The Impact of Digital Finance on Enhancing the Spatial Effects of Heterogeneous Environmental Regulations in Supporting Agricultural Green Total Factor Productivity

Abstract

Improving agricultural green total factor productivity (AGTFP) is the key to achieving sustainable agricultural development and empowering agricultural modernization. Based on the panel data of 30 provincial levels in China from 2011 to 2021, AGTFP is measured using the non-expected MinDS super-efficiency—MetaFrontier Malmquist model, and the impact of environmental regulation (ER) and digital finance on AGTFP is analyzed using the spatial Durbin model (SDM). The results show the following: (1) ER can increase local AGTFP and has a positive spatial spillover effect. Command-based ER has the highest impact on AGTFP, followed by market-incentive and public-voluntary ER. (2) Digital finance has a direct promotional effect on local AGTFP, while it has an inhibitory effect on AGTFP in neighboring regions due to the siphon effect. (3) Digital finance is an important regulatory variable affecting AGTFP concerning command-based, market-incentive and public-voluntary ER. Digital finance plays a significantly moderating role in the effectiveness of the three ERs on AGTFP, with the market-incentive ER being the highest in eastern China. Nonetheless, digital finance has a significantly moderating effect on the effectiveness of command-based and public-voluntary ER on AGTFP, with command-based ER being higher in central China. Meanwhile, digital finance only plays a significantly moderating role in the effectiveness of command-based environment regulation on AGTFP in western China. This study provides valuable reference for policymakers concerning agriculture green production in varied regions.

1. Introduction

Agricultural green development is an essential aspect of building a strong agricultural country and a critical requirement for constructing China’s modern economic system [1,2]. For a long time, China’s agricultural development was mainly driven by traditional production factors. Although the agricultural economy has achieved rapid growth, the negative impact of agricultural environmental pollution caused by the extensive production mode of “high input, high pollution and low efficiency” has become increasingly prominent. Coupled with resource and ecological constraints, the pressure of agricultural green development has piled up [3,4]. In China’s 20th National Congress Report, it was clearly put forward that firmly establish and practice the concept that clean waters and lush mountains are invaluable assets, and plan development from the perspective of harmonious coexistence between man and nature nature. The fundamental goal of environmental regulation (ER) from the government is to realize the growth of the agricultural economy and promote the green transformation of agriculture. In 2023, the State Council promulgated the development of Pratt & Whitney’s financial quality implementation opinion, emphasizing that inclusive finance should play a role in supporting green and low-carbon development and promoting the development of green ecological agriculture. Empowering rural financial transformation through digitalization and promoting financial elements to embed in agricultural green development has become the focus of rural revitalization in the new era. Digital finance can help ease credit restrictions and facilitate access to information, thereby meeting financing needs to advance green technologies [5]. Therefore, under the background of rapid development of digital finance, it is of great practical importance to study whether ER can improve agricultural green total factor productivity (AGTFP) for transforming agricultural production modes and improving agricultural modernization levels.

AGTFP is a comprehensive index, which takes into account all factor inputs (such as labor, capital, land, etc.) and all expected outputs (such as crop yields, etc.) in agricultural production but also includes non-expected outputs such as environmental pollution and ecological damage [6]. Therefore, on the basis of traditional total factor productivity, GTFP adds the concept of environmental protection and sustainable development, aiming to measure the production efficiency of agriculture under the reduction of negative environmental impacts. Specifically, it measures the balance between increases in agricultural output and losses from environmental degradation for a given resource input [7]. Different countries and international organizations have slightly different definitions and indicators of AGTFP. For example, China believes that agricultural green total factor productivity not only includes the efficiency of input and output of production factors such as labor and capital in the traditional sense but also puts special emphasis on energy consumption and environmental impact in the process of agricultural production. This requires that while improving agricultural output and efficiency, we should also pay attention to the rational use of resources and environmental protection. The United Nations Development Programme (UNDP) defines green total factor productivity as the efficiency relationship between the comprehensive input and output of production factors after taking into account energy and resource factors [8]. Developing countries may be more concerned with increasing agricultural output and reducing poverty, while developed countries may be more concerned with environmental protection and sustainable use of resources, but the core purpose is to evaluate the overall efficiency of agricultural production and promote sustainable development of agriculture [9]. When constructing AGTFP indicators, countries also make adjustments based on their national circumstances and available statistics. For example, China is likely to focus more on the efficiency of fertilizer and pesticide use, water resource management and agricultural waste disposal, as these issues are particularly prominent in Chinese agriculture [6]. Other countries are likely to focus more on sustainable land use, biodiversity conservation and climate change adaptation.

Because the resource consumption of agricultural production has external diseconomy, ER has become one of the important measures to alleviate the constraint of agricultural resources and ecological construction. According to the environmental policy of the agricultural sector, different countries implement various measures to promote the sustainable development of agriculture according to their own national conditions and environmental problems. For example, in the EU’s Common Agricultural Policy (CAP), farmers must meet environmental requirements to encourage agricultural production, such as soil protection and water resource management, and reduce the use of fertilizers and pesticides, in order to obtain EU agricultural subsidies [10]. The U.S. government supports green agricultural practices through legislation such as the U.S. Farm Bill (CRP), and the Department of Agriculture implements crop insurance programs and environmental incentives to encourage farmers to protect land and water resources and reduce pesticide and fertilizer use [11]. The ASEAN is committed to adopting sustainable agricultural policies and water management policies to promote organic and ecological agriculture and providing subsidies and policy support to encourage farmers to switch to more environmentally friendly farming practices [12]. The Indian government has implemented a “green revolution” policy, which aims to improve the efficiency of agricultural production while focusing on environmental protection. Measures include promoting water-saving irrigation techniques, improving crop varieties and reducing the use of fertilizers and pesticides [13].

Existing studies are still divided on the effectiveness of ER on AGTFP. According to the cost compliance theory, unsound agricultural ER leads to the inadaptability of point and agricultural non-point source pollution, which increases investment cost, crowding out resources and technology investment to restrain the agricultural growth rate and green innovation [14,15]. Porter’s hypothesis holds that appropriate ER can prompt agricultural producers to expand technological input and use green technologies, thus producing an innovation compensation effect to offset agricultural production costs and promote agricultural green development [16,17,18]. The uncertainty theory holds that ER inhibits the green development of agriculture in the short term, but in the long run its innovation effect will compensate for the cost compliance effect and promote agricultural development, so there is a nonlinear U-shaped relationship between the two [19]. With the advent of the digital age, digital finance with inclusive features comes into being. Its characteristics of easy access, low cost and liquidity provide a new way to promote AGTFP. On the one hand, with the help of new media and e-commerce platforms, digital finance can effectively broaden the channels for agricultural producers to obtain information and provide them with green production technology consultation and services at the same time, thus promoting agricultural green development [20,21]. On the other hand, relying on Internet channels, digital finance can greatly improve the digitization level of the agricultural chain, break regional restrictions to expand the extension of financial inclusive services and promote the cross-regional flow of agricultural resources to improve efficiency, thereby promoting the gradual extension of financial resources and services to agriculture and rural areas, stimulating the upgrading of agricultural demand and supply transformation to stimulate agriculture green development potential [22,23].

