A
The Impact of New-generation Artificial Intelligence Policies on the Green Innovation Efficiency of FirmsAbstract
This paper takes A-share listed companies from 2010 to 2023 as samples and adopts the multi-period difference-in-differences (DID) method to examine the impact of the new-generation artificial intelligence pilot zone policy on corporate green innovation efficiency. The results show that this policy significantly improves corporate green innovation efficiency, and exhibits significant positive effects in both green R&D efficiency and green achievement transformation efficiency. Mechanism tests indicate that the policy effectively promotes the improvement of corporate green innovation efficiency by alleviating financing constraints and increasing analysts' and media attention. The analysis of moderating effects reveals that public environmental attention and corporate information transparency can strengthen the promoting effect of the policy on green R&D efficiency, while the level of human capital can enhance the promoting effect of the policy on green achievement transformation efficiency. Heterogeneity analysis shows that the policy has a more significant effect on enterprises located in first-tier cities, non-resource-based enterprises, non-polluting enterprises, enterprises in the growth stage, as well as technology-intensive and labor-intensive enterprises. This study provides empirical evidence and policy implications for promoting the synergetic development of artificial intelligence policies and green innovation efficiency.
Key words:
New-generation Artificial Intelligence Pilot Zone Policy
Green R&D Efficiency
Green Achievement Transformation Efficiency
Difference-in-Differences Method
A
1. Introduction
Green innovation is a key driving force for achieving coordinated development of the economy and the environment, promoting sustainable development of enterprises and green upgrading of industries[1]. China attaches great importance to the development of green innovation. The report of the 20th National Congress of the Communist Party of China in 2022 proposed that promoting the green and low-carbon development of the economy and society is a key link in achieving high-quality development. At present, the key technology system that supports Chinese enterprises in enhancing their capabilities at different stages of green innovation, including technology research and development and application, is still relatively weak[2]. According to the data disclosed by the National Intellectual Property Administration: From 2016 to 2023, the proportion of green and low-carbon patents authorized in the total number of invention patents authorized in China has remained relatively low and shown a downward trend, dropping from 6.2% in 2016 to 5.4% in 2023. This data indicates that many domestic enterprises have failed to fully take into account resource and environmental constraints in their technological research and development processes, and their pace of green transformation has significantly lagged behind the overall technological innovation process. Therefore, manufacturing enterprises that urgently need green transformation should give priority to using energy-saving materials in the R&D process, accelerate the development of green and environmentally friendly technologies, and reduce energy consumption and strictly control pollutant emissions in the production process. However, compared with traditional innovation, green innovation has the risk characteristics of large capital demand, long profit cycle and high uncertainty of returns [3, 4], which leads to insufficient motivation for enterprises to carry out green innovation [5]. Therefore, in order to better improve the green technology innovation system, enterprises need to constantly break through technological bottlenecks and take enhancing the efficiency of green innovation as an important direction for green development.
As a disruptive technology, artificial intelligence has the potential to drive social progress and restructure industrial paradigms, and has gradually become the core engine of the global technological revolution. At present, countries around the world are accelerating the development of the artificial intelligence industry by formulating national strategies [6]. China has also established artificial intelligence as a national strategic high ground, driving the dual-wheel interaction of technological innovation and industrial upgrading through a policy coordination system between the central and local governments, and accelerating the construction of a new model of artificial intelligence development with global competitiveness. In 2019, the Ministry of Science and Technology of China issued the "Guidelines for the Construction of National New Generation Artificial Intelligence Innovation and Development Pilot Zones", officially launching the policy for artificial intelligence pilot zones and incorporating AI into the national strategic level for overall planning. This policy relies on cities with solid digital foundations and obvious industrial advantages. Through institutional innovation, platform construction and scenario demonstration, it creates a policy environment conducive to the research and development and application of AI technology. By the end of 2021, 18 cities had been approved to build national new-generation artificial intelligence innovation and development pilot zones, basically forming a spatial pattern of coordinated development among the eastern, central and western regions and the northeastern region. The policy focuses on supporting the integration of industry, academia and research, the research and development of common technologies, the construction of computing power infrastructure, and typical scenarios such as "AI + manufacturing" and "AI + urban governance", promoting AI to empower industrial upgrading and modernization of governance, and providing institutional guarantees and technical support for the intelligent transformation and green development of enterprises.
This paper takes A-share listed companies from 2010 to 2023 as the research object. Based on the policy background of the construction of the new generation of artificial intelligence pilot zones, it adopts the multi-period difference-in-differences (DID) method to systematically analyze the impact of this policy on the green innovation efficiency of enterprises. In terms of research design, this paper divides the efficiency of green innovation into two parts: green R&D efficiency and green technology transfer efficiency, thereby more accurately depicting the input-output performance of enterprises in the process of green innovation. Meanwhile, the internal transmission path of policy effects is explored through mechanism tests, with a focus on mediating variables such as the alleviation of financing constraints, the increase in analysts' attention, and the enhancement of media attention. By introducing external and internal conditions such as public environmental attention, enterprise information transparency and human capital level through moderating effect analysis, the differences in their impacts on policy effects are revealed. Heterogeneity analysis is carried out in combination with enterprise characteristics, industry characteristics and regional characteristics to achieve multi-dimensional characterization of policy effects.
The marginal contributions of this article are mainly reflected in four aspects: First, by choosing the policies of the new generation of artificial intelligence pilot zones as the research context and combining artificial intelligence with green innovation efficiency, it enriches the policy-driven research perspective on green innovation efficiency. Second, the efficiency of green innovation is subdivided into R&D efficiency and the efficiency of technology transfer. This not only examines the investment stage of enterprises' green innovation but also pays attention to the stage of technology transfer and implementation, thus expanding the measurement dimensions of green innovation research. Thirdly, starting from intermediary mechanisms such as financing constraints and information attention, and introducing regulatory factors such as environmental attention, information transparency and human capital level, it reveals the multi-level transmission chain of policy influence. Fourth, through the heterogeneity analysis of regional, industry and enterprise characteristics, targeted policy optimization suggestions were put forward, providing more operational empirical evidence for the coordinated development of artificial intelligence and green innovation.
2. literature review
Green innovation efficiency is a comprehensive indicator reflecting the core competitiveness of an enterprise in green innovation. It systematically measures the efficiency of innovation input and output on the basis of considering environmental impact[7]. The research on the evaluation of green innovation efficiency constructs an index system based on the dual attributes of the environment and innovation[8, 9]. At present, the measurement methods for green innovation efficiency mainly include: (1) the integration method of innovation elements[10, 11]. (2) Stochastic Frontier Analysis (SFA)[12]. (3) Data Envelopation Analysis (DEA)[13–15]. In this paper, DEA will be selected to measure the efficiency of green innovation, and the efficiency of green innovation will be divided into green R&D efficiency and green technology transfer efficiency.
The influencing factors of green innovation efficiency are mainly studied from four perspectives: environmental regulation, digital economy, industrial agglomeration, and industrial structure. Firstly, from the perspective of environmental regulations, Chen Bin et al. (2020)[16] hold that strict environmental regulations and a favorable market environment can encourage enterprises and research institutions to carry out scientific and technological innovation activities, thereby reducing pollution emissions and enhancing the efficiency of green innovation. Li Chuang et al. (2023)[17] pointed out that carbon emission rights trading under market-incentive-driven environmental regulations can significantly enhance the green technological innovation capabilities of enterprises, and this promoting effect is more pronounced in state-owned enterprises. Secondly, from the perspective of the digital economy, Lu Yanwei et al. (2022)[18] pointed out that digital finance promotes the improvement of green innovation efficiency by facilitating financial development, stimulating consumer demand, and enhancing human capital. Xin Daleng (2023)[19] and Yu Zhihui (2023)[20] hold that the construction of information infrastructure will achieve the optimal allocation of urban innovation resources through three major effects: financial development, talent aggregation, and informatization improvement, thereby enhancing the level and efficiency of urban green innovation. Furthermore, from the perspective of industrial agglomeration, different types of industrial agglomeration have different impacts on innovation efficiency. Wu Chuanqing et al. (2019)[21] analyzed using the Super-SBM model that the impact of equipment manufacturing agglomeration on green innovation efficiency shows a significant N-shaped relationship and has a dual threshold effect. Finally, from the perspective of industrial structure, Dong Huizhong et al. (2021)[22] pointed out that industrial structure has an inhibitory effect on the improvement of green innovation efficiency and can promote the development of green innovation efficiency in neighboring regions. Huang Huan et al. (2024)[23] found that the upgrading of industrial structure can accelerate the flow of green innovation elements, promote the aggregation of talents, and deepen green innovation cooperation among enterprises to enhance the efficiency of green innovation.
A
The economic effects of artificial intelligence are mainly studied from four perspectives: income distribution, technological innovation, employment, and the integration of intelligence and reality. Firstly, from the perspective of income distribution, Bessen (2018)[24], Jiang Yonghong et al. (2021)[25], and scholars at home and abroad unanimously believe that artificial intelligence, as an important force leading the new generation of technological revolution, will have an impact on income distribution. Secondly, from the perspective of technological innovation, Deng Yue and Jiang Wanyi (2022)[26] pointed out that enterprises adopting industrial robots promote technological innovation by enhancing management efficiency and digital capabilities and optimizing the structure of human capital. The research results of Ren Yinghua et al. (2023)[27] and Yang Jin (2023)[28] indicate that disruptive AI innovation promotes the efficient and rational operation of production and business, and accelerates the intelligent transformation of enterprises. Furthermore, from the perspective of employment patterns, Wu Qiang (2023)[29] pointed out that artificial intelligence replacing low-skilled procedural work in production significantly affects workers of different occupations, wage levels, ages, and educational backgrounds. Sun Zao and Hou Yulin (2022) [30] found that in the southeast coastal areas of China, the excessively high cost of living has a crowding-out effect on low-educated workers, while industrial intelligence has intensified the replacement of low-educated labor by equipment, highlighting the problem of regional imbalance. Finally, from the perspective of the integration of intelligence and the real economy, Ren Baoping and Wang Xin (2024)[31] proposed that the formation of new quality productivity can be achieved by promoting the innovation of core artificial intelligence technologies, implementing the digital and intelligent transformation of the real economy, supporting the digital transformation of enterprises, and building a digital and intelligent ecosystem. Fan Dezhi and Yu Shui (2024)[32] believe that artificial intelligence can promote the real economy to move towards high-quality development by enhancing production efficiency, driving industrial transformation, and innovating business models.To sum up, the existing relevant literature mainly conducts research on the influencing factors of green innovation efficiency from four perspectives: environmental regulations, digital economy, industrial agglomeration, and industrial structure. Research on the economic effects of artificial intelligence mainly focuses on income distribution, technological innovation, employment, and the integration of intelligence and real economy. However, there are obvious deficiencies in the cross-research between the two: neither has the correlation between the policies of the new generation of artificial intelligence pilot zones and the efficiency of green innovation been incorporated into the analytical framework, nor has the mechanism of action and boundary conditions been deeply explored, and there is a lack of systematic examination of the policy effects in differentiated scenarios. It can be seen from this that the internal path of how the policies of the new generation of artificial intelligence pilot zones act on the efficiency of green innovation through mechanisms such as alleviating financing constraints and enhancing information attention has not yet been fully tested. Meanwhile, the moderating effects of external environmental attention and internal governance levels on policy outcomes have rarely been explored. More importantly, there is a lack of systematic comparison of the differentiated manifestations of policy effects under different regional, industry and enterprise characteristics. This article makes up for the above deficiencies from aspects such as policy context, efficiency multi-dimensional measurement, mechanism and regulatory effects, as well as heterogeneity differences.
