Mandatory Environmental Regulation, Enterprise Labor Demand and Green Innovation Transformation: A Quasi-Experiment from China’s New Environmental Protection Law

Abstract

To improve the human living environment and maintain the balance of the ecosystem, the Chinese government implemented a new environmental protection law (NPL) in 2015. Based on data for Chinese A-share listed companies and prefecture-level cities from 2005 to 2020, a difference-in-difference model is used to empirically explore the impact of the mandatory environmental regulation on labor demand (LD) and green innovation transformation (GIT) for heavy pollution enterprise (HPE). The results indicate that NPL leads HPE to reduce LD and achieve GIT, compared to non-HPE. This finding still holds by a series of robustness tests. Lower financial constraints and higher fintech can alleviate the negative impact of the NPL on the LD of HPE and enhance the positive impact of the NPL on the GIT of HPE. From regional heterogeneity, NPL causes HPE to increase their labor in the eastern region but reduce labor in the middle and western regions. NPL positively affects the GIT and shows a “U” shape from the east-middle-west regions. From enterprise heterogeneity, NPL mainly has a significant dampening effect on the LD for old and high staff cost enterprises and has a greater positive impact on the GIT for these both types of enterprises. Meanwhile, there is a gradually increasing lag in the impact of NPL on LD and GIT. Our findings provide new perspectives for the government to implement the policy of NPL and for enterprises to transform development.

1. Introduction

The economic dividend generated by industrialization development is the result of countries gaining at the cost of rough resource utilization and environmental pollution [1]. Since the reform and opening up in 1978, high energy consumption with the goal of economic development has caused many pollution problems in China [2]. In other words, China is currently both the largest developing country and the most polluting emission country [3]. Specifically, China’s primary energy consumption accounted for 24.3% of global energy consumption, the main force of global energy consumption in 2019 [4] (Data source: https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp-stats-review-2020-full-report.pdf (Accessed on: 20 April 2023)). Data from the People’s Republic of China’s 2021 National Economic and Social Development Statistical Bulletin (Data source: http://www.stats.gov.cn/tjsj/zxfb/202202/t20220227_1827960.html (Accessed on: 20 April 2023)) show that GDP reached 1,143,670 billion yuan in 2021, up 8.1% over the previous year. Total energy consumption in the same period was 5.24 billion tons of standard coal, an increase of 5.2% over the previous year. However, The combined efficiency of energy processing and conversion is 81.0%, and the efficiency of thermal power generation is 42.0% in 2020 (Data source: http://www.stats.gov.cn (Accessed on: 21 April 2023)). This means that China’s economic development is dependent on crude energy consumption. The development of enterprise mode determines the level of environmental pollution as the main body in the process of economic development. There is no doubt that enterprises can optimize the allocation of labor, capital, and technology from the input side as well as regulate pollutant emissions from the production side. However, the incentive for enterprises to carry out reforms stems from economic interests and governmental drive. In the absence of government intervention, enterprises exacerbate environmental pollution by being unwilling to reform automatically in pursuit of market scale effects and initial factor of production advantages. Therefore, it is imperative to exert the power of competent government to encourage and support heavy pollution enterprises to shift from a labor-intensive to a green and innovative development model [5,6].

As major pollutant emitters, the production activities of enterprises are inevitably affected by environmental regulations. For example, Sarkodie and Adams (2018) found that the political system is an important factor in improving the quality of the environmental environment [7]. This suggests that environmental regulation can mitigate the emergence of a range of problems caused by environmental pollution such as diseases [8], inefficient economic growth [9], and social stability [10]. This is a deviation from the Sustainable Development Goals. Thus, environmental regulation serves as an effective tool to address the negative externalities of environmental pollution [11,12,13,14]. In recent years, the Chinese government has improved environmental quality through a dual approach of environmental protection policy and governance investment. For example, in terms of policies, the Chinese government has successively issued a series of policies such as the Air Pollution Prevention and Control Action Plan (2013), the Strategic Action Plan for Energy Development (2014–2020), the Three-Year Action Plan to Win the Blue Sky Defense War (2018), and the Action Plan for Carbon Peaking by 2030 (2022). For investment in environmental governance, the Ministry of Finance of the People’s Republic of China granted 53.2 billion yuan, 52.3 billion yuan, and 57.2 billion yuan for special protection of the ecological environment from 2019 to 2021, respectively. It has risen to 62.1 billion yuan in 2022. These policies and investments are aimed at upgrading pollution treatment facilities, accelerating the development of environmental prevention technologies, and supporting enterprises to save energy and reduce emissions, which promotes a virtuous cycle among society, economy, environment, and ecosystem.

The environmental protection policies introduced by the Government can alleviate environmental pollution [15,16]. Declining pollutant emission intensity is inextricably linked to changes in enterprises’ development patterns. Specifically, strict environmental regulations increase the production costs of polluting enterprises [17]. Under financing constraints, enterprises have a higher propensity to choose environmental technology investments [18], crowding out labor inputs. Among them, environmental regulation has a crowding-out effect on low-skilled labor in the region and a crowding-in effect on high-skilled labor. However, It has a crowding-out effect on both low-skilled and high-skilled labor in neighboring regions, and the negative impact on high-skilled labor is more obvious [19]. This effect is mainly observed in heavily polluting industries represented by textile printing [20] and manufacturing [21]. This implies that environmental regulation is an important factor influencing enterprises’ labor demand.

Furthermore, enterprises meet their emission reduction targets by changing their propensity to invest in both labor and green innovation under environmental regulations. Sustainable development depends on decoupling economic growth from resource use [22]. Thus, improving the efficiency of enterprise resources in all aspects becomes the optimal choice to decouple from economic growth, including energy efficiency [9], green investment [23], technological innovation [24,25,26], and greening [24]. Based on this background, does the new Environmental Protection Law of the People’s Republic of China (Hereinafter collectively referred to as the new environmental protection law) implemented in China cause heavy pollution enterprises (HPE) to change their labor demands and transform to green innovation? Scholars have not answered the question of the logical relationship between the new environmental protection law (NPL), labor demand (LD), and green innovation transformation (GIT). To promote social development and improve people’s living standards, we study the effect of NPL on the transformation and development of heavily polluting enterprises, which has important theoretical value and practical significance for the green development of enterprises and society.

The main contributions of this study are shown as follows: First, scholars studied the relationship between environmental regulation and LD [21], or between environmental regulation and green technology innovation [27]. However, labor and green technology innovation are two indispensable factors in the process of enterprise development. On the one hand, scholars explored the relationship between environmental regulation and LD, or green technological innovation. They ignore the inherent logical relationship between the three. On the other hand, in the process of profit-seeking enterprises, limited capital requires enterprises to continuously optimize resource allocation and improve production efficiency. In contrast, we explored command environmental regulation and LD, and further analyzed the relationship between command environmental regulation, LD, and GIT. The three variables are included in the same research framework and are systematically studied rather than explored in a compartmentalized manner. Second, we use NPL as a quasi-natural experiment and perform a difference-in-difference test. HPE such as textiles and chemicals are defined as the experimental group, and enterprises such as electronics and instrumentation, excluding service-oriented enterprises such as finance and education, are defined as the control group. This not only improves the empirical accuracy but also alleviates the bias caused by endogeneity. Third, we provide insight into the impact of NPL on the LD of HPE in terms of regions and enterprises, as well as differences in motivations for GIT in response to changes in labor demand. Meanwhile, the moderating effects of financial constraints and fintech on the NPL affecting LD and GIT of HPE are examined. This is of great practical significance to the government’s implementation of environmental protection policies tailored to local conditions, as well as for enterprises to fully utilize their comparative advantages to improve productivity. Finally, this study fills the gaps in existing research while providing lessons not only for the Chinese government to optimize its environmental policies and for firms to adjust their capital investment structure, but also for developing countries similar to China.

The remainder of this paper is presented as follows. The literature review is shown in Section 2. The theoretical mechanism is presented in Section 3. Section 4 describes the empirical model and variables. The empirical results of NPL affecting LD are analyzed in Section 5. The empirical results of GIT by enterprises reducing LD under policy constraints of NPL are discussed in Section 6. The research findings and policy implications are exhibited in Section 7.

2. Literature Review

2.1. Environmental Regulation and Employment

The mainstream views on studies related to environmental regulation and labor demand include the promotion theory, the suppression theory, and the no-effect theory. In terms of promotion theory, the employment growth effect of environmental regulation comes from the absorption effect of industrial structure upgrading [28]. Meanwhile, although technological innovation can create jobs, this effect applies mainly to technology researchers at the production end and end-of-pipe governance technicians [19]. Sun et al. (2019) investigated the impact of two control area policies on employment in China using a difference-in-difference model. They found that the implementation of this policy led to a flow of labor from large cities to small cities, which raises the size of employment in the primary and tertiary sectors in small cities [29]. Among them, the employment-enhancing effect mainly stems from investment-based environmental regulations [30]. In addition, both the LD and employment structure are positively affected by the carbon emissions trading pilot policy. R&D innovation and investment are the main channels through which the carbon emissions trading pilot policy boosts the highly educated labor force [31]. The policy mainly enhances the LD by guiding or forcing enterprises to participate in front-end environmental governance, which in turn increases LD. This is particularly evident among high-carbon-emitting firms and low-financing-constrained firms [32].

In terms of disincentive theory, Walker (2011) explored the relationship between environmental regulation and employment at the county level. Following the implementation of the Clean Air Act Amendments in 1990, environmental regulations reduced the size of employment in the regulated sector by 15% over 10 years [33]. Ferris et al. (2014) used panel data from fossil-fueled power plants to study the employment impacts of a sulfur dioxide trading program. The dampening effect of this project on employment is more pronounced in the first stage of the plant and there is a decaying effect over time [34]. Li and Lin (2022) took the Clean Air Initiative policy implemented in China from 2013 to 2017 as a perspective and found that the policy reduced haze pollution. However, air quality improvement comes at the expense of industrial employment and there is a significant lag [35]. Raff and Earnhart (2022) distinguished the workforce into productive labor and environmental labor. Environmental regulation is the reason for the decline in production employment and environmental employment, and the decrease in production is an important channel for this effect [36]. Output and substitution effects are the main mechanisms by which environmental regulations inhibit labor employment [37]. Government control of SO₂ and acid rain has reduced demand for secondary and tertiary employment [29]. This suggests that the negative impact of environmental regulation on labor is mainly in the industrial sector, with significant regional differences [38]. However, it can lead to an increase of LD in the finance, technology, and environmentally friendly sectors [39]. In addition, environmental regulations lead HPE to optimize the allocation of capital and labor by reducing labor and increasing capital, which strengthens the capital-labor ratio [40]. There is also a threshold effect of reduced labor supply, where environmental regulation leads to a decline in labor supply only when the air pollution level is higher than 30 ug/m³ [41].