The existing literature about digital financial and the impact of ER on AGTFP research is relatively rich, but there is still room for further extension: First, most of the existing measurements of AGTFP are based on parametric stochastic frontier production function (SFA) and non-parametric data envelopment analysis (DEA), which cannot avoid the bias caused by the heterogeneity of regional agricultural technology level. Moreover, most studies only used agricultural carbon emissions as undesired output to calculate AGTFP, failing to fully reflect the types of agricultural pollution. Second, existing studies have, respectively, discussed the impact of digital finance and ER on AGTFP, but there are very few studies on whether digital finance can have a regulatory effect on the impact of ER on AGTFP. Third, most of the existing studies use proxy indicators to study the impact of ER on AGTFP, ignoring spatial correlation, which is not conducive to the implementation of ER measures.

Thus, the non-expected MinDS super-efficiency—MetaFrontier Malmquist model is adopted to measure AGTFP in the three major economic zones of eastern, western and central China in this study, and agricultural non-point source pollution is included in the non-expected output, so as to improve the measurement accuracy to meet the requirements of agricultural green development. Then, the impact of ER on AGTFP is discussed under the regulation of digital finance. Finally, the spatial Durbin model is used to empirically analyze the spatial effects of different types of ER on AGTFP using the panel data of 30 provinces (autonomous regions) in China from 2011 to 2021.

2. Research Hypothesis

2.1. Effects of ER on AGTFP

ER as a binding government intervention means to limit the excessive use of environmental resources by agricultural producers, and it is widely used to solve the negative externalities of ER [24,25]. ER is the direct and indirect intervention from the government in non-market ways to manage the use of environmental resources, which can also extend to economic means and the utilization of market policy. Meanwhile, public environmental protection consciousness gradually improves with the development of informatics, and environmental quality has been improved through consultation with polluting enterprises or pressure on local governments. Thus, this study subdivides ER into command-based, market-incentive and public-voluntary [26] and discusses the influence of heterogeneous ER on AGTFP.

2.1.1. The Effect of Command-Based ER on AGTFP

Command-based ER means that the government restrains agricultural producers who do not meet the emission standards based on environmental laws, regulations and statutes [27,28]. On the one hand, it standardizes the behavior of producers through strict pollution index restrictions, guidelines and corresponding penalties, forcing them to carry out pollution control and technological innovation and encouraging them to optimize green production techniques, adopt more environmentally friendly production methods, such as reducing the use of chemical fertilizers and pesticides and adopting biological pesticides and organic fertilizers, and improve agricultural management practices, such as precision agriculture technology, irrigation technology, etc., to improve AGTFP. On the other hand, strict ER may also send a signal to the market and consumers to protect the environment, thereby increasing the market demand for agricultural products that meet environmental standards and incentivize agricultural production to develop in a greener and more sustainable direction. In addition, the adoption of environmentally friendly technologies may reduce pollution emissions in neighboring areas, or the promotion of green production practices may improve the environmental quality of the entire region, and this impact may cross geographical boundaries, thereby indirectly increasing the agricultural green total factor productivity of neighboring areas. Therefore, hypothesis 1a is put forward as follows:

Hypothesis 1a. 

Command-based ER has a positive promoting effect on the improvement of local AGTFP and has a positive spillover effect on the improvement of AGTFP in other regions.

2.1.2. The Effect of Market-Incentive ER on AGTFP

Market-incentive ER means that the government changes the costs or benefits of agricultural producers based on market policy, such as the carbon emission trading scheme and overcapacity reduction policy, implements the principle that polluter pays and encourages them to internalize environmental costs to achieve green technological innovation [29,30,31]. Furthermore, the government can link environmental benefits with economic benefits through market transactions, encourage agricultural producers to actively promote green transformation from passively meeting legal requirements and enhance the momentum of green transformation. On the other hand, under the guidance of the market, capital and resources may be more inclined to flow to green agricultural production methods to support agricultural projects and enterprise development that can reduce pollution emissions and improve resource use efficiency, so as to realize resource reallocation to improve agricultural green total factor productivity. In addition, market-motivated ER may stimulate agricultural technological innovation, such as the development of green pesticides and the promotion of energy-saving irrigation technologies. These technological innovations occur first in the region where regulation is implemented and may subsequently affect neighboring regions through knowledge diffusion and technology diffusion. At the same time, in order to better achieve environmental goals, regional cooperation may be strengthened, such as through regional ecological compensation projects and joint research and development of green agricultural technology, which will help to increase the AGTFP of neighboring regions. Therefore, hypothesis 1b is put forward as follows:

Hypothesis 1b. 

Market-incentive ER has a positive promoting effect on the improvement of local AGTFP and has a positive spillover effect on the improvement of AGTFP in other regions.

2.1.3. The Effect of Public-Voluntary ER on AGTFP

Public-voluntary ER is a non-compulsory and soft environmental constraint based on the public’s conscious environmental awareness [32,33]. Through petitioning, self-media and other means, pressure is exerted on agricultural producers who do not meet emission standards, thus damaging their reputation and economy, forcing them to choose green transformation behaviors to meet the social requirements for green agricultural development. Heterogeneous ER forces agricultural producers to carry out green technology innovation based on external pressure and internal incentives, thus promoting local AGTFP. At the same time, the flow of agricultural production factors between regions leads to a spillover effect of technology and knowledge and promotes the development of green technology in neighboring regions to improve AGTFP. Even if a public-voluntary ER policy in a region is not directly applicable to the neighboring region, its successful cases may inspire policymakers and agricultural producers in the neighboring region to adopt similar policies and actions, and this policy spillover effect will help improve the agricultural green total factor productivity in the neighboring region. Therefore, hypothesis 1c is put forward as follows:

Hypothesis 1c. 

Public-voluntary ER has a positive promoting effect on local AGTFP and positive spillover effect on AGTFP in other regions.

2.2. The Impact of Digital Finance on AGTFP

Financial resources play a policy-oriented role in promoting the green transformation of agriculture [34]. Due to the lack of traditional financial credit supply and low service efficiency, the agricultural sector is facing serious financial exclusion [35]. With the characteristics of low cost, wide coverage and high accessibility, digital finance has gradually extended financial resources and services to agriculture and rural areas, promoting the upgrading of demand and the transformation of supply to stimulate the agriculture green development.

2.2.1. The Impact of Digital Finance on Local AGTFP

On the one hand, the long-tail effect of digital finance can broaden the channels of financial services, solve the long-term problems such as the dispersion of traditional financial demand subjects and the lack of effective supply [23], improve the accessibility of financial services for agricultural producers, provide sustainable financing channels for the agriculture green development, and thus, increase the R&D investment in new technologies to accelerate the innovation of agricultural green technologies [36]. Meanwhile, digital finance promotes traditional finance to reduce service costs and expand service scope through the catfish effect, assists agricultural producers to break through the financial threshold, and promotes the progress of agricultural green technology to improve AGTFP. On the other hand, digital finance can effectively reduce information asymmetry, promote agricultural producers to obtain market demand in time, rationally allocate capital, labor, land and other input factors to improve agricultural technical efficiency [37], and promote the transformation of agricultural production mode from extensive to green intensive. Therefore, hypothesis 2a is proposed as follows:

Hypothesis 2a. 