3. Theoretical analysis and research hypotheses
The policies of the new generation of artificial intelligence pilot zones can enhance the efficiency of enterprises' green innovation through multiple channels: First, it can alleviate the financing constraints on enterprises, broaden the sources of funds for green research and development, and reduce the financial pressure on technology development and transformation. Second, enhance media attention, strengthen external supervision and the effect of information dissemination, and encourage enterprises to invest more resources in green innovation. Third, enhance the attention of analysts. Through professional interpretation and expectation guidance in the capital market, improve the financing environment for enterprises and investor confidence, thereby forming a synergy in terms of funds, reputation and technology, and promoting the continuous improvement of green innovation efficiency.
From the perspective of financing constraints, on the one hand, the policies of the experimental zone effectively reduce the financing pressure on enterprises by guiding market capital towards the field of artificial intelligence and optimizing the financing environment [33]. Meanwhile, enterprises in the experimental zone can obtain the credit default status and supply and demand data of their upstream and downstream partners in a timely manner by leveraging artificial intelligence technology, thus more conveniently alleviating their financial pressure through commercial credit. Furthermore, the easing effect of artificial intelligence policies on financing constraints has significantly reduced the financial asset allocation level of manufacturing enterprises[34]. On the other hand, the high financing constraints faced by green technology innovation activities will restrict enterprises' continuous R&D investment, thereby affecting the level and efficiency of green innovation[35]. Therefore, alleviating green financing constraints and improving the operating conditions of enterprises are the key channels for digital empowerment of enterprises' green innovation and development.
From the perspective of media attention, on the one hand, the media plays a core role in the information transmission of the capital market, constituting the main channel for communication and interaction between enterprises and external investors. At the same time, it has the motivation and ability to deeply explore enterprise information. This dual attribute enables it to simultaneously enhance public attention[36] and the transparency and credibility of enterprises[37]. Therefore, under the impetus of the pilot policies, the news media have flourished, and at the same time, through the role of information transmission, they have promoted the implementation of the policies in the pilot cities[38]. On the other hand, the increase in media attention will strengthen external supervision[39], alleviate the information asymmetry between enterprises and external stakeholders[40], promote enterprises to enhance the transparency and quality of environmental information disclosure, and at the same time help them build a good reputation and enhance public recognition, thereby promoting green innovation of enterprises[41, 42].
From the perspective of analysts' attention, on the one hand, the establishment of the pilot zone has attracted a great deal of market attention and resource aggregation. External market attention can to a certain extent enhance the external supervision of listed companies[43]. External analysts conduct a comprehensive assessment of the operating conditions and industry trends of enterprises, helping them better seize market opportunities. Meanwhile, the evaluations of analysts also help enterprises establish a good market image and enhance the confidence of investors and consumers[44]. On the other hand, analysts play a significant role as information providers and external supervisors in the capital market. With their professional knowledge, they continuously provide various aspects of information about listed companies to participants in the capital market, enabling investors to understand the value of long-term risky investment in enterprises and alleviate information asymmetry in innovation activities. This will further reduce financing costs, increase innovation investment and enhance innovation capabilities[45], which is conducive to optimizing innovation efficiency.
In conclusion, this paper proposes Hypothesis 1 and Hypothesis 2:
H1: The policies of the new generation of artificial intelligence pilot zones can enhance the efficiency of enterprises' green innovation
H2: The policies of the new generation of artificial intelligence pilot zones can enhance the green innovation efficiency of enterprises by reducing the financing constraints on enterprises, increasing media attention, and raising the attention of analysts
(2) The moderating effects of public environmental attention, human capital level and enterprise information transparency
With the increase in public attention to the environment, the power of social public opinion supervision will be significantly enhanced[46]. To avoid negative public opinion, enterprises will strictly follow the artificial intelligence policies, precisely invest in the innovative projects encouraged by the policies, and thereby enhance the positive effect of the policies on the innovation efficiency of enterprises. Secondly, the increase in public attention to the environment will change consumers' consumption preferences and choices[47]. When the public's attention to enterprises' green innovation in artificial intelligence and the improvement of product quality rises, in order to meet market demands, enterprises will be more proactive in leveraging artificial intelligence policy support to accelerate innovation, making the policy's promoting effect on enterprises' innovation efficiency more significant.
Human capital accumulation includes the improvement of employees' innovative awareness[48] and knowledge reserves[49]. Employees with a high level of human capital are more likely to integrate the policy orientation of artificial intelligence, explore green innovation directions, transform policy incentives into actual innovation actions, and amplify the policy's promotion of innovation efficiency. At the same time, the implementation of artificial intelligence policies relies on professional talents and a high level of human capital. This means that enterprise employees have a stronger understanding and application ability of artificial intelligence technology, can quickly integrate the intelligent technologies supported by policies into the green innovation process, avoid the "mismatch" between technology and human resources, and effectively release the policy's effectiveness.
The improvement of enterprise information transparency can alleviate information asymmetry within enterprises[50]. The implementation of policies requires smooth two-way communication: enterprises should understand the policies and departments should be aware of the progress. Transparency helps to precisely allocate subsidies and tax incentives, avoiding misallocation. Enterprises can also promptly align with policies, optimize innovation paths, and efficiently transform policies into green driving forces. In addition, information transparency enhances external supervision and market trust, allowing investors and consumers to clearly understand the green investment and effectiveness of enterprises. To safeguard their reputation, enterprises proactively invest policy dividends in green technologies, transparently reduce investment risks, attract funds and drive traffic, and amplify the policy's promotion of green innovation efficiency.
In conclusion, this paper proposes Hypothesis 3:
H3: Public environmental attention, human capital level and enterprise information transparency can enhance the positive impact of the policies of the new generation of artificial intelligence pilot zones on the green innovation efficiency of enterprises
4. Research Design
4.1 Model Settings
4.1.1Benchmark regression model
To examine the impact of policies in the new generation of artificial intelligence pilot zones on the green innovation efficiency of enterprises, the following regression model is constructed:
In the above formula, i represents the enterprise, t represents the year, and the explained variable is the enterprise's green innovation efficiency, which is specifically expressed by the enterprise's green technology research and development (GTE) and the enterprise's green technology transfer efficiency (GAE). The explained variable is the policy of the new generation artificial intelligence pilot zone (inter).
are the relevant control variables.
represents the fixed effect of the enterprise,
represents the time effect, and
is a random disturbance term.
4.1.2 Mediating effect model
To explore the mechanism by which the policies of the new generation of artificial intelligence pilot zones affect the green innovation efficiency of enterprises, this paper adopts the mediation effect analysis method. Given that the traditional stepwise regression method with mediating effects has the problem of endogenous estimation bias (Jiang Ting, 2022), this paper examines the influence of the core explanatory variable on the mediating variable by replacing the explained variable with the mediating variable, in order to test the existence and effectiveness of the mediating path. The mediating effect model is set as follows:
As mediating variables, WW represents financing constraints, Analyst represents analyst attention, and Media represents media attention. The remaining variables are consistent with equations (1) and (2).
4.1.3 Moderating effect model
To examine the influence of each moderating variable on the results, the following model was constructed for in-depth discussion:
To adjust the variables, specifically including public environmental attention (PEC), human capital level (HC), and information transparency (TRANS), the remaining variables are consistent with those in Equations (1) and (2).
4.2 Variable selection
4.2.1 The explained variable
The green innovation efficiency of enterprises: Referring to the practice of Xiao Renqiao et al. (2022)[51], the DEA-SBM model is used to measure the green technology transformation and achievement transformation of enterprises. In the stage of green technology research and development, the number of R&D personnel and R&D expenditure of the enterprise are taken as the initial investment, and the number of green patent applications and authorizations of the enterprise are taken as the intermediate output to calculate the green R&D efficiency (GTE) of the enterprise. During the application stage of green achievements, the number of green patent applications and authorizations is taken as intermediate inputs, and sales revenue, pollution emission index and energy consumption index are regarded as final output indicators to calculate the green achievement transformation efficiency (GAE) of enterprises.
4.2.2 Explanatory variable
The policy of the New Generation Artificial Intelligence Pilot Zone (inter), inter is equal to Treat×Post, where Treat is a dummy variable for policy grouping. The enterprises matched by the pilot cities of the new generation artificial intelligence pilot zone are set as 1, and the other enterprises are set as 0. Post is a policy time dummy variable. The year when a city is approved as a pilot for the new generation of artificial intelligence experimental zone and subsequent years are set to 1, while other years are set to 0.
4.2.3 Mediating variable
Financing constraints (WW) It is expressed by the WW index. Based on the relevant research on financial distress prediction and financing constraint measurement, by integrating multiple financial ratios, the degree of financing restrictions faced by enterprises is comprehensively evaluated, providing an effective quantitative tool for analyzing the financing environment of enterprises[52]. The specific formula is: WW = -0.091 × (operating cash flow/total assets) − 0.062 × Dummy variable of dividend payment + 0.021 × (long-term liabilities/total assets) − 0.044 × (natural logarithm of total assets) + 0.102 × Industry sales growth rate − 0.035 × Growth rate of sales revenue.
Analyst Attention (Analyst). Analyst attention reflects the degree to which securities analysts track and study listed companies. It can not only provide incremental information for the market and improve information transparency, but also form external supervision over enterprises. Referring to the research of Fang Qiaoling et al. (2024)[53], the natural logarithm of the number of analysts tracking the company + 1 was used to measure the attention paid by analysts to the enterprise. The larger this value, the higher the attention paid by analysts to the enterprise. Analysts' attention may have a dual effect: on the one hand, it can improve the quality of accounting information through information interpretation and restrain earnings management; On the other hand, it may induce short-sighted behavior by the management and even form a conspiracy relationship, which is detrimental to the long-term development of the enterprise.
Media attention (Media). It is expressed as the logarithm of the number of media reports plus one. This measurement indicator mainly comes from media effects or relevant reports of public opinion. Numerical processing can eliminate the skewness of data distribution, making the results easier to interpret and analyze[54]. Specifically, the original data of this indicator is derived from various reports about the enterprise by the media, covering the frequency of the enterprise being mentioned in public channels such as newspapers, periodicals, online news platforms, and industry information websites. It includes both positive and neutral as well as negative reports, comprehensively reflecting the overall popularity of the enterprise in terms of media attention.
4.2.4 Moderating variable
Public Environmental Concern (PEC) : This article uses Baidu's Haze search Index to measure the degree of public concern for environmental issues. This indicator captures public search behavior through big data technology, providing a new perspective for assessing environmental awareness. With the increasingly prominent problem of environmental pollution, the public's attention to environmental issues has continued to rise[55]. This indicator can dynamically reflect the sensitivity of people in different periods and regions to environmental quality, providing an important reference for the study of environmental governance.
Human capital level (HC) Human capital usually refers to the sum total of knowledge, skills, abilities and qualities that are condensed in workers through education, training and learning and can create economic and social value. It is the main factor influencing economic growth[56]. The level of human capital defined by the educational attainment of workers = the number of personnel with a university degree or above/the number of employees.
