalization between different-sized enterprises, thereby differentially impacting their green transformation efforts.

Estimation results from group regression based on average enterprise size are presented in

Table 7. Heterogeneity of firm size.

Variable		(1)	(2)	(3)	(4)
		Pollution Emissions	Pollution Emissions	Carbon Emissions	Carbon Emissions
		Small-Scale Enterprises Group	Larger Enterprise Group	Small-Scale Enterprises Group	Larger Enterprise Group
Digital Transformation		0.668 ***	−0.006	0.109	0.146 **
		(0.245)	(0.217)	(0.091)	(0.067)
Digital Transformation Squared		−0.250 ***	0.006	−0.055 *	−0.028
		(0.085)	(0.072)	(0.030)	(0.023)
Constant		19.008	3.325	10.585 ***	2.430
		(12.353)	(9.211)	(3.879)	(2.835)
Control Variable		YES	YES	YES	YES
Observed Values		489	490	489	490
Firm Fixed Effects		YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES
R2		0.330	0.351	0.942	0.963
U-test	Extreme Point	1.335	0.495	0.993	2.558
	Interval	[0, 3.738]	[0, 3.738]	[0, 3.738]	[0, 3.738]
	Slope	[0.668, −1.203]	[−0.005, 0.037]	[0.109, −0.301]	[0.146, −0.671]
	p > |t|	0.003	0.490	0.116	0.276

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. Column (1) reveals an inverted U-shaped relationship between digitalization and pollution emissions among smaller firms, with the U-test indicating a rightward shift in the extreme point to 1.355 from 1.272 in the benchmark regression. This shift suggests that the mitigating effect of digital transformation on pollution emissions is delayed in smaller enterprises, where digitalization occurs more slowly, and the cost pressures during transformation are more pronounced. This effect is not observed in larger agribusiness firms. Columns (3) and (4) indicate that firm size does not significantly alter the inverted U-shaped relationship between digitalization and carbon emissions.

5.2.2. Heterogeneity in the Nature of Enterprises

The majority of agriculture-related enterprises are privately owned. In the study’s sample of agriculture-related listed companies, state-owned enterprises (SOEs) constitute approximately 33.74 percent but remain predominantly private. On one hand, compared to private enterprises, SOEs benefit from greater resource availability and have more flexibility in making cost investments for digital transformation, such as easier access to digitalization-related financing. On the other hand, SOEs often bear a greater burden of social responsibility, including environmental protection, potentially leading them to be more proactive in green transformation initiatives. Thus, the nature of ownership may influence the relationship between digitization and green transformation in agribusinesses. The results of regression analyses based on the nature of ownership are presented in

Table 8. Heterogeneity in the nature of firms.

Variable		(1)	(2)	(3)	(4)	(5)	(6)
		Pollution Emissions	Pollution Emissions	Carbon Emissions	Carbon Emissions	Pollution Emissions	Carbon Emissions
		Private Enterprise Group	State-Owned Enterprises	Private Enterprise Group	State-Owned Enterprises	State-Owned Enterprises	State-Owned Enterprises
Digital Transformation		0.593 ***	−0.022	0.098 *	0.092	−0.104	−0.018
		(0.180)	(0.221)	(0.058)	(0.085)	(0.088)	(0.034)
Digital Transformation Squared		−0.195 ***	−0.031	−0.033 *	−0.042
		(0.057)	(0.073)	(0.017)	(0.026)
Constant		1.675	19.811	0.321	−0.792	20.175	−0.299
		(6.109)	(12.846)	(2.030)	(5.100)	(12.859)	(5.119)
Control Variable		YES	YES	YES	YES	YES	YES
Observed Values		677	348	677	348	348	348
Firm Fixed Effects		YES	YES	YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES	YES	YES
R2		0.201	0.213	0.943	0.935	0.213	0.934
U-test	Extreme Point	1.522	−0.364	1.500	1.110
	Interval	[0, 3.738]	[0, 3.738]	[0, 3.738]	[0, 3.738]
	Slope	[0.593, −0.862]	[−0.022, −0.251]	[0.098, −1.147]	[0.092, −0.218]
	p > |t|	0.000	/	0.046	0.138

Note: Robust standard errors in parentheses; *** p < 0.01, * p < 0.1.

.