In terms of nullification theory, Berman and Bui (2001) investigated the effects of government air regulation on employment using a Los Angeles perspective. They found that the policy is effective in reducing NOx emissions but did not have a substantial disincentive effect on employment. After the disruptions of plant exit and entry are removed, environmental regulation continues to have a less significant effect on employment [42]. The main reason is that government-regulated plants are capital-intensive industries. Cole and Elliott (2007) found that environmental regulation does have no statistically significant effect on employment, whether the cost of environmental regulation is exogenous or endogenous [43]. The above studies show inconsistent findings on the impact of environmental regulations on employment. Innovation, capital, and industrial structure are the main channels through which environmental regulation affects employment. In addition, scholars found that there is also variation in the impact of institutional factors on employment under environmental regulatory constraints based on differences in research perspective, methodology, and data levels (provincial, city, and firm).

2.2. Environmental Regulation and Green Innovation

The mainstream views on studies related to environmental regulation and green innovation include facilitation theory and nonlinear theory. In terms of facilitation theory, Luo et al. (2021) found that command environmental regulation and informal environmental regulation each have a significant “Porter effect” on green innovation. Market-based environmental regulations have a disincentive effect on green innovation [44]. Enterprises’ willingness to green innovation is the intermediate mechanism variable for this effect, and industrial agglomeration is the positive moderating variable in the transmission channel [45]. Yang et al. (2021) found that environmental regulations significantly improved urban green innovation capacity based on a pilot policy for water ecological civilization cities in China. This policy is more effective for small cities in terms of innovation and emission reduction effects [46]. Fang et al. (2021) studied by using a difference-in-difference model based on the new Environmental Protection Law implemented in China in 2015. The results show that government environmental regulation can effectively enhance green innovation in heavy pollution industries. The quality of information disclosure plays a mediating role [27]. In addition, heterogeneous environmental regulatory instruments can improve green innovation through a combination of hind legs and compensatory effects [47]. This effect is more prominent among firms with poorer internal and external governance and helps to drive long-term firm growth [48]. Although environmental regulations can encourage enterprises to improve fossil energy and reduce end-of-pipe pollution emissions, and promote green innovation among the HPE, they do not significantly affect new energy green innovation [49]. Meanwhile, scholars found that there is an asymmetry in the green innovation effects of differences in environmental standards [50]. Cleaner production standards can effectively promote green innovation; pollution emission standards have a negative impact on green technology innovation.

From a non-linear perspective, Fan et al. (2021) found that urban green innovation efficiency exhibits an inverted “U” shaped regional characteristic from the eastern to the middle to the western regions. Environmental regulation has a “U” shaped impact on urban green innovation industries [9]. That effect exhibits enhancement, cessation, weakening, U-shaped and inverted U-shaped features [51]. Similar conclusions were obtained by Li and Du (2021), and there is a spatial spillover effect. Shorter official tenure and lower corporate profitability can significantly weaken this effect [52]. In contrast, Pan et al. (2021) used Chinese patent census data to study the quantity and quality of green innovations. The results show that the relationship between environmental regulation and green innovation exhibits an inverted “U” shape and is insignificant in the middle, western, and high pollution levels regions [53]. Zhang et al. (2021) investigated the impact of environmental regulations on green innovation in the construction industry. They found that there is a “U” shape relationship between mandatory environmental regulations and green innovation. However, the effect of market-based environmental regulation on green innovation has an inverted “U” shape [54]. There is only a negative linear relationship between voluntary environmental regulation and green innovation.

In summary, The impact of employment and innovation on environmental quality, as inevitable factors driving economic development in various countries, become a hot issue of academic concern [55,56,57,58]. The performance of government functions is the main tool for regulating the direction of market economy development. In the wave of global industrialization and urbanization reform, China joined it, autonomously or involuntarily, to develop its economy and improve the living standards of its people. This action has placed China among the larger pollution emission countries while increasing the rate of economic growth. Thus, China officially implemented a new environmental protection law on 1 January 2015, taking on the responsibility of providing a “green home” for people around the world by reducing environmental pollution. Scholars delved into the impact of environmental regulation on industry employment or green technology innovation, respectively [53,59]. Academics analyzed the impact of the NPL on industry labor demand through green innovation mechanisms. The contributions of the existing research include the following two aspects. On the one hand, the impact of environmental regulation on employment or green innovation is debated, respectively. On the other hand, technological innovation is an intermediate variable in the impact of environmental regulation on employment, and the effects arising from the transmission process are equally debated. Unlike existing studies, we take a new perspective to explore the impact of the NPL on the labor demand of HPE, and whether HPE undergo the GIT in the face of reduced LD. While filling the gap of existing research, this study provides a reference for the Chinese government to optimize its environmental protection policy, and for enterprises to adjust their production structure and improve their production mode. More importantly, this study can also provide empirical evidence for developing countries similar to China.

3. Theoretical Mechanisms

3.1. Policy Background

The Environmental Protection Law of the People’s Republic of China was adopted and implemented at the 11th meeting of the Standing Committee of the Seventh National People’s Congress on 26 December 1989. However, the penalties imposed by the Ministry of Ecology and Environment of the People’s Republic of China are limited by the original law and do not exert environmental regulatory capacity [55]. Meanwhile, the earlier “promotion tournament” targeting economic growth further weakened the effectiveness of the policy’s implementation. Thus, on 24 April 2014, the Standing Committee of the 12th National People’s Congress amended the Environmental Protection Law of the People’s Republic of China and provided for its official implementation on 1 January 2015. The new “Environmental Protection Law of the People’s Republic of China” has been amended to increase the total number of articles from 47 to 70. Its penalties for polluting companies have risen to an all-time high [60]. The policy focuses on the comprehensive deployment of multi-dimensional aspects such as regulation development, multi-dimensional co-regulation, joint prevention and control, and environmental information transparency. The differences and implementation effects of the policy before and after the revision have been outlined in detail by Zhou et al. (2021) [55]. In addition, to better protect and improve the environment and promote the construction of ecological civilization, the Chinese government further amended the New Environmental Protection Law in 2023. Government departments are not only interested in expanding the scope of treatment of polluted environments but also greatly advocate the development of new energy sources and the development model of resource recycling. The policy also delivers environmental protection content to schools, cultivating students’ environmental awareness and realizing universal participation. In addition, protection subsidies, scientific and technological innovation, taxation, and regulatory requirements have been further refined (Data sources: https://cjjg.mee.gov.cn/xxgk/zcfg/202302/t20230208_1015860.html (Accessed on: 13 July 2023)).

3.2. Mechanisms of NPL Affecting LD

The mechanism of the influence of NPL on the LD can be divided into cost effect and output effect. First, in terms of cost effect. On the one hand, although this policy reduces pollutant emissions by alleviating collusion between government and business, environmental violations, and enforcement limitations, it also directly raises additional production costs for enterprises [55]. Specifically, the NPL strengthens enterprises’ emission reduction through daily penalties, seizure and seizure, production restrictions, and information disclosure. Enterprises are punished by fines for the transitional discharge of pollutants. If the enterprise fails to meet the emission standard, the administrative organ will impose a continuous penalty by the original penalty amount in units of days (Data sources: https://www.mee.gov.cn/ywgz/fgbz/fl/201404/t20140425_271040.shtml (Accessed on: 28 April 2023)). This pattern of cumulative penalties has resulted in higher sewage charges, Falling labor demand due to rising costs. For example, Li et al. (2022) explored the effect of emission charges on LD in manufacturing enterprises. The results show that high emission charges lead enterprises to reduce their LD [61]. This means that profitability constraints are another reason for enterprises to shrink their labor requirements. On the other hand, the intensity of environmental regulations that enterprises within the same industry can withstand varies because of differences in production scale, human capital structure, and profitability. For smaller production scale, insufficient capital, and backward process equipment enterprises (especially small enterprises), the higher marginal cost of production makes it impossible to afford the additional cost of pollution control, forcing such enterprises to shut down production and exit the market [62,63]. Meanwhile, the government forced the closure of outdated, low-end, and inefficient production capacity and discontinued units, resulting in a decline in LD. Thus, NPL raises the cost of pollution control for enterprises, leading to a decrease in their ability to absorb employment.

Output effect. Zhao and Zhang (2022) found that strict environmental regulations compel enterprises to curtail primary pollution factor inputs based on the Cobb Douglas function. Fossil energy (coal, oil, etc.) is the driving force behind the operation of HPE. Strict environmental regulations force enterprises to reduce fossil energy consumption and constrain production scale-up. A decline in the scale of production by firms usually affects profitability [64]. Of course, if enterprises want to maintain the original production scale, they should increase the cost of pollution control, resulting in higher production costs. Decreasing employee salaries and increasing product prices are the main ways that enterprises can shift additional costs. However, a decline in employee salaries directly reduces the motivation of the workforce, and some of the workforces will leave because of the salary reduction. Enterprises transfer the cost of pollution control by raising the price of their products, leading to a decline in market demand and lowering the production scale. Enterprises cut production as a reason for the decreased LD [21]. In addition, the NPL strengthens the central government’s environmental performance assessment of local governments, which in turn puts pressure on enterprises to reduce pollution. Faced with the risk of high pollution emission penalties, enterprises increase their demand for pollution control materials. This means that mandatory environmental regulation drives up the price of materials and products used to treat pollution, leading to lower profits and LD for HPE [64]. The willingness to demand labor has subsequently weakened. Therefore, hypothesis H1 is proposed.

H1.

The NPL raises the production cost of enterprises and reduces the scale of production, leading to a decrease in the LD of HPE.