Corporate Information Transparency (TRANS). As the convergence point of a series of contracts, the contract design of enterprises is usually based on accounting figures, such as the evaluation of company operating performance and the design of managers' compensation, etc. However, due to the existence of bounded rationality and transaction costs, a contract cannot be complete. Due to the incompleteness of corporate contracts and information asymmetry, management may manage earnings through the option of accounting policies. Therefore, the quality of earnings disclosure directly affects the transparency of accounting information. This paper refers to the research of Xin Qingquan et al. (2014)[57] and adopts earnings aggressiveness and earnings smoothness as the measurement indicators of information transparency.
4.2.5 Control variables
Referring to previous studies, the following variables that may affect the green innovation resilience of enterprises are selected: CASH holdings (CASH) : expressed as the net cash flows generated from operating activities of the enterprise divided by total assets; Enterprise SIZE: It is expressed as the logarithm of the total assets of the enterprise at the end of the year. Debt-to-asset ratio (LEV) : It is expressed by dividing a company's total liabilities at the end of the year by its total assets at the end of the year. Profit on total assets (ROA) : It is expressed by dividing net profit by total assets. Whether the enterprise is a DUAL position: If the chairman and general manager of the enterprise are the same person, it is 1; otherwise, it is 0. Enterprise AGE (AGE) : It is equal to the logarithm of the year of the current year minus the year of the company's establishment plus 1. The shareholding ratio of the largest shareholder (TOP) : It is expressed by dividing the shareholding of the largest shareholder by the total share capital of the company. Regional gross domestic product (GDP) The regional industrial upgrading situation (IS) is expressed by dividing the added value of the tertiary industry by that of the secondary industry in the region.
4.3 Data sources and processing
This paper takes the A-share listed companies on the Shanghai and Shenzhen stock exchanges from 2010 to 2023 as the research samples. It excludes the samples of financial institution listed companies, ST and *ST samples, and samples with missing data on enterprise green innovation efficiency. All the data are truncated by 1%, and finally 31,623 observations are obtained. The main explanatory variable, the list of pilot zones for the new generation of artificial intelligence, was sourced from the Ministry of Science and Technology. The enterprise-level data were obtained from WIND and CSMAR databases of Guotai 'an. The data processing and regression analysis were completed using Stata18.0. The descriptive statistics of the variables are detailed in Table 1.
Variables | N | Mean | SD | Min | Median | Max | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
GTE | 31623 | 0.5494 | 0.2136 | 0.1486 | 0.5377 | 0.9869 | ||||||||
GAE | 31623 | 0.5494 | 0.2130 | 0.1490 | 0.5377 | 0.9880 | ||||||||
inter | 31623 | 0.3398 | 0.4736 | 0.0000 | 0.0000 | 1.0000 | ||||||||
WW | 31623 | -0.9054 | 0.3331 | -1.2301 | -1.0096 | 0.0000 | ||||||||
Media | 31623 | 4.9661 | 0.9946 | 2.7726 | 4.8752 | 7.9244 | ||||||||
Analyst | 31623 | 1.3600 | 1.1987 | 0.0000 | 1.3863 | 3.8067 | ||||||||
PEC | 31623 | 4.0724 | 0.8291 | 1.4093 | 4.3344 | 5.0734 | ||||||||
HC | 31623 | 0.2202 | 0.1841 | 0.0000 | 0.1793 | 0.7549 | ||||||||
TRANS | 31623 | 0.3050 | 0.1944 | 0.0050 | 0.2867 | 0.9067 | ||||||||
CASH | 31623 | 0.2198 | 0.3552 | -0.5032 | 0.1414 | 1.8417 | ||||||||
SIZE | 31623 | 22.3535 | 1.3022 | 19.9272 | 22.1625 | 26.3978 | ||||||||
LEV | 31623 | 0.4399 | 0.2036 | 0.0630 | 0.4338 | 0.9206 | ||||||||
ROA | 31623 | 0.0370 | 0.0639 | -0.2275 | 0.0355 | 0.2223 | ||||||||
DUAL | 31623 | 0.2646 | 0.4411 | 0.0000 | 0.0000 | 1.0000 | ||||||||
AGE | 31623 | 2.3361 | 0.6665 | 1.0986 | 2.3979 | 3.4012 | ||||||||
TOP | 31623 | 0.3356 | 0.1480 | 0.0816 | 0.3110 | 0.7367 | ||||||||
GDP | 31623 | 10.6429 | 0.7391 | 8.4799 | 10.6840 | 11.8180 | ||||||||
IS | 31623 | 2.4839 | 0.1471 | 2.2272 | 2.4555 | 2.8461 | ||||||||
5. Empirical results
5.1 Benchmark regression results
The relationship between the policies of the new generation of artificial intelligence pilot zones and the green innovation efficiency of enterprises was examined using a bidirectional fixed-effect model. As shown in Table 2, whether in the base model or after adding a series of control variables, the coefficient of the policy core variable (inter) remains significantly positive at the 1% level[58]. Specifically, this policy not only effectively enhances the green R&D efficiency (GTE) of enterprises, but also significantly promotes the green technology transfer efficiency (GAE). In terms of the efficiency of technological research and development, environmentally adaptive artificial intelligence, under the constraints of resources and the environment, can independently adjust parameters, screen and combine green knowledge and technologies, helping enterprises to reserve and enhance research and development efficiency. In terms of the efficiency of green technology transfer, it can adjust production strategies, monitor pollution, reduce losses, promote transformation, and enhance transfer efficiency[59]. This result indicates that the policies of the artificial intelligence pilot zone have empowered the green innovation activities of enterprises through dual paths.
Variables | (1) | (2) | (3) | (4) |
|---|---|---|---|---|
GTE | GAE | GTE | GAE | |
inter | 0.0174*** | 0.0178*** | 0.0175*** | 0.0184*** |
(4.839) | (4.900) | (4.812) | (5.010) | |
CASH | 0.0009 | 0.0038 | ||
(0.253) | (1.119) | |||
SIZE | -0.0029 | 0.0023 | ||
(-1.401) | (1.100) | |||
LEV | -0.0057 | -0.0124 | ||
(-0.648) | (-1.393) | |||
ROA | -0.0270 | -0.0184 | ||
(-1.522) | (-1.027) | |||
DUAL | 0.0021 | 0.0006 | ||
(0.732) | (0.198) | |||
AGE | -0.0027 | -0.0000 | ||
(-0.509) | (-0.008) | |||
TOP | 0.0007 | 0.0142 | ||
(0.048) | (1.005) | |||
GDP | 0.0047 | -0.0139 | ||
(0.322) | (-0.951) | |||
IS | 0.0098 | -0.0264 | ||
(0.212) | (-0.565) | |||
_cons | 0.2326*** | 0.2325*** | 0.2333 | 0.3851** |
(57.028) | (56.542) | (1.222) | (2.001) | |
FE | Yes | Yes | Yes | Yes |
Year | Yes | Yes | Yes | Yes |
N | 31623 | 31623 | 31623 | 31623 |
R2 | 0.6358 | 0.6278 | 0.6359 | 0.6279 |
| Standard errors in parentheses | ||||
| * p < 0.1, ** p < 0.05, *** p < 0.01 | ||||
5.2 Parallel trend test
The establishment of the difference-in-differences model should satisfy the parallel trend assumption, that is, before being affected by policies, there should be no significant differences between the experimental group and the control group; otherwise, the difference-in-differences method may misestimate the effect of policies[58]. Therefore, this paper takes the year before the policy implementation as the base period for parallel trend testing. As can be seen from Fig. 1, before the policy was implemented, the estimated coefficients of both groups of samples, whether in terms of green R&D efficiency or green technology transfer rate, fluctuated around the zero axis without showing a significant trend difference. This indicates that the overall development trend of the treatment group and the control group was consistent before the policy was introduced, which conforms to the parallel trend assumption required by the difference-in-differences model.
During and after the implementation of the policy, the coefficients of both indicators showed significant positive upward jumps, and this positive effect persisted in the following years. This "intuitively indicates that the policies of the new generation of artificial intelligence pilot zones have had a positive promoting effect on the green innovation efficiency of enterprises."
Fig. 1
Parallel trend test
A
5.3 Placebo testIn reality, policy effects are influenced by various observable and unobservable factors. To further eliminate the influence of certain possible random factors on the regression results, this paper adopts the permutation test method to conduct a placebo test[60]. By randomizing the processing assignment of enterprises and years, it aims to eliminate the possible interference of potential unobservable factors on the regression results. The dual pilot cities and the time nodes for policy implementation were randomly set. To ensure the reliability of the placebo test results, this paper repeated the above process 500 times, and the results are shown in Fig. 2.
In Fig. 2, the distribution of estimators generally presents a symmetrical bell-shaped curve, with the majority concentrated around 0, indicating that there is no significant policy effect under random conditions. However, the estimated values obtained by the actual regression in this paper significantly deviate from the main region of the distribution and do not fall into its high-density interval, indicating that the original results were not caused by random disturbances or model Settings. This result further validates the robustness of the core conclusion, that is, the impact of the policies of the new generation of artificial intelligence pilot zones on the green innovation efficiency of enterprises has a practical policy effect, rather than being driven by unobservable factors or sample selection biases.
Fig. 2
Placebo test
5.4 Robustness test
5.4.1Propensity Score Matching method
To further alleviate the possible impact of sample selection bias on the benchmark regression results, this paper introduces propensity score matching (PSM) to re-match the samples based on the basic characteristic variables of enterprises. The specific approach is to take the aforementioned control variables as the basis for matching, implement 1:3 nearest neighbor matching for enterprises in the treatment group and the non-treatment group[61], reconstruct the control group on this basis, and re-estimate using the matched samples. The policy of the New Generation Artificial Intelligence Pilot Zone (inter) has a significant positive impact on both the green technology transfer efficiency (GTE) and the green achievement transfer efficiency (GAE) of enterprises. Among them, in columns (1) and (2) of Table 3, the inter coefficients are 0.0245 and 0.0247 respectively, and they are significant at the 1% level. In columns (3) and (4), the regression coefficients of inter for the green achievement conversion rate are 0.0233 and 0.0234 respectively. This indicates that the policy effect of the new generation of artificial intelligence pilot zones remains robust after sample matching, verifying the reliability of the regression results mentioned earlier.
(1) | (2) | (3) | (4) | ||
|---|---|---|---|---|---|
GTE | GTE | GAE | GAE | ||
inter | 0.0245*** | 0.0247*** | 0.0233*** | 0.0234*** | |
(3.852) | (3.855) | (3.605) | (3.605) | ||
CASH | -0.0017 | 0.0081 | |||
(-0.316) | (1.456) | ||||
SIZE | -0.0045 | -0.0032 | |||
(-1.096) | (-0.768) | ||||
LEV | -0.0002 | -0.0006 | |||
(-0.011) | (-0.035) | ||||
ROA | -0.0527* | -0.0304 | |||
(-1.838) | (-1.047) | ||||
DUAL | 0.0038 | -0.0012 | |||
(0.815) | (-0.258) | ||||
AGE | -0.0053 | -0.0044 | |||
(-0.527) | (-0.431) | ||||
TOP | -0.0025 | 0.0184 | |||
(-0.092) | (0.680) | ||||
GDP | 0.0312 | -0.0296 | |||
(0.517) | (-0.485) | ||||
IS | 0.0545 | 0.0341 | |||
(0.446) | (0.276) | ||||
_cons | 0.2252*** | -0.1141 | 0.2294*** | 0.5156 | |
(20.259) | (-0.152) | (20.362) | (0.676) | ||
FE | Yes | Yes | Yes | Yes | |
Year | Yes | Yes | Yes | Yes | |
N | 15434 | 15434 | 15434 | 15434 | |
R2 | 0.6119 | 0.6122 | 0.6001 | 0.6002 | |
5.4.2 Change the regression method
To compare the estimation differences between the bidirectional fixed effects model and the mixed regression model, the results in columns (1) and (2) of Table 4 show that among the two explained variables, namely the green R&D efficiency of enterprises and the green technology transfer rate, the regression coefficients of the policies for the new generation of artificial intelligence pilot zones are 0.0662 and 0.0625, respectively. And all were significantly positive at the 1% level. This indicates that regardless of the model setting adopted, the policy has significantly enhanced the efficiency of enterprises' green innovation, verifying the robustness of the regression model setting.
