Green Credit Policy, Environmental Investment, and Green Innovation: Quasi-Natural Experimental Evidence from China

Abstract

In order to explore whether green credit policy can guide the green transformation of heavily polluting firms, we examine the influence of green credit policy on green innovation. Further, we analyze the mediating effect of environmental investment and the moderating effect of type of ownership and green finance development level in this relationship. Findings from the DID model indicate that the Green Credit Guidelines led to a significant increase in green innovation at heavily polluting enterprises, both quantitatively and qualitatively, with environmental investment acting as partial mediators. Further, the positive influence of green credit policy is more substantial in state-owned firms and in regions with high levels of green finance development. Findings are robust and remain valid after different sensitivity tests, including the improved PSM-DID model and the elimination of interference from some samples to address the sample selection bias existing in the DID model.

1. Introduction

Chinese President Xi Jinping announced in September 2020 that China will achieve peak carbon by 2030 and carbon neutrality by 2060, paving the way for China to embrace the green transformation of the manufacturing sector more aggressively. With this as a backdrop, China has issued the Green Credit Guidelines (hereafter referred to as the Guidelines), in February 2012, which has drawn researchers’ attention to examine the Guidelines’ impact on firm performance. However, the available literature on the effect of the Guidelines mainly focuses on the disincentive impact on financial performance [1,2]. There is a lack of research on whether the Guidelines may lead to a green transformation and the effects of such transformation. This study aims to fill this gap by exploring the relationship between the Guidelines and green innovation. Green innovation, defined as “technological innovation activity that promotes green technology and improves the ecological performance”, is crucial to evaluate firms’ environmental performance [3,4].

The role of green credit in green innovation is mostly based on two theoretical mechanisms: the “compliance cost effect” and the Porter’s hypothesis. The compliance cost effect is that environmental regulations will increase production costs and pollution control costs, which will have a crowding-out effect on R&D investment activities, and thus reduce green innovation, while Porter’s hypothesis suggests that reasonable environmental regulations can encourage green innovation, which in turn will improve production management efficiency and enterprise competitiveness, offsetting the increased costs brought about by environmental regulations [5,6]. These two contradictory theories inspired much research on the impact of the Guidelines.

The Guidelines, which are China’s first regulatory guideline on green credit, mandate that banks and financial institutions use enterprises’ environmental and social risks as the credit operation threshold and actively encourage green credit. In particular, the Guidelines use the incentive and constraint effects of credit to internalize the external costs associated with business environmental emissions. First, the credit constraint effect increases the threshold restrictions and transaction costs for backward manufacturing to receive credit financing; second, the credit incentive effect is to give interest rate assistance for green innovation involving green transformation and increase clean investment. As opposed to earlier environmental regulatory policies characterized by administrative penalties such as sewage charges, the most essential purpose of the Guidelines is to encourage heavily polluting firms to engage in green innovation through the allocation of credit resources.

Lately, researchers are interested to study the mediating mechanism between the Guidelines and green innovation. R&D investment continues to be a go-to mediator in academic research [7]. However, R&D investment is considered more from the perspective of technical innovations involving backward production, making it an unsuitable metric for green innovation. Environmental investment, which may promote pollution control, emissions reduction, cleaner production, and green technology creation, is a reasonable choice compared to R&D investment. However, environmental regulations, and how strict they are, affect every investment decision a firm makes [8]. Therefore, as the Guidelines will inevitably affect corresponding changes in environmental investment by heavily polluting enterprises, will this effect be facilitative or inhibitive? Will facilitated or hindered environmental investment promote or diminish green innovation in heavily polluting enterprises? In other words, does environmental investment serve as a mediator between the Guidelines and green innovation? Is it completely mediated or partially mediated?

This study uses a DID model and a sample of Chinese listed industrial enterprises from 2009 to 2017 to empirically investigate the impact of the Guidelines on green innovation in heavily polluting enterprises, as well as the mechanisms via which this impact is exerted. The result of baseline regression reveals that the Guidelines have a significantly positive effect on a firm’s green innovation. The dynamic regression results show the volatility of green innovation while testing the parallel trend hypothesis. The findings remain consistent even after addressing any endogeneity issues through different regression techniques, particularly the improved PSM-DID method. The mediating effects model is used to reveal the partial mediating role of environmental investment. Additionally, the Moderation model results indicate that the positive influence of green credit policy is more substantial in state-owned firms and in regions with high levels of green finance development.

This study may have the following marginal contributions: First, environmental investment’s role as a mediator between the Guidelines and green innovation in heavy polluters is disclosed for the first time, and our findings expand on the green credit literature. Second, an examination of heterogeneity demonstrates that the effect of the Guidelines is influenced by the type of ownership and the level of regional green financial development, which helps to identify the variation in the effects of the Guidelines. Ultimately, our findings inform the development of green credit policy to encourage green innovation and advance sustainable development.

Following a literature review in Section 2, the research hypothesis is presented in Section 3, the foundational models, variables, and data are introduced in Section 4, the empirical analysis is described in Section 5, the mediating mechanism is analyzed in Section 6, the moderating effect is analyzed in Section 7, and the implications and future directions are presented in Section 8.