3.3. Declining LD and GIT Response under the Policy Constraints of NPL

Command-based environmental regulation forces HPE to free up more capital by reducing LD. This helps to increase investment in clean technology research and introduce advanced digital and intelligent production equipment. This can improve production efficiency and reduce pollution at source. For example, industries such as mining, steel processing, textiles, and paper, which require a low-skilled labor force, can be easily replaced by smart automation technologies [65]. On one hand, Green technological innovations can both help HPE gain a higher market share and alleviate the environmental pollution penalties they face. HPE invest their capital in innovations that can bring long-term benefits, rather than in labor. In other words, technological innovation has a substitution effect on low-skilled labor under environmental regulation [66,67]. Some studies showed that environmental regulations can lead to technological innovation in HPE, which in turn creates more specialized skill-based jobs [68]. However, the substitution effect of innovation on low-skilled labor is greater than the creation effect on medium- and high-skilled labor, leading to a decline in the overall LD [69]. This implies that command environmental regulation makes HPE optimize their production mode to improve high-skilled labor, such as technology research and development, technology implementation, and technology maintenance, and reduce the demand for low-skilled labor. The upgrading of enterprise labor structure leads to technological progress.

On the other hand, enterprises reduce LD under policy constrains of NPL, impacting productivity. Enterprises choose to invest more capital in technological innovation and corporate R&D to maintain or increase their current productivity. Meanwhile, the NPL has improved the transparency of information by forcing enterprises to disclose the types of pollution emissions and emissions on time. At this point, if the enterprises choose to increase their labor force to improve productivity, this will inevitably lead to higher energy consumption, resulting in higher pollution emissions and putting them at a disadvantage. Thus, to obtain strong market competitiveness and share, enterprises prefer technological innovation to improve production efficiency or use clean technology to replace other production factor inputs such as labor [59]. This helps enterprises reduce production and operating costs and gain a better market share. In addition, the NPL likewise strengthens central government regulation of local governments through environmental protection and economic growth targets. Local officials exerted more stringent pressure on local industrial enterprises for environmental management to gain promotion opportunities. Rational enterprise managers choose technological innovation to optimize their energy consumption structure and production processes rather than lower the cost of labor to avoid costly environmental penalties [41]. Furthermore, local governments will boost environmental and innovation subsidies for HPE to support their green innovation under the central government’s environmental pressure. HPE rely too much on primary labor, leading to their single production structure and poor innovation and technology absorption capacity [70]. This reflects the greater risk in the process of innovation and transformation of enterprise production and pollution control technologies [71]. Government financial subsidies can reduce the cost and risk of enterprise environmental management, improve the precautionary psychology of green innovation transition and stimulate the incentive of green innovation investment. This indicates that under the policy impact of the New Environmental Protection Law, local governments increase environmental protection capital investment and lower taxes, leading to HPE preferring to reduce LD and realize GIT, therefore, hypothesis H2 is proposed.

H2.

HPE respond to the decline in LD through GIT under the policy constraints of NPL.

3.4. Mechanism Analysis

3.4.1. The Moderating Effect of Financial Constraints

The financial constraints play a moderating role in the NPL affecting the LD of HPE and the GIT, respectively. On the one hand, principal-agent problems and information asymmetry are important reasons for the emergence of financial constraints [72]. After the implementation of the NPL, HPE were included in the key emission list and forced to disclose detailed information on the types of pollutants and emissions. This can alleviate enterprise information asymmetry and financial constraints. Compared to HPE with high financial constraints, HPE with low financial constraints will have easier access to financing opportunities. This will help HPE to enhance their market competitiveness and market share and to obtain higher economic benefits. As the financial constraints decrease, the market competitiveness and market share of HPE rise, and higher economic benefits are obtained. Thus, HPE will have more funds to support investment in human capital, while enhancing investment in green innovation. In other words, Under lower financing constraints, green innovation activities that do not come at the expense of reducing LD become the goal of HPE to adjust their development model.

On the other hand, enterprises, as the main body of integrating market resources, absorb production factors while providing products and services for the market. Under the influence of the global green development concept, green products, and services are highly preferred by midstream and downstream enterprises and consumers [73]. The NPL can strengthen enterprises and public awareness of environmental protection. When HPE face lower financing constraints, they are more capable of increasing their scientific and technological staff and green R&D capital investment and providing green products and services to attract environmentally conscious consumers. In addition, HPE with lower financial constraints have a more flexible investment propensity compared to HPE with higher financial constraints. Lower financial constraints can mitigate the negative impact of the NPL on the LD and enhance the positive impact on the GIT of HPE. For example, Wang et al. (2022) explained the moderating effect of financial constraints on environmental regulations affecting green technology innovation from two perspectives: equity financial constraints and debt financial constraints. They found that a decline in financial constraints enhances the impact of environmental regulations on green technology innovation [74]. In conclusion, for HPE with low financial constraints, the NPL puts less pressure on them to face a capital crunch. This suggests that there is a declining incentive for such heavily polluting enterprises to release investment funds for pollution treatment by reducing labor demand. Meanwhile, HPE are more capable of the GIT under lower financial constraints. Therefore, hypothesis H3 is proposed.

H3.1.

Other things equally, the lower financial constraints mitigate the negative relationship between the NPL and the LD of HPE.

H3.2.

Other things equally, the lower financial constraints enhance the positive relationship between the NPL and the GIT of HPE.

3.4.2. Moderating Effect of Fintech

Under the wave of the development of the digital economy, emerging technologies represented by blockchain, artificial intelligence, and big data have penetrated the traditional financial field and have given rise to fintech. This not only meets the needs of the times but also forms a positive feedback mechanism that has a huge impact on the traditional financial ecological landscape [75]. According to the document “Financial Technology Development Plan (2022–2025)” released by the People’s Bank of China in 2022, fintech has gone from a starburst to a hundred barges, from basic support to driving change, powerfully enhancing the quality of financial services. This suggests that FinTech affects the relationship between the NPL and the LD and GIT of HPE by optimizing the quality of financial services, respectively. First, the NPL forces HPE to reduce pollution emissions and meet emission standards. Green innovation is the most efficient way to reduce pollution for HPE. It is generally characterized by a higher risk R&D process, long lead time, and difficult credit review. However, fintech has improved the accessibility of financial services through information technology tools, in turn, enhances the efficiency of financial institutions in collecting information on enterprises and improves the speed of capital placement in financial structures. Fintech shortens the cycle time for enterprises to obtain special funds for environmental protection [76] and eases financing constraints of HPE. In addition, fintech relies on the deep integration of visual technologies such as 5G high bandwidth and hybrid realization (MR) with banking scenarios to drive the upgrade of physical branches to multimodal and interactive smart branches. As a result, the flexibility and breadth of financial services are enhanced [77], and the operational efficiency of financial companies is increased. Meanwhile, big data technology promotes the accuracy and effectiveness of environmental capital allocation by reducing the information asymmetry between financial institutions and enterprises. The probability of borrowers diverting funds for pollution control to other non-pollution purposes decreases as the lenders’ monitoring efficiency and risk control over the use of funds increases [78]. This not only improves the action of HPE to transfer labor input funds to pollution control but also can reduce the caution of HPE to invest in green innovation.

Second, with the advancement of financial technology, the convenience of financial services has enabled residents to break through the geographical restrictions of credit funding and weakened the problem of adverse selection that exists in the traditional financial system. Buchak et al. (2018) suggest that fintech can enhance financial services that banks cannot provide under the traditional financial framework, facilitating borrowers by breaking the financial constraints of time and space [79]. This can smooth intertemporal consumption and ease consumption budget constraints for low- and middle-income groups, prompting firms to increase green product inputs. Especially after the implementation of the NPL, HPE provide more green products and optimize the production structure by continuously improving the demand for green innovative R&D personnel. Certainly, the negative impact of environmental pollution on the population increases the risk of disease [80]. Employees of enterprises migrate outward to avoid the negative externalities of environmental pollution. To compensate for the harm caused by environmental pollution, the middle- and high-skilled labor force will require HPE to pay the necessary “environmental injury compensation” to protect their rights and interests, which raises the financial pressure on HPE. Fintech can provide HPE with easy access to financing, making it easier not only to maintain a psychological cost-benefit balance for workers by adjusting their compensation structure but also to mitigate their incentive to release funds for pollution control by reducing their workforce. The level of green innovation in HPE will be raised in parallel. Therefore, hypothesis H4 is proposed.

H4.1.

Other things equally, the stronger level of fintech mitigates the negative relationship between the NPL and the LD of HPE.

H4.2.

Other things equally, the stronger level of fintech enhances the positive relationship between the NPL and the GIT of HPE.

4. Variable Description and Model Design

4.1. Sample Descriptions and Data Sources

Heavy pollution industries listed in A-shares from 2005 to 2020 are used as the experimental group, and other industries are used as the control group. According to the “List of Listed Industries for Environmental Verification and Management” issued by the Ministry of Ecology and Environment of the People’s Republic of China, the steel, mining, cement, coal, electrolytic aluminum, metallurgy, chemical, building materials, petrochemical, pharmaceutical, light industry, textile, tannery, thermal power, and transportation industries are defined as heavy pollution industries. The rest of the industries excluding finance, education, etc., are defined as non-heavy pollution industries. Meanwhile, the core enterprises regulated by the NPL are heavy pollution industries. According to the “Industry Classification Guidelines for Listed Companies” revised by the China Securities Regulatory Commission in 2012 (Data sources: http://www.csrc.gov.cn/csrc/c101864/c1024632/content.shtml (Accessed on: 28 April 2023)), the above industries are defined as the experimental group [55]. The remaining industries such as furniture, electronics, instruments, and equipment are defined as the control group. In addition, to ensure the correctness of the regression results, the samples are processed according to the following method: ① Removing enterprises with strictly missing data; ② Removing ST and *ST enterprises; ③ Removing enterprises with three years of discontinuity; ④ Removing enterprises such as finance, insurance, and education; ⑤ Continuous variable tailing is performed by winsor2 for 1% and 99%. Enterprises data are obtained from the CSMAR database, WIND database, Diebold database, and State Intellectual Property Office. To control the influence of macro factors on the empirical results, we introduce prefecture-level city data as control variables. The data of prefecture-level cities are obtained from the China City Statistical Yearbook.