5.4.3 Excluding the impact of the epidemic
The COVID-19 pandemic, as an extreme external shock, has led to abnormal fluctuations in global economic activities and policy environments. Under this influence, the production, supply chain and energy usage patterns of enterprises were forced to undergo abnormal changes during the epidemic. Therefore, this paper excluded the samples from 2020 to eliminate the interference of the COVID-19 pandemic on the results. The results in columns (3) and (4) of Table 4 show that the regression coefficients of the policies of the new generation of artificial intelligence pilot zones for GTE and GAE are 0.0190 and 0.0163 respectively, which remain significantly positive, indicating that after excluding the factor of the pandemic, The policy effect still exists steadily.
5.4.4 Add control variables
To further control the possible interfering factors brought about by the corporate governance structure and business conditions, this paper introduces the board size (MBOARD) and the GROWTH potential of the enterprise as additional control variables in the regression. Among them, the size of the board of directors is expressed by the logarithm of the number of directors in the enterprise[62], and the growth of the enterprise is the growth rate of the main business[63]. Columns (5) and (6) of Table 4 show that the regression coefficients of the policies of the new generation of artificial intelligence pilot zones on the green R&D efficiency and green technology transfer rate of enterprises are 0.2143 and 0.3285 respectively. The significance remains robust, indicating that the research conclusions are not disturbed by the introduction of additional control variables.
(1) | (2) | (3) | (4) | (5) | (6) | |
|---|---|---|---|---|---|---|
GTE | GAE | GTE | GAE | GTE | GAE | |
inter | 0.0662*** | 0.0625*** | 0.0190*** | 0.0163*** | 0.0175*** | 0.0184*** |
(22.559) | (21.112) | (4.629) | (3.924) | (4.822) | (5.017) | |
CASH | 0.0122*** | 0.0149*** | -0.0000 | 0.0051 | 0.0008 | 0.0038 |
(3.274) | (3.982) | (-0.014) | (1.475) | (0.249) | (1.112) | |
SIZE | -0.0041* | 0.0015 | -0.0018 | 0.0012 | -0.0031 | 0.0021 |
(-1.818) | (0.653) | (-0.875) | (0.562) | (-1.493) | (1.015) | |
LEV | -0.0263*** | -0.0335*** | -0.0060 | -0.0113 | -0.0058 | -0.0123 |
(-2.727) | (-3.445) | (-0.674) | (-1.250) | (-0.654) | (-1.383) | |
ROA | -0.0703*** | -0.0602*** | -0.0238 | -0.0185 | -0.0276 | -0.0183 |
(-3.616) | (-3.070) | (-1.294) | (-0.996) | (-1.553) | (-1.021) | |
DUAL | 0.0000 | -0.0017 | 0.0029 | 0.0020 | 0.0022 | 0.0007 |
(0.000) | (-0.526) | (1.003) | (0.687) | (0.783) | (0.233) | |
AGE | 0.0137** | 0.0113** | -0.0053 | -0.0005 | -0.0026 | 0.0001 |
(2.474) | (2.035) | (-0.986) | (-0.094) | (-0.478) | (0.013) | |
TOP | 0.0336** | 0.0449*** | 0.0001 | 0.0141 | 0.0014 | 0.0146 |
(2.174) | (2.875) | (0.005) | (0.983) | (0.099) | (1.028) | |
GDP | 0.3377*** | 0.3259*** | -0.0006 | -0.0109 | 0.0048 | -0.0140 |
(40.450) | (38.671) | (-0.040) | (-0.739) | (0.327) | (-0.955) | |
IS | 0.4713*** | 0.5053*** | 0.0078 | 0.0140 | 0.0091 | -0.0268 |
(12.004) | (12.750) | (0.169) | (0.299) | (0.197) | (-0.573) | |
MBOARD | 0.0067 | 0.0056 | ||||
(0.846) | (0.699) | |||||
GROWTH | 0.0002 | -0.0000 | ||||
(0.639) | (-0.050) | |||||
_cons | -4.1764*** | -4.2566*** | 0.2710 | 0.2825 | 0.2236 | 0.3778** |
(-56.229) | (-56.767) | (1.415) | (1.460) | (1.170) | (1.960) | |
FE | Yes | Yes | Yes | Yes | Yes | Yes |
Year | No | No | Yes | Yes | Yes | Yes |
N | 31623 | 31623 | 28792 | 28792 | 31623 | 31623 |
R2 | 0.5565 | 0.5454 | 0.6554 | 0.6458 | 0.6359 | 0.6279 |
6. Analysis of the Mechanism of Action
6.1 Analysis of Mediating Effects
6.1.1 Financing constraints (WW)
From the results in column (1) of Table 5, the coefficient of the policy variable is -0.0131, which is significantly negative, indicating that the policies of the new generation artificial intelligence pilot zone are conducive to alleviating the financing constraints of enterprises. The policy of the new generation of artificial intelligence pilot zones effectively reduces the financing pressure on enterprises by guiding market capital to tilt towards the field of artificial intelligence and optimizing the financing environment[33], lowers the level of financial asset allocation[34], supports R&D investment, and enhances the level and efficiency of green innovation[35]. Therefore, Alleviating the constraints of green financing and improving the business conditions of enterprises are the key channels for digital empowerment to promote the green and innovative development of enterprises[64].
6.1.2 Analyst
In column (2) of Table 5, the regression coefficient of the new generation artificial intelligence pilot zone policy on analysts' attention is positive at the 1% level, indicating that the pilot policy has significantly increased the attention of enterprise analysts. The policies of the new generation of artificial intelligence pilot zones, by attracting the attention of analysts and leveraging their information transmission and supervision functions in the capital market[43], alleviate the information asymmetry in innovation activities, thereby reducing financing costs, increasing innovation investment and enhancing innovation capabilities[45], and further promoting the improvement of enterprises' green innovation efficiency. This indicates that analysts' attention constitutes an important transmission hub for the impact of artificial intelligence policies on the green innovation efficiency of enterprises.
6.1.3 Media attention (Media)
The results in column (3) of Table 5 show that the policies of the new generation of artificial intelligence pilot zones have significantly enhanced the media attention of enterprises, with a regression coefficient of 0.0275, and have passed the 5% significance test. Under the impetus of pilot policies, the media, leveraging its core position in information transmission and information mining capabilities, has enhanced public attention[36] and the transparency and credibility of enterprises[37]. Meanwhile, the increase in media attention can strengthen external supervision[39], alleviate information asymmetry[40], encourage enterprises to optimize environmental information disclosure, establish a good reputation, and thereby promote the efficiency of green innovation. Therefore, media attention plays an intermediary role in the impact of the policies of the new generation of artificial intelligence pilot zones on the green innovation efficiency of enterprises.
Variables | (1) | (2) | (3) | |||||
|---|---|---|---|---|---|---|---|---|
WW | Analyst | Media | ||||||
inter | -0.0131** | 0.0299* | 0.0275** | |||||
(-2.019) | (1.732) | (1.962) | ||||||
CASH | -0.0105* | -0.0059 | 0.0289** | |||||
(-1.737) | (-0.368) | (2.211) | ||||||
SIZE | -0.1147*** | 0.4962*** | 0.1800*** | |||||
(-30.783) | (50.040) | (22.387) | ||||||
LEV | -0.2021*** | -0.4026*** | 0.1891*** | |||||
(-12.792) | (-9.576) | (5.546) | ||||||
ROA | -0.1508*** | 3.4148*** | 0.7205*** | |||||
(-4.754) | (40.460) | (10.526) | ||||||
DUAL | 0.0056 | 0.0419*** | -0.0092 | |||||
(1.092) | (3.081) | (-0.838) | ||||||
AGE | 0.0362*** | -0.3444*** | 0.0901*** | |||||
(3.804) | (-13.585) | (4.380) | ||||||
TOP | 0.0046 | -0.2400*** | -0.5066*** | |||||
(0.184) | (-3.591) | (-9.345) | ||||||
GDP | 0.0469* | 0.2055*** | -0.3442*** | |||||
(1.802) | (2.968) | (-6.131) | ||||||
IS | -0.1388* | -0.5246** | -0.1563 | |||||
(-1.672) | (-2.376) | (-0.873) | ||||||
_cons | 1.4496*** | -8.9605*** | 4.4065*** | |||||
(4.250) | (-9.871) | (5.986) | ||||||
FE | Yes | Yes | Yes | |||||
Year | Yes | Yes | Yes | |||||
N | 31623 | 31623 | 31623 | |||||
R2 | 0.5214 | 0.7384 | 0.7501 | |||||
6.2 Analysis of moderating effect
6.2.1 Environmental Concern
It can be seen from columns (1) and (2) of Table 6 that the coefficients of the interaction term between environmental attention and policy variables are 0.0051 and 0.0034 respectively, and they are significantly positive in the regression of GTE. This indicates that in regions where environmental issues receive more public attention, the policies of the new generation of artificial intelligence pilot zones have a stronger promoting effect on the green innovation efficiency of enterprises. High attention strengthens social public opinion supervision[46], compelling enterprises to precisely invest in green innovation in accordance with policy guidance and amplifying the positive effects of policies; At the same time, the increased attention will change consumer preferences[47], prompting enterprises to take the initiative to utilize policies to accelerate innovation. Therefore, public environmental attention, through the guiding role of external supervision and market demand, regulates the intensity of the impact of the policies of the new generation of artificial intelligence pilot zones on the efficiency of green innovation.
6.2.2 Human capital level
The regression coefficients of the interaction term between human capital level and policy in columns (3) and (4) of Table 6 are both positive, and they have passed the 1% significance test in the efficiency of green technology transfer. This indicates that the richer the human capital, the more obvious the positive impact of the policies of the new generation of artificial intelligence pilot zones on the green innovation efficiency of enterprises. Employees with a high level of human capital are more likely to integrate the policy orientation of artificial intelligence, explore green innovation directions, transform policy incentives into actual innovation actions, and amplify the policy's promotion of innovation efficiency. At the same time, enhance the application capacity of artificial intelligence technology, avoid the mismatch between technology and human resources, and effectively release the effectiveness of policies. Therefore, enhancing the regional human capital level can strengthen the promoting effect of the policies of the new generation of artificial intelligence pilot zones on the green innovation efficiency of enterprises.
6.2.3 Information Transparency
The coefficients of the terms between enterprise information transparency and policy interaction in columns (5) and (6) of Table 6 are 0.0183 and 0.0152 respectively, both of which are significantly positive, indicating that the higher the degree of information disclosure, the stronger the policy incentive effect. The improvement of enterprise information transparency can alleviate information asymmetry within enterprises[50]. Enhanced transparency can facilitate the two-way flow of policy information, optimize innovation paths, and efficiently transform policy effectiveness. Additionally, information transparency strengthens external supervision and market trust, driving enterprises to direct policy dividends towards green technologies, while reducing investment risks to attract funds, thereby amplifying the policy's role in promoting the efficiency of green innovation. Therefore, information transparency has a positive moderating effect on artificial intelligence policies and the green innovation efficiency of enterprises.