2. Literature Review

2.1. The Microscopic Impact of Green Credit Policies on Enterprises

Current study on the link between green credit policies and heavily polluting enterprises has centered on analyzing the impacts on a firm’s financing behavior and technical innovation. Academics have different views on whether it affects a firm’s financing. Liu (2020) [9] discovered that the Guidelines greatly lowered the debt financing ability of highly polluting firms, with this negative effect being especially evident for state-owned enterprises and financially disadvantaged regions. Xu and Li (2020) [10] contend that the Guidelines has raised the cost of debt financing for “two high” firms. In contrast, Wang et al. (2019) [11] found that no correlation between heavy polluters’ disclosure of environmental information and their financial health. In terms of technical innovation, Yu (2021) [7] found, using the DID model, that firms in heavily polluting industries invest less in R&D than those in other industries due to the “crowding out effect” of environmental regulations, which restricts their technological innovation. Zhang et al. (2022) [12] discovered that the Guidelines greatly promoted technical innovation among energy-intensive and polluting enterprises, and that profitability and equity concentration strengthened this incentive.

2.2. Environmental Regulation and Environmental Investment

Environmental investments are also known as eco-friendly investments, socially or sustainably responsible investments, and green investments [13]. By boosting environmental investments, firms can not only reduce their energy consumption and carbon emissions, but also enhance their financial performance by increasing their visibility, operational efficiency, and maximizing new opportunities [14]. Most scholars believe that environmental regulation can effectively motivate firms to invest in environmental protection. Saygili (2016) [15] and Liu and Xie (2020) [16] argue that the amount of capital firm investment in environmental management, production equipment renewal, and renovation is related to the level of government regulation. Costa-Campi et al. (2017) [17] suggests a mix of environmental, energy, and technology regulation policies to support enterprise environmental R&D. Kvach et al. (2020) [18] compared the environmental investments, efficiency, and effectiveness of environmental taxes in Ukraine and EU countries and concluded that the high rates at which these taxes are levied ensure that firms will be motivated to pursue environmental investment.

2.3. Environmental Regulation and Green Innovation

Dependent on the nature and intensity of environmental regulations, the link between them and green innovation may vary. Del Río González (2009) [19] highlights environmental regulation and corporate image as the primary drivers of the Spanish paper industry’s adoption of clean technologies. Johnstone et al. (2017) [20] discovered that flexibility in environmental regulation is a crucial determinant of low-carbon technical innovation but might have a negative impact when stringency reaches a particular threshold. China is in the beginning of the suppression phase before the inflexion point of the U-shaped connection between environmental tax and innovation in green technology [21,22]. According to Hojnik and Ruzzier [23], command policies, market policies, and corporate structures are considered to be the most important factors affecting green innovation. Emissions limitation programs, energy legislation, and enforcement of environmental rules are examples of mandatory regulatory programs [24,25]; carbon dioxide trade and ecological rights trade are examples of market-oriented regulatory policies market regulatory policies [26,27,28]. Enterprise organizational structures incorporate corporate governance systems and pressure from stakeholders [29,30,31].

3. Research Hypothesis

3.1. Green Credit and Green Innovation

As a key initiative to guide the green allocation of credit resources, green credit encourages enterprises to innovate sustainably at the micro-level and improve energy efficiency and alter the energy structure at the macro-level, achieving energy conservation, emission reduction, and green development [32].

Unlike command- and market-based regulatory policies, the Guidelines will examine a firm’s environmental, social, and governance (ESG) more strictly. Firms with superior ESG performance have lower debt financing [33]. For heavy polluters, the Guidelines result in the cost of environmental regulation far outweighing the benefits of their sloppy development, and higher financing thresholds and costs when ESG performance is poor [34,35].

In the face of expensive environmental legislation, based on the theory of Porter’s hypothesis, this study holds that heavy polluting firms rely on green technology innovation to realize green transformation. Liu et al. (2022) [36] discovered that green credit aided a low-carbon transition by means of environmentally induced R&D. The capital costs of energy efficiency and emission reduction projects were found to be greatly decreased according to the Guidelines, and the return on investment in innovative green technologies was shown to be significantly enhanced thanks to the Guidelines as well [37].

In addition, green credit has a stronger supervisory function than conventional credit in general, allowing banks to act as “big lender supervisors” due to the lower cost of access to information on corporate environmental and social responsibility performance [38], and the supervisory function of green credit is long-lasting and rigid [39]. Faced with the strong supervision of green credit, heavy polluters will also make more efforts to carry out green technological innovation and timely green transformation. Thus, this paper proposes research hypothesis H1.

Hypothesis 1:

The Guidelines encourages heavy polluters to engage in more green innovation.

3.2. Environmental Investment and Green Innovation

Environmental regulations may encourage business environmental investment and green innovation, as was mentioned before. So, how does environmental investment relate to the Guidelines and green innovation? Xiang et al. (2020) [40] examined environmental investment as a metric of environmental information disclosure and concluded that it may improve access to funding, boost the sales of products, and garner greater media coverage, all of which would be beneficial to the development of green innovations. Environmental investment was proven to be a mediator between buyer-driven knowledge transfer operations and innovative green procedures for businesses in Pakistan by Awan et al. (2021) [41]. Macro-studies also exist that examine the effect of environmental legislation on green innovation by directly using variables such as the proportion of environmental investment and GDP [42].