4.2. Variable Description

4.2.1. Core Variables

Explained variables: Labor Demand (LD). Following the method used by Liu et al. (2021) [21], the total number of persons employed by each enterprise was used as a proxy variable (Unit: person); Green Innovation Transformation (GIT). The innovation labor ratio is used as a proxy variable calculated through the ratio of the number of utility model patent applications to the number of employees. The number of patents granted and the number of patent applications are used as two indicators by scholars to measure green innovation (Unit: units). The number of patent applications is timely to reflect the willingness and motivation of enterprises for green innovation. Therefore, the number of corporate utility model patent applications is used as a proxy variable for green technology innovation [21].

Core explanatory variable: Mandatory environmental regulation (NPL). We took 1 January 2015, as the time when the new environmental protection law policy was officially implemented. The time dummy variable Post takes the value of 1 in 2015 and beyond, and 0 before 2015. For individual dummy variables, Treat denotes the treatment and control groups. If the enterprise belongs to the heavy pollution industries, Treat takes the value of 1, otherwise, it takes the value of 0 [60].

Moderating variables: ① Financial constraints (Fic). To avoid endogeneity effects, we construct the SA index using only firm size and firm age, which are highly exogenous variables, following the [81]: (−0.737*Size) + (0.043*Size²) − (0.040*Age). Where Size represents the enterprise size, expressed as the natural logarithm of the enterprises’ total asset size, and Age represents the enterprise age, expressed as the value of the firm’s observed value year minus the year of establishment. Further, the calculated SA index is divided into two groups according to the median. Since the SA index is negative, a higher absolute value indicates higher financial constraints. As a result, samples with original values above the median are defined as weak financial constraints and take the value of 1. Samples with original values below the median are defined as strong financial constraints and take a value of 0. 0–1 dummy variables are used as proxy variables for financial constraints. ② Fintech (Fte). There are currently no variables that directly measure fintech indicators, we refer to [82,83] and use text analysis techniques to obtain important keywords related to fintech in the target prefecture-level cities. The specific method is as follows: we collect 48 core words based on the policy documents related to science and technology innovation, big data, and fintech issued by the central government (Figure 1). Further, we count the total number of fintech-related words with the help of Baidu’s advanced search function. A higher total word count indicates a higher level of fintech. Similarly, based on the median of the total number of fintech, we define the sample larger than the median as a stronger fintech region (assigned a value of 1), otherwise, it is defined as a weaker fintech region (assigned a value of 0).

4.2.2. Control Variables

Enterprise level: concerning existing literature [84,85], Bodie’s internal control index, book-to-market ratio, operating income, operating costs, gearing ratio, and cash holdings are controlled. At the prefecture-level city level: the advanced industrial structure, the intensity of fiscal science and technology expenditures, the informatization level, and the financial development level are controlled. The definitions of all the variables are presented in

Table 1. Variable definitions.

Type	Variable Name	Symbols	Explanation
Explained variables	Labor Demand	LD	Ln (The number of enterprise employees)
	Green technology innovation	GRI	Ln (Utility model patent applications + 1)
	Green Innovation Transformation	GIT	GRI/LD
Explanatory variables	Mandatory environmental regulation (NPL)	Treat	The sample is taken as 1 when the sample is a heavy pollution enterprise, otherwise, it is taken as 0.
		Post	The value is 1 if the sample belongs to 2015 and later, otherwise, it is 0.
		Treat × Post	The sample takes the value 1 when it is a heavy pollution industry and after the implementation of the new environmental protection law policy, otherwise, it takes the value 0.
Moderating variables	Financial constraints	Fic	0–1 dummy variables calculated using SA index as proxy variables
	Fintech	Fte	0–1 dummy variables using fintech calculations as proxy variables
Enterprise control variables	Bodie’s internal control index	Bic	Ln (the internal control index from the Bodie database)
	book-to-market ratio	Omc	Owner’s equity to market capitalization ratio
	operating income	Opi	Ln (enterprise’s total operating revenue)
	operating costs	Opc	Ln (enterprise’s total operating costs)
	Gearing ratio	Ger	The ratio of total liabilities to total assets
	cash holdings	Cah	The ratio of corporate annuity and cash equivalents balances to total assets
Prefecture-level city control variables	advanced industrial structure	Ais	The ratio of the output value of the tertiary sector to the secondary sector
	the intensity of fiscal science and technology expenditures	Fse	The proportion of government expenditure on science and education to GDP
	informatization level	Inf	Computer software industry employees accounted for the proportion of the total number of urban employees.
	financial development level	Fdl	The loan balance of financial institutions as a percentage of GDP

.

4.3. Model Design

First, the difference-in-difference model, as a more mature approach to evaluating quantitative models [86], has been widely applied in many subfields of economics for policy evaluation [87]. This approach verifies policy effectiveness by constructing interaction terms for individual and time dummy variables to compare the differences between the experimental and control groups before and after policy implementation. Meanwhile, this method also mitigates the endogeneity that exists among continuous variables and improves the accuracy of the quantitative analysis results. Therefore, we construct the following difference-in-difference model of the impact of NPL on LD, following the approach used by Liu et al. (2021) [59] and Liu et al. (2023) [88]:

$$L{D}_{it}={\alpha }_0+{\alpha }_{1}Trea{t}_i\times Pos{t}_t+Contro{l}_{it}+{\lambda }_i+{\eta }_t+{\mu }_i+{\epsilon }_{it}$$

(1)

where i and t denote enterprises and year, respectively; LD denotes enterprise labor demand; Treat is an individual dummy variable. If the enterprise belongs to the heavy pollution industries, Treat takes the value of 1 for the experimental group, otherwise, it takes the value of 0 for the control group; Post is a time dummy variable. A time range of 2005–2014 takes a value of 0. The time range 2015–2020 takes a value of 1; Control is a set of enterprise- and city-level control variables. α₁ is the estimated coefficient of the impact of the NPL on the LD. α₁ is significantly negative if the NPL makes LD of HPE decrease, otherwise, it is significantly positive; λ is the firm fixed effect; η is the year fixed effect; μ is the city fixed effect and ε is the random error term.

Second, to explore the green innovation transformation of HPE in response to the decline in LD caused by the NPL, we construct the following empirical model:

$$GI{T}_{it}={\beta }_1+{\beta }_{2}Trea{t}_i\times Pos{t}_t+Contro{l}_{it}+{\lambda }_i+{\eta }_t+{\mu }_i+{\epsilon }_{it}$$

(2)

where GIT is the enterprise green innovation transformation. Based on the literature review and mechanistic analysis, we construct the innovation-to-labor ratio for discussion. β₂ is significantly positive if HPE perform more innovation activities due to the decrease in LD caused by the NPL. The meaning of the remaining variables is interpreted in the same way as in Equation (1).

Third, to test the moderating effects arising from financial constraints and fintech, we construct the following moderating effect model.

$$\begin{array}{c}L{D}_{it}(GI{T}_{it})={\gamma }_1+{\gamma }_{2}Trea{t}_i\times Pos{t}_t+{\gamma }_{3}Trea{t}_i\times Pos{t}_t\times Fi{c}_{it}\left(Ft{e}_{it}\right)+{\gamma }_{4}Fi{c}_{it}(Ft{e}_{it})\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +Contro{l}_{it}+{\lambda }_i+{\eta }_t+{\mu }_i+{\epsilon }_{it}\hfill \end{array}$$

(3)

where Fic and Fte denote financial constraints and fintech intensity, respectively. γ_i denotes the regression coefficient of each variable. The rest of the variables are explained in Equation (1).

4.4. Statistical Description of Variables

Table 2. Descriptive statistics.

Type	Variables	Average	S.D.	25th	Median	75th
Explained variables	LD	7.705	1.146	6.917	7.658	8.454
	GRI	1.838	1.649	0	1.792	3.091
	GIT	0.233	0.202	0	0.238	0.403
Explanatory variables	Treat	0.295	0.456	0	0	1
	Post	0.528	0.499	0	1	1
	Treat × Post	0.145	0.353	0	0	0
Moderating variables	Fic	0.500	0.500	0	0.5	1
	Fte	0.288	0.453	0	0	1
Enterprise control variables	Bic	6.492	0.150	6.445	6.514	6.562
	Omc	0.301	0.146	0.190	0.278	0.389
	Opi	21.479	1.308	20.549	21.365	22.271
	Opc	21.119	1.409	20.125	20.986	21.976
	Ger	0.439	0.192	0.287	0.438	0.586
	Cah	0.154	0.109	0.075	0.125	0.204
Prefecture-level city control variables	Ais	1.378	0.855	0.867	1.113	1.533
	Fse	0.202	0.039	0.173	0.199	0.226
	Inf	0.028	0.029	0.009	0.017	0.039
	Fdl	1.496	0.612	0.979	1.481	1.973

demonstrates the statistical descriptive results for each variable. The observed values of each variable are 24,186. The mean, 25th percentile, and 75th percentile values for LD are 7.705, 6.917, and 8.454, respectively. The results indicate that there are differences in LD among the enterprises. The mean, 25th percentile, and 75th percentile values for green technology innovation are 1.838, 0, and 3.091, respectively. The mean, 25th percentile, and 75th percentile values for GIT are 0.233, 0, and 0.403, respectively. This means that green technology innovation and green innovation transformation also vary across enterprises. The mean value of Treat is 0.295, indicating that HPE account for 29.500% of the full sample. The mean value of Post is 0.528, indicating that the sample stands at 52.8% of the total sample after the implementation of the new environmental protection law policy. All other variables are within a reasonable range of values. Regarding the SOURCE + Statistical Software, Excel and Stata software are used to process the data.

5. Empirical Results of Mandatory Environmental Regulations Impacting Enterprise Labor Demand

5.1. Benchmark Regression Results

Table 3. Main regression results.