Variables | (1) | (2) | (3) | (4) | (5) | (6) |
|---|---|---|---|---|---|---|
GTE | GAE | GTE | GAE | GTE | GAE | |
inter | 0.0168*** | 0.0178*** | 0.0171*** | 0.0174*** | 0.0175*** | 0.0183*** |
(4.592) | (4.829) | (4.655) | (4.697) | (4.813) | (5.006) | |
CASH | 0.0008 | 0.0038 | 0.0008 | 0.0038 | 0.0009 | 0.0038 |
(0.235) | (1.109) | (0.249) | (1.116) | (0.262) | (1.115) | |
SIZE | -0.0030 | 0.0022 | -0.0029 | 0.0022 | -0.0032 | 0.0026 |
(-1.432) | (1.062) | (-1.395) | (1.031) | (-1.439) | (1.197) | |
LEV | -0.0058 | -0.0125 | -0.0057 | -0.0124 | -0.0053 | -0.0129 |
(-0.660) | (-1.406) | (-0.648) | (-1.386) | (-0.602) | (-1.435) | |
ROA | -0.0258 | -0.0176 | -0.0261 | -0.0161 | -0.0258 | -0.0175 |
(-1.455) | (-0.985) | (-1.470) | (-0.900) | (-1.451) | (-0.978) | |
DUAL | 0.0021 | 0.0006 | 0.0021 | 0.0005 | 0.0021 | 0.0006 |
(0.727) | (0.194) | (0.722) | (0.176) | (0.729) | (0.196) | |
AGE | -0.0027 | 0.0000 | -0.0027 | 0.0003 | -0.0018 | 0.0005 |
(-0.499) | (0.002) | (-0.503) | (0.062) | (-0.335) | (0.084) | |
TOP | 0.0007 | 0.0143 | 0.0007 | 0.0143 | 0.0003 | 0.0140 |
(0.052) | (1.008) | (0.046) | (1.012) | (0.024) | (0.991) | |
GDP | 0.0054 | -0.0138 | 0.0051 | -0.0130 | 0.0049 | -0.0138 |
(0.368) | (-0.937) | (0.347) | (-0.888) | (0.339) | (-0.939) | |
IS | 0.0274 | -0.0145 | 0.0128 | -0.0181 | 0.0134 | -0.0256 |
(0.580) | (-0.306) | (0.274) | (-0.384) | (0.287) | (-0.546) | |
PEC | 0.0013 | 0.0022 | ||||
(0.324) | (0.544) | |||||
PEC*inter | 0.0051** | 0.0034 | ||||
(2.084) | (1.355) | |||||
HC | -0.0009 | 0.0049 | ||||
(-0.090) | (0.468) | |||||
HC*inter | 0.0083 | 0.0201* | ||||
(0.742) | (1.792) | |||||
TRANS | -0.0023 | -0.0057 | ||||
(-0.293) | (-0.724) | |||||
TRANS*inter | 0.0183* | 0.0059 | ||||
(1.863) | (0.597) | |||||
_cons | 0.1806 | 0.3479* | 0.2220 | 0.3566* | 0.2269 | 0.3756* |
(0.939) | (1.794) | (1.159) | (1.848) | (1.187) | (1.948) | |
FE | Yes | Yes | Yes | Yes | Yes | Yes |
Year | Yes | Yes | Yes | Yes | Yes | Yes |
N | 31623 | 31623 | 31623 | 31623 | 31623 | 31623 |
R2 | 0.6359 | 0.6279 | 0.6359 | 0.6279 | 0.6359 | 0.6279 |
7. Heterogeneity analysis
7.1 City Grade
To further explore the impact of urban development levels on policy effects, this paper classifies the samples into three categories - first-tier, second-tier and third-tier cities - based on the positioning of the cities where enterprises are located in the national urban system. Examine the differences in the role of policies for the new generation of artificial intelligence pilot zones at different urban levels. Columns (1) to (3) of Table 7 show the impact of policies on the efficiency of enterprises' green technology transformation. Since the higher the city grade, the greater the degree of resource aggregation[65], in first-tier cities, the policy regression coefficient is 0.0344, which is significantly positive, indicating that the policy has significantly promoted the absorption and research and development capabilities of enterprises' green technologies. In second-tier and third-tier cities, the policy coefficients were 0.0087 and 0.0104 respectively, neither reaching a significant level, indicating that the promoting effect of policies in lower-tier cities was relatively limited.
In terms of the efficiency of enterprises' green technology transfer, columns (4) to (6) of Table 7 show that the coefficients of the policies for the new generation of artificial intelligence pilot zones in first-tier and second-tier cities are 0.0284 and 0.0343 respectively, both significantly positive, while in third-tier cities they are − 0.0051, which is not significant. It is indicated that the policy is more effective in promoting the market transformation and output benefits of green patents in first - and second-tier cities.
Variables | (1) | (2) | (3) | (4) | (5) | (6) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
GTE | GTE | GTE | GAE | GAE | GAE | |||||||||
First-tier | Second-tier | Third-tier | First-tier | Second-tier | Third-tier | |||||||||
inter | 0.0344*** | 0.0087 | 0.0104 | 0.0284*** | 0.0343*** | -0.0051 | ||||||||
(5.416) | (1.173) | (1.417) | (4.429) | (4.589) | (-0.689) | |||||||||
CASH | 0.0028 | 0.0074 | -0.0064 | 0.0045 | 0.0096 | -0.0026 | ||||||||
(0.591) | (0.974) | (-0.998) | (0.953) | (1.264) | (-0.405) | |||||||||
SIZE | -0.0010 | -0.0054 | -0.0059 | 0.0036 | 0.0045 | -0.0021 | ||||||||
(-0.327) | (-1.249) | (-1.432) | (1.202) | (1.036) | (-0.498) | |||||||||
LEV | -0.0094 | 0.0158 | -0.0197 | -0.0163 | -0.0114 | -0.0073 | ||||||||
(-0.752) | (0.836) | (-1.171) | (-1.296) | (-0.599) | (-0.431) | |||||||||
ROA | -0.0096 | -0.0506 | -0.0354 | -0.0304 | -0.0193 | 0.0057 | ||||||||
(-0.380) | (-1.327) | (-1.056) | (-1.198) | (-0.505) | (0.168) | |||||||||
DUAL | -0.0016 | 0.0016 | 0.0107* | -0.0007 | -0.0051 | 0.0087 | ||||||||
(-0.387) | (0.275) | (1.905) | (-0.172) | (-0.859) | (1.539) | |||||||||
AGE | -0.0029 | -0.0176 | 0.0129 | -0.0016 | -0.0003 | 0.0054 | ||||||||
(-0.404) | (-1.544) | (1.137) | (-0.227) | (-0.023) | (0.475) | |||||||||
TOP | -0.0056 | -0.0144 | 0.0290 | 0.0110 | 0.0044 | 0.0340 | ||||||||
(-0.274) | (-0.491) | (1.104) | (0.535) | (0.150) | (1.282) | |||||||||
GDP | 0.0156 | -0.0141 | 0.0078 | -0.0884** | -0.0173 | 0.0160 | ||||||||
(0.409) | (-0.522) | (0.386) | (-2.299) | (-0.641) | (0.782) | |||||||||
IS | -0.0744 | 0.1063 | 0.0971 | 0.0013 | -0.1145 | 0.0096 | ||||||||
(-0.916) | (1.118) | (1.202) | (0.016) | (-1.200) | (0.117) | |||||||||
_cons | 0.2916 | 0.2681 | 0.0373 | 1.0411** | 0.5815 | 0.0871 | ||||||||
(0.629) | (0.725) | (0.138) | (2.224) | (1.569) | (0.318) | |||||||||
FE | Yes | Yes | Yes | Yes | Yes | Yes | ||||||||
Year | Yes | Yes | Yes | Yes | Yes | Yes | ||||||||
N | 16593 | 7464 | 7566 | 16593 | 7464 | 7566 | ||||||||
R2 | 0.6398 | 0.6251 | 0.6385 | 0.6309 | 0.6200 | 0.6299 | ||||||||
7.2 Types of Urban Resources
Based on the resource endowment characteristics of the cities where the enterprises are located, this study divides the samples into resource-based cities and non-resource-based cities to investigate the differences in the impact of the policies for the new generation of artificial intelligence pilot zones among different types of cities. It can be seen from columns (1) and (2) of Table 8 that in terms of the transformation of green technology, the coefficient of the policy in resource-based cities is 0.0147, which is positive but does not reach a significant level. In non-resource-based cities, the policy coefficient is 0.0194 and is significantly positive, indicating that policies can more effectively stimulate enterprises' green R&D investment and transformation capabilities in non-resource-based cities.
In terms of the transformation of green achievements, the results in columns (3) and (4) of Table 8 show that the regression coefficient of the policy for the new generation of artificial intelligence pilot zones in resource-based cities is -0.0104, which is negative and not significant, while in non-resource-based cities, the coefficient is 0.0222, which is significantly positive. This indicates that in resource-based cities, enterprises lack the motivation to transform green innovation achievements.
Variables | (1) | (2) | (3) | (4) | ||
|---|---|---|---|---|---|---|
GTE | GTE | GAE | GAE | |||
Resource-based | Non-resource-based | Resource-based | Non-resource-based | |||
inter | 0.0147 | 0.0194*** | -0.0104 | 0.0222*** | ||
(1.346) | (4.908) | (-0.944) | (5.554) | |||
CASH | -0.0155 | 0.0030 | -0.0075 | 0.0052 | ||
(-1.562) | (0.826) | (-0.742) | (1.443) | |||
SIZE | -0.0068 | -0.0025 | -0.0091 | 0.0036 | ||
(-1.127) | (-1.143) | (-1.491) | (1.614) | |||
LEV | -0.0146 | -0.0051 | -0.0410 | -0.0085 | ||
(-0.571) | (-0.541) | (-1.590) | (-0.898) | |||
ROA | -0.0531 | -0.0243 | -0.0030 | -0.0203 | ||
(-1.053) | (-1.280) | (-0.059) | (-1.062) | |||
DUAL | -0.0008 | 0.0024 | 0.0131 | -0.0010 | ||
(-0.095) | (0.790) | (1.468) | (-0.341) | |||
AGE | -0.0042 | -0.0032 | 0.0014 | -0.0001 | ||
(-0.248) | (-0.568) | (0.080) | (-0.025) | |||
TOP | 0.0618 | -0.0080 | 0.0369 | 0.0138 | ||
(1.551) | (-0.533) | (0.917) | (0.910) | |||
GDP | 0.0277 | 0.0013 | 0.0256 | -0.0212 | ||
(0.800) | (0.082) | (0.733) | (-1.304) | |||
IS | 0.0336 | 0.0111 | 0.1115 | -0.0239 | ||
(0.297) | (0.212) | (0.978) | (-0.453) | |||
_cons | 0.0318 | 0.2588 | -0.0554 | 0.4204** | ||
(0.072) | (1.222) | (-0.124) | (1.969) | |||
FE | Yes | Yes | Yes | Yes | ||
Year | Yes | Yes | Yes | Yes | ||
N | 3392 | 28231 | 3392 | 28231 | ||
R2 | 0.6421 | 0.6353 | 0.6284 | 0.6281 | ||
7.3 Degree of Pollution
To examine whether the pollution attributes of the industries in which enterprises are located affect the policy effects, this paper divides the samples into two categories based on the degree of environmental impact of the industries: polluting industry enterprises and non-polluting industry enterprises, and analyzes the implementation effects of the policies of the new generation of artificial intelligence pilot zones under different environmental attributes. It can be seen from columns (1) and (2) of Table 9 that in terms of the efficiency of green technology research and development, this policy has a positive impact on both types of enterprises. However, the regression coefficient (0.0198) for non-polluting industry enterprises is higher and more significant, while the coefficient for polluting industry enterprises is 0.0127. Although it is also significant, the effect is relatively weak. In terms of the patent efficiency of green innovation achievements, the results in columns (3) and (4) of Table 9 show that the new generation of artificial intelligence policies have not demonstrated a significant promoting effect on enterprises in polluting industries (with a coefficient of -0.0019), while their promoting effect on non-polluting industries is relatively significant (with a coefficient of 0.0260).