There are two options for the environmental management of heavy polluters: end-of-pipe treatment or green transformation. The former has limited effect and will affect production efficiency, while the latter can be used to promote the efficiency of resource use and reduce pollution emissions at source through raw material substitution, as well as the green transformation of production processes and recycling, which is a fundamental solution to environmental problems. Following the approach of Shu and Liao (2022) [43], environmental investments are divided into two categories according to whether they materially affect the business structure or production process: green transition investment that green the production process and end-of-pipe treatment investment that treat pollutants at the end of production. Clearly, for heavily polluting firms, green transition investment is a source of funding for the quality of green innovation. Thus, this study proposes research hypothesis H2.

Hypothesis 2:

The Guidelines require heavy polluters to support green innovation via environmental investment, especially green transition investment.

3.3. Enterprise, Regional Heterogeneity, and Green Innovation

The desire of businesses is also important in determining whether or not the Guidelines may be utilized to guide a green revolution of highly polluting companies. State-owned enterprises need to play a “leading role” in the government’s strategy to transform and upgrade the industrial structure. In addition, compared with non-state enterprises, state-owned enterprises are more concerned with environmental and social risks in their production and operation activities, and they have more advantages in credit financing, so they are more willing to make a green transformation. This study proposes research hypothesis H3a.

Hypothesis 3a:

The Guidelines more positively encourage green innovation in state-owned firms.

The level of regional green financial development will influence the determination and strength of regional green credit to support enterprises’ investment in environmental protection. In regions with a high level of green financial development, the environmental governance policies that accompany the Guidelines are more complete, and the government and society will pay more attention to the effectiveness of green credit implementation, and thus green credit will have a greater effect on the green innovation. Therefore, this study proposes research hypothesis H3b.

Hypothesis 3b:

The Guidelines put more emphasis on green innovation in places where green finance is growing quickly.

4. Research Design and Statistical Analysis

4.1. Sample Selection and Data Sources

This study includes all listed industrial firms between 2009 and 2017 as its initial sample, excluding those having gearing ratios of less than 0 or larger than 1, non-length arm’s listings (including ST, ST*, and PT), and those with missing relevant data. We matched the patent classification numbers acquired from the China Research Data Service Platform (CNRDS) with the International Patent Classification Green List produced by WIPO in 2010. Patents are sorted into two groups, “green” and “non-green”, based on the match outcome, with the former broken down further into invention and utility model patents. The main financial data comes from the CSMAR database and the shareholder data are from the RESSET database. A total of 7911 yearly data were gathered after applying tail reduction at the 1% and 99% levels to the major continuous variables.

4.2. Variable

Corporate green innovation (Patent). Referring to Li and Lu (2015) [44], this study uses the number of green patent applications to measure green innovation. The total number of green innovations (Total) is calculated by adding the numbers of patent applications for green inventions and green utility models; the former provides a measure of the green innovation’s quality (Inva) while the latter provides an indication of its quantity (Uma). To fix the data’s right-skewed distribution, we take the natural logarithm to get LnTotal, LnInva, and LnUma, respectively.

An enterprise’s dummy variable (Treat). According to the Guidelines, the former CBRC has specified the types of environmental and social risks in the Key Evaluation Indicators for the Implementation of Green Credit, and the industries with environmental and social risks of Category A are identified as heavily polluting industries in this study. The nuclear power industry, hydroelectric power industry, water conservancy, inland river and port engineering construction, coal mining and washing, oil and gas mining, ferrous metal mining and selection, non-ferrous metal mining and selection, non-metallic mining and selection, and other mining industries make up the Class A industries. If an enterprise belongs to the above 9 industries, it is identified as a heavily polluting enterprise with $Treat=1$ (experimental group), otherwise $Treat=0$ (control group).

A period dummy variable (Post). The Guidelines were officially published in February 2012, and the years before 2012 are used as the experimental period in this study. If the sample year is after 2012, Post = 1, otherwise $Post=0$.

Control variables. Following previous studies [45,46], in this study, enterprise size (Size), the ratio of assets to liabilities (Debt), proportion of fixed assets (Ppe), return on assets (ROA), growth (Growth), cash flow (Cash), equity level (Topl), enterprise age (Age), employee count (Employee), independence of the board (Board), and Tobin Q (TQ) are selected as control variables.

Table 1. Summary statistics.

Variables	Mean	Std. Dev	Min	Median	Max
LnTotal	0.77	1.07	0	0	5.97
LnInva	0.51	0.86	0	0	5.63
LnUma	0.51	0.85	0	0	4.77
Treat	0.30	0.46	0	0	1
Post	0.67	0.47	0	1	1
Size	22.28	1.33	18.16	22.14	28.51
Debt	0.50	0.20	0.01	0.50	0.99
Ppe	0.26	0.18	0	0.23	0.97
ROA	0.04	0.06	−0.45	0.03	0.48
Cash	0.05	0.08	−0.66	0.05	0.77
Growth	2.66	169.64	−0.98	0.09	14,883.06
Employee	7.86	1.36	2.30	7.91	13.22
Age	2.71	0.36	0	2.77	3.64
Topl	36.64	15.36	4.16	35.02	89.09
Board	35.25	9.73	6.67	33.33	83.33
TQ	2.26	2.02	0.74	1.71	64.47

displays descriptive statistics for the primary variables. Appendix A explains in detail what each variable means.

