Does green credit affect the green innovation performance of high-polluting and energy-intensive enterprises? Evidence from a quasi-natural experiment

Taking the green credit policy in 2012 as a quasi-natural experiment, this paper applies the methods of propensity score matching and Difference-in-Difference (PSM-DID) to investigate the relationship between green credit policy and enterprises’ green technology innovation performance based on Chinese industrial enterprises database and green patent database. The results show that the implementation of “green credit guidelines” policy has significantly improved the green innovation performance of high-polluting and high-energy consuming enterprises, which indicates that the incentive effect of green credit policy on enterprises exceeds the constraint effect and leads to “Porter effect.” Moreover, the green credit policy has significantly increased the number of non-invention patents rather than invention patents. In addition, the green credit policy has a more significant effect on the green innovation performance of high-polluting and energy-intensive enterprises that are state-owned and have weak market power. Mechanism test shows that green credit policy can change the credit financing constraints and R&D investment allocation to affect the green innovation performance of high-polluting and energy-intensive enterprises.


Introduction
The severe challenges of resources and environment have aroused widespread concern, and accelerating green technology innovation has become a priority agenda. The green innovation activities of enterprises are inseparable from sufficient capital investment and financial resources. In China, bank loan accounts for a high proportion of the total social financing scale, and it is one of the main external financing sources for enterprises. In particular, high-polluting and energy-intensive industries have always been the key industries that benefit from China's credit resource allocation and bank loans. The credit process of banks in the past mainly focuses on the clients' business profits and collateral adequacy, ignoring or underestimating environmental factors in the loan standards and requirements. In recent years, green credit policies, which have dual functions of financial resource allocation and environmental regulations, are increasingly abundant. In 2012, the China Banking Regulatory Commission issued the "Green Credit Guidelines," which institutionalized the binding of China's green finance policy and enterprises' environmental performance, put forward clear requirements for banks to carry out green credit policy, and took it as an important market means of environmental protection. As a result, the innovation behavior of high-polluting and energyintensive enterprises is gradually internalized in the implementation of financial institutions' policies, and their innovation decisions are also more likely to be affected by the policy implementation (Liu et al. 2017). The green credit policy is a valuable supplement to the traditional environmental regulation policy and has gradually become an important marketoriented financial regulation means of environmental governance in China. Theoretically, after the implementation of green credit policy, through the multiple constraint mechanisms such as environmental access threshold and credit quota control, the credit resources allocation of high-polluting and energy-intensive enterprises would be adjusted. Thus, the investment structure of credit resources and technological innovation decision-making of enterprises would be affected (Horbach 2008;Li et al. 2018). However, in practice, whether the green credit policy can obtain actual effect depends on multiple factors such as reasonableness and strictness of the policy, effective implementation by banks, and the strategies of enterprises. Therefore, this paper takes China's situation as a typical representative of the transition economy, discusses whether the green credit policy can affect the allocation of credit resources and the decision-making of enterprises' green innovation, and examines whether the green credit policy and "Porter Hypothesis" are applicable in transition countries with imperfect market incentive tool of environmental regulation. It is not only of theoretical significance to understand the microeconomic consequences of green finance policy, but also of practical value to further improve the policy system of green finance and green innovation in developing countries. This paper is closely related to two bunches of literature. First, part of the literature reviews the practical effect of green credit policy from different perspectives. According to neoclassical economics theory, environmental regulation is likely to have a negative impact on enterprises' green innovation performance since it raises the threshold of investment access and the cost of pollution control (Gray and Shadbegian 1995;Jaffe et al. 2004). Due to the punitive interest rate and credit threshold imposed by green credit policies on the financing activities for high-polluting and energy-intensive industries, the financing costs significantly increase, while the total amounts of financing and investment of those industries are significantly reduced (Liu et al. 2017;Xu and Li 2020). When enterprises face a shortage of funds, they would invest less in projects with greater risk uncertainty and longer return period, especially for green innovation activities that may further aggravate their financing constraints (Cecere et al. 2020;Demirel and Parris 2015;Kapoor et al. 2011). In this way, the green credit policy may reduce the financing sources of highpolluting and energy-extensive enterprises and restrict capital investment for their green innovation activities. Thus, this paper mainly investigates whether the high-polluting and energy-intensive enterprises comply with the requirements of green governance through innovation activities to obtain the comparative advantage in the process of credit resource allocation after the implementation of green credit policy. Second, the essence of green credit policy is a kind of environmental regulation. Some literature has studied the innovation effect caused by environmental regulation, mainly focusing on whether "Porter effect" is tenable. Theoretically, "Porter Hypothesis" suggests that moderate environmental institutional constraints show "innovation compensation effect," which means that technological innovation of enterprises can offset the rising cost of environmental policies and achieve a win-win situation between economic interests and environmental protection. Also, studies on whether "Porter Hypothesis" can be established in the field of green credit and whether China's green credit policy can form an effective incentive for the enterprises' green innovation are crucial for developing countries to evaluate the effect of green credit policy and optimize the following policy.
The existing research can be expanded in the following aspects: (1) The existing literature mainly focuses on the impact of green credit policy on enterprise investment and financing resource allocation, but rarely discusses the impact of green credit policy on the enterprises' green innovation performance. In fact, the green credit policy is the application of environmental regulation in the allocation of financial credit resources. However, there is a lack of theoretical research on how green credit policy affects enterprises' green innovation performance. In addition, the "Porter Hypothesis" has already been verified in developed countries, but whether it is valid in the field of "green credit regulation" in developing countries remains to be verified. (2) There is a lack of reliable empirical research on the micro effect of green credit policy by overcoming endogenous problems. In transition economies with imperfect market-incentive environmental regulation, there is few empirical studies on the causal relationship and mechanism between green credit policy and enterprises' green innovation. Most quantitative studies concern the average treatment effect of environmental policy, seldom discussing the asymmetric impact of policy on heterogeneous enterprises, which cannot provide a reference for differentiated green credit policy.
The marginal contribution of this paper lies in the following: (1) It enriches studies on various types of environmental regulation policies. Existing research on environmental policy mainly focuses on the emission reduction effect of marketoriented policies such as order-oriented policy, emission fee, and emission trading system, while fewer concerns about the role of the green credit policy, which is an important new environmental regulation tool. (2) It expands the empirical framework and empirical evidence of green credit policy. Due to the difficulty of obtaining green patent data, related research on the innovation effect of green credit policy at the micro-level is extremely scarce. Thus, this paper sorts out the green patent database of China's listed enterprises and constructs a quasi-natural experiment based on the exogenous event of the implementation of the "Green Credit Guidelines" in 2012 to alleviate the endogenous problem and further applies the PSM-DID method to identify the effect of China's green credit policy on enterprises' green innovation. Furthermore, the mediation model examines the impact mechanisms of green credit policy, namely, "credit constraint effect" and "R&D incentive effect." Then, we examine the asymmetric effect on heterogeneous enterprises. Thus, this paper not only provides reliable empirical evidence for the causal relationship between green credit policy and enterprises' green innovation, but also provides a reference for developing countries to optimize the green finance policy and green innovation strategy.