Variables	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)
Treat × Post	−0.053 *** (0.014)	−0.031 *** (0.012)	−0.046 *** (0.015)	−0.186 *** (0.014)	−0.025 ** (0.012)
Bic		−0.118 *** (0.020)		−0.193 *** (0.034)	−0.117 *** (0.020)
Omc		0.290 *** (0.031)		0.107 *** (0.039)	0.292 *** (0.031)
Opi		0.597 *** (0.021)		1.001 *** (0.019)	0.602 *** (0.021)
Opc		−0.062 *** (0.020)		−0.298 *** (0.018)	−0.066 *** (0.020)
Ger		0.397 *** (0.034)		−0.393 *** (0.036)	0.400 *** (0.034)
Cah		−0.290 *** (0.035)		−0.371 *** (0.049)	−0.284 *** (0.035)
Ais			0.025 * (0.015)	−0.068 *** (0.009)	0.073 *** (0.012)
Fse			0.706 *** (0.150)	−0.177 (0.127)	0.551 *** (0.122)
Inf			0.437 * (0.236)	0.178 (0.258)	0.243 (0.192)
Fdl			−0.007 (0.017)	−0.136 *** (0.010)	−0.017 (0.014)
Cons	7.713 *** (0.004)	−3.251 *** (0.163)	7.532 *** (0.045)	−5.718 *** (0.219)	−3.477 *** (0.168)
Enterprise fixed effect	YES	YES	YES	NO	YES
Time fixed effect	YES	YES	YES	NO	YES
Prefectural fixed effect	YES	YES	YES	NO	YES
R2	0.861	0.907	0.861	0.567	0.907
N	24,150	24,150	24,150	24,186	24,150

Notes: * p < 0.1, ** p < 0.05, *** p < 0.01; Standard errors are in parentheses.

presents the empirical results of the NPL impacting the LD. Model (1)–Model (3) show the estimation results without the inclusion of control variables, with the inclusion of firm-level control variables only, and with the inclusion of prefecture-level city control variables only, in that order. Model (4) captures the regression results without controlling for enterprises, time, and prefecture fixed effects. Model (5) shows the regression results controlling for both enterprises and prefecture-level influences, as well as enterprises, time, and prefecture-fixed effects. The regression coefficients of the above models are all negative and significant, indicating that the NPL effectively suppresses LD in HPE, compared to non-HPE. Hypothesis H1 is tested. Specifically, the regression coefficient of Treat × Post is −0.025 and significant at the 5% level in Model (5). This means that after the implementation of the NPL, the average LD in HPE is reduced by 55 people (exp(7.705) × (−0.025)). The main reason is that the NPL forced enterprises to invest a lot of capital in pollution control, resulting in higher production costs and a decline in production scale [29,53]. Furthermore, this causes a reduction of LD of HPE.

5.2. Robustness Tests

5.2.1. Parallel Trend Testing

The regression results in Table 3 hold on the premise that the difference-in-difference model satisfies the parallel trend assumption. If there are different trends in the enterprise labor demand of the experimental and control groups over time, the difference between the experimental and control groups before and after the implementation of the NPL may have a heterogeneous effect that is difficult to observe and is not caused by the NPL. Thus, we tested the parallel trends of the experimental and control groups before and after the implementation of the NPL. The model is constructed as follows.

$$L{D}_{it}={\alpha }_0+{\displaystyle \sum _{\mathrm{j}=-8}^5{\beta }_{j}Trea{t}_i\times Pos{t}_{2015+j}}+Contro{l}_{it}+{\lambda }_i+{\eta }_t+\mu +{\epsilon }_{it}$$

(4)

where Post_2015+j denotes the year dummy variable. The value of j ranges from −8 to 5. j < 0 means before the implementation of the NPL; j = 0 means in the year of implementation of the NPL; j > 0 means after the implementation of the NPL. The result of the parallel trend test is presented in Figure 2. According to Figure 2, the estimated results of all Treat × Post are insignificant before the implementation of the NPL. This means that there is an insignificant difference in the trend of LD change between the experimental and control groups. The estimated result of Treat × Post is significantly negative after the implementation of the NPL. This indicates that the implementation of the NPL leads to significant differences in the LD of the experimental and control groups. Thus, the result satisfies the parallel trend assumption and the estimation results are robust in Table 3.

5.2.2. Placebo Testing

Although we control for some of the characteristic variables at the enterprise and prefecture level in the main regression model, we still cannot exclude the influence of some unobserved characteristic factors on the regression results in Table 3. We further use a placebo testing method to identify the contingency of the NPL, referring to Lim et al. (2010) [89]. First, 3517 samples are randomly selected from the 24,186 observations as the virtual experimental group, and the remaining samples are used as the control group. Second, we construct the pseudo-interaction term (Treatfalse i × Post_t) for the enterprise individual and time. The estimation results after the sample are repeatedly sampled 1000 times are shown in Figure 3. According to Figure 3, the estimated coefficients of Treatfalse i × Post_t are distributed around 0 and are much smaller than the absolute values of the baseline regression coefficients (−0.025) in Model (5) of Table 3. In addition, the distribution pattern of the estimated coefficients resembles a normal distribution and most of the p-values are greater than 0.10 and insignificant at the 10% level. This suggests that the results of the NPL leading to the reduction of LD in HPE are not coincidental, corroborating the robustness of the main regression results in Table 3.

5.2.3. High-Dimensional Fixed Testing

The main regression results (Table 3) exclude the effects of differences in the enterprise’s own, time, and prefecture-level development on LD. Industry effect does not be ruled out. Thus, based on Equation (1), the industry fixed effect is controlled. The regression results are presented in Model (1) of

Table 4. Robustness testing.

Variables	High-Dimensional Fixed Testing	Removing Observations in 2015	Removing Provincial Capitals and Municipalities	Lag Period Method
	Model (1)	Model (2)	Model (3)	Model (4)
Treat × Post	−0.039 *** (0.012)	−0.049 *** (0.013)	−0.027 * (0.015)	−0.029 ** (0.014)
Control variables	YES	YES	YES	YES
Cons	−3.665 *** (0.166)	−3.774 *** (0.174)	−3.598 *** (0.207)	−0.628 *** (0.210)
Enterprise fixed effect	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
Prefectural fixed effect	YES	YES	YES	YES
Industry fixed effect	YES	YES	YES	YES
R2	0.912	0.912	0.923	0.908
N	24,149	22,354	12,852	19,532

Notes: * p < 0.1, ** p < 0.05, *** p < 0.01; Standard errors are in parentheses.

. The regression coefficient of Treat × Post is −0.039 and significant, confirming the robustness of the regression results in Table 3.

5.2.4. Removing Observations in 2015

The 2015 observations are removed from the full sample to test the robustness of the regression results (Table 3). The main reasons are as follows: ① Although the NPL was officially implemented in January 2015, local governments are constrained by institutional rigidity, behavioral habits, and conversion costs, and there is a certain time lag in the policy implemented. At the same time, after HPE are informed of the mandatory environmental regulations, the effectiveness of adjusting factor allocation, increasing investment in R&D and innovation, and optimizing production patterns are not immediately apparent. ② China’s Labor Contract Law, implemented in 2008, protects employees’ professional interests. This means that the rate of labor reduction in HPE is limited by the “labor contract” under the NPL and cannot be achieved in the current period. Of course, there is a fact in reality. Some of the low-skilled laborers sign labor contracts with the enterprises and some of the workers do not sign labor contracts with the enterprises. Due to the relatively low awareness of self-advocacy of this labor, HPE can fire low-skilled labor through lower default costs and transfer labor capital investment to emission control. Given this, we removed observations in 2015 and then used Equation (1) for estimation. The results are presented in Model (2) in Table 4. The regression coefficient of Treat × Post is −0.049 and significant, indicating that the regression results are robust in Table 3.

5.2.5. Controlling Urban Characteristics

The full sample of provincial capital cities and municipalities directly under the central government are included for estimation in Table 3, which may make the estimation results biased. The main reason is that China’s population, economy, resources, environment, and social development vary greatly from East to West region, and the impact of the NPL on LD can also vary. Meanwhile, provincial capitals and municipalities will be subject to the policy bias of the central and local governments, making HPE in these areas a better natural competitive advantage (such as manpower, capital, and technology). Therefore, 27 provincial capitals and four municipalities are removed from the full sample (The removed cities include: Beijing, Shanghai, Chongqing, Tianjin, Harbin, Changchun, Shenyang, Hohhot, Shijiazhuang, Urumqi, Lanzhou, Xining, Xi’an, Yinchuan, Zhengzhou, Jinan, Taiyuan, Hefei, Changsha, Nanjing, Chengdu, Guiyang, Kunming, Nanning, Tianjin, Lhasa, Hangzhou, Nanchang, Guangzhou, and Fuzhou City, Haikou City. Hong Kong, Macau and Taiwan provinces are not included in the sample.). The regression results are presented in Model (3) of Table 4. The regression coefficient of Treat × Post is −0.027 and significant, indicating that the regression is robust in Table 3.

5.2.6. The Lag Period of Enterprise Labor Demand

The difference-in-difference model can alleviate the endogeneity problem to some extent. LD is used as the explanatory variable and it is difficult to influence the core explanatory variable (NPL). This can alleviate the endogenous bias caused by reverse causation. In addition, to ensure the robustness of regression results in Table 3, the explanatory variable (LD) is lagged by one period to mitigate endogeneity. The results are shown in Model (4) of Table 4. The regression coefficient of Treat × Post is −0.029 and significant at the 1% level. This indicates that the main regression results are robust after accounting for the endogeneity issue.

5.2.7. PSM-DID Model Testing

Furthermore, we use PSM-DID models to explore the endogeneity problem caused by factors such as reverse causality and sample selection bias, referring to Fan and Zhang (2021) [90]. There may be systematic differences in the trends of changes between heavy pollution and non-heavy pollution enterprises, which affect the estimation results of the DID model. PSM-DID models can overcome the “selection bias” caused by biased estimation and sample “self-selection” [91]. The specific analysis steps for propensity score matching are as follows: ① Enterprise-and prefecture-level control variables are selected as covariates; ② We apply the Logit model to calculate the score value of the propensity to demand labor in HPE; ③ Propensity score matching method selection. Commonly used matching methods in academia include K-nearest-neighborhood matching, caliper matching, and kernel matching [92]. K-nearest-neighborhood matching is performed by finding the K different individuals with the closest propensity scores to match. We set K to 4 and perform 1:4 matching to minimize the mean square error. The caliper match is the absolute distance that limits the propensity score. We calculate the caliper value to be 0.029. Kernel matching selects the default kernel density function and broadband. To test the sample matching effect, the average treatment effect (ATT) of the NPL on LD is displayed in

Table 5. Average treatment effects for propensity score matching.