4.3. Model

In order to effectively mitigate the problems of endogeneity and omitted variables, this study analyzes the impact by constructing a DID model, as follows:

$$Paten{t}_{it+1}=\alpha +{\beta }_{1}Trea{t}_{it}\times Pos{t}_{it}+{\beta }_2{X}_{it}+{\gamma }_i+{\mu }_t+{\epsilon }_{it}$$

(1)

In Equation (1), $i$ denotes the firm and $t$ denotes the year. With enterprises needing time to file patent applications, we choose Patent with a lag of one period as our explanatory variable. The control variable is represented by X, the firm fixed effect by $\gamma$, the year fixed effect by $\mu$, and the random disturbance term by $\epsilon$. ${\beta }_1$ reflects the impact of the Guidelines on green innovation in heavy polluters.

5. Empirical Results and Analysis

5.1. Parallel Trend Test

Figure 1 illustrates the trend of green innovation across enterprises in both high- and low-polluting industries from 2009 to 2017. The solid dark blue line marked by a triangle is the average value of the green innovation of the heavily polluting enterprises, while the solid green line marked by a circle represents the non-heavily polluting industries. It can be seen that, whether it is a green invention patent or green utility model patent, the trend of green innovation over time between the heavily polluting and control group industries is basically the same, but the difference in the number of green innovations between the two has widened significantly after the Guidelines. This is basically in line with parallel trend hypothesis of the DID model.

5.2. Baseline Regression Analysis

Based on Equation (1), the effect of the Guidelines with and without controlling for the firm and year fixed effects is reported.

Table 2. Baseline regression results.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	LnTotal	LnTotal	LnInva	LnInva	LnUma	LnUma
$Treat\times Post$	0.356 ***	0.295 ***	0.228 ***	0.153 ***	0.185 ***	0.228 ***
	(0.029)	(0.032)	(0.025)	(0.027)	(0.024)	(0.027)
Size	0.174 ***	0.229 ***	0.164 ***	0.179 ***	0.124 ***	0.160 ***
	(0.014)	(0.027)	(0.012)	(0.023)	(0.012)	(0.022)
Debt	−0.354 ***	−0.215 **	−0.353 ***	−0.153 *	−0.172 ***	−0.180 **
	(0.069)	(0.094)	(0.058)	(0.080)	(0.056)	(0.079)
Ppe	−0.634 ***	0.183 *	−0.563 ***	0.008 *	−0.358 ***	0.320 ***
	(0.063)	(0.108)	(0.051)	(0.088)	(0.050)	(0.089)
ROA	−0.231	0.669 ***	−0.283 *	0.391 ***	−0.192	0.522 ***
	(0.213)	(0.178)	(0.172)	(0.148)	(0.004)	(0.145)
Cash	−0.306 **	−0.175 *	−0.097	−0.058	−0.292 **	−0.126
	(0.133)	(0.095)	(0.104)	(0.077)	(0.106)	(0.078)
Growth	−0.001 ***	0.001	−0.001 ***	0.001	−0.001 **	0.001
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
TQ	0.033 ***	−0.001	0.030 ***	0.002	0.020 ***	−0.005
	(0.006)	(0.004)	(0.005)	(0.004)	(0.004)	(0.003)
Employee	0.210 ***	0.041 **	0.139 ***	0.033 **	0.159 ***	0.027 **
	(0.010)	(0.017)	(0.008)	(0.014)	(0.008)	(0.013)
Age	−0.126 ***	0.154 *	−0.094 ***	0.136 *	−0.070 ***	0.052
	(0.033)	(0.102)	(0.026)	(0.076)	(0.026)	(0.094)
Topl	−0.003 ***	−0.005 ***	−0.002 ***	−0.005 ***	−0.001	−0.005 ***
	(0.001)	(0.002)	(0.001)	(0.001)	(0.001)	(0.001)
Board	0.003 ***	0.001	0.003 ***	0.001	0.002 ***	−0.001
	(0.001)	(0.001)	(0.001)	(0.001)	(0.224)	(0.001)
Constant	−4.224 ***	−4.904 ***	−3.783 ***	−3.947 ***	−3.225 ***	−3.236 ***
	(0.271)	(0.585)	(0.236)	(0.495)	(0.224)	(0.498)
Firm FE	N	Y	N	Y	N	Y
Year FE	N	Y	N	Y	N	Y
N	7911	7911	7911	7911	7911	7911
${\mathrm{R}}^2$	0.184	0.732	0.163	0.705	0.152	0.692

Notations: * p < 0.10, ** p < 0.05, *** p < 0.01; Y = fixed effect controlled, N = uncontrolled; Robust standard errors are reported in brackets.

displays the outcomes. The explanatory variables in columns (1) and (2) are LnTotal, columns (3) and (4) are LnInva, and columns (5) and (6) are LnUma. Columns (1), (3), and (5) do not control for firm fixed effects and year fixed effects, while columns (2), (4), and (6) do.

The coefficient of $Treat\times Post$ in columns (1) and (2) is significantly positive at the 1% confidence level, and after controlling for firm and year fixed effects, its coefficient is 0.295, which implies there was a 29.5% jump in the green patent applications by heavily polluting enterprises after the Guidelines, indicating that the Guidelines significantly increased the green innovation output of heavily polluting enterprises. The Guidelines have significantly improved the quality and quantity of green innovation of heavily polluting enterprises, as shown by the coefficients of 0.153 and 0.228 for $Treat\times Post$ in columns (4) and (6), which mean that 15.3% and 22.8% average increase in the number of green invention patent applications and green utility model patent applications, respectively.