Policy background
In July 2007, China's State Environmental Protection Administration (SEPA), the People's Bank of China (PBOC), and the China Banking Regulatory Commission (CBRC) jointly issued the document of "Opinions on the Implementation of Environmental Protection Policies and Regulations to Prevent Credit Risks," marking green credit as a potential tool for environmental protection and emission reduction in China. However, this document does not formulate specific implementation measures on how to carry out green credit policies. Subsequently, in February 2012, the CBRC issued the "Green Credit Guidelines" (CBRC [2012] No. 4) as the new core of China's green credit policy system. It provides feasible guidance for financial institutions, mainly commercial banks, on how to carry out green credit policy and promote the green transformation of the high-polluting and energy-intensive industries. This policy is likely to influence the financing costs, financing scale, and financing maturity of enterprises especially those with high pollution and high energy consumption and further affect their green innovation performance.
The rules of "Green Credit Guidelines" can be summarized into three aspects. First, commercial banks should set more strict conditions for access to financing, embed environmental factors into the risk management system of credit business, and refuse to grant loans to enterprises with poor environmental performance. Second, the policy guides financial institutions to allocate more credit funds to enterprises with great green innovation capacity. Banks are encouraged to reduce credit scale or charge a punitively high interest rate for the high-polluting and energy-intensive enterprises, while in contrast increasing financial support for enterprises with great green innovation capacity. Third, financial institutions should dramatically strengthen their management ability of environmental risks. After issuing the loans, banks need to scrutinize the use of credit funds strictly to ensure that more credit funds can configured to the green patent innovation activities.