Matching Method	Average Treatment Effect	Standard Error	t-Test Value
K-nearest-neighborhood matching (K = 4)	−0.038 **	0.019	−1.98
Caliper Matching (caliper = 0.029)	−0.041 **	0.017	−2.40
Nuclear matching	−0.033 *	0.017	−1.92
Average value	−0.037

Notes: * p < 0.1, ** p < 0.05.

. The estimation results obtained after we applied the three methods of matching are generally consistent, indicating that the matched sample data are robust. The average net effect of the NPL on LD is −0.037. This demonstrates that after accounting for covariate bias, the NPL causes a 3.7% decrease in LD of HPE.

After the samples are K-nearest-neighborhood matching, caliper matching, and kernel matching, we perform regression using Equation (1). The results are presented in

Table 6. Regression results of the PSM-DID model.

Variables	K-Nearest-Neighborhood Matching (K = 4)	Caliper Matching (Caliper = 0.029)	Nuclear Matching
	Model (1)	Model (2)	Model (3)
Treat × Post	−0.038 *** (0.012)	−0.039 *** (0.012)	−0.039 *** (0.012)
Control variables	YES	YES	YES
Cons	−3.654 *** (0.168)	−3.653 *** (0.168)	−3.654 *** (0.168)
Enterprise fixed effect	YES	YES	YES
Time fixed effect	YES	YES	YES
Prefectural fixed effect	YES	YES	YES
Industry fixed effect	YES	YES	YES
R2	0.912	0.912	0.912
N	24,146	24,144	24,146

Notes: *** p < 0.01; Standard errors are in parentheses.

. After K-nearest-neighborhood matching and kernel matching, the control group has three observations that are not the common support domain, respectively, and the matched observations are 24,183. After caliper matching, the two observations of the experimental group and the three observations of the control group are not in the common support domain, and the matched observations are 24,181. The estimated coefficients of Treat × Post for K-nearest-neighborhood matching, caliper matching, and kernel matching are −0.038, −0.039, and −0.039, respectively, and all are significant at the 1% level. This confirms that the NPL indeed leads to a reduction in LD for HPE.

5.3. Heterogeneous Analysis

5.3.1. Regional Heterogeneity

Due to the large variation in geographic, resource, and economic factors in China [85], the sample is divided into three subsamples, East, Middle, and West, to analyze the spatial heterogeneity of the impact of NPL on the LD of HPE. Regression results are shown in

Table 7. Regional heterogeneity (Eastern China: Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, Hainan; Middle China: Shanxi, Inner Mongolia Autonomous Region, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei, Hunan; Western China: Xinjiang Uygur Autonomous Region, Tibetan Autonomous Region, Qinghai, Yunnan, Guizhou, Ningxia Hui Autonomous Region, Gansu, Guangxi Zhuang Autonomous Region, Yunnan, Sichuan, Chongqing).

Variables	Eastern China	Middle China	Western China
	Model (1)	Model (2)	Model (3)
Treat × Post	0.029 * (0.015)	−0.061 ** (0.024)	−0.148 *** (0.033)
Control variables	YES	YES	YES
Cons	−3.227 *** (0.212)	−4.723 *** (0.348)	−2.539 *** (0.437)
Enterprise fixed effect	YES	YES	YES
Time fixed effect	YES	YES	YES
Prefectural fixed effect	YES	YES	YES
R2	0.906	0.923	0.891
N	16,721	4399	3079

Notes: * p < 0.1, ** p < 0.05, *** p < 0.01; Standard errors are in parentheses.

by using Equation (1). The impact of the NPL on the LD of HPE exhibits a spatial characteristic that is promoted in the eastern region and inhibited in the middle and western regions. The estimated coefficient of Treat × Post is 0.029 and significant at a 1% level in the east. The main reason is that the eastern region is the concentration area of quality factors such as China’s economy, medium and high-skilled personnel, and information. After the implementation of the NPL, HPE can adjust their production patterns more quickly with the advantage of the region, leading to an increase in LD. The estimated coefficients of Treat × Post are −0.061 and −0.148 and are significant in the middle and western regions, respectively. This means that the NPL has a significant dampening effect on LD for HPE in the middle and western regions. The main reason is that HPE are limited by factor endowments and cannot flexibly adjust the production structure as the eastern region with the advantage of economic scale under strict constraints of NPL. To reduce pollution emissions, HPE may prefer to reduce LD for increasing investment in science and technology innovation and R&D technology or green technology introduction within their limited financial capacity. The suppressive effect of the NPL on LD for HPE is higher in the western region than in the middle region. This suggests that HPE in the West with the lowest absolute advantage in factor endowments have a greater motivation to reduce capital investment in terms of labor. This confirms that HPE decrease labor due to the absolute disadvantage of factor endowment under the NPL.

5.3.2. Enterprise Heterogeneity

(1)

New and old enterprises

The time of enterprise establishment reflects the production scale of the enterprise to some extent. The longer the enterprise is established, the larger the production scale and the higher the market share is likely to be [93]. After the implementation of the NPL, HPE established at different times may adapt to different measures to cope with pollution control. Therefore, we use the median of the sample of enterprise establishment time as the dividing line and divide the sample into two groups: old enterprises (long establishment time) and new enterprises (short establishment time). The results are presented in Model (1) and Model (2) of

Table 8. Enterprises Heterogeneity.

Variables	New Enterprises	Old Enterprises	Low Staff Cost	High Staff Cost	Short-Term	Long-Term
	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)
Treat × Post	0.002 (0.016)	−0.057 *** (0.019)	0.021 (0.017)	−0.056 *** (0.016)	0.023 (0.015)	−0.021 * (0.012)
Control variables	YES	YES	YES	YES	YES	YES
Cons	−3.318 *** (0.218)	−2.897 *** (0.259)	−1.131 *** (0.234)	−2.183 *** (0.256)	−3.493 *** (0.168)	−3.480 *** (0.168)
Enterprise fixed effect	YES	YES	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES	YES	YES
Prefectural fixed effect	YES	YES	YES	YES	YES	YES
R2	0.928	0.902	0.833	0.872	0.907	0.907
N	12,755	11,317	11,926	11,955	24,150	24,150

Notes: * p < 0.1, *** p < 0.01; Standard errors are in parentheses.

. The estimated coefficient of Treat × Post is 0.002 and insignificant in the new enterprise group. The estimated coefficient of Treat × Post is −0.057 and significant in the old enterprise group. This suggests that there is an obvious difference in the impact of the NPL on the LD of new and old HPE. In other words, the NPL reduces the LD of the old HPE but does not affect the LD of the new HPE. The main reason is that old HPE have better shock resistance through their long economic accumulation. They can adjust their production structure more flexibly to reduce pollutants under the strict policy constraints of the NPL. Specifically, the old HPE can increase investment in environmental treatment to meet pollution emission standards, while foregoing the economic effects generated by labor factor inputs. As new HPE are in the production scale expansion phase, the possible incentive to reduce labor remains low under strict environmental governance constraints.

(2)

Enterprise labor cost

Regardless of the length of time when HPE are established, the cost of labor payments determines the enterprises’ willingness to adjust their production patterns to some extent [94]. Based on the median cash paid by enterprises to employees, the sample is divided into two groups: low staff cost and high staff cost. The results are presented in Model (3) and Model (4) of Table 8. The regression coefficient of Treat × Post is 0.021 and insignificant in the group of low staff cost. The regression coefficient of Treat × Post is −0.056 and significant in the group of low staff cost. This reflects the difference in the impact of the NPL on the LD of HPE in the low and high labor cost groups. The NPL leads to a decrease in LD for HPE in the high labor cost group but failed to effectively influence LD for HPE in the low labor cost group. Therefore, this reflects that when HPE are faced with higher employee wage costs under the policy constraints of NPL, they are more inclined to increase investment in pollution emission treatment by reducing LD. On the contrary, if the cost of employee wages borne by HPE is low, their tendency to reduce LD in the process of raising capital investment in pollution control is low.

5.3.3. Dynamic Effects

To explore the dynamic effects of the NPL on the LD of HPE, we investigate the effects of the NPL on the LD referring to Liu et al. (2022) [94] in the short-run shocks and long-run shocks. The methodology is presented as follows: in the short-term shock, the time dummy variable (Post) is defined as 1 in the first 2 years (2015–2016) after the implementation of the NPL, otherwise, it is defined as 0. In the long-term shock, the time dummy variable (Post) is defined as 1 for the first 4 years (2015–2018) after the introduction of the NPL, otherwise, it is defined as 0. The other time dummy variables are taken as 0 for either the long or short term. Model (5) and Model (6) in Table 8 show the effects of the NPL affecting the LD of HPE. The regression coefficient of Treat × Post is 0.023 and insignificant for short-term shocks. The regression coefficient of Treat × Post is −0.021 and significant under long-term shocks. This implies that there is a lag in the reflection of the reduction in LD of HPE along with the advancement of NPL. Furthermore, the full-sample regression coefficient of Model (5) in Table 3 is −0.025 and significant, indicating that the negative impact of the NPL on the LD of HPE gradually increases.

5.4. Mechanism Analysis

5.4.1. Examination of the Moderating Effect of Financial Constraints

According to the regression results of Model (1) in

Table 9. Regression results of the moderating effect of financial constraints and fintech.

Moderating Variables	Financial Constraints		Fintech
Variables	Model (1)	Model (2)	Model (3)	Model (4)
Treat × Post	−0.074 *** (0.014)	−0.087 *** (0.016)	−0.037 *** (0.012)	−0.048 *** (0.014)
Treat × Post × Fic	0.118 *** (0.019)	0.130 *** (0.021)
Fic	−0.029 *** (0.011)	−0.025 ** (0.012)
Treat × Post × Fte			0.071 *** (0.022)	0.082 *** (0.023)
Fte			−0.006 (0.008)	−0.008 (0.008)
Control variables	YES	YES	YES	YES
Cons	−3.411 *** (0.168)	−3.521 *** (0.176)	−3.442 *** (0.168)	−3.554 *** (0.176)
Enterprise fixed effect	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
Prefectural fixed effect	YES	YES	YES	YES
R2	0.907	0.906	0.907	0.906
N	24,110	22,341	24,110	22,341

Notes: ** p < 0.05, *** p < 0.01; Standard errors are in parentheses.