Columns (1) and (3) of Table 8 demonstrate an inverted U-shaped relationship between digitization and both pollution and carbon emissions in private agriculture-related enterprises. The U-test reveals that the extreme points in Columns (1) and (3) are 1.522 and 1.500, respectively, which are right-shifted compared to the benchmark regression’s extreme point of 1.272. This indicates that the mitigating effects of digitization on emissions in the private sector are delayed due to cost constraints and disadvantages in resource acquisition. In contrast, for SOEs, the U-test does not confirm a non-linear relationship between digitization and green transformation concerning both pollution and carbon emissions. Consequently, a linear relationship was further tested, and the results are displayed in Columns (5) and (6) of Table 8. The findings suggest a negative impact of digitization on emissions in SOEs; however, this impact is not statistically significant at this stage, which may be attributed to the complex governance structure of SOEs, resulting in higher internal control costs. Initially, digitization in SOEs tends to reduce internal control costs and improve productivity rather than significantly lowering pollution and carbon emissions.

5.2.3. Heterogeneity of Supply Chain Characteristics

Given that agribusinesses are intrinsically linked with agriculture, their production and operational outcomes are significantly influenced by the supply and demand dynamics of agricultural products. Consequently, characteristics of the supply chain are crucial to both the production outcomes and the relationship between digitalization and green transformation in agribusinesses. To assess the impact of supply chain stability, this study utilizes the average of the combined proportion of purchases and sales to the top five suppliers and customers of agriculture-related listed companies as a proxy indicator of supply chain stability. Using the average of all sampled enterprises as a benchmark, a group regression test was conducted, with results presented in

Table 9. Heterogeneity analysis of supply chain stability.

Variable		(1)	(2)	(3)	(4)
		Pollution Emissions	Pollution Emissions	Carbon Emissions	Carbon Emissions
		Low Supply Chain Stability	High Supply Chain Stability	Low Supply Chain Stability	High Supply Chain Stability
Digital Transformation		0.310	0.586 ***	0.041	0.133 *
		(0.246)	(0.226)	(0.079)	(0.077)
Digital Transformation Squared		−0.137 *	−0.215 ***	−0.028	−0.064 ***
		(0.079)	(0.074)	(0.022)	(0.024)
Constant		17.582	2.288	3.070	−1.943
		(11.773)	(10.575)	(3.842)	(2.971)
Control Variable		YES	YES	YES	YES
Observed Values		483	494	483	494
Firm Fixed Effects		YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES
R2		0.327	0.344	0.951	0.950
U-test	Extreme Point	1.129	1.364	0.721	1.036
	Interval	[0, 3.738]	[0, 3.738]	[0, 3.738]	[0, 3.738]
	Slope	[0.310, −0.716]	[0.586, −1.019]	[0.041, −0.172]	[0.133, −0.347]
	p > |t|	0.104	0.005	0.301	0.043

Note: Robust standard errors in parentheses; *** p < 0.01, * p < 0.1.

. Columns (2) and (4) indicate a significant inverted U-shaped relationship between digitalization and green transformation when supply chain stability is higher. This finding suggests that digitalization enhances supply chain management in agribusiness enterprises. Moreover, a more stable supply chain within a firm facilitates the capture of spillover effects from digitalization, benefiting both customer and supplier enterprises, which, in turn, supports the green transformation of the enterprise by reducing pollution and carbon emissions.

5.2.4. Heterogeneity of Environmental Regulation

Environmental regulation by the government plays a crucial role in influencing firms’ green transition. Supporting Porter’s hypothesis, appropriate environmental regulations can enhance firms’ innovation and productivity, effectively offsetting the costs associated with green transformation behaviors [86]. Given that the production behaviors of some agribusinesses generate negative environmental externalities, these activities are a focal point for government oversight. To enforce corporate environmental responsibility, in June 2008, China’s Ministry of Environmental Protection issued the “List of Listed Companies Classified and Managed by Industry for Environmental Verification”, categorizing 14 industries, including thermal power and iron and steel, as heavily polluting. These heavily polluting listed companies, due to their significant environmental impact, face heightened pressures for environmental protection and regulation.

In this study, samples from heavily polluting enterprises are designated with a value of 1, while others are assigned a value of 0. Group regression results are displayed in

Table 10. Heterogeneity of environmental regulation.