Mechanism analysis
Green credit policy is an important practice to guide the green transition of enterprises through the allocation of financial resources. From the perspective of reallocation of credit resources, the implementation of green finance or green credit policy probably reduces the debt financing scale, increases the debt cost, and then affects the capital investment structure and green patent innovation performance of the high-polluting and energy-intensive enterprises. With green credit policy, the debt financing environment faced by related enterprises is becoming more severe. According to "Porter Hypothesis," high-polluting and energy-intensive enterprises tend to expand investment in green patent R&D innovation when facing more stringent environmental regulations (Porter and Vanderlinde 1995a, b). This is because a higher capacity for green innovation in enterprises can not only improve pollution control, but also help them to eliminate outdated technologies through innovative production processes, thus reducing the negative impact on the environment. Therefore, the highpolluting and energy-intensive enterprises have stronger motivation to improve the "green" content of their products through green technology innovation, so as to offset the adverse impact of "tight financial regulation policy" caused by strong environmental regulation on their operation and business activities.
Furthermore, compared with ordinary patents, the R&D process of green patents has the characteristics of higher technical requirements, longer pay-off cycle, more investment, and higher risk (Aghion et al. 2012). Therefore, enterprises must rely on a well-developed financial market to carry out green technology innovation activities. Debt financing from the banking system is still the most important external financing source for developing country enterprises to carry out R&D activities (Ayyagari et al. 2011). However, the traditional financial market tends to avoid risks for the sake of profitability and focuses on whether the investment project is profitable or not, ignoring the resource and environmental factors in the investment project. This kind of loan granting mechanism that neglects environmental benefits may inhibit the allocation of external funds to green technology R&D activities and hinder the enthusiasm of enterprises to carry out green innovation activities.
From the views of policy practice, the green credit policy gives credit resource priority to enterprises with good environmental performance, while refrains credit support on enterprises with high environmental risk, thereby internalizing enterprises' external environmental costs and strengthening the "innovation compensation effect." It is conducive to affect the financing resource allocation and guide the bank's credit investment to green technology innovation activities of enterprises. Specifically, after the implementation of the green credit policy, banks are guided to consider more of the environmental risks involved in enterprises when conducting loan approval. For high-polluting and energy-intensive enterprises, due to the signal display mechanism, the pressure of social responsibility and public opinion, managers of such enterprises may expect their loans would become more difficult, because they are in the category of key monitored industries. In addition, after the implementation of green credit policy, banks raise the lending threshold in order to improve their own green rating, so the financing costs of these related enterprises are rise in response (Lemmon and Roberts 2010;Liu et al. 2019). To some extent, green credit policy intensifies the possibility of financing constraints of high-polluting and energyintensive enterprises (Ghisetti et al. 2017;Wang et al. 2019). To reverse this adverse effect, the high-polluting and energy-intensive enterprises are forced to promote the green transformation and improve the green innovation performance, so as to obtain continuous and sufficient debt financing support from the banking institutions (Hou et al. 2019;Tsai and Liao 2017;Zhang et al. 2020). Therefore, we propose the following hypothesis.
Hypothesis: The "Green Credit Guidelines" policy is conducive to promoting the green innovation behavior of highpolluting and energy-intensive enterprises.