The results show that the Guidelines have significantly improved the green innovation performance of heavily polluting enterprises, both in terms of quality and quantity. However, compared to the quantity of green innovation, the Guidelines still do not improve the quality of green innovation enough. Research hypothesis H1 passed the test. Similarly, Yu et al. (2019) [47] confirms Porter’s hypothesis in his study of the relationship between environmental taxes and corporate green innovation. However, we compare the differences in the quantity and quality of green innovation after the Guidelines, so our conclusions are more comprehensive.

5.3. Analysis of Dynamic Effects

Valid estimation of the DID model presupposes that the experimental and control groups meet the assumption of parallel trends. Drawing on Jacobson et al. (1993) [48], this study uses an event study approach to further analyze the dynamic effect of the Guidelines on green innovation while testing the assumption of parallel trends and constructing the following model:

$$Paten{t}_{it+1}=\alpha +{{\displaystyle \sum }}_{t=2009}^{2017}{\beta }_{t}Trea{t}_i\times {\mu }_t+\lambda X_{it}+{\gamma }_i+{\mu }_t+{\epsilon }_{it}$$

(2)

In Equation (2), the first year, 2009, is taken as the base year, ${\beta }_t$ denotes the estimated coefficient from 2009 to 2017, and the other variables are defined the same as in Equation (1).

Figure 2 plots the estimated results of ${\beta }_t$ at a 95% confidence interval and finds that, regardless of whether the explanatory variable is LnInva or LnUma, ${\beta }_t$ is insignificant from 2009 to 2011, satisfying the parallel trend assumption. ${\beta }_t$ is positive after 2012, suggesting that the Guidelines has an enhancing effect on both the quality and quantity of green innovation in heavily polluting firms. Compared to the year-on-year expansion of green utility model patents, the estimated coefficient of green invention patents had a downward phase from 2014 to 2016, indicating greater volatility in the quality of green innovation after the Guidelines, reflecting the difficulty for heavy polluters to achieve a green production transformation.

5.4. Robustness Tests

5.4.1. Improved PSM-DID Test

Due to estimation errors arising from sample selection bias, this study uses the PSM-DID model for the robustness test. To avoid the “self-matching” problem of mixed matching [49] and the instability of the control group caused by matching from period to period, this study draws on the covariate averaging method of Jia et al. (2018) [50], in which the covariates of each period for each individual are averaged before treatment, and then the averages are used for the PSM. All control variables were selected as characteristic variables, and the probability of each sample being selected for the experimental group was estimated using a logit model, and then nearest neighbor matching was used to match the experimental group with a reasonable control group.

Table 3. Balance test of the improved PSM.

Variables		Mean		Deviation		T-Test
		Treat	Control	%Bias	%Reduct	t	$\mathbf{P}>\left|\mathbf{t}\right|$
Size	U	22.273	21.674	47.5		6.72	0.000
	M	22.273	21.839	34.4	27.6	3.90	0.000
Debt	U	0.5036	0.4797	13.0		1.76	0.078
	M	0.5036	0.4799	13.5	1.1	2.49	0.082
Ppe	U	0.3667	0.2342	82.4		10.91	0.000
	M	0.3667	0.3327	21.1	74.3	2.39	0.017
ROA	U	0.0541	0.0487	9.4		1.35	0.178
	M	0.0541	0.0601	−10.3	−10.0	−1.14	0.255
Cash	U	0.0723	0.0528	29.1		3.90	0.000
	M	0.0723	0.0706	2.5	91.4	0.29	0.775
Growth	U	0.2153	2.3312	−9.1		−1.04	0.297
	M	0.2153	0.1971	0.1	99.1	0.79	0.431
TQ	U	2.1513	2.2813	−10.1		−1.40	0.162
	M	2.1513	2.3649	−16.6	−64.3	−1.77	0.077
Employee	U	8.199	7.3614	63.1		8.45	0.000
	M	8.199	7.9404	19.5	69.1	2.42	0.016
Age	U	2.4468	2.5064	−15.7		−2.10	0.036
	M	2.4468	2.4315	4.0	74.3	0.44	0.661
Topl	U	39.337	36.974	15.3		2.11	0.035
	M	39.337	37.013	15.0	1.7	1.73	0.085
Board	U	34.502	34.934	−5.0		−0.67	0.503
	M	34.502	34.933	−5.0	0.2	−0.57	0.568

Notes: U means before the match, while M means after the match.

shows the results of the balance test of the improved PSM. It can be found that the deviation rate of each characteristic variable showed a decrease between the experimental and control groups after matching, and most of the characteristic variables were not significantly different, especially the t-test of Cash, Age, and Topl turned from significant to insignificant, indicating that the improved PSM has a good matching effect. On this basis, the DID model was again used to test hypothesis H1 and the results are shown in columns (1), (2), and (3) of

Table 4. Robustness test results.