Digital finance has a positive promoting effect on the improvement of AGTFP in one region.

2.2.2. The Impact of Digital Finance on Other Neighboring AGTFP

Firstly, digital finance has a strong positive externality, which can improve capital flow distribution and industrial structure optimization, promote the cross-regional flow of finance, technology and knowledge, improve the economic and green technology development level of neighboring regions through the trickle-down effect, and then have a spillover effect on their AGTFP. Secondly, through big data analysis and other means, digital finance can more accurately assess the risks and benefits of agricultural projects, improve the efficiency of loans and investments, and promote the flow of capital and resources to efficient areas to improve agricultural production efficiency in neighboring regions. Finally, digital finance contributes to the efficiency of policy transmission, and policymakers can understand the actual needs and problems of agricultural development more quickly and accurately through digital platforms to formulate and implement well-targeted policies and strengthen support for green agricultural development. However, if the development of digital finance is uneven among neighboring regions, it may lead to unfair information dissemination, allowing some regions to enjoy the benefits of digital finance faster, while others may be marginalized, thus affecting their AGTFP. At the same time, digital finance aims to expand the reach of financial services, especially in areas underserved by traditional financial institutions. However, if there is a large difference in the level of digital financial development between these regions, it may exacerbate the spatial imbalance of financial services and further expand the gap in agricultural production efficiency between regions. Therefore, hypothesis 2b is proposed as follows:

Hypothesis 2b. 

Digital finance may have a spillover effect or siphon effect on other neighboring AGTFP.

2.3. The Regulatory Effect of Digital Finance on ER Affecting AGTFP

Digital finance plays an important role in the influence of ER on AGTFP. First of all, by providing precise financial services, such as agricultural supply chain financing and rural e-commerce, digital finance helps to improve the efficiency of agricultural production and reduce production costs, so as to realize the green development of agriculture under the framework of ER. Secondly, digital finance provides more financial support and services for agricultural technology innovation, such as providing financial support for agricultural science and technology innovation projects through Internet crowdfunding platforms and providing scientific decision-making support for agriculture through big data analysis to improve AGTFP. Finally, the development of digital finance helps to strengthen agricultural environmental supervision, such as monitoring and early warning of agricultural pollution through big data technology, and whole-process supervision of the agricultural industry chain through blockchain technology to improve AGTFP. Different types of ER lead to different regulatory effects of digital finance.

2.3.1. The Regulatory Effect of Digital Financial Type in the Process of Command-Based ER Affecting AGTFP

Command-based ER forces enterprises to increase innovative financing demand through the government’s high standards for agricultural producers, and digital finance can effectively reduce the financial service threshold to ease financing constraints [38]. Meanwhile, governments in different regions have different efforts to implement ER due to the asymmetry of information and different levels of economic development [39]. On the one hand, numbers based on financial data across time and space of the financial allocation efficiency can effectively reduce regional differences in ER to increase AGTFP. On the other hand, financial subsidies or tax breaks are provided for agricultural activities that meet green production standards through digital finance, and penalties are imposed for violations of environmental protection regulations to encourage agricultural producers to adopt more environmentally friendly production methods. Therefore, hypothesis 3a is proposed as follows:

Hypothesis 3a. 

Digital finance has a positive moderating effect in the process of Command-based ER affecting AGTFP.

2.3.2. The Regulatory Effect of Digital Financial Type in the Process of Market-Incentive ER Affecting AGTFP

Market-incentive ER limits pollution by raising pollution charges and imposing environmental taxes [40,41], and the higher cost forces agricultural producers to increase investment in green technologies to reduce pollution levels. On the one hand, digital finance can build efficient information communication channels, break geographical distance restrictions and product information barriers, reduce the delay and uncertainty of information transmission to reduce transaction costs and encourage agricultural producers to cultivate green technology advantages to improve AGTFP. On the other hand, through big data, cloud computing and other means, digital finance can more accurately evaluate the environmental and economic benefits of agricultural projects, so as to guide funds to green and efficient agricultural practices and encourage agricultural producers to adopt more environmentally friendly production methods to improve AGTFP. Therefore, hypothesis 3b is proposed as follows:

Hypothesis 3b. 

Digital finance has a positive moderating effect in the process of Market-incentive ER affecting AGTFP.

2.3.3. The Regulatory Effect of Digital Financial Type in the Process of Public-Voluntary ER Affecting AGTFP

Public-voluntary ER puts pressure on environmental offenders by raising public awareness of environmental protection and supervises them when removing environmental hazards [42,43,44]. On the one hand, digital finance relies on Internet platforms to broaden the channels for social subjects to participate in environmental governance and at the same time improve their environmental awareness so that farmers and investors can more easily obtain information about the environmental impact of agricultural production. Information disclosure helps to improve public recognition and investment in environmentally friendly agricultural technologies and to encourage more public participation in environmental governance to force agricultural operators to carry out green production. On the other hand, digital finance can enhance the awareness and participation of farm-level agricultural enterprises in environmental protection through mobile applications, online courses and other forms, so as to better cooperate with public voluntary ER. Therefore, hypothesis 3c is put forward as follows:

Hypothesis 3c. 

Digital finance has a positive moderating effect in the process of public-voluntary ER affecting AGTFP.

The relevant theoretical framework of this study is shown in Figure 1.

3. Materials and Methods

3.1. Study Area and Data Source

Considering that the starting year of Peking University digital finance index is 2011, this study selected 30 provincial-level panel data in China (except Tibet and Hong Kong, Macao and Taiwan) from 2011 to 2021. Among them, the original data of AGTFP are mainly from China Rural Statistical Yearbook and China Environmental Yearbook. The original data of ER mainly come from the Peking University website, China Carbon Trading Market and provincial statistical yearbooks. The digital finance index was taken from the Peking University Pratt & Whitney financial index (2011–2021). The remaining data are from China Statistical Yearbook and China Agricultural Yearbook. The missing data were supplemented by linear interpolation, and the indexes with large values were logged to reduce the heteroscedasticity.