Empirical methods
In recent years, the Difference-in-Difference (DID) model has been widely used to assess policy effects. In this paper, we treat the "Green Credit Guidelines" policy as an exogenous shock and construct a quasi-natural experiment, in which the high-polluting and energy-intensive enterprises are regarded as the treatment group and enterprises that are neither high-polluting and energyintensive nor green enterprises are regarded as the control group. Then, we can quantitatively measure the policy effects by comparing the green innovation behavior of the treatment group and control group before and after the policy implementation.
Due to the heterogeneity such as business scale and financial situation among different enterprises, if conducting the DID model directly, the empirical results may be biased and cannot reflect the real policy effects. In light of this, the propensity score matching method (PSM) and the Difference-in-Difference method (DID) are combined to solve this problem. PSM model is applied to match different types of enterprises to ensure that the treatment group and the control group can meet randomness and homogeneity requirements, and then the matched samples are used for DID analysis. The baseline model is as follows: In model (1), invtotal ij denotes the number of green patent grants in i enterprise in year j. treat is the policy dummy variable, where treat is equal to 1 if the sample enterprise is high-polluting and energy-intensive, otherwise treat is equal to 0. time is the dummy variable for the treatment period. Since the "Green Credit Guidelines" policy was issued in 2012, time equals to 1 denotes the years 2012 and later, and time equals to 0 denotes the years before 2012. The interaction term (namely treat*time) is the core variable, and its coefficient σ measures the policy effects on the green innovation of high-polluting and energy-intensive enterprises. X ij denotes a set of characteristic variables of enterprises over time. α i , δ j , and ε ij denote individual effects, time effects, and the error term, respectively. Specifically, as the null hypothesis was rejected in the Hausman test (p value = 0.022) and to control for unobservable differences in years and enterprises, we use a two-way fixed effects regression model to reduce the potential endogeneity.
Besides, to investigate the impact mechanism of the "Green Credit Guidelines" policy on the green innovation of highpolluting and energy-intensive enterprises, the following mediation model is constructed with reference to Baron and Kenny (1986): In models (2), (3), and (4), M denotes the mediation variable. The indirect effect exists where the coefficients a, b, and c are all significantly not equal to 0. Therefore, if the coefficient c' is significantly not equal to 0, it is suppression effect when c' is opposite in sign to a*b, and partial mediation effect when c' is in the same direction as a*b. Otherwise, it indicates a complete mediation effect.

Dependent variable
The number of green patent grants (invtotal ij ) is the dependent variable. Existing literature typically uses R&D investment or the number of patents to measure enterprises' innovation capacity since there are no uniform standards. Given that R&D investment emphasizes the pre-innovation stage and cannot accurately measure enterprises' innovation output (Cruz-Cazares et al. 2013;Tumelero et al. 2019), green patents are introduced to reflect the innovation performance and actual innovative capacity of enterprises more accurately. Referring to Guan et al. (2009) and Zhang et al. (2019), this paper uses the number of green patent grants to denote the enterprises' innovation capacity.
Further, green patents can be subdivided into green invention patents and green utility patents. Green invention patents represent substantial green technological innovation, while green utility patents represent non-substantial incremental innovation. Therefore, we use green invention patent grants (inv ij ) and green utility patent grants (invpra ij ) as dependent variables for the robustness test and further refine the policy effects on enterprises' green innovation behavior.

Independent variable
High-polluting and energy-intensive enterprise under "Green Credit Guidelines" policy (treat*time) is the independent variable. The "Green Credit Guidelines" policy is a vital component in promoting green finance development in China. It sets out the policy boundaries, management methods, and assessment criteria for financial institutions, especially commercial banks, to carry out green credit business.
Therefore, this paper regards the "Green Credit Guidelines" policy as an exogenous shock to explore the policy effects on the green innovation performance of highpolluting and energy-intensive enterprises. The interaction term of the policy treatment group's dummy variable (treat) and the time dummy variable (time) represents the specific samples of high-polluting and energy-intensive enterprises under the period of implementation of the "Green Credit Guidelines" policy.

Control variables
Referring to Liu et al. (2019) and Wang et al. (2019), this paper includes the following variables in the empirical analysis to avoid estimation bias errors due to omitted variables: (1). The independence and objectivity of decision-making (indr): The percentage of independent directors measures the independence and objectivity of the enterprise's decision-making. Following Levinson (1999), the intensity of environmental regulations is measured by a synthetic index of environmental regulations.
(6). Enterprise supervision system (dual): The enterprise supervision system is measured by the chairman and general manager's concurrent appointment, with dual equal to 1 if the chairman does not also serve as general manager, otherwise dual equal to 0. (7). The cumbersomeness of enterprise's decision-making (board): The logarithm of the number of shareholders me asure s th e ente rprise ' s de cision-mak i ng cumbersomeness.
The following variables are introduced as mediator variables in the mechanism test: (1) Financing constraints (sa): Referring to Hadlock and Pierce (2010), the financing constraints of the enterprise is measured by the SA index 1 .