Variables	Improved PSM-DID			Elimination of Some Samples
	(1)	(2)	(3)	(4)	(5)	(6)
	LnTotal	LnInva	LnUma	LnTotal	LnInva	LnUma
$Treat\times Post$	0.272 ***	0.143 ***	0.186 ***	0.268 ***	0.136 ***	0.199 ***
	(0.036)	(0.031)	(0.026)	(0.032)	(0.027)	(0.027)
Constant	−5.172 ***	−4.236 ***	−4.851 ***	−4.967 ***	−3.983 ***	−3.333 ***
	(0.852)	(0.689)	(0.291)	(0.586)	(0.496)	(0.497)
Control	Y	Y	Y	Y	Y	Y
Firm FE	Y	Y	Y	Y	Y	Y
Year FE	Y	Y	Y	Y	Y	Y
N	4734	4734	4734	7668	7668	7668
${\mathrm{R}}^2$	0.736	0.719	0.203	0.728	0.700	0.690

Notations: *** p < 0.01; Y = fixed effect controlled, N = uncontrolled; Robust standard errors are reported in brackets.

. It can be found that, after the improved PSM, the research hypothesis H1 remained robust.

5.4.2. Elimination of Interference from Some Samples

In addition to heavy polluters, the Guidelines also support the growth of green industries, such as energy conservation and environmental protection, which can lead to a selection bias in the estimation. Therefore, this study refers to Qi et al. (2017) [51], excludes from the initial sample those whose business involves energy conservation and environmental protection, recycling, and new energy, and conducts a baseline regression according to Equation (1). Table 4 shows that, at the 5% level of significance, the coefficients of $Treat\times Post$ in columns (4), (5), and (6) are all positive, providing robust support for the research hypothesis H1.

6. The Mediating Effect of Environmental Investment

This study examines the possible environmental investment channel behind the Guidelines to encourage green innovation in highly polluting enterprises, and with a significant ${\beta }_1$ of model (1), constructs a mediating effect model below:

$$E_{it}=\varphi +{\phi }_{1}Trea{t}_{it}\times Pos{t}_{it}+{\phi }_2{X}_{it}+{\gamma }_i+{\mu }_t+{\epsilon }_{it}$$

(3)

$$Paten{t}_{it+1}=k+{\psi }_{1}Trea{t}_{it}\times Pos{t}_{it}+{\psi }_2{E}_{it}+{\psi }_3{X}_{it}+{\gamma }_i+{\mu }_t+{\epsilon }_{it}$$

(4)

$E_{it}$ is the mediating variable, denoting environmental investment, and total environmental investment and green transition investment are denoted by TEi and GEi, respectively. Drawing on [44], TEi and GEi were manually collected from the debit additions related to environmental protection in the construction-in-progress detail item table of the corporate annual report. GEi includes both investments in green industries and investment in green upgrades to traditional production. Green industry investment is judged according to whether the investment project belongs to the green industries and their subheadings in the Green Industry Guidance Catalogue (2019 Edition) issued by the National Development and Reform Commission; green upgrading investment is defined according to the Green Manufacturing Project Guidelines (2016–2020) issued by the Ministry of Industry and Information Technology on the green upgrading of traditional manufacturing. Therefore, enterprises’ investments in five areas, including green industries, clean production, efficient and low-carbon use of energy, efficient use of water resources, and recycling of resources, are all considered green transition investments.

When both ${\phi }_1$ and ${\psi }_2$ in Equations (3) and (4) are significant, if ${\psi }_1$ is also significant, and the absolute value of ${\psi }_1$ is smaller than the absolute value of ${\beta }_1$, then there is a partial mediating effect of $E_{it}$; if ${\psi }_1$ is not significant, then there is a full mediating effect of $E_{it}$; if at least one of ${\phi }_1$ and ${\psi }_2$ is not significant, a Bootstrap test is required; if the test result is significant, the mediating effect is considered significant; otherwise, it is not significant.

6.1. Total Environmental Investment

The results of the test for the mediating effect of TEi are shown in

Table 5. Test for mediating effect of TEi.

Variables	TEi	LnTotal	LnInva	LnUma
	(1)	(2)	(3)	(4)
$Treat\times Post$	0.8698 ***	0.2850 ***	0.1447 ***	0.2189 ***
	(0.1420)	(0.0319)	(0.0269)	(0.0270)
Tei		0.0119 ***	0.0098 ***	0.0098 ***
		(0.0036)	(0.0031)	(0.0032)
Constant	−6.1489 ***	−4.8305 ***	−3.8870 ***	−3.1753 ***
	(1.9200)	(0.5821)	(0.4925)	(0.4951)
Control	Y	Y	Y	Y
Firm FE	Y	Y	Y	Y
Year FE	Y	Y	Y	Y
N	7911	7911	7911	7911
${\mathrm{R}}^2$	0.6673	0.7322	0.7055	0.6929

Notations: *** p < 0.01; Y = fixed effect controlled, N = uncontrolled; Robust standard errors are reported in brackets.

. The coefficient of $Treat\times Post$ in column (1) is significantly positive, indicating that the Guidelines significantly increase the TEi of heavily polluting enterprises. Meanwhile, the coefficients of $Treat\times Post$ in columns (2), (3), and (4) are all significantly positive at the 1% significant level, provided that the TEi coefficient is significant, indicating that the Guidelines promote green innovation through TEi. Moreover, the absolute value of the $Treat\times Post$ coefficient is smaller than the absolute value of ${\beta }_1$, indicating that the mediating impact of TEi is only partial.

6.2. Green Transition Investment

The results of the test for the mediating effect of GEi are shown in

Table 6. Test for mediating effect of GEi.