Variables | (1) | (2) | (3) | (4) | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
GTE | GTE | GAE | GAE | |||||||
Pollution | Non-polluting | Pollution | Non-polluting | |||||||
inter | 0.0127* | 0.0198*** | -0.0019 | 0.0260*** | ||||||
(1.781) | (4.639) | (-0.269) | (6.025) | |||||||
CASH | -0.0048 | 0.0026 | 0.0032 | 0.0041 | ||||||
(-0.706) | (0.656) | (0.477) | (1.044) | |||||||
SIZE | -0.0064 | -0.0013 | 0.0017 | 0.0025 | ||||||
(-1.576) | (-0.528) | (0.425) | (1.013) | |||||||
LEV | -0.0022 | -0.0052 | -0.0099 | -0.0119 | ||||||
(-0.124) | (-0.504) | (-0.562) | (-1.148) | |||||||
ROA | 0.0112 | -0.0426** | -0.0224 | -0.0159 | ||||||
(0.320) | (-2.050) | (-0.640) | (-0.759) | |||||||
DUAL | 0.0076 | 0.0004 | 0.0013 | 0.0002 | ||||||
(1.290) | (0.128) | (0.213) | (0.055) | |||||||
AGE | -0.0089 | -0.0001 | 0.0071 | -0.0010 | ||||||
(-0.793) | (-0.025) | (0.628) | (-0.159) | |||||||
TOP | -0.0392 | 0.0181 | -0.0230 | 0.0285* | ||||||
(-1.511) | (1.079) | (-0.883) | (1.675) | |||||||
GDP | 0.0188 | -0.0031 | 0.0001 | -0.0210 | ||||||
(0.765) | (-0.173) | (0.003) | (-1.147) | |||||||
IS | -0.0372 | 0.0399 | -0.0034 | -0.0206 | ||||||
(-0.452) | (0.700) | (-0.041) | (-0.359) | |||||||
_cons | 0.3053 | 0.1941 | 0.2064 | 0.4328* | ||||||
(0.935) | (0.823) | (0.631) | (1.818) | |||||||
FE | Yes | Yes | Yes | Yes | ||||||
Year | Yes | Yes | Yes | Yes | ||||||
N | 7705 | 23918 | 7705 | 23918 | ||||||
R2 | 0.6413 | 0.6338 | 0.6396 | 0.6240 | ||||||
7.4 Life Cycle
To examine the impact of different development stages of enterprises on policy effects, this paper divides the samples into three groups - the growth stage, the mature stage and the decline stage - based on the characteristics of the enterprise life cycle, and conducts regression analyses respectively. It can be seen from columns (1) to (3) of Table 10 that in terms of the efficiency of green technology research and development, the policy has the most significant impact on growth-stage enterprises (with a coefficient of 0.0236), followed by mature enterprises (with a coefficient of 0.0112), while it also has a certain positive effect on recessional-stage enterprises (with a coefficient of 0.0179), but it is relatively weak.
In terms of the efficiency of green technology transfer, the results from columns (4) to (6) of Table 10 show that the policy has played a positive role in promoting both growth-stage and mature enterprises (0.0251 and 0.0241 respectively), but has no significant effect on declining enterprises (coefficient − 0.0023).
Variables | (1) | (2) | (3) | (4) | (5) | (6) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
GTE | GTE | GTE | GAE | GAE | GAE | |||||||||
Growth stage | Mature stage | Decline stage | Growth stage | Mature stage | Decline stage | |||||||||
inter | 0.0236*** | 0.0112* | 0.0179*** | -0.0023 | 0.0251*** | 0.0241*** | ||||||||
(2.935) | (1.652) | (2.756) | (-0.283) | (3.667) | (3.699) | |||||||||
CASH | -0.0006 | 0.0019 | -0.0043 | 0.0097 | -0.0042 | 0.0022 | ||||||||
(-0.074) | (0.293) | (-0.721) | (1.286) | (-0.630) | (0.363) | |||||||||
SIZE | -0.0052 | -0.0058 | 0.0004 | 0.0037 | 0.0041 | 0.0036 | ||||||||
(-1.196) | (-1.436) | (0.107) | (0.838) | (0.997) | (0.876) | |||||||||
LEV | -0.0104 | -0.0057 | -0.0128 | -0.0147 | -0.0153 | -0.0180 | ||||||||
(-0.545) | (-0.326) | (-0.770) | (-0.767) | (-0.874) | (-1.085) | |||||||||
ROA | -0.0823** | -0.0366 | -0.0270 | -0.0708* | -0.0001 | -0.0358 | ||||||||
(-2.105) | (-1.074) | (-0.758) | (-1.803) | (-0.003) | (-1.002) | |||||||||
DUAL | 0.0022 | 0.0059 | 0.0029 | -0.0034 | 0.0042 | 0.0005 | ||||||||
(0.348) | (1.103) | (0.550) | (-0.548) | (0.784) | (0.094) | |||||||||
AGE | 0.0023 | -0.0080 | 0.0150 | 0.0042 | 0.0065 | -0.0075 | ||||||||
(0.198) | (-0.752) | (1.281) | (0.351) | (0.603) | (-0.639) | |||||||||
TOP | 0.0002 | 0.0134 | 0.0097 | 0.0033 | 0.0079 | 0.0386 | ||||||||
(0.006) | (0.509) | (0.358) | (0.109) | (0.298) | (1.421) | |||||||||
GDP | 0.0128 | 0.0017 | 0.0222 | 0.0174 | -0.0181 | -0.0320 | ||||||||
(0.346) | (0.062) | (0.886) | (0.467) | (-0.653) | (-1.274) | |||||||||
IS | 0.1058 | -0.0016 | -0.0032 | -0.1043 | -0.1338 | 0.0980 | ||||||||
(0.958) | (-0.019) | (-0.039) | (-0.940) | (-1.524) | (1.186) | |||||||||
_cons | -0.0302 | 0.3594 | -0.0182 | 0.2392 | 0.6421* | 0.2458 | ||||||||
(-0.063) | (0.983) | (-0.055) | (0.495) | (1.745) | (0.748) | |||||||||
FE | Yes | Yes | Yes | Yes | Yes | Yes | ||||||||
Year | Yes | Yes | Yes | Yes | Yes | Yes | ||||||||
N | 9442 | 11181 | 11000 | 9442 | 11181 | 11000 | ||||||||
R2 | 0.7108 | 0.7113 | 0.6797 | 0.7035 | 0.7056 | 0.6798 | ||||||||
7.5 Industry Intensity
To analyze the heterogeneous effects of the industry types in which enterprises operate on the policy outcomes of the new generation of artificial intelligence pilot zones, this paper classifies the samples into three types of enterprises - technology-oriented, capital-oriented, and labor-oriented - based on the characteristics of industry elements, and conducts regression analyses respectively. Columns (1) to (3) of Table 11 show the regression results of the green R&D efficiency of enterprises. It can be seen that in terms of the efficiency of green technology research and development, the positive effect of policies on technology-based enterprises and labor-intensive enterprises is relatively significant, with regression coefficients of 0.0264 and 0.0164 respectively, indicating that these enterprises are more sensitive to policy incentives. In contrast, the regression coefficient of capital-oriented enterprises is only 0.0071 and does not have statistical significance, indicating that the policy has a relatively weak impact on their R&D end.
In terms of the efficiency of green technology transfer, the results from columns (4) to (6) of Table 11 show that technology and labor enterprises still have a significant positive impact (with regression coefficients of 0.0324 and 0.0134 respectively), while the estimated coefficient for capital enterprises is negative and not significant (-0.0028). This further confirms that its policy-driven effect in the green transformation is not strong.
Variables | (1) | (2) | (3) | (4) | (5) | (6) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
GTE | GTE | GTE | GAE | GAE | GAE | |||||||||
Technology | Assets | Labor | Technology | Assets | Labor | |||||||||
inter | 0.0264*** | 0.0071 | 0.0164*** | 0.0324*** | -0.0028 | 0.0134** | ||||||||
(4.569) | (0.874) | (2.734) | (5.541) | (-0.342) | (2.217) | |||||||||
CASH | 0.0021 | -0.0052 | 0.0025 | 0.0040 | 0.0135* | 0.0016 | ||||||||
(0.401) | (-0.684) | (0.442) | (0.755) | (1.731) | (0.277) | |||||||||
SIZE | -0.0082** | -0.0051 | 0.0027 | 0.0073* | 0.0025 | -0.0039 | ||||||||
(-2.244) | (-1.045) | (0.765) | (1.957) | (0.492) | (-1.084) | |||||||||
LEV | -0.0006 | 0.0254 | -0.0209 | -0.0306** | -0.0037 | 0.0024 | ||||||||
(-0.042) | (1.251) | (-1.396) | (-2.047) | (-0.179) | (0.157) | |||||||||
ROA | -0.0078 | 0.0178 | -0.0515* | -0.0158 | -0.0635 | 0.0036 | ||||||||
(-0.285) | (0.447) | (-1.691) | (-0.567) | (-1.570) | (0.119) | |||||||||
DUAL | -0.0041 | 0.0092 | 0.0080 | 0.0021 | 0.0039 | -0.0037 | ||||||||
(-0.946) | (1.324) | (1.628) | (0.491) | (0.554) | (-0.759) | |||||||||
AGE | 0.0029 | -0.0241* | 0.0017 | 0.0066 | 0.0098 | -0.0108 | ||||||||
(0.343) | (-1.839) | (0.185) | (0.787) | (0.737) | (-1.154) | |||||||||
TOP | -0.0001 | -0.0336 | 0.0384 | 0.0418* | 0.0021 | -0.0057 | ||||||||
(-0.005) | (-1.070) | (1.645) | (1.658) | (0.066) | (-0.244) | |||||||||
GDP | 0.0298 | -0.0206 | 0.0036 | -0.0084 | -0.0304 | -0.0061 | ||||||||
(1.060) | (-0.692) | (0.159) | (-0.297) | (-1.006) | (-0.266) | |||||||||
IS | 0.1605** | 0.0190 | -0.1110 | -0.0755 | -0.0428 | 0.0193 | ||||||||
(1.991) | (0.194) | (-1.458) | (-0.926) | (-0.430) | (0.253) | |||||||||
_cons | -0.2818 | 0.5426 | 0.4011 | 0.3272 | 0.5624 | 0.3567 | ||||||||
(-0.786) | (1.357) | (1.333) | (0.904) | (1.386) | (1.181) | |||||||||
FE | Yes | Yes | Yes | Yes | Yes | Yes | ||||||||
Year | Yes | Yes | Yes | Yes | Yes | Yes | ||||||||
N | 14447 | 5965 | 11211 | 14447 | 5965 | 11211 | ||||||||
R2 | 0.6233 | 0.6457 | 0.6521 | 0.6082 | 0.6420 | 0.6497 | ||||||||
8. Conclusion
8.1 Research Conclusions
Based on the data of A-share listed companies from 2010 to 2023, this paper uses the multi-period difference-in-differences method to analyze the impact of the policies of the new generation of artificial intelligence pilot zones on the green innovation efficiency of enterprises, and draws the following main conclusions: Firstly, the implementation of the policies for the new generation of artificial intelligence pilot zones has significantly enhanced the green innovation efficiency of enterprises, which includes both R&D efficiency and the efficiency of green technology transfer. Secondly, mechanism tests show that the policies of the new generation of artificial intelligence pilot zones enhance the efficiency of enterprises' green R&D and green technology transfer by easing financing constraints, increasing the attention of analysts and the media. Furthermore, the analysis of the moderating effect indicates that public environmental attention and the transparency of enterprise information can enhance the positive impact of the policies for the new generation of artificial intelligence pilot zones on the green R&D efficiency of enterprises, and the level of human capital of enterprises can enhance the positive impact of the policies for the new generation of artificial intelligence pilot zones on the efficiency of green technology transfer of enterprises. Finally, heterogeneity analysis indicates that the policies of the new generation of artificial intelligence pilot zones have a greater promoting effect on the green R&D efficiency and green technology transfer efficiency of enterprises in first-tier cities, non-resource, non-pollution, growth-stage, technology and labor-intensive enterprises.