Descriptive statistics
This paper is mainly based on the balanced panel data of 1171 listed enterprises from 2006 to 2018, and the summary of the main variables is reported in Table 1. It shows that after implementing the "Green Credit Guidelines" policy, a number of high-polluting and energy-intensive enterprises' green patent grants, green invention patent grants, and green utility patent grants have increased in varying degrees. Therefore, we verify whether there is a causal relationship between the "Green Credit Guidelines" policy and enterprises' innovation behavior in the following sections.

Empirical results and discussion
Propensity score matching In propensity score matching (PSM) model, seven observable variables, including the independence and objectivity of decision-making (indr), the growth stage of the enterprise (growth), the profitability (roa), the size of the enterprise (size), the intensity of environmental regulation (env), the enterprises' supervision system (dual), and the cumbersomeness of enterprise's decision-making (board), are selected as the matching indexes of the PSM model. We use the Probit model to estimate the propensity scores and then match enterprises by the nearest matching method. Finally, we get 575 enterprises matched in total, including 323 from the treatment group and 252 from the control group.
We implement a balance test to ensure no significant difference between the treatment and control groups after matching. Table 2 shows the balance test results. Before propensity score matching, independent decision-making ability (indr), profitability (roa), and environmental regulatory intensity (env) all differ at the 5% level of significance, while all variables do not differ significantly after matching. This suggests that the matched enterprises have similar characteristics in 2011, the year before the policy was implemented, meeting randomness and homogeneity requirements for the Difference-in-Difference approach.

Parallel trend assumption test
A vital prerequisite of the DID model is that the treatment and control groups have similar trends before the exogenous policy shock (Bertrand et al. 2004), which means that other exogenous factors that may affect enterprises' green innovation performance should be excluded.
To test whether the treatment and control groups have a parallel trend before the policy shock, we plot the trend of the average number of green patent grants for treatment group enterprises and control group enterprises from 2006 to 2018. Figure 1 shows that before the implementation of the "Green Credit Guidelines," an average number of green patent grants in the treatment and control groups maintain a similar upward trend. After implementing the policy, the treatment group's overall increase is significantly larger than the control group. Figure 1 indicates the treatment and control groups which meet the parallel trend assumption. Therefore, it can tentatively infer that the "Green Credit Guidelines" policy improves high-polluting and energy-intensive enterprises' green innovation performance.
Further, we set the year before the policy implementation, namely 2011, as a benchmark year, and set up an empirical equation for the parallel trend test (Chen et al. 2020;Liu and Qiu 2016). The equation is as follows:  We define time t as a dummy variable. time t is equal to 1 when the year is t; otherwise time t is equal to 0. Other variables are consistent with the baseline regression (1). The results of the regression are shown in Fig. 2. It can be seen that the regression coefficients are not significant before 2012, while the regression coefficients become significant and show an increasing trend after 2012. It suggests that before implementing the "Green Credit Guidelines" policy, the trends of treatment and control groups have no significant difference. After 2012, the number of green patents in the treatment and control groups shows a significant difference. Therefore, the empirical results are consistent with the inference of Fig. 1, which proves that the research sample conforms to the parallel trend assumption.

Average treatment effects and marginal treatment effects
According to the baseline regression (1), we can measure the policy's average treatment effects to assess the overall impact of the "Green Credit Guidelines" policy on highpolluting and energy-intensive enterprises' green innovation performance. However, the marginal effect of this policy cannot be assessed. To further explore whether the policy impact is persistent, Eq. (1) is extended as follows (Fan et al. 2012): Based on the research sample matched through PSM, we evaluate the policy effects of the "Green Credit Guidelines" based on Eqs. (1) and (6), and the results are reported in columns (1) and (2) of Table 3, respectively. In column (1), the coefficient of the interaction term treat*time is positive at the 1% level of significance. In column (2), the interaction term's coefficients from 2012 to 2018 are all positive at the 5% level of significance.
The empirical results suggest that the implementation of the "Green Credit Guidelines" policy can promote the green technology innovation of high-polluting and energy-intensive enterprises overall. Furthermore, the empirical results also show that the "Green Credit Guidelines" can persistently promote high-polluting and energy-intensive enterprises to carry out green technology innovation activities, rather than intermittently and temporarily.