Variables	GEi	LnTotal	LnInva	LnUma
	(1)	(2)	(3)	(4)
$Treat\times Post$	0.8028 ***	0.2850 ***	0.1450 ***	0.2190 ***
	(0.1375)	(0.0319)	(0.0269)	(0.0270)
GEi		0.0128 ***	0.0102 ***	0.0105 ***
		(0.0038)	(0.0032)	(0.0034)
Constant	−6.1877 ***	−4.8243 ***	−3.8837 ***	−3.1705 ***
	(1.8439)	(0.5819)	(0.4925)	(0.4947)
Control	Y	Y	Y	Y
Firm FE	Y	Y	Y	Y
Year FE	Y	Y	Y	Y
N	7911	7911	7911	7911
${\mathrm{R}}^2$	0.6557	0.7323	0.7056	0.6929

Notations: *** p < 0.01; Y = fixed effect controlled, N = uncontrolled; Robust standard errors are reported in brackets.

. The coefficient of $Treat\times Post$ in column (1) is significantly positive, indicating that the Guidelines significantly increase the GEi of heavily polluting enterprises. Meanwhile, the coefficients of $Treat\times Post$ in columns (2), (3), and (4) are all significantly positive at the 1% significant level, provided that the GEi coefficient is significant, indicating that the Guidelines encourages green innovation via GEi. Same as the analysis as for TEi, the partial mediating effect of GEi is demonstrated.

In summary, both in terms of the quality and quantity of green innovation, the Guidelines contribute to heavy polluters via TEi and GEi partially. Research hypothesis H2 is confirmed. This implies that the effect of the Guidelines on green innovation further affects green innovation through the indirect effect of environmental investment, in addition to the direct effect. Similarly, Ma et al. (2022) [52] also verified that environmental investment plays a partially mediating role in the relationship between internal control and corporate green innovation. Our added value lies in the splitting of environmental investments into TEi and GEi, with a better research framework.

7. Moderating Effects of Enterprise and Regional Heterogeneity

7.1. The Type of Enterprise Ownership

To dissect the moderating role of the enterprise ownership type, based on model (4), we add the type of enterprise ownership and its interaction term with the dummy variables Treat and Post to construct a moderating effect model with mediation, as follows:

$$\begin{array}{cc}\hfill Paten{t}_{it+1}=& {\eta }_0+{\eta }_{1}Trea{t}_{it}\times Pos{t}_{it}+{\eta }_2{E}_{it}+{\eta }_{3}Class\hfill \\ & +{\eta }_{4}Class\times Trea{t}_{it}\times Pos{t}_{it}+{\eta }_5{X}_{it}+{\gamma }_i+{\mu }_t+{\epsilon }_{it}\hfill \end{array}$$

(5)

Class in Equation (5) is an indicator variable for the type of ownership, 1 for state-owned enterprises and 0 otherwise. If ${\eta }_4$ is significant, this indicates that Class has a moderating effect on the influence of the Guidelines on green innovation; based on this, if the coefficient ${\eta }_2$ of $E_{it}$ is significant, then this indicates that the mediating effect of $E_{it}$ between the Guidelines and green innovation is moderated by the type of enterprise ownership.

The test results of the moderating effect of the type of enterprise ownership are shown in

Table 7. Moderating mechanism with a mediator: nature of enterprise ownership.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	LnTotal	LnInva	LnUma	LnTotal	LnInva	LnUma
$Class\times Treat\times Post$	0.329 ***	0.303 ***	0.225 ***	0.328 ***	0.302 ***	0.224 ***
	(0.076)	(0.062)	(0.063)	(0.076)	(0.062)	(0.063)
$Treat\times Post$	0.071	−0.038	0.055	0.071	−0.038	0.055
	(0.062)	(0.048)	(0.050)	(0.062)	(0.048)	(0.050)
GEi	0.019 ***	0.015 ***	0.014 ***
	(0.004)	(0.004)	(0.004)
TEi				0.018 ***	0.014 ***	0.014 ***
				(0.004)	(0.004)	(0.004)
Class	−0.161 ***	−0.110 **	−0.125 ***	−0.161 ***	−0.110 **	−0.126 ***
	(0.060)	(0.046)	(0.047)	(0.060)	(0.046)	(0.047)
Constant	−6.233 ***	−4.924 ***	−4.393 ***	−6.241 ***	−4.928 ***	−4.397 ***
	(0.460)	(0.407)	(0.375)	(0.460)	(0.408)	(0.375)
Control	Y	Y	Y	Y	Y	Y
N	7911	7911	7911	7911	7911	7911
${\mathrm{R}}^2$	0.155	0.146	0.128	0.155	0.145	0.128

Notations: ** p < 0.05, *** p < 0.01; Y = fixed effect controlled, N = uncontrolled; Robust standard errors are reported in brackets.

. The moderating effect of Class between the Guidelines and green innovation is significant at the 1% significance level for both LnInva and LnUma, and the coefficients of the interactions are significantly positive, indicating that state-owned heavy polluters have a more significant effect on the quantity and quality of green innovation compared to non-state-owned enterprises. In addition, the coefficients of GEi in columns (1), (2), and (3) are significantly positive, indicating that the mediating effect of GEi between the Guidelines and green innovation is moderated by Class; the coefficients of TEi in columns (4), (5), and (6) are also significant, indicating that the mediating effect of TEi between the Guidelines and green innovation is also moderated by Class. The research hypothesis H3a is confirmed. The result suggests that state-owned polluters place more emphasis on the mediating role of environmental investment than non-state-owned firms. Because the government is a major stakeholder, state-owned enterprises are more involved in environmental and social concerns than non-state-owned firms.