Variable		(1)	(2)	(3)	(4)
		Pollution Emissions	Pollution Emissions	Carbon Emissions	Carbon Emissions
		Non-Heavily Polluting Enterprises	Heavily Polluting Enterprises	Non-Heavily Polluting Enterprises	Heavily Polluting Enterprises
Digital Transformation		0.331	0.382 *	0.071	0.142 *
		(0.201)	(0.223)	(0.061)	(0.077)
Digital Transformation Squared		−0.128 **	−0.153 **	−0.031	−0.053 **
		(0.064)	(0.077)	(0.019)	(0.022)
Constant		5.884	17.806	−4.086	−0.098
		(11.017)	(12.450)	(3.304)	(3.917)
Control Variable		YES	YES	YES	YES
Observed Values		566	442	566	442
Firm Fixed Effects		YES	YES	YES	YES
Year Fixed Effects		YES	YES	YES	YES
R2		0.278	0.311	0.946	0.959
U-test	Extreme Point	1.286	1.250	1.134	1.334
	Interval	[0, 3.738]	[0, 3.738]	[0, 3.738]	[0, 3.738]
	Slope	[0.331, −0.630]	[0.381, −0.759]	[0.071, −0.163]	[0.142, −0.256]
	p > |t|	0.051	0.044	0.124	0.033

Note: Robust standard errors in parentheses; ** p < 0.05, * p < 0.1.

. Columns (1) and (2) show that an inverted U-shaped relationship between digital transformation and pollution emissions persists among agriculture-related enterprises, regardless of their classification as heavily polluting. Notably, the U-test reveals that the extreme point for heavily polluting firms, at 1.250, is lower than that for non-heavily polluting firms, indicating a leftward shift from the benchmark regression’s extreme point of 1.272. This shift suggests that environmental regulations may accelerate the pollution reduction benefits of digitization. Furthermore, Columns (3) and (4) demonstrate that the inverted U-shaped relationship between digitization and carbon emissions is significant among heavily polluting firms but not in non-heavily polluting firms, indicating that environmental regulation supports the reduction of carbon emissions through digital transformation in more regulated industries.

6. Conclusions and Outlook

6.1. Conclusions of the Study

This paper examines the impact of digital transformation on the green transformation of agriculture-related enterprises using a double fixed-effect model and focusing on listed agriculture-related companies from 2013 to 2021. Through a comprehensive approach that includes a literature review, theoretical analysis, descriptive statistics, and regression analysis, several key findings emerge:

Firstly, the relationship between digitization and green transformation in agribusiness enterprises is non-linear, characterized by an inverted U-shaped pattern. This indicates that initially, the adoption and application of digital technologies increase energy consumption and costs, which may slow the progress of green transformation. However, as digitization reaches a certain threshold, it begins to facilitate green transformation, evidenced by reductions in both pollution and carbon emissions, thus achieving synergistic environmental benefits. The findings on agricultural-related enterprises are inconsistent with the existing conclusions from a macroeconomic perspective that evaluate the impact of the digital economy on green agricultural development, which generally conclude that digitalization promotes green agricultural development [13,14,15,16]. However, our observations from the perspective of enterprises reveal an initial adverse impact of digital transformation on green transformation, which aligns with the current development reality of agricultural-related enterprises. As noted in the literature [5,6,7,8,9,10], the overall level of green agricultural development in China is uneven across regions, the concept of green development is not deeply ingrained, agricultural production methods remain extensive, and the supply and brand competitiveness of high-quality green agricultural products are insufficient. These factors may further cause a delay in the impact of digital transformation on green development for enterprises.

Secondly, the dynamics of the inverted U-shaped relationship vary across different contexts. Factors such as enterprise scale, ownership, supply chain stability, and environmental regulation influence this relationship. For example, smaller and private agriculture-related enterprises experience a delayed effect of digital transformation on green transformation. Conversely, enterprises with stable supply chain relationships show a pronounced inverted U-shaped relationship, indicating that supply chain stability supports green transformation. This also answers, from another perspective, why our conclusions differ from the existing literature. The digital transformation of agricultural-related enterprises has a non-linear impact on green transformation because the relationship between the two is influenced by heterogeneous factors such as industry and enterprise size. The existing study rarely focuses on agricultural-related enterprises, with most evaluations being conducted on all listed companies or specifically on the manufacturing industry [59]. These studies overlook the unique characteristics of the agricultural sector and firms. Compared to manufacturing, agricultural-related enterprises are generally smaller in scale and younger [87]. Undertaking both digital and green transformations can increase production costs, adversely affecting the enterprises in the initial stages. Additionally, enterprises subjected to government environmental regulations exhibit a weakened non-linear relationship, supporting Porter’s hypothesis that effective environmental regulation can enhance green transformation. However, our study only considers government-mandated environmental regulations. Due to data limitations, we have not adequately analyzed another important factor: the impact of consumer environmental awareness on our conclusions. Khuc et al. [11] have found that environmental culture has become a key factor affecting green development.