Impact mechanism
As a kind of macro policy, the impact of green credit policy needs to be conducted through enterprises, because the effect of the policy depends on how the enterprises react to the policy. To this end, this paper further tests the response behavior of enterprises when facing the green credit policy, which helps to better understand the micro transmission mechanism of green credit policy on enterprise green technology innovation. According to Eqs. (2), (3), and (4), we construct a mediation model to investigate the impact mechanism of the "Green Credit Guidelines" policy on enterprises' green innovation behavior. Moreover, the bootstrap method is introduced to ensure that the empirical results are reliable (Zhao et al. 2010). The estimation results are reported in Table 4.
Through causal steps regression, the interaction term's coefficients in columns (1)-(4) are positive at the 1% level of significance, and the interaction term's coefficient in column (5) is not significant. The coefficient of SA index is negative, while the coefficient of R&D investment is positive, both significant at the 1% level. Moreover, in the bootstrap test, the direct effect coefficient is 0.4340, and the indirect effect coefficient is -0.2444, both significant at the 1% level. Besides, in the R&D incentive mechanism, only the indirect effect is significant at the 1% level, while the direct effect is not significant.
The empirical results obtained by the causal steps regression method and the bootstrap method are consistent. It suggests that the mechanism of financing constraints has a suppression effect, suppressing part of direct effect of "Green Credit Guidelines" policy, whereas the R&D incentive mechanism has a positive complete mediation effect.
The results reveal that "Green Credit Guidelines" policy stimulates high-polluting and energy-intensive enterprises to develop green technologies, but in the meanwhile, it also leads to a shortage of funds for R&D on green innovation by aggravating their financing constraints, which weakens the effectiveness of the policy to some extent. Moreover, since the R&D is a complete mediator while financing constraints have a suppression effect, it suggests that both mechanisms affect enterprises' green innovation behavior through changing the enterprises' R&D investment essentially.

Robustness test
In this subsection, we present a series of robustness tests on the above empirical results. Specifically, we check the robustness by clustering standard error at different levels, replacing the variables, altering the observation period, and adding other policy dummy variables.

Clustered standard error
Considering the potential serial correlation may exist in panel data, we use clustered standard errors to ensure that the regression above is reliable. We cluster the robust standard error at different levels. The results are reported in Table 5, and the standard errors in columns (1)-(3) are respectively clustered at enterprise, industry, and city level. As the table clearly shows, the interaction term's coefficients are significantly positive regardless of the level of clustering standard errors used. This suggests that "Green Credit Guideline" policy has a positive impact on the green innovation behavior of highpolluting and energy-intensive enterprises; the empirical results above are robust.  Notes: *p < 0.10, **p < 0.05, ***p < 0.01.

Replacing variables
Referring to Liu et al. (2019), we replace the dependent variable green patent grants (invtotal ij ), using green invention patent grants (inv ij ) and green utility patent grants (invpra ij ) as proxy variables for enterprises' green innovation behavior. Table 6 shows that the average treatment effect of the "Green Credit Guidelines" policy on green innovation performance of high-polluting and energy-intensive enterprises remains significantly positive regardless of the dependent variable is green invention patent grants or green utility patent grants. It can be seen that the test results are consistent with the previous results, suggesting that our findings are robust.
Furthermore, we find that the average treatment effect coefficient in column (1) of Table 6 is 0.104, smaller than 0.188 in column (2). This suggests that the "Green Credit Guidelines" policy has different impacts on different types of green innovation behavior. Therefore, under the incentive of green credit policy, the number of non-invention patents increases more obviously than that of invention patents. The possible reason is that Chinese commercial banks currently lack the fine definition of high-quality green patents, which may prompt enterprises to implement the quantity competition strategy of green patents for seeking more financial support and credit resources. Enterprises are more likely to develop non-substantive green utility patents, probably because substantive green innovation patents have longer pay-off periods, higher uncertainty risk, and greater execution difficulty. Thus, high-polluting and energy-intensive enterprises are less willing to undertake substantive green innovation in the costbenefit trade-off.
In summary, under the "Green Credit Guidelines" policy, to obtain more green credit support, enterprises have motivations to adopt speculative innovation strategy, manifesting the scale of green patents, especially non-invention patents increase significantly.