3.2. Variable Selection

In this study, AGTFP was used as the explained variable. The input and output indicators of AGTFP were selected as follows: (1) Input variables. Based on the five-element theory model, the production factors including land, labor, energy, water resources and capital were selected as the input variables [45]. Land input was measured by the total planting area of crops; the annual number of planting employees was used to measure labor input; energy input was measured by the agricultural machinery total power; water resources input was measured by agricultural effective irrigated area; the capital input was represented by the amount of fertilizer, agricultural film and pesticide. (2) Expected output. The expected output is represented by the total output value of agriculture, forestry, animal husbandry and fishery, and the base period is 2011. (3) Undesirable output. Agricultural carbon emission and agricultural non-point source pollution were selected as non-expected outputs. Among them, agricultural non-point source pollution is the compound oxygen demand (COD), total phosphorus (TP) and total nitrogen (TN) residues in water and soil, which are measured by uniform investigation and evaluation method [46], and the entropy weight method is used to combine water pollution and soil pollution into the comprehensive index of agricultural pollution. Furthermore, greenhouse gases (GHGs) emitted in six agricultural production activities are uniformly converted into agricultural carbon emissions [47].

The core explanatory variable is ER, which is subdivided as follows: Command-based ER (CER) is made by the administrative department of legislation to directly influence the polluters to choose the behaviors beneficial to the environment, so the variable is calculated based on the number of yearly environmental policies released at the provincial level; market-incentive ER (MER) is market-based and aims to guide the emission behavior of agricultural enterprises with the help of market signals. Therefore, it is represented by whether the provincial level initiates carbon emission trading, and the value is 1 indicating the initiation of the local carbon emission trading market and 0 otherwise. Public-voluntary ER (PER) refers to the voluntary participation of enterprises in the implementation of ER, exerting influence on the government through raising public environmental awareness and putting forward constructive suggestions to strive for the government to implement improved environmental policies. Therefore, it is expressed as the ratio of provincial agricultural environmental protection investment to regional GDP in that year.

The moderating variable is digital finance (DF). Digital finance can broaden the channels for all social entities to participate in environmental governance through its coverage, use depth and convenience, so as to affect agricultural green total factor productivity. Therefore, the provincial digital finance index published by Peking University Digital Finance Center since 2011 was used. This index is the most influential, most frequently used and most widely used digital financial inclusion index in China. Many scholars have used this index to study the impact of digital finance on agricultural production and operation [48,49,50]. It contains 3 first-level, 12 s-level and 33 third-level indicators, which reflect its multi-level and diversified connotation, and can be more objective, comprehensive and actually reflect the level of financial development in various provinces.

In addition, the following control variables were selected referring to previous studies [51,52,53]: (1) agricultural structure (STR), which is measured by the ratio of the added value of planting industry to that of agriculture, forestry, animal husbandry and fishery; (2) agricultural machinery density (MAC), which is determined by the ratio of the total power of agricultural machinery to the total specific gravity area; (3) income distribution (IND), which is measured by the ratio of local urban per capita disposable income to rural per capita disposable income; (4) agricultural disaster rate (ADR), which is determined by the ratio of the affected area of agricultural production to the total sown area of crops; (5) human capital (HCA), which is measured by the per capita education years of rural labor force; (6) urbanization rate (URB), which is determined as the ratio of non-farm population to total population; and (7) rural road density (RDE), which is measured by the ratio of rural road mileage to regional land area.

Table 1. Descriptive statistics of variables.

Variable Name	Variable Symbol	Variable Meaning and Assignment	Observation	Mean	Standard
Agricultural green total factor productivity	AGTFP	AGTFP Index	330	1.011	0.049
Digital finance index	DF	Total digital finance index (logarithm)	330	5.283	0.669
Command ER	CER	Number of environmental protection policies by each province at the end of each year	330	1.446	1.574
Market-incentive ER	MER	One after the provincial carbon emission trading market is launched, otherwise zero	330	0.218	0.414
Public-voluntary ER	PER	Investment in environmental protection as a percentage of regional GDP	330	6.414	0.637
Agricultural structure	STR	Added value of plantation industry/added value of agriculture, forestry, animal husbandry and fishery	330	0.526	0.084
Agricultural machinery density	MAC	Total power of agricultural machinery per unit sown area	330	0.651	0.239
Income distribution	IND	Urban per capita disposable income/rural per capita net income	330	2.610	0.429
Agricultural disaster rate	ADR	Crop affected area/crop sown area	330	0.147	0.118
Human capital	HCA	Per capita years of education in rural areas	330	7.809	0.614
Urbanization rate	URB	Non-farm population/total population	330	0.596	0.121
Rural road density	RDE	Rural road mileage/regional land area	330	0.130	0.170

shows variable definitions and descriptive statistics.

3.3. Research Method

3.3.1. Non-Expected MinDS Super-Efficiency—MetaFrontier Malmquist Model

The existing methods for measuring AGTFP include SFA and DEA. Compared with SFA, DEA can effectively measure the efficiency of multiple input indicators and multiple output indicators and distinguish the technical efficiency level of decision-making units at the frontier, so it is widely used in the measurement of AGTFP. Meanwhile, the non-expected SBM model constructed by Tone (2003) can further avoid the radial and bias caused by angle measurement, which can effectively reflect the authenticity of efficiency evaluation [54]. But the efficiency value measured by SBM model has the feature of non-negative truncation, which cannot solve the problem of ratio between input–output target and actual value. In addition, most measurements of AGTFP are based on decision units (DMUs) with a common production frontier. However, the varied level of agricultural technology in various regions is different in reality due to the variance of resources endowment and economic development, so the efficiency measurement based on the common production frontier is prone to bias.

In view of this, O’Donnell et al. (2008) proposed the MetaFrontier production function, which divides DMU into different groups according to specific criteria and can effectively calculate the efficiency difference between the common boundary and the group boundary [55]. Furthermore, agricultural non-expected output involves a variety of pollution types. Most existing studies only use agricultural non-point source pollution or agricultural carbon emission as non-expected output to measure AGTFP [19,47], which may lead to differences in the measurement results. In this study, two pollution types of agricultural non-point source pollution and carbon emission are both included. Then, AGTFP is thought to be measured more comprehensively. Thus, this study uses the non-expected MinDS super-efficiency—MetaFrontier Malmquist model to measure AGTFP referring to previous studies [56], which can more accurately evaluate the heterogeneity of AGTFP in different provinces. The MetaFrontier method is applied to the global reference Malmquist model, and the MetaFrontier Malmquist model is decomposed into the MinDS super-efficiency—MetaFrontier Malmquist–Luenberger productivity index, as follows:

First, it is supposed that there are $n$ decision evaluation units DMU, each DMU contains $m$ input indicators$y_{rk}(r=$1, 2, …, $q_1$), and $E$ is the effective set of non-oriented, non-radial, variable RTS and non-expected SBM super-efficiency model DMUs. Based on model (1), the following is determined:

$$\begin{gathered} min\rho =\frac{1+\frac{1}{m}\sum _{i=1}^m\frac{s_i^{-}}{x_{ik}}}{1-\frac{1}{q_1+q_2}\left(\sum _{r=1}^{q_1}\frac{s_r^{+}}{y_{rk}}+\sum _{r=1}^{q_1}\frac{s_b^{+}}{y_{bk}}\right)} \\ \mathrm{s}.\mathrm{t}.{\sum }_{j=1,j\ne k}^n{x}_{ij}{\lambda }_j-s_i^{-}\le x_{ik};{\sum }_{j=1,j\ne k}^n{y}_{rj}{\lambda }_j+s_r^{-}\ge y_{rk};{\sum }_{j=1,j\ne k}^n{y}_{bj}{\lambda }_j+s_b^{-}\ge y_{bk}; \\ \sum _{j=1,j\ne k}^n{\lambda }_j=1;\lambda ,s^{-},s^{+}\ge 0; \\ i=1, 2, \dots , m; r=1, 2, \dots , q; b=1, 2, \dots , q;j=1, 2, \dots , n(j\ne k) \end{gathered}$$

(1)

In Formula (1), $\rho$ is the efficiency evaluation index, $s_i^{-}$, $s_r^{+}$, $s_b^{+}$ are input relaxation, expected output relaxation and non-expected output relaxation variables, respectively, and ${\lambda }_j$ represents DMU weight. In the super-efficient SBM model, DMU is effective when $\rho \ge$1.