8.2 Policy Recommendations
First, it is suggested that a special credit guarantee fund be established to provide interest-subsidized loans and green bond issuance convenience for enterprises in the artificial intelligence pilot zone. By optimizing the financing environment, the financial pressure on enterprises' green innovation can be reduced, thereby enhancing innovation efficiency. The government can join hands with commercial banks and venture capital institutions to formulate differentiated credit assessment standards for green technology research and development enterprises, appropriately relax mortgage requirements, extend loan terms, and offer interest rate discounts. Meanwhile, a green innovation risk compensation fund can be established to share the lending risks of financial institutions, guide more social capital to flow into the field of environmental protection technology, and build a multi-level and multi-channel green financial support system.
Second, it is suggested that enterprises in the pilot zone be required to disclose special reports on green innovation, and a green rating incentive mechanism for analysts be established. By enhancing the transparency of market information, capital can be guided to flow into the field of green technology, and the information asymmetry of innovation activities can be reduced. Regulatory authorities may require listed companies to separately list key indicators such as green R&D investment, the number of patents and emission reduction achievements in their annual reports, and encourage securities research institutions to incorporate ESG factors into valuation models. Provide research funding subsidies or tax reductions to analyst teams that continuously track green innovation enterprises, promote the formation of a value investment concept in the capital market that focuses on long-term environmental benefits, and enhance the market valuation and financing capabilities of green technology enterprises.
Third, it is suggested that a national-level media dissemination platform be established to regularly release rankings of enterprises' environmental performance. By leveraging public opinion supervision and reputation effects, enterprises can be encouraged to increase their investment in green research and development, and the role of external supervision in promoting innovation efficiency can be strengthened. This platform can integrate mainstream media resources and set up a "Green Innovation Observation" column, which not only exposes the environmental violations of high-pollution and high-energy-consuming enterprises but also promotes benchmark cases of energy conservation and emission reduction. At the same time, a linkage mechanism between the media and environmental protection departments should be established to conduct in-depth tracking and reporting on major environmental incidents, creating social public opinion pressure. This will prompt enterprises to incorporate environmental responsibility into their strategic decisions and proactively increase investment in green technologies to maintain their brand image.
Fourth, it is suggested that environmental education be incorporated into the compulsory education curriculum and a green innovation monitoring system for public participation be developed. By enhancing social environmental awareness and creating a market demand-driven mechanism, the positive impact of policies on green innovation can be magnified. Add environmental protection knowledge modules such as climate change and circular economy to the curriculum of primary and secondary schools to cultivate teenagers' sustainable development concepts. At the same time, develop a "Green Consumption" mobile application, allowing consumers to check the carbon footprint and environmental protection certification of products, and support green enterprises through purchasing behavior. The government can also regularly hold public open days and hearings with environmental protection themes to broaden the channels for the public to participate in environmental supervision and create a favorable atmosphere in which the whole society pays attention to green innovation.
Fifth. It is suggested that the interdisciplinary subject of "AI + Green Technology "be added to the universities in the experimental zone, and a joint training program between universities and enterprises be implemented. Through professional talent cultivation, the response ability of enterprises to policies can be enhanced, and the mismatch between technology and human resources can be avoided. Colleges and universities can integrate resources from disciplines such as computer science and environmental engineering to offer emerging professional directions like intelligent environmental protection and green manufacturing. Enterprises offer internship positions and research and development projects, allowing students to participate in actual environmental protection technology breakthroughs. The government has established special scholarships and employment subsidies to attract outstanding talents to devote themselves to the field of green innovation. At the same time, it provides continuing education opportunities for on-the-job personnel, systematically enhancing the ability of practitioners to apply AI technology to environmental protection practices.
Sixth, it is suggested that blockchain technology be applied to environmental information disclosure, and tax incentives be given to enterprises with high transparency. By reducing information asymmetry, the accuracy of policy implementation can be enhanced, and the allocation efficiency of green innovation resources can be optimized. Key emission enterprises are required to use the blockchain platform to upload key data such as energy consumption and pollutant discharge in real time, ensuring that the information is unalterable and traceable. Establish an environmental information disclosure rating system and link the rating results to the credit and financing costs of enterprises. For enterprises with high-quality information disclosure, incentives such as environmental protection tax reduction and exemption and priority in green procurement will be provided to form a virtuous cycle of "transparent disclosure - policy support - innovation and improvement", enhancing market confidence in green technologies.
Appendix
Variable type | Variable | Variable explanation | Variable description |
|---|---|---|---|
The explained variables | GTE | Green R&D efficiency | DEA-SBM model calculation |
GAE | The efficiency of green technology transfer | ||
Core explanatory variable | inter | Policies for the new generation of artificial intelligence pilot zones | The year when a region implements the pilot policy for the new generation of artificial intelligence experimental zones and subsequent years is counted as 1; otherwise, it is counted as 0 |
Mediating variables | WW | Financing constraints | WW index |
Analyst | Analyst attention | ln (1 + number of analysts) | |
Media | Media attention | Ln (Number of media reports + 1 | |
Moderating variables | PEC | Public concern about the environment | Baidu's search index for smog |
TRANS | Enterprise information transparency | The earnings aggressiveness and earnings smoothness are adopted as the measurement indicators of information transparency | |
HC | Human capital level | The number of personnel/employees with a university degree or above | |
Control variables | SIZE | Company scale | Ln(Total Assets of the Enterprise at the end of the Year) |
TOP | The shareholding ratio of the largest shareholder | The shareholding ratio of the largest shareholder/the total share capital of the company | |
LEV | Asset-liability ratio | The presentation of the total liabilities/total assets at the end of the year of an enterprise | |
ROA | Profit rate on total assets | Net profit/total assets | |
AGE | Enterprise age | Ln (1 + equals the year of the current year minus the year of the company's establishment) | |
CASH | Cash holdings | Net cash flows from operating activities of the enterprise/total assets | |
GDP | The level of urban economic development | Ln (Urban GDP) | |
IS | Regional industrial upgrading | Added value of the tertiary industry/added value of the secondary industry in the region | |
DUAL | duality | If the chairman and general manager of an enterprise are the same person, it is 1; otherwise, it is 0 |
A
Funding
No funding was received for this study.
Data Availability
Data sharing is not applicable to this research as no data were generated or analysed.
Ethics Statements not Applicable
This article does not contain any studies with human participants performed by any of the authors
A
Author Contribution
In the research process of the paper *The Impact of Next Generation Artificial Intelligence Policies on the Green Innovation Efficiency of Firms*, the division of labor between Wang Jingbin and Chen Xiangge is clearly defined, with each undertaking core tasks that align with their respective roles and strengths. Wang Jingbin, serving as the supervisor, took the lead in shaping the **theoretical framework** of the study—a foundational work that underpins the entire research. His responsibilities included: 1. Determining the core research direction by integrating the academic frontier of "next-generation artificial intelligence policies" and the practical demand for "corporate green innovation efficiency", ensuring the study’s theoretical value and practical relevance. 2. Constructing the overall analytical logic, including clarifying the theoretical mechanisms through which AI pilot zone policies affect green innovation efficiency (e.g., the transmission paths of financing constraints, analyst attention, and media attention), and designing the hierarchical research framework covering benchmark tests, mechanism analysis, moderating effect tests, and heterogeneity analysis. 3. Guiding the formulation of research hypotheses (H1-H3) by drawing on existing literature on environmental regulation, digital economy, and AI economic effects, ensuring the hypotheses are theoretically rigorous and logically consistent with the research context. 4. Reviewing and optimizing the theoretical rationality of research design, such as the selection of multi-period DID method (to address the phased implementation of AI pilot zone policies) and the subdivision of green innovation efficiency into "green R&D efficiency (GTE)" and "green achievement transformation efficiency (GAE)", to avoid potential flaws in theoretical design. Chen Xiangge, as the student, assumed the primary responsibility for **data empirical work** and played a key role in supplementing and refining the study. His main tasks involved: 1. Data collection, sorting, and processing: Systematically collecting data of A-share listed companies from 2010 to 2023 (including enterprise-level indicators from WIND and CSMAR databases, and AI pilot zone policy data from the Ministry of Science and Technology), excluding samples of financial institutions, ST/*ST enterprises, and those with missing key data, and conducting 1% winsorization to ensure data validity—ultimately obtaining 31,623 valid observations. 2. Empirical calculation and model operation: Using Stata 18.0 to implement the multi-period DID model, including benchmark regression (to verify the positive impact of AI policies on green innovation efficiency), parallel trend test (to validate the DID model’s premise assumption), placebo test (to rule out random interference), and robustness tests (propensity score matching, regression method adjustment, excluding epidemic samples, adding control variables). He also calculated green innovation efficiency using the DEA-SBM model, with precise measurement of intermediate inputs/outputs for both GTE and GAE stages. 3. Mechanism and moderating effect verification: Specifically testing the mediating roles of financing constraints (WW index), analyst attention (ln(1+number of tracking analysts)), and media attention (ln(1+number of media reports)), as well as the moderating effects of public environmental attention (Baidu Haze Search Index), human capital level (proportion of employees with university degrees or above), and corporate information transparency (earnings aggressiveness and smoothness). 4. Heterogeneity analysis and result refinement: Conducting subgroup regressions based on city grades (first-tier/second-tier/third-tier), urban resource types (resource-based/non-resource-based), enterprise pollution attributes (polluting/non-polluting), life cycles (growth/mature/decline stages), and industry intensity (technology-intensive/capital-intensive/labor-intensive), and supplementing the analysis with descriptive statistics (Table 1) and result interpretation to ensure the comprehensiveness and depth of empirical findings.
References
1.