Altering the observation period
To avoid the potential randomness error caused by a single observation period, we replace the experimental and control groups' observation periods with 2007-2017, 2008-2016, and 2009-2015, respectively. Then, we re-evaluate the "Green Credit Guidelines" policy effects on the number of green patents of high-polluting and energy-intensive enterprises. Table 7 shows the results during varied observation periods. The interaction term's coefficient is significantly positive at the 1% level of significance despite the observation periods. It indicates that the "Green Credit Guidelines" policy effects on green innovation performance of high-polluting and Notes: *p < 0.10, **p < 0.05, ***p < 0.01

Adding other policy dummy variables
Since the green innovation behavior of high-polluting and energy-intensive enterprises can also be affected by other related policies, it may lead to estimation bias if we misjudge other policies' effects as the "Green Credit Guidelines" policy's effects. To identify and resolve this problem, we set the other two green credit-related policies during the sample observation period as policy dummy variables policy1 2 and policy2 3 , respectively, and add them into the regression equation for robustness test.
The results in Table 8 show that the interaction term's coefficients have little change with the introduction of other policy dummy variables and they are all significant at the 1% level, while the coefficients of the dummy policy variables are not significant. This result indicates that the other relevant policies have little interference effects with the "Green Credit Guidelines" policy. It confirms that our findings are reliable and robust.

Heterogeneous effects
For enterprises with various ownership patterns and market power, the sources of financing channels and financing constraints they faced to carry out technological innovation activities are different (Chen and Schwartz 2013;Liu et al. 2019). Therefore, the policy effects on the innovation activities of these different types of enterprises may not be the same. To discuss the heterogeneity, it is necessary to explore the impact of the "Green Credit Guidelines" policy on the green innovation behavior of high-polluting and energy-intensive enterprises with different ownership patterns and market power.

Heterogeneous effects with respect to ownership pattern
We report the results for the SOEs and non-SOEs subsamples in columns (1) and (2) of Table 8 to discuss whether the green innovation behavior response to the "Green Credit Guidelines" policy differs by ownership patterns. The results show that the interaction term's coefficient for state-owned enterprises is significantly positive at the 1% level, while the interaction term's coefficient for non-state enterprises is insignificant, and Chow-test indicates a significant group difference. It suggests that the "Green Credit Guidelines" policy can drive the state-owned high-polluting and energyintensive enterprises to carry out green innovation, but the policy effects on non-state-owned enterprises are not pronounced.
Due to the different patterns of ownership, the bankenterprise relationship and the proportion of bank debt financing of enterprises are different. Therefore, different types of enterprises also have different responses and coping strategies to the implementation of green credit policies. The existing literature shows that state-owned enterprises have a comparative advantage in obtaining bank credit resources, while private enterprises usually face credit discrimination due to insufficient collateral and other reasons (Cull and Xu 2003). When the green credit policy was promulgated, private enterprises with narrow debt financing channels cannot maintain a high level of sensitivity to policy implementation and thus lack the motivation to carry out green innovation activities. In contrast, the share of debt financing of state-owned enterprises is usually high, leading to a greater impact of green financial policies on their green innovation activities. Therefore, state-owned enterprises have stronger green patent  Notes: *p < 0.10, **p < 0.05, ***p < 0.01. Robust standard errors are reported in parentheses motives to respond to green credit policies and thus obtain more policy dividends .