Second, mixed integer linear programming is calculated to determine the non-expected MinDS super-efficiency model values of non-oriented, non-radial and variable returns to scale.

$$\begin{gathered} max\rho =\frac{\frac{1}{m}\sum _{i=1}^m(1-\frac{s_i^{-}}{x_{ik}})}{\frac{1}{s_1+s_2}\left(\sum _{r=1}^{s_1}\frac{s_r^{+}}{y_{rk}}+\sum _{b=1}^{s_2}\frac{s_b^{+}}{y_{bk}}\right)} \\ \mathrm{s}.\mathrm{t}.{\sum }_{j=1,j\ne k}^n{x}_{ij}{\lambda }_j-s_i^{-}\le x_{ik};{\sum }_{j\in E,j\ne k}^n{y}_{rj}{\lambda }_j-s_r^{+}\ge y_{rk};{\sum }_{j\in E,j\ne k}^n{\lambda }_j=1;s_i^{-},s_r^{+},s_b^{-}\ge 0; \\ -\sum _{i=1}^m{v}_i{x}_{ij}+\sum _{r=1}^{s_1}{u}_r{y}_{rj}+\sum _{b=1}^{s_2}{w}_b{y}_{bj}+d_j=0,j\in E;v_i,u_r,w_b\ge 1; \\ i=1, 2, \dots , m; r=1, 2, \dots , q; b=1, 2, \dots , q;j=1, 2, \dots , n(j\ne k) \\ {d}_j\le M{b}_j,{\lambda }_j\le M\left(1-b_j\right),b_j\in \left\{0,1\right\},d_j\ge 0,j\in E;u_{0}free \end{gathered}$$

(2)

In Formula (2), $M$ is a sufficiently large positive number, $\rho$ is an efficiency evaluation index, $s_i^{-}$, $s_r^{+}$, $s_b^{+}$ are input, expected output relaxation and non-expected output relaxation variables, respectively, and ${\lambda }_j$ is the DMU weight. According to the classification of provinces in the three major economic zones, the undesired MinDS super-efficiency of non-oriented, non-radial and variable RTS is divided into three groups. Each group takes itself as the reference set for intra-regional self-assessment, and then the technology gap ratio (TGR) is obtained. Based on the results and the global reference Malmquist model, AGTFP is calculated and MetaFrontier index is decomposed.

3.3.2. Dynamic Spatial Durbin Model

Classical linear regression model cannot analyze and identify the spatial effects of factors. In order to measure the spatial effects of ER and digital finance on AGTFP, a spatial Durbin model (SDM) was constructed with reference to Fischer et al. (2008) [57]. The specific model is set as follows:

$$\begin{gathered} \begin{array}{cc}\hfill {AGTFP}_{it}& =\rho {\displaystyle \sum _{j=1}^N}{W}_{ij}{AGTFP}_{it}+{\beta }_1{\displaystyle \sum _{j=1}^N}{W}_{ij}{ER}_{it}+{\beta }_2{\displaystyle \sum _{j=1}^N}{W}_{ij}{DF}_{it}\hfill \\ & +{\beta }_3{\displaystyle \sum _{j=1}^N}{W}_{ij}{DF}_{it}\times {ER}_{it}+{{\gamma }_{1}ER}_{it}+{\gamma }_2{DF}_{it}+\varphi {\displaystyle \sum _{j=1}^N}{W}_{ij}{X}_{it}+\mathsf{\kappa }{X}_{it}+{\alpha }_i\hfill \\ & +{\mu }_t+{\epsilon }_{it}\hfill \end{array} \\ {\epsilon }_{it}=\lambda \sum _{j=1}^N{W}_{ij}{\epsilon }_{it}+{\upsilon }_{it} \end{gathered}$$

(3)

In Formulas (3) and (4), ${AGTFP}_{it}$ is the AGTFP of province $i$ in year $t$, and $W_{ij}$is the spatial distance space weight matrix. ${ER}_{it}$is the ER index of $t$ year $i \mathrm{p}\mathrm{r}\mathrm{o}\mathrm{v}\mathrm{i}\mathrm{n}\mathrm{c}\mathrm{e},{ DF}_{it}$ is the digital finance index of $t$ year $i \mathrm{p}\mathrm{r}\mathrm{o}\mathrm{v}\mathrm{i}\mathrm{n}\mathrm{c}\mathrm{e},X_{it}$is the explanatory variable, ${\alpha }_i$is the spatial fixed effect, ${\mu }_t$ is the time fixed effect, ${\epsilon }_{it}$ represents the random error term and $\beta$, $\gamma$, $\varphi$, $\mathsf{\kappa }$ are their respective regression coefficients.

4. Results and Discussion

4.1. AGTFP Spatial Correlation Test

In order to determine whether AGTFP has spatial correlation, the Moran’I index was used to test it. Although Hainan Province is geographically located in no neighboring provinces, it is still affected by other non-neighboring regions. In this study, a geographical distance matrix was selected to reveal the evolution characteristics of AGTFP. As can be seen from

Table 2. AGTFP’s Moran’I index.

Year	Moran’I	z	Year	Moran’I	z
2011	0.127 ***	2.661	2017	0.209 ***	2.629
2012	0.191 ***	2.668	2018	0.215 ***	3.474
2013	0.202 ***	2.729	2019	0.226 ***	3.579
2014	0.210 ***	2.752	2020	0.234 ***	3.658
2015	0.231 ***	3.370	2021	0.230 ***	3.235
2016	0. 193 *	1.932

Note: * means 10% significance level; *** means 1% significance level.

, AGTFP’s Moran’I index from 2011 to 2019 was significantly positive, indicating that AGTFP among provinces does not have mutual independence, and a spatial econometric model needs to be introduced for empirical test. Meanwhile, the scatter plot of the Moran’I index for the representative year AGTFP is plotted. As shown in Figure 2, the proportion of typical provinces in high–high agglomeration and low–low agglomeration was 56.7% in 2011 and increased to 80.0% in 2021, indicating that the positive spatial self-correlation of AGTFP was strengthened.