, the estimated coefficient of Treat × Post is −0.074 and significant. This indicates that HPE with strong financial constraints are putting capital into pollution treatment by reducing LD under the pressure of the NPL. The estimated coefficient of Treat × Post × Fic is 0.118 and significant. This result suggests that HPE with weaker financial constraints have a decreasing incentive to reduce LD under the NPL compared to HPE with stronger financial constraints. This implies that lower financial constraints can mitigate the negative impact of the NPL on the LD of HPE. Hypothesis H3.1 is tested. The main reason is that compared to the HPE with strong financial constraints, the HPE with low financial constraints achieve pollution emission standards through rapid financing after the implementation of the NPL, thereby lessening the behavior of reducing LD. To ensure the robustness of the empirical results, we follow the 5.2.4 robustness tests and remove observations in 2015. The regression results are presented in Model (2) of Table 9. The sign of the regression coefficients for each variable is consistent with that of Model (1), indicating robust results.

5.4.2. Examination of the Moderating Effects of Fintech

According to the regression results in Model (3) of Table 9, the regression coefficient of Treat × Post is −0.037 and significant. This suggests that the LD of HPE in areas with weaker fintech is reduced by the implementation of NPL. In areas where financial technology is weak, the implementation of NPL leads to greater difficulty in accessing financial services for HPE due to poor digital technology. Reducing labor becomes one of the options for HPE to improve their ability to treat pollution. The regression coefficient of Treat × Post × Fte was 0.071 and significant. This means that in regions with a high level of financial technology, financial institutions rely on big data, artificial intelligence, cloud computing, and other information technology to accurately collect information on HPE and improve the speed of special environmental funds. This suggests that the behavior of HPE to reduce LD in areas with higher levels of fintech is improved due to the NPL. Hypothesis H4.1 is tested. To test the robustness of the results in Model (3) of Table 9, we use the 5.2.4 robustness test and remove observations in 2015. The results are presented in Model (4) of Table 9. The sign of the estimated coefficients of Model (4) indicates that the regression results of Model (3) are robust in Table 9.

6. Reducing Enterprise Labor Demand and Green Innovation Transition Response under the Policy Constraints of New Environmental Protection Law

6.1. Benchmark Regression Results

Furthermore, the NPL affecting the GIT of HPE is examined. First, the empirical results in Section 5 indicate that the NPL makes HPE reduce their labor. According to the theoretical mechanism analysis, the NPL makes HPE promote the level of green technology innovation as an important measure to respond to the reduction of LD. Thus, we examine the impact of the NPL on the green technology innovation of HPE (Model (1) in

Table 10. Regression results of reduction in enterprise labor demand and response to green innovation transition.

Explained Variables	Enterprise Green Innovation	Enterprise Green Innovation Transformation
Variables		Main Regression Results	High-Dimensional Fixation	Removing Observations in 2015	Removing Provincial Capitals and Municipalities	Lag Period Method	K-Nearest-Neighborhood Matching (K = 4)
	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)	Model (7)
Treat × Post	0.298 *** (0.027)	0.039 *** (0.003)	0.039 *** (0.003)	0.046 *** (0.004)	0.026 *** (0.005)	0.025 *** (0.004)	0.039 *** (0.003)
Control variables	YES	YES	YES	YES	YES	YES	YES
Cons	−5.709 *** (0.370)	−0.350 *** (0.047)	−0.403 *** (0.048)	−0.419 *** (0.049)	−0.308 *** (0.067)	−0.231 *** (0.057)	−0.401 *** (0.048)
Enterprise fixed effect	YES	YES	YES	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES	YES	YES	YES
Prefectural fixed effect	YES	YES	YES	YES	YES	YES	YES
Industry fixed effect	YES	NO	YES	YES	YES	YES	YES
R2	0.790	0.764	0.770	0.772	0.761	0.779	0.770
N	24,149	24,150	24,149	22,354	12,582	19,532	24,146

Notes: *** p < 0.01; Standard errors are in parentheses.

). The results show that the NPL has a positive impact on the green technology innovation of HPE, compared to non-heavy pollution enterprises. Second, we test the impact of the NPL on the GIT by using Equation (2). The regression results are presented in Model (2)–Model (7) of Table 10. The estimated coefficient of Treat × Post in Model (2) is 0.039 and significant. This indicates that under the policy constraint of the NPL, HPE choose to reduce labor, engage in more green innovation activities and shift to green technology-intensive [42]. Hypothesis H2 is verified. This means that while the NPL implemented in 2015 negatively influences the LD of HPE, it also prompts the HPE to achieve GIT. This also helps to promote the development of China’s green economy.

To ensure that the regression results are robust, we use the method of high-dimensional fixation, removing observations in 2015, removing observations in four municipalities and provincial capitals, one-period lags of the explanatory variables, and PSM-DID to test. Since the specific steps of the robustness test are the same as the robustness test method in Section 5, they are not repeated here. From high-dimensional fixation, removing observations in 2015, removing observations in four major municipalities and provincial capitals, one-period lags of the explanatory variables, and PSM-DID, the regression coefficients are positive and significant in Model (3)–Model (7). This indicates that the conclusion still holds after excluding a series of potential influencing factors.

6.2. Heterogeneity Analysis

6.2.1. Regional Heterogeneity

The regression results in

Table 11. Regional heterogeneity.

Variables	Eastern China	Middle China	Western China
	Model (1)	Model (2)	Model (3)
Treat × Post	0.044 *** (0.004)	0.026 *** (0.007)	0.034 *** (0.009)
Control variables	YES	YES	YES
Cons	−0.270 *** (0.059)	−0.507 *** (0.110)	−0.449 *** (0.117)
Enterprise fixed effect	YES	YES	YES
Time fixed effect	YES	YES	YES
Prefectural fixed effect	YES	YES	YES
R2	0.766	0.759	0.756
N	16,721	4399	3079

Notes: *** p < 0.01; Standard errors are in parentheses; The regional division is the same as Table 7.

demonstrate the spatial differences between the reduction of LD and the GIT response under the policy constraints of NPL. According to the regression results in Table 11, the NPL significantly contributes to the GIT of HPE in Eastern, Middle, and Western China. This effect shows a “U” shaped spatial characteristic with the eastern region being the largest, followed by the western region and the middle region being the smallest. Specifically, from the eastern region, the regression coefficient of Treat × Post is 0.044 and significant, implying that the NPL provides an incentive for HPE to achieve GIT. The NPL significantly promotes the increase of LD in the HPE in the eastern region in Model (1) of Table 7. This indicates that HPE under the constraints of the NPL leverage the region’s high-quality factor endowment and rich operational experience to enable them to not only absorb more labor but also synchronize their GIT in the eastern region. For the middle and western regions, the impact of the NPL on the GIT of HPE is 0.026 and 0.034, respectively. According to the regression results of Model (2) and Model (3) in Table 7, the NPL significantly leads to an LD reduction of HPE in the middle and eastern regions, respectively, and has the largest dampening effect on the west. Unlike the eastern region, this means that the GIT of HPE for reducing pollution emissions needs to be conducted at the cost of reducing labor under the policy constraints of the NPL in the middle and western regions. Of course, the NPL has the greatest effect of reducing labor for HPE in the West, and the corresponding greatest effect of promoting the GIT. Of course, according to the reality of regional development differences, China’s economic development has a decreasing phenomenon from the East to the center and finally to the West. The eastern region has higher economic development potential and market size, which generates higher wage premiums and attracts laborers from central and western regions to move in. This provides sufficient human capital for the development of enterprises in the eastern region. When the new environmental protection law policy was implemented, the HPE in the eastern region choose to increase the demand for medium- and high-skilled labor to promote green innovation and improve productivity. On the contrary, for the central and western regions, the implementation of the new environmental protection law forces HPE to reduce pollutants. The lack of labor leads to higher labor costs, raising firms’ production costs. Enterprises choose to introduce advanced technologies and invest more in green innovation R&D in order to improve production efficiency. This not only improves operational efficiency but also improves environmental conditions.

6.2.2. Enterprise Heterogeneity

(1)

New and old enterprises

The results of the NPL affecting GIT in new and old enterprises are shown in Model (1) and Model (2) of

Table 12. Enterprise Heterogeneity.

Variables	New Enterprises	Old Enterprises	Low Staff Cost	High Staff Cost	Short-Term	Long-Term
	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)
Treat × Post	0.030 *** (0.005)	0.042 *** (0.005)	0.024 *** (0.006)	0.046 *** (0.004)	−0.007 (0.004)	0.006 * (0.003)
Control variables	YES	YES	YES	YES	YES	YES
Cons	−0.179 ** (0.070)	−0.458 *** (0.067)	−0.232 *** (0.079)	−0.420 *** (0.071)	−0.331 *** (0.047)	−0.334 *** (0.047)
Enterprise fixed effect	YES	YES	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES	YES	YES
Prefectural fixed effect	YES	YES	YES	YES	YES	YES
R2	0.776	0.774	0.743	0.814	0.762	0.762
N	12,755	11,317	11,926	11,955	24,150	24,150

Notes: * p < 0.1, ** p < 0.05, *** p < 0.01; Standard errors are in parentheses.

. The division method of old and new enterprises is the same as in Table 8. The regression coefficient of Treat × Post is 0.030 and significant in the new enterprise group. The regression coefficient of Treat × Post is 0.042 and significant in the old firm group. The results show that the NPL causes both new and old HPE to the GIT and has a greater impact on the old HPE. The main reason is that the old HPE already have a good production scale and can easily transform themselves into green innovation by reducing labor investment and transferring more capital to R&D and innovation in the face of strict pollution emission standards. However, new HPE may be in the growth phase. The GIT with no reduction of labor is the optimal decision under the pressure of the NPL to reduce emissions. Regression results in Model (1) and Model (2) of Table 8 verify the LD motivation adopted by new and old HPE in the face of the NPL.

(2)

Enterprise labor cost

Following the methodology in Table 8, the sample is divided into two groups: low staff cost and high staff cost. The regression results are presented in Model (3) and Model (4) of Table 12. The regression coefficient of Treat × Post is 0.024 and significant in the group of low staff cost. The regression coefficient of Treat × Post is 0.046 and significant in the group of high staff cost. The results indicate that the NPL prompts both types of HPE (low staff cost and high staff cost) to transition to green innovation, and this effect is more pronounced in the group of high staff cost. This reflects that compared to the low staff cost group, HPE in the high staff cost group have a higher GIT effect by reducing labor to relieve capital pressure for green innovation, allowing them to cope with the high emission standards of the NPL. HPE prefer to make GIT without reducing labor when they reduce emission pollution under the policy constraints of NPL in the group of low staff cost. However, its GIT is less efficient compared to the high cost of the labor group. The main reason is that the HPE do not reduce labor (Model (3) of Table 8) under the NPL, sharing part of the funds allocated to green technology innovation in the low labor cost group.