Thirdly, the mechanisms through which digitization influences green transformation include scale effects, technology effects, and structural effects. Digital transformation fosters economies of scale by enhancing production efficiency and reducing average costs. While these economies of scale can initially impede green transformation by encouraging production expansion, ongoing digital advancement leads to green technological innovation. This innovation facilitates cleaner production and reduces emissions through technological progress. As digitization advances, it also promotes structural changes in production methods, further aiding green transformation through these structural effects.

6.2. Policy Recommendations

Based on the findings of this study, the following policy recommendations are proposed to enhance the digital and green transformation of agriculture-related enterprises:

Firstly, it is crucial to robustly support the digital transformation of agriculture-related enterprises through government policy initiatives and by facilitating the involvement of financial institutions. This support should aim to alleviate technical and financial challenges encountered during the transformation process, thereby enhancing the digital transformation level and motivating green transformation initiatives, which will, in turn, accelerate the green transformation of these enterprises.

Secondly, the digital transformation of small and medium-sized enterprises (SMEs) and private enterprises should be actively encouraged and guided. Given that SMEs and private enterprises constitute the majority of agriculture-related businesses, increasing their digital transformation efforts will elevate the overall level of digitization within the agricultural sector and foster greater digital integration across the industry.

Additionally, the role of environmental regulation should be leveraged to enhance the ecological awareness of agriculture-related enterprises. Effective environmental regulation can act as a catalyst for enterprises to adopt more sustainable practices, aligning with global green transformation goals.

Finally, it is recommended that agriculture-related enterprises actively adopt and practice Environmental, Social, and Governance (ESG) principles. By taking responsibility for environmental protection and modifying their production methods through digital transformation, enterprises can harness the benefits of green technological innovation and structural upgrades. Enterprises can utilize digital platforms to establish green incentive mechanisms, such as environmental points systems, to encourage employees to actively participate in environmental activities, thereby fostering a positive corporate environmental culture. This shift will not only reduce pollution and carbon emissions but also elevate the overall level of green transformation within the industry.

6.3. Research Limitations and Future Perspectives

The limitations of this study mainly include the following aspects. Firstly, this study primarily focuses on listed agricultural companies as representatives of the agricultural sector, which inevitably excludes non-listed agricultural-related enterprises. Despite efforts to mitigate sample selection bias using the Heckman two-step method, residual biases may still affect the precision of the estimation results. Future research could extend its scope by incorporating data from unlisted agriculture-related enterprises, potentially gathered through surveys or other means, to more comprehensively address this issue.

Secondly, the relationship between digitalization and green transformation is multifaceted, and while this paper examines the scale effect, technology effect, and structural effect, other influencing mechanisms remain unexplored. These additional factors merit deeper investigation in future studies to fully understand the dynamics at play in the digital and green transformation of the agricultural sector.

Thirdly, although we have focused on the impact of environmental regulation on the relationship between agribusinesses and green transformation, this topic is only considered as part of the heterogeneity analysis in this paper. Furthermore, in our article, environmental regulation is viewed as government-mandated, and we did not take into account the influence of environmental culture. The research by Khuc et al. emphasizes the importance of environmental culture in policy-making and urban green development [11]. The concept of culture tower put forward by Khuc [43] shows that culture will affect the individual decisions of enterprises and consumers through mutual influence. On one hand, in the digital age, consumers’ values are crucial to the direction of corporate innovation. Changes in environmental culture will drive companies to produce cleaner products by influencing consumer behavior. For example, thanks to the development of digital technologies such as social media, mobile applications, and online platforms, companies can fully understand consumers’ views on the environment and thus update their green product strategies. On the other hand, changes in consumer environmental awareness will promote the formation and shaping of environmental culture [11], which will influence government environmental regulation policies and industrial policies, such as the formulation of policy plans guiding green agricultural development. This is also a topic worthy of further discussion.

Finally, with the evolution of economics, analyzing the impact of digital technology adoption by agribusinesses on green transformation purely from a techno-economic paradigm may not fully reflect the decision-making logic of enterprises. For example, we believe that due to cost factors, the adoption rate of digital technology by agribusinesses is low because our analysis still follows the neoclassical economic assumptions of rational decision-makers and profit maximization. However, recent advancements in economic research, such as studies on Corporate Social Responsibility (CSR) and Environmental, Social, and Governance (ESG) criteria, may be challenging these assumptions. Moreover, digitalization has altered corporate decision-making mechanisms [87]. As Khuc [88] put forward, using the mindspongeconomics analytical framework to delve into the issues discussed in this paper could be a significant research direction in the future.