4.2. Spatial Model Selection

In order to select a suitable spatial measurement model, further tests are needed, and the results are shown in

Table 3. Spatial panel model test.

Inspection Name	Coefficient	p-Value	Inspection Name	Coefficient	p-Value
LM-SLM	48.186	0.000	LR-SAR	30.80	0.002
LM-SEM	48.298	0.000	LR-SEM	32.11	0.003
Robust LM-SLM	0.222	0.000	Hausman	57.98	0.000
Robust LM-SEM	0.334	0.000	LR-area	102.41	0.000
Wald	33.660	0.000	LR-time	142.71	0.000

. Firstly, the Lagrange multiplier (LM) and Robust LM tests of SLM and SEM models show that both statistics pass the 1% significance test, and the tests should be continued at this time. Secondly, both Wald and LR tests passed the significance test, indicating that the SDM model cannot be degraded into an SEM model and SLM model. Finally, the result is significant at the 1% level using the Hausman model test, showing that the spatio-temporal dual fixed effect model is better. Therefore, the spatio-temporal double fixed effect space Durbin model was selected for empirical analysis.

4.3. Baseline Regression Analysis

The spatial lag model baseline regression results from

Table 4. Estimation results of spatial Durbin model.

Variable	CER	MER	PER	Variable	CER	MER	PER
	(1)	(2)	(3)		(4)	(5)	(6)
ER	0.032 *** (0.011)	0.031 ** (0.016)	0.029 ** (0.014)	W $\times$ ER	0.028 ** (0.013)	0.026 * (0.017)	0.015 * (0.008)
DF	0.026 ** (0.012)	0.019 ** (0.010)	0.011 ** (0.006)	W $\times$ DF	−0.016 ** (0.008)	−0.015 * (0.009)	−0.006 ** (0.003)
ER $\times$ DF	0.009 ** (0.005)	0.006 * (0.004)	0.004 * (0.003)	W $\times \mathrm{E}\mathrm{R} \times$ DF	0.008 ** (0.004)	0.005 * (0.003)	0.003 * (0.002)
STR	0.016 (0.011)	0.024 ** (0.012)	0.020 ** (0.010)	W $\times$ STR	0.008 (0.059)	0.053 (0.064)	0.041 ** (0.010)
MAC	−0.004 ** (0.002)	−0.006 *** (0.002)	−0.005 ** (0.002)	W $\times$ MAC	−0.020 (0.014)	−0.037 ** (0.015)	−0.050 *** (0.013)
IND	−0.001 (0.002)	−0.001 (0.002)	−0.002 (0.002)	W $\times$ IND	−0.006 (0.011)	−0.010 (0.012)	−0.005 (0.010)
ADR	−0.003 *** (0.001)	−0.002 ** (0.001)	−0.002 ** (0.001)	W $\times$ ADR	−0.005 *** (0.002)	−0.004 ** (0.002)	−0.008 ** (0.004)
HCA	0.000 (0.001)	0.000 (0.001)	0.001 * (0.000)	W $\times$ HCA	0.003 (0.006)	0.004 (0.007)	0.003 * (0.002)
URB	0.058 ** (0.025)	0.084 *** (0.028)	0.090 *** (0.025)	W $\times$ URB	0.104 (0.137)	0.134 (0.152)	0.118 *** (0.042)
RDE	0.055 *** (0.020)	0.048 * (0.029)	0.043 * (0.023)	W $\times$ RDE	0.075 *** (0.030)	0.059 (0.064)	0.054 (0.052)
Province FE	YES	YES	YES	Time FE	YES	YES	YES
$\rho$	0.282 *** (0.016)	0.259 *** (0.023)	0.234 *** (0.015)	R2	0.371	0.357	0.302

Note: * means 10% significance level; ** means 5% significance level; *** means 1% significance level; standard error in parentheses.

can be seen, and the regression coefficients of ER and their spatial lag terms are significantly positive, indicating that ER can not only increase local AGTFP but also have a promoting effect on AGTFP in neighboring regions. The regression coefficient of the digital finance index is significantly positive, and the regression coefficient of its spatial lag term is significantly negative, indicating that digital finance can improve local AGTFP but produce negative spillover effects on neighboring regions. The coefficients of the interaction terms and spatial lag terms of ER along with digital finance are significantly positive, indicating that digital finance has a significant moderating effect on the impact of ER on AGTFP.

Since the spatial lag term of some independent variables is significantly non-zero, its regression coefficient is no longer the partial derivative of the dependent variable to the independent variable, so the regression coefficient of the independent variable cannot be directly regarded as the spatial spillover effect. When there is a spatial lag term in the spatial metrology model, the total effect can be decomposed into direct effect and indirect effect by the calculus method [58], and the specific results are shown in

Table 5. Spatial Durbin decomposition effect.

Variable	CER			MER			PER
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Direct	Indirect	Total Effect	Direct	Indirect	Total Effect	Direct	Indirect	Total Effect
ER	0.036 *** (0.011)	0.014 *** (0.003)	0.050 *** (0.019)	0.031 ** (0.015)	0.008 * (0.005)	0.039 ** (0.020)	0.026 ** (0.012)	0.007 ** (0.003)	0.033 ** (0.015)
DF	0.025 ** (0.011)	−0.008 ** (0.004)	0.017 ** (0.008)	0.015 ** (0.007)	−0.005 * (0.003)	0.010 ** (0.004)	0.013 ** (0.006)	−0.004 ** (0.002)	0.009 ** (0.004)
ER $\times$ DF	0.017 ** (0.008)	0.002 * (0.001)	0.019 ** (0.009)	0.008 ** (0.004)	0.009 * (0.005)	0.017 ** (0.009)	0.005 ** (0.013)	0.003 * (0.002)	0.008 ** (0.004)
STR	0.033 *** (0.012)	0.016 ** (0.007)	0.049 * (0.031)	0.042 *** (0.016)	0.011 * (0.005)	0.053 ** (0.025)	0.039 *** (0.013)	0.007 * (0.004)	0.045 * (0.025)
MAC	−0.014 *** (0.005)	−0.001 (0.003)	−0.015 *** (0.005)	−0.036 *** (0.009)	−0.003 (0.005)	−0.039 *** (0.005)	−0.013 *** (0.005)	−0.002 (0.003)	−0.015 *** (0.004)
IND	−0.004 (0.006)	−0.002 (0.002)	−0.006 (0.006)	−0.004 (0.005)	−0.005 (0.009)	−0.009 (0.010)	−0.001 (0.003)	−0.005 (0.005)	−0.006 (0.005)
ADR	−0.001 (0.001)	−0.001 (0.003)	−0.002 (0.006)	−0.002 (0.005)	−0.002 (0.010)	−0.004 (0.010)	−0.001 (0.003)	−0.001 (0.006)	−0.002 (0.005)
HCA	0.006 * (0.003)	0.001 (0.001)	0.007 * (0.004)	0.002 ** (0.001)	0.005 (0.005)	0.007 * (0.004)	0.005 * (0.003)	0.002 (0.002)	0.007 ** (0.003)
URB	0.140 ** (0.064)	0.059 ** (0.030)	0.199 *** (0.065)	0.218 ** (0.098)	0.116 *** (0.005)	0.334 *** (0.105)	0.082 *** (0.031)	0.042 ** (0.020)	0.124 ** (0.052)
RDE	0.104 *** (0.033)	0.002 (0.013)	0.106 *** (0.034)	0.097 * (0.061)	0.007 (0.005)	0.104 ** (0.050)	0.063 * (0.038)	0.006 (0.013)	0.069 * (0.041)

Note: * means 10% significance level; ** means 5% significance level; *** means 1% significance level; standard error in parentheses.