7.2. Regional Green Finance Development Level

Referring to Wang and Pan (2019) [53], this study selects the local green financial development index to measure regional green financial development, using the median green financial development index as the criterion and identifying provinces with a high level of green financial development if they are above the median. To analyze the moderating effect of the regional green finance development level, the moderating effect model with mediation is constructed as follows

$$\begin{array}{cc}\hfill Paten{t}_{it+1}=& {\tau }_0+{\tau }_{1}Trea{t}_{it}\times Pos{t}_{it}+{\tau }_2{E}_{it}+{\tau }_{3}GFd\hfill \\ & +{\tau }_{4}GFd\times Trea{t}_{it}\times Pos{t}_{it}+{\tau }_5{X}_{it}+{\gamma }_i+{\mu }_t+{\epsilon }_{it}\hfill \end{array}$$

(6)

GFd in Equation (6) is an indicator variable for the regional green finance development level, with a value of 1 assigned if the enterprise is registered at a high level and 0 otherwise.

The GFd results are shown in

Table 8. Moderating mechanism with a mediator: regional green financial development.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	LnTotal	LnInva	LnUma	LnTotal	LnInva	LnUma
$GFd\times Treat\times Post$	0.297 ***	0.185 ***	0.165 ***	0.296 ***	0.202 ***	0.164 ***
	(0.072)	(0.073)	(0.056)	(0.072)	(0.0633)	(0.056)
$Treat\times Post$	0.011	0.048	0.003	−0.011	0.004	−0.004
	(0.064)	(0.056)	(0.073)	(0.083)	(0.073)	(0.068)
GEi	0.018 ***	0.014 ***	0.014 ***
	(0.004)	(0.004)	(0.004)
TEi				0.017 ***	0.013 ***	0.014 ***
				(0.004)	(0.004)	(0.004)
GFd	0.011	0.042	−0.035	0.012	−0.034	0.043
	(0.057)	(0.044)	(0.045)	(0.057)	(0.045)	(0.044)
Constant	−6.253 ***	−4.959 ***	−4.384 ***	−6.262 ***	−4.389 ***	−4.962 ***
	(0.462)	(0.409)	(0.377)	(0.463)	(0.378)	(0.410)
Control	Y	Y	Y	Y	Y	Y
N	7911	7911	7911	7911	7911	7911
${\mathrm{R}}^2$	0.147	0.137	0.122	0.147	0.122	0.137

Notations: *** p < 0.01; Y = fixed effect controlled, N = uncontrolled; Robust standard errors are reported in brackets.

. As shown in Table 7, the interaction term coefficient and the coefficients of the mediating variables GEi and TEi are all substantially positive, indicating that regions with a high level of green financial development promote green innovation among heavy polluters more effectively. The mediating effect of GEi and TEi between the Guidelines and green innovation is also moderated by GFd. The results suggest that the intermediation of environmental investment is more significant in regions with high levels of green finance development. This may be because, when the level of green financial development is low, the credit incentive effect of the Guidelines cannot be effectively exercised and heavy polluters have little incentive to make environmental investments.

8. Conclusions and Policy Implications

This research experimentally explored the connection between the Guidelines and green innovation using data from publicly listed Chinese industrial enterprises from 2009 to 2017 and analyzed the mediation role of environmental investment. The following conclusions are drawn. First, the Guidelines can effectively encourage green innovation in heavily polluting enterprises, and the increase in quantity is more significant compared to quality. Second, environmental investment, and particularly investment in green transition, partially mediates the Guidelines’ influence on green innovation. This implies that, in addition to the direct effect of the Guidelines on green innovation, the indirect effect of environmental investment also has an impact on green innovation. Third, the moderating effect of enterprise ownership and the regional green finance development level is verified, and they enhance the implementation effect of the Guidelines.

This paper theoretically enriches the literature on the effects of green credit policy and the factors influencing corporate green innovation, and also provides important insights for improving the implementation of green credit guidelines. First, strengthen the Guidelines’ current implementation. It is more important to correctly identify green initiatives than different categories of businesses. Assistance should be provided to significantly polluting firms who want to make environmental investment for the green transformation of the industrial structure. Second, take into account the diversity of enterprises and geographic areas to moderate the uneven impact of the Guidelines. Green credit should take enterprises of different ownership and regions into account, such as extending to regions with low green financial development, private enterprises, and highly competitive industries.

Based on the findings of this study, the following are the avenues for future research. Firstly, while this study has explored the mediating impact of environmental investment, future research should explore the impact of other factors on green innovation, such as green agency costs, which arise from the conflicting interests of environmental regulation between the principal as shareholder and the agent as manager. Secondly, this paper simply extracts the green transition investment from the total environmental investment to represent how much heavy polluters are investing in pollution control at the source. Future research could find a better way of separating environmental investment into green transition investment and end-of-pipe treatment investment so as to distinguish if firms are favoring pollution prevention at the source instead of end-of-pipe treatments. Lastly, this study has focused on the Chinese context, and future research could explore whether the findings can be generalized to other countries with different political and economic systems.