Heterogeneous effects with respect to market power
Since the market power of enterprises can affect the impact of green credit policy, we report the subsample estimation results for enterprises with strong and weak market power in columns 3 and 4 of Table 9. The results show that the coefficient of the interaction term for weak market power enterprises is 0.355, which is at the 1% level of significance. In contrast, the coefficient of the interaction term for strong market power enterprises is not significant. Also, the results of Chow-test suggest that there is a significant difference between groups. This indicates that the "Green Credit Guidelines" policy can promote the green innovation of weak market power enterprises but has no pronounced effect on strong market power enterprises. The differential impact of green credit policies on enterprises with various market powers may be caused by the different dependence of enterprises on bank financing channels. Enterprises with strong market power usually own more financing channels and multiple financing options in addition to credit financing. Therefore, the innovation incentive effect of the green credit policy on such enterprises is not obvious. However, enterprises with weaker market power typically face a tougher financing environment and limited financing options. As a basic source of financing, enterprises are highly dependent on financing methods through bank loans. As a result, the incentive effect of the green credit policy on such enterprises becomes more pronounced. Therefore, enterprises with weaker market power are more motivated to seek to obtain preferential policy benefits through green innovations to alleviate financial pressure.

Discussion
Based on "Porter Hypothesis", this paper uses a quasi-natural experiment method to examine the impact of green credit policy on green innovation. The most important finding is that the "Green Credit Guidelines" policy significantly improves the performance of high-polluting and energy-intensive enterprises. The results are significant at least in following respects. From the perspective of financial credit regulation, this paper provides empirical evidence for the applicability of "Porter Hypothesis" in China. Moreover, it contributes to clarify the relationship between green financial policy and green technological innovation behavior in the context of China's current environmental governance, as well as to provide a new solution for enterprises to achieve a win-win situation from the perspective of implementing endogenous environmental governance and obtaining credit resource.
Although this study identifies the green innovation effect of green credit policy, several questions remain unanswered at present. For instance, we cannot distinguish the roles of bank or enterprise manager's characteristics and motivations. In future investigations, it might be possible to further extend a theoretical model to explore how the bank or enterprise manager's characteristics and motivations affect the green credit policy and enterprises' innovation. Also, due to the lack of detailed and comprehensive enterprise pollution emission data and green credit data, it is difficult to conduct a more detailed classification discussion on green credit policy types, green patent quality, and other indicators. These questions suggest potential topics for further research to overcome the limitations of micro data and identification strategies and further improve the accuracy of empirical results.

Conclusions and policy implications
To sum up, this paper applies the PSM-DID method and China's listed enterprises' data to study the effects of the "Green Credit Guidelines" policy on the green innovation performance of high-polluting and energy-intensive enterprises. The conclusions are as follows. First, the "Green Credit Guidelines" policy significantly promotes the green innovation performance of high-polluting and energyintensive enterprises, and the policy effect is persistent. Second, the "Green Credit Guidelines" policy has a more Notes: *p < 0.10, **p < 0.05, ***p < 0.01 prominent impact on non-substantive green innovations samples than substantive green innovations. Third, the policy effect is heterogeneous due to the enterprises' ownership and market power, respectively, with a more pronounced effect for state-owned enterprises and weaker market power enterprises. Finally, the "Green Credit Guidelines" policy affects the green innovation behavior of high-polluting and energy-intensive enterprises through the mechanisms of financing constraints and R&D incentives, while R&D incentives have a complete mediation effect, and financing constraints show a suppression effect. Based on the above conclusions, we propose the following policy implications. First, it is necessary to further carry out refined classification and differentiated management of the green patent achievement standards, clarify the support objects and the technical scopes, and guide the flow of funds to environment-friendly patents and green transition of highpolluting and energy-intensive enterprises. Secondly, we should take into account the heterogeneity of enterprises and avoid a "one size fits all" approach to policy implementation. The relevant departments should strengthen the green credit policy support for non-state-owned enterprises and small and medium-sized enterprises (SMEs), while adhering to the fairness, openness and transparency of the green credit policy, and further moderately regulate the green credit resources obtained by the state-owned enterprises and large enterprises of high-polluting and energy-intensive industries. Finally, the government should create a refined policy system for the implementation of green credit policy. Specifically, it needs to expand the financing sources for green technology innovation, ease enterprises' financing constraints, and increase investment in green technologies, thereby accelerating the transformation of high-polluting and energy-intensive enterprises into intelligent, green, and service-oriented enterprises.