.

4.3.1. Analysis of the Effect of ER on AGTFP

Table 5 shows that the direct effects, indirect effects and total effect coefficients of command-based, market-incentive and public-voluntary ER are significantly positive. ER encourages agricultural producers to carry out green technology innovation to improve local AGTFP through external pressure and internal incentives and can indirectly improve AGTFP in neighboring regions through the knowledge spillover effect. This finding is consistent with a previous study [19]. Instead of using only a single ER indicator, we examined the different effects of heterogeneous ER on AGTFP. It was found that the total effect coefficient of the command-based ER is 0.050, and it is significantly positive at the 1% level. The ER coefficients of market-incentive and public-voluntary ER were 0.039 and 0.033, respectively, which were significantly positive at the 5% level. This indicates that China is still dominated by command-based ER, and its pollution control costs force agricultural producers to carry out green technology innovation to produce an innovation compensation effect, thus offsetting the initial investment cost [25,59]. Based on the principle that the polluter pays and other market mechanisms, market-incentive ER transforms the social cost of pollution into the internal cost of agricultural producers and promotes their choice of green technological innovation. However, there are only eight local carbon trading pilots in China at present, such as Beijing, Shanghai and Shenzhen, and its impact on AGTFP is still in the early stage. The improvement of public awareness of environmental protection encourages citizens to participate more in environmental governance, which has a positive promoting effect on AGTFP. Although the social responsibility of agricultural producers has been improved, its ultimate goal is still to maximize profits, so the impact of voluntary ER on AGTFP is still in the development stage.

4.3.2. Moderating Effects of Digital Finance on ER affecting AGTFP

As can be seen from Table 5, the direct and total effects of the digital finance index on agricultural green total factor growth rate are significantly positive, while the indirect effects are significantly negative. The reason is that digital finance has the advantages of low threshold, low cost and wide coverage, which can realize the accurate portrait of the financing characteristics of agricultural producers and the efficient matching of capital supply and demand, so as to effectively alleviate the financing constraints of agricultural producers, provide them with sufficient financial support and convenient financial services and accelerate the spatial flow of agricultural green technology innovation elements. Meanwhile, digital inclusive finance can promote the rational allocation of resources, guide agricultural production towards the direction of green development, help to give full play to the resource guidance function of finance, promote capital, technology and talent flow to those regions with perfect green industry layout and rapid development of green agriculture and then improve the environmental awareness of agricultural producers to improve AGTFP. While the development of digital finance increases local AGTFP, it could easily lead to a new digital divide and digital inequality, resulting in a large outflow of capital, technology and labor resources in neighboring regions [60], resulting in a siphon effect and hindering the development of AGTFP in neighboring regions.

By comparison, it was found that the total effect of the coefficient of digital finance on the AGTFP of command-based, market-incentive and public-voluntary ER is 0.019, 0.017 and 0.008, respectively. The reason is that digital finance can provide financial support for agricultural producers and then promote green technology innovation to promote the innovation compensation effect of command-based ER and achieve positive interaction. Meanwhile, digital finance can reduce information asymmetry to reduce the cost of market-incentive ER, but the domestic digital finance policy for agricultural carbon trading has not yet been perfected, and its impact on AGTFP is still in the early stage. Public-voluntary ER can broaden the channels for social subjects to participate in environmental governance through digital finance to enhance their environmental awareness. There are clear regulations on the content, frequency and method of information disclosure at present, resulting in limited public access to environmental quality information. There are both similarities and differences between this study and Hong’s [61], where they emphasize the promotional effect of digital finance on AGTFP. On this basis, this study analyzes the moderating effect of digital finance on ER affecting AGTFP. In addition, the spatial feedback effects of the interaction terms of command-based, market-incentive and public-voluntary ER and digital finance are 0.008, 0.002 and 0.001, respectively, indicating that the impact of local ER and digital financial development on AGTFP is transmitted to neighboring regions and then positively fed back to local AGTFP.

In terms of control variables, agricultural structure has a positive and significant impact on the total effect of AGTFP in local and neighboring regions under the three types of ER, indicating that the planting proportion is conducive to optimizing factor allocation, thereby increasing agricultural ecological benefits to increase AGTFP, which is in accordance with a previous study [62]. Agricultural machinery density has a significant negative impact on the total effect of local AGTFP under the three types of ER, because agricultural machinery density may increase greenhouse gas emissions while improving agricultural technical efficiency, and thus reduce AGTFP. Human capital has a positive and significant impact on the total effect of local AGTFP, indicating that the high education level of agricultural producers can help improve their production skills and then reduce agricultural pollution emissions to increase AGTFP. Urbanization rate has a positive and significant impact on the total AGTFP effect of the local and neighboring areas, indicating that a high urbanization rate can effectively provide economic and technical support for rural areas and then increase AGTFP. Rural road density has a positive and significant impact on the total effect of local AGTFP, indicating that the higher the rural road density, the higher the agricultural and non-agricultural correlation, providing financial and technical support to improve AGTFP.

4.4. Robustness Test

In order to ensure the reliability and robustness of the regression results, the robustness test was carried out in the following two ways. First, the spatial weight matrix was replaced by the economic geography nested matrix to re-evaluate the impact of ER and digital finance on AGTFP [63]. The second was to replace the measure method of explained variable and re-measure AGTFP with non-radial directional distance function (NDDF) [64].

Table 6. Robustness test.

Variable	Economic Matrix			Change the Explained Variable
	(1)	(2)	(3)	(4)	(5)	(6)
	CER	MER	PER	CER	MER	PER
ER	0.019 ** (0.005)	0.016 ** (0.007)	0.015 ** (0.001)	0.011 ** (0.006)	0.010 ** (0.003)	0.008 * (0.005)
DF	0.016 ** (0.009)	0.011 * (0.006)	0.007 ** (0.004)	0.009 ** (0.004)	0.008 ** (0.004)	0.004 * (0.003)
ER $\times$ DF	0.009 ** (0.004)	0.008 ** (0.004)	0.005 * (0.003)	0.010 ** (0.005)	0.008 * (0.005)	0.006 * (0.004)
$\rho$	0.252 *** (0.017)	0.220 *** (0.025)	0.205 *** (0.018)	0.223 *** (0.014)	0.211 *** (0.027)	0.202 *** (0.016)
Control variables	YES	YES	YES	YES	YES	YES
Province FE	YES	YES	YES	YES	YES	YES
Time FE	YES	YES	YES	YES	YES	YES
R2	0.332	0.315	0.307	0.309	0.302	0.287

Note: * means 10% significance level; ** means 5% significance level; *** means 1% significance level; standard error in parentheses.

shows that the total effects of ER, digital finance and their interaction terms are significantly positive, which proves that the conclusions in this study are robust.