6.2.3. Dynamic Effects

Using the methodology in Table 8, the sample is divided into two discussion groups for short-term shocks and long-term shocks to the implementation of the NPL. The regression results are presented in Model (5) and Model (6) of Table 12. The regression coefficient of Treat × Post is −0.007 and insignificant in the short term of the implementation of the NPL. The regression coefficient of Treat × Post is 0.006 and significant in the long term of the implementation of the NPL. This result indicates that there is a lag in the impact of the NPL on the GIT of HPE. Meanwhile, the full-sample regression result of the NPL affecting the TIG is 0.039 and significant in Model (2) of Table 10, indicating that the effect of the NPL on the GIT shows a dynamic and increasing trend over time. This means that after the implementation of the NPL, HPE are constrained by the original production structure, and cannot achieve GIT in the short term. This lagging characteristic is also present in the NPL affecting the LD of HPE (Model (5) in Table 3 and Model (5) and Model (6) in Table 8). As the NPL advances, HPE gradually adjust their production structure to achieve GIT by reducing labor input and increasing capital investment in green innovation research. This is in line with the reality of the process of optimizing the development model of the enterprise.

6.3. Mechanism Analysis

6.3.1. Examination of the Moderating Effect of Financial Constraints

According to the regression results of Model (1) in

Table 13. Regression results of the moderating effect of financial constraints and fintech.

Moderating Variables	Financial Constraints		Fintech
Variables	Model (1)	Model (2)	Model (3)	Model (4)
Treat × Post	0.030 *** (0.004)	0.037 *** (0.004)	0.037 *** (0.004)	0.044 *** (0.004)
Treat × Post × Fic	0.023 *** (0.005)	0.024 *** (0.006)
Fic	−0.001 (0.003)	−0.000 (0.003)
Treat × Post × Fte			0.013 ** (0.006)	0.013 * (0.007)
Fte			−0.001 (0.002)	−0.000 (0.002)
Control variables	YES	YES	YES	YES
Cons	−0.344 *** (0.047)	−0.351 *** (0.049)	−0.347 *** (0.047)	−0.355 *** (0.049)
Enterprise fixed effect	YES	YES	YES	YES
Time fixed effect	YES	YES	YES	YES
Prefectural fixed effect	YES	YES	YES	YES
R2	0.764	0.767	0.764	0.767
N	24,110	22,341	24,110	22,341

Notes: * p < 0.1, ** p < 0.05, *** p < 0.01; Standard errors are in parentheses.

, the estimated coefficient of Treat × Post is 0.030 and significant. This means that the NPL prompts HPE with high financial constraints to transition to green innovation. Specifically, according to the regression results in Model (1) of Table 9, the negative effect of the NPL on LD is greater for HPE with high financial constraints. This indicates that such HPE shift the capital invested in labor to technology development, enhancing green innovation. The regression coefficient of Treat × Post × Fic is 0.023 and significant. This result indicates that the effect of transitioning to green innovation is enhanced for HPE with weaker financial constraints compared to HPE with stronger financial constraints due to the NPL. In other words, the lower financial constraints enlarge the space for HPE to achieve GIT. Hypothesis H3.2 is tested. The main reason is that HPE with low financial constraints have a good source of financing to reduce pollution after the implementation of the NPL, and the incentive to reduce LD is weakened (Model (1) in Table 9). Therefore, such HPE rely on the advantages of capital and labor to reduce production costs and improve production efficiency. They also have higher market competitiveness and profits, and greater opportunities for green innovation. To ensure that the regression results are robust, we use the robustness test of 5.2.4 and remove observations in 2015. The results of the regressions are presented in Model (2) of Table 13. The sign of the regression coefficients for each variable in Model (2) is consistent with Model (1), verifying the robustness of the results.

6.3.2. Examination of the Moderating Effects of Fintech

According to the regression results in Model (3) of Table 13, the estimated coefficient of Treat × Post is 0.037 and significant. In regions with weaker levels of fintech, the NPL prompts HPE to transform to green innovation. Meanwhile, the Treat × Post coefficient in Model (3) of Table 9 is −0.037, indicating that the NPL leads to a reduction in LD of HPE. This implies that the transition of HPE to green innovation by reducing LD becomes an important development model in regions with weaker fintech. The estimated coefficient of Treat × Post × Fte is 0.013 and significant. This suggests that the effect of the transformation of HPE to green innovation due to the implementation of NPL is enhanced in regions with higher levels of fintech compared to regions with weaker fintech. Hypothesis H4.2 is tested. The main reason is that in regions with higher levels of financial technology, HPE have easier access to “environmental special” funds based on information technology “dividends”. This mitigates the tendency of HPE to reduce the workforce due to the implementation of the NPL (Model (3) in Table 9). Thus, the effect of the transformation of HPE to green innovation is greater after the implementation of NPL in areas with a stronger level of fintech empowerment. To test the robustness of Model (3) in Table 13, we use the 5.2.4 robustness test, and the regression results removing the observations in 2015 are presented in Model (4) of Table 13. The results indicate that the regression results of Model (3) are robust.

7. Conclusions and Policy Implications

Based on theoretical analysis, panel data of A-share listed companies from 2005 to 2020 are used for empirical testing. A difference-in-difference model is used in quantitative studies. The research conclusions are presented as follows: ① The mandatory environmental regulation represented by the NPL significantly enables HPE to reduce labor and prompts a simultaneous transformation and upgrade to green innovation. This conclusion holds after a series of tests such as high-dimensional fixation, removing observations in 2015, removing observations in municipalities and the provincial capital, one-period lags of the explanatory variables, and the PSM-DID method. ② For the eastern region, the NPL significantly prompts HPE to increase their labor and simultaneously transform and upgrade to green innovation. For the middle and western regions, the NPL has led to the HPE reducing their labor and transitioning to green innovations, which is most evident in the West. ③ For new firms and firms with low labor costs, the NPL has no significant effect on LD, but both are effective in inducing a transition to green innovation. In contrast, for old enterprises and enterprises with low labor costs, the NPL has a significant negative impact on LD and a positive impact on the GIT. ④ The negative impact of the NPL on the LD and GIT of HPE is not significant in the short term and only manifests itself in the long term. This indicates that there is a lag in the impact effect of the new environmental protection law. Based on the above findings, we have the following policy implications.

First, while this policy improves environmental quality, it also has a negative impact on the overall LD of HPE (Table 3). Local governments have made it possible for HPE to be penalized in the short term for transitional emissions of pollutants in the process of implementing the NPL. HPE adjust their production patterns to avoid administrative penalties. Meanwhile, the NPL forces HPE to improve its environmental disclosure. The public reports on enterprises when their pollution emissions exceed the standards. This strengthens local governments’ supervision of enterprises. In the face of strict environmental regulation, HPE move the capital released by reducing LD to pollution control. For this reason, the government should not rush in the process of pollution control but needs to consider the consequences of pollution control and the countermeasures that cause negative externalities. Specifically, the government systematically analyzes and evaluates the potential reduction in labor for each HPE to meet the emission standards. This negative impact is detected by dynamic monitoring over time. Based on the pollution and emission reduction standards, the environmental system will be continuously optimized and improved. Training platforms are set up to improve the adaptability of the workforce to the job. Opportunities for cross-fertilization of the workforce from various skill settings are created to enhance the skill level of the low-skilled workforce and mitigate the negative impact of the policy.

Second, enterprises will transform to green innovation to meet pollution emission standards under the shock of the NPL (Table 9). The government should give full play to the role of guiding HPE in GIT. For example, the green innovation legal system is improved. A safe innovation environment stimulates enterprises’ willingness to be green innovation. The positive impact of green finance on green innovation is significantly enhanced in regions with higher environmental regulations [95]. This means that the government strengthens environmental controls and simultaneously upgrades financial services at the same time. Especially in the current stage of integration of finance and technology, a diversified financial services industry should be built. The government works with financial institutions to monitor enterprises dynamically with the help of technology to ensure that they spend their pollution control funds only on environmental treatment. As a result, green financial services can improve the impetus of environmental regulation on enterprise pollution control and green technology innovation. Furthermore, the government uses lending institutions to enhance the effect of GIT. This can not only solve the problem of slow and difficult financing for enterprises but also alleviate the shortage of funds by reducing LD due to financing constraints in the process of improving environmental quality.

Third, there is significant regional heterogeneity in the impact of the NPL on LD, and the efficiency of GIT of HOE. The main reason is that there are large differences in labor quality, salary treatment, enterprise, and economic structure in various regions. For example, because the eastern region has an advantage in all aspects, the NPL not only allows HPE to absorb the labor but also promotes GIT. For the middle and western regions, the NPL prompts HPE to transform itself into green innovations in response to declining LD. Thus, as the NPL is being pushed forward, the government needs to guide the rational flow of resources to promote the convergence of regional resource endowment differences. The government should improve public services such as medical coverage, transportation facilities, and children’s education. This can alleviate the dilemma of talent shortage faced by enterprises transformation and upgrading due to the loss of highly skilled labor. In addition, enterprises’ structural differences in reacting to the NPL produce a larger response. The government should combine the characteristics of different types of HPE to “prescribe the right medicine” and guide them to achieve GIT without reducing LD as much as possible.

Based on the NPL implemented in China, we analyze the impact of mandatory environmental regulations on LD, and enterprises’ motivation to GIT. This fills the gap in existing studies. However, there are still some limitations in this study. Specifically, although we exclude some of the influencing factors from the prefecture-level and firm-level in our empirical study, factors such as regional social security services, enterprise compensation structure, and the educational background of executives are not excluded. This may lead to bias in the empirical results. In addition, we only explore that the NPL leads to a reduction in LD of HPE and GIT. However, it has not been discussed whether the NPL leads to a movement of labor between regions or industries. Finally, we do not further explore the impact of the NPL on the LD and GIT of HPE in the segment. This is a direction for our future research.