Haijian L, Huaguang H, Shushan Z et al (2023) The Green Innovation Effect of Supply Chain Digitalization [J]. Journal of Finance and Economics, 49(3): 4 to 18
2.
Jingyu B, En X, Cheng Lulu (2024) Research on the Realization Mode of Enterprise Green Innovation from the Perspective of R&D Alliance Combination [J]. Manage Rev 36(11):70–83
3.
Hao X, Miao E, Sun Q et al (2024) The impact of digital government on corporate green innovation: evidence from China[J]. Technol Forecast Soc Chang 206:123570
4.
Xu Yingzhi Z, Ruijie SW (2021) Green Technology Innovation, Factor Market Distortion and Industrial Structure Upgrading [J]. Res Dev Manage 33(6):75–86
5.
Wang Fengzheng Z, Yuxia XJ (2022) Heterogeneous Environmental Policies, Executive Risk Preferences and Green Technological Innovation: An Empirical Study Based on Heavily Polluting Listed Companies in China [J]. Sci Res Manage 43(11):143–153
6.
Chen Ming W, Qiaobao, Tang Wenqiao (2020) Research on Strategic Intelligence Assurance Capacity for Cyberspace Security [J]. J Intell 39(4):127–131
7.
Zhou D, Lu Z, Qiu Y (2023) Do carbon emission trading schemes enhance enterprise green innovation efficiency? Evidence from China’s listed firms[J]. J Clean Prod 414:137668
8.
Horbach J, Rammer C, Rennings K (2012) Determinants of eco-innovations by type of environmental impact—the role of regulatory push/pull, technology push and market pull[J]. Ecol Econ 78:112–122
9.
Tian, Hongbin (2020) Hao Wenwen. FDI, Environmental Regulation and Green Innovation Efficiency [J]. China Soft Sci, (8): 174–183
10.
Xu, Yujie (2022) Liu Shuguang Evolution of Spatial Pattern of Green Innovation Efficiency in the Yellow River Basin and its Influencing factors [J]. J Nat Resour 37(3):627–644
11.
Ma Zongguo Z, Qianqian JY (2022) Evaluation of Green and High-Quality Development of National Independent Innovation Demonstration Zones [J], vol 32. China Population, Resources and Environment, pp 118–127. 2
12.
Yang, Huan, Li Xiangju (2023) Research on the Impact Effect and Mechanism Identification of Government Innovation Subsidies on Enterprise Innovation Efficiency [J]. J Manag 20(4):558–567
13.
Tone K (2001) A slacks-based measure of efficiency in data envelopment analysis[J]. Eur J Oper Res 130(3):498–509
14.
Xiaofeng Q, Hongying Y Chinese-style Short Selling Mechanism and Corporate Innovation: A Natural Experiment Based on Phased Expansion of Margin Trading and Short Selling. [J] Manage World, 2017(1): 128–144, 187–188
15.
Zhao Yulin L, Chao P, Maomao (2022) R&D Resource Mismatch and Loss of Green Innovation Efficiency: An Empirical Analysis Based on China's High-tech Industry [, vol 39. J] Science & Technology Progress and Policy, pp 49–59. 4
16.
Chen Bin L (2020) Have fiscal decentralization and environmental regulations promoted green technological innovation in China? [J]. Stat Res 37(6):27–39
17.
Li C, Zhijia W (2023) The Impact of Carbon Emission Trading Policy on Green Technological Innovation of Enterprises: A Test Based on Instrumental Variables and Triple Difference [J]. Sci Sci Manage Sci Technol 44(5):15–33
18.
Lu Yanwei Z (2022) Research on the Impact of Digital Finance on Green Innovation Efficiency and Its Transmission Mechanism [J]. Financial Forum 27(8):52–62
19.
Xin Daleng P, Zhiyuan (2023) The Impact of the Pilot Policy of the Broadband China Strategy on Urban Green Innovation [J], vol 33. Population, Resources and Environment of China, pp 159–170. 9
20.
Yu Zhihui HC (2023) Can Information Infrastructure Construction Enhance the Green Innovation Level of Cities: An Empirical Analysis Based on 108 Prefecture-level Cities in the Yangtze River Economic Belt [J] Resources and Environment in the Yangtze River Basin. 32(9):1834–1848
21.
Chuanqing W (2019) Shen Yuqi. Research on the Impact Effect of Agglomeration of China's Equipment Manufacturing Industry on Green Innovation Efficiency [J]. Sci Technol Progress Policy 36(5):54–63
22.
Dong Huizhong L, Xuan Z (2021) Analysis of Spatio-temporal Characteristics and Driving Factors of Green Innovation Efficiency in the Guangdong-Hong Kong-Macao Greater Bay Area [J]. Econ Geogr 41(5):134–144
23.
Huang Huan H, Hui X, Yi et al (2024) Upgrading of Industrial Structure, Government Attention to Ecological Environment and Green Innovation Efficiency: Evidence Based on 115 Resource-based Cities in China [J]. J Nat Resour 39(1):104–124
24.
Bessen JE (2017) AI and jobs: the role of demand[J]. SSRN Electronic Journal
25.
Jiang Yonghong Z (2021) Benxiu. Research on the Mechanism and Countermeasures of Artificial Intelligence Affecting Income Distribution [J]. J Humanit, (7): 58–68
26.
Deng, Yue Jiang Wanyi. Industrial Robots, Management Capabilities and Enterprise Technological Innovation [J]. China Soft Sci, 2022(11): 129–141
27.
Ren Yinghua L, Yuzhao LH (2023) Artificial Intelligence Technological Innovation and Total Factor Productivity of Enterprises [J]. Economic Manage 45(9):50–67
28.
Yang Jin Promoting the Intelligent Transformation (2023) and Upgrading of Manufacturing through Disruptive Technological Innovation [J]. People's Forum · Acad Front, (17): 37–44
29.
Wu Qiang The (2023) Impact of New Technologies such as Artificial Intelligence on Occupational Substitution Risks [J]. Jianghuai Forum, (4): 69–78
30.
Zao S (2019) Hou Yulin. How Industrial Intelligence Reshapes the Employment Structure of the Labor force [J]. China Industrial Econ, (5): 61–79
31.
Baoping R (2024) (7) Wang Xin. The Framework and Path for the Deep Integration of Artificial Intelligence and the Real Economy to Form New Quality Productive Forces [J] Social Sciences, : 120–127
32.
Dezhi F (2024) J Yunnan Minzu Univ (Philosophy Social Sci Edition) 41(1):152–160Yu Shui. Generative Artificial Intelligence Large Models Boost High-Quality Development of the Real Economy: Theoretical Mechanism, Practical Foundation and Policy Path [J]
33.
Sun Haoying L (2023) The Economic Effects of Artificial Intelligence Policies: A Quasi-Natural Experiment Based on 2015 Artificial Intelligence Policies [J]. J Shanxi Univ Finance Econ 45(12):16–29
34.
Feng Wanxin Artificial Intelligence and Enterprise Financial Asset Allocation (2024) Empirical Evidence from National Artificial Intelligence Innovation Application Pilot Zones [J]. Contemp Finance Econ, (4): 141–152
35.
Ye Cuihong (2021) Financing Constraints, Government Subsidies and Green Innovation of Enterprises [J]. Stat Decis 37(21):184–188
36.
Zhao Can C, Shihua C (2020) Wei. Internet + Information Disclosure: Substantive Statements or Strategic Speculation - Evidence Based on the Risk of Stock Price Crash [J]. China Industrial Econ, (3): 174–192
37.
Li X, Jianbing H (2025) How can Digital Economy Policies Empower the Development of New Quality Productive Forces [J]. Economic Manage 47(2):5–26
38.
Liu Huihong Z (2024) Zheyuan. Can the pilot policies of smart cities improve the quality of ESG information disclosure? -- Empirical Research Based on Quasi-Natural Experiments [J]. J Nanjing Univ Finance Econ, (4): 56–66
39.
Pan, Chulin (2025) Jiang Yufeng Ethical Decision-making Mechanism of Digital Responsibility in Artificial Intelligence Enterprises. An Empirical Study Based on Throughput Model [J] Soft Science
40.
Wong JB, Zhang Q (2022) Stock market reactions to adverse ESG disclosure via media channels[J]. Br Acc Rev 54(1):101045
41.
Han, Jun, Ge Chunrui. Research on the Impact of National Low-Carbon Industrial Park Pilot Policies on Green Innovation of Enterprises [J]. J Jiangxi Univ Finance Econ, 2025(3): 23–36
42.
Liu Yiwen C, Xijun G, Jinglin et al (2023) Media Attention and Green Technological Innovation in Heavily Polluting Enterprises [J]. China Soft Sci, (9): 30–40
43.
Hongbo P, Haixia Y (2022) Does Stakeholder Innovation Focus Promote Enterprise Innovation? Evidence from Shenzhen Stock Exchange's Interactive Easy [J]. Nankai Bus Rev 25(3):85–96
44.
Yang, Qiuping (2024) Liu Hongzhong. Capital Market Openness and Enterprise Innovation: From the Perspective of Corporate Governance [J]. World Economic Res, (4): 73–90, 135–136
45.
Yu Minggui Z, Huijie F (2017) Analyst Focus and Enterprise Innovation: Empirical Evidence from China's Capital Market [J]. Economic Manage 39(3):175–192
46.
Siqi Z, Guanghua W, Weizeng S et al Public Demands and Urban Environmental Governance [J]. Manage World, 2013(6): 72–84
47.
Dong Kangyin L, Yang D, Xiucheng (2025) The Impact of Green Bonds on Regional Carbon Emission Reduction: From the Perspective of Environmental Concern [J]. J Beijing Inst Technol (Social Sci Edition) 27(4):42–53
48.
Zhang Kuan H (2019) Trade Openness, Human Capital and Independent Innovation Capacity [J]. Finance Trade Econ 40(12):112–127
49.
Yang C, Ridong H (2024) Enterprise Digital Transformation and Employee Income Gap [J]. Stat Decis 40(2):177–182
50.
DeBoskey DG, Gillett PR (2013) The impact of multi-dimensional corporate transparency on us firms’ credit ratings and cost of capital[J]. Rev Quant Financ Acc 40(1):101–134
51.
Renqiao X, Xiaoting C, Li Q (2022) Heterogeneous Environmental Regulation, Government Support and Enterprise Green Innovation Efficiency: A Two-Stage Value Chain Perspective [J]. Finance Trade Res 33(9):79–93
52.
Ju Xiaosheng L, Di Y (2013) Financing constraints, Working Capital Management and Sustainability of Enterprise Innovation [J]. Econ Res J 48(1):4–16
53.
Qiaoling F, Nutao Y (2024) Xu Hui. Research on the Governance Effects of Digital Transformation: From the Perspective of Accounting Information Quality [J]. Acc Res, (3): 34–50
54.
Zhao Wannan C (2025) Yan, Zhang Tiankai. Pollutant Discharge Permit System and ESG Greenwashing of Enterprises [J]. Macroeconomic Res, (4): 112–127
55.
Wu Libo Y, Meimin S, Keguo (2022) The Impact of Public Environmental Concern on Environmental Governance by Enterprises and Governments [J], vol 32. China Population, Resources and Environment, pp 1–14. 2
56.
Yang Jianfang G, Liutang ZQ (2006) Human Capital Formation and Its Impact on Economic Growth: An Endogenous Growth Model Including Education and Health Inputs and Its Test [J]. Manage World, (5): 10–18