4.5. Heterogeneity Analysis

Due to geographical environment, economic foundation and resource endowment, ER intensity and digital financial conditions are different in varied regions of China, so the impact on AGTFP may also have regional heterogeneity. Therefore, this study divides the sample provinces into three regions according to geographical location: eastern, central and western, and the grouping regression results are shown in

Table 7. Heterogeneity analysis.

Variable	East			Middle			West
	CER	MER	PER	CER	MER	PER	CER	MER	PER
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
ER	0.024 ** (0.012)	0.062 *** (0.018)	0.015 ** (0.007)	0.006 ** (0.003)	0.001 (0.033)	0.002 ** (0.001)	0.004 ** (0.001)	——	0.015 (0.019)
DF	0.020 ** (0.009)	0.038 ** (0.008)	0.014 * (0.008)	0.005 ** (0.002)	0.002 (0.002)	0.002 ** (0.001)	0.003 * (0.002)	0.006 (0.013)	0.002 (0.023)
ER $\times$ DF	0.008 ** (0.003)	0.021 ** (0.010)	0.002 * (0.001)	0.003 * (0.002)	0.000 (0.005)	0.001 * (0.000)	0.001 * (0.001)	——	0.003 (0.002)
$\rho$	0.219 *** (0.021)	0.241 *** (0.018)	0.217 *** (0.016)	0.230 *** (0.018)	0.196 *** (0.022)	0.211 *** (0.017)	0.163 *** (0.025)	0.132 *** (0.023)	0.115 ** (0.057)
Control variables	YES	YES	YES	YES	YES	YES	YES	YES	YES
Province FE	YES	YES	YES	YES	YES	YES	YES	YES	YES
Time FE	YES	YES	YES	YES	YES	YES	YES	YES	YES
R2	0.268	0.334	0.238	0.298	0.237	0.256	0.259	0.100	0.151

Note: * means 10% significance level; ** means 5% significance level; *** means 1% significance level; standard error in parentheses.

. In eastern China, the effects of the digital financial index and interaction terms on AGTFP are significantly positive concerning the total effect of command-based, market-incentive and public-voluntary ER. It was found that market-incentive ER and its interaction with the digital finance index have a higher impact on AGTFP. The reason is that there are seven local carbon trading pilots in the eastern region, accounting for 87.50% of the total number of local carbon trading pilots. Meanwhile, the development level of digital finance in the eastern region is higher, which is more conducive to promoting green technology innovation to improve AGTFP. The effects of the digital finance index and interaction terms on AGTFP in central China are significantly positive considering the total effect of command-based and public-voluntary ER. Since only Hubei province in the central region has a local carbon trading pilot, the impact on AGTFP is limited. Furthermore, the economic level and the development of digital finance in the central region show a catch-up trend, and the gap between the central region and the eastern region is narrowing significantly, which promotes the gradual improvement of public awareness of environmental protection and therefore presents an ER model dominated by command and supplemented by the public volunteering. In the western region, the effects of the digital financial index and interaction terms on AGTFP are significantly positive only for the total effect of command-based ER. Compared with eastern and central China, western China has a weaker economic development foundation without a carbon trading pilot, resulting in a lower development level of carbon trading market, public awareness of environmental protection and digital finance. Therefore, command-based ER is still the main way to promote agriculture green development.

5. Conclusions and Policy Enlightenment

Under the background of digital financial development, it is important to explore its agricultural green value and improve ER for the sustainable development of agriculture. Based on the panel data of 30 provincial levels in China from 2011 to 2021, this study used the SDM model to explore the spatial effects of heterogeneous ER and digital finance on AGTFP. The findings are as follows:

(1)

Command-based, market-incentive and public-voluntary ER can increase local AGTFP and have positive spatial spillover effect. Command-based ER has the highest effect on AGTFP, followed by market-incentive and public-voluntary ones.

(2)

Digital finance has a direct promotional effect on local AGTFP, and it has an inhibitory effect on AGTFP in neighboring regions due to the siphon effect.

(3)

Digital finance is an important moderating variable of the three ERs affecting AGTFP.

(4)

The heterogeneous analysis found that digital finance had a significant moderating effect on the impact of command-based, market-incentive and public-voluntary ER on AGTFP with the market-incentive being highest in eastern China. Digital finance has a significant moderating effect on the impact of command-based and public-voluntary ER on AGTFP with the command-based being higher in central China. In western China, digital finance only plays a significant moderating role in the impact of command-based regulation on AGTFP.

We further proposed policy implications based on the results and conclusions in this study:

(1) Develop diversified ER policies. The government should strengthen the implementation of policy and implement the environmental investment policy to play a positive role in command-based ER. Meanwhile, an effective mechanism to realize the value of agricultural carbon sinks has been developed for the agricultural carbon trading market to improve the market incentive ER. In addition, it is necessary to improve the government information disclosure system and the role of the media reputation mechanism to promote the promotion of public-voluntary ER for AGTFP. In addition, it is necessary to scientifically use diversified policy tools, dynamically adjust the intensity of ER, apply the right medicine, appropriately loosen restrictions, pay attention to the coordinated operation of ER and other policies and give more play to market forces to avoid excessive government intervention in the economic activities of agricultural enterprises.

(2) Improve digital finance according to heterogeneous ER. The government should expand the coverage of digital finance and give play to the advantages of digital finance inclusion to ease the financial constraints of agricultural producers. Carbon financial derivatives compatible with agricultural carbon trading should be developed in a timely manner, and the management ability of digital financial agricultural carbon sink loans needs to be improved. At the same time, the channels for obtaining digital financial information should be broadened, and the public’s awareness of environmental protection should be enhanced to enhance AGTFP.

(3) Implement differentiated ER policies according to local conditions. Varied ER policies should be implemented according to the level of economic quality growth in different regions. In eastern China, market-incentive ER is the main one, and the command-based and public-voluntary ones are the auxiliary, so the market-incentive role should be given full play to promote the development of agriculture by green technology. In central China, command-based ER is the main one, and the public-voluntary one is secondary. Environmental laws and regulations should be further improved, and systems such as the content, frequency and mode of environmental information disclosure should be clarified to increase the public’s access to environmental quality information. The western region should strengthen the top-level design and perfect the environmental tax and subsidy system to standardize the command-based ER policy. At the same time, each region should constantly improve and adjust the intensity of ER according to its own actual development, so as to avoid the loss of production efficiency caused by blind pursuit of governance effects. While guiding the green development of agriculture in various regions through scientific and reasonable ER means, resource allocation efficiency and technology utilization efficiency are further improved, so as to play the role of heterogeneous ER in promoting agricultural green total factor productivity.