Can Green Finance Policy Achieve Collaborative Governance of Air Pollution? Evidence from Prefecture-Level Cities in China

Abstract

Green finance policy represents a critical market-oriented instrument that channels financial resources toward environmentally sustainable development. This study employs the difference-in-differences (DID) model to empirically analyze the panel data of 276 cities in China spanning 2011 to 2020, using China’s 2017 green finance policy as a quasi-natural experiment. The results demonstrate that after passing a series of robustness tests, green finance policy can effectively alleviate air pollution. In addition, the mechanism test shows that green finance policy can significantly reduce air pollution through the resource allocation effect and green innovation effect. According to heterogeneity analysis, the effects of policy are more noticeable in Western cities, resource-based cities, and cities with higher levels of financial development. These findings provide scientific support for the policy pathways through which green finance facilitates coordinated pollution reduction and high-quality development, offering valuable insights for developing countries in advancing sustainable urban governance.

1. Introduction

Extreme natural disasters have become increasingly frequent due to climate change, drawing heightened attention from governments and scholars worldwide to ecological and environmental challenges [1]. These conditions continue to threaten the Earth’s ecosystems and human survival, significantly impacting the global economy and the sustainable development of society [2]. Controlling air pollution emissions has become an urgent priority for humanity to proactively mitigate the impacts of climate change [3]. However, any initiative to reduce air pollution requires massive capital investment. The role of financial instruments, particularly green finance, in air management is receiving increasing attention [4].

To channel financial resources toward environmental protection, China has actively begun the exploration of green finance [5]. At the United Nations Conference on Sustainable Development in 2012, China first put forward the idea of building a green financial system. It elucidated the state’s policy of providing support for the advancement of green finance [6]. In 2017, the State Council decided to make Zhejiang, Guangdong, Jiangxi, Guizhou, and Xinjiang the first pilot regions for green finance policy [7]. The principal aim of establishing a pilot green finance zone is to utilize financial instruments to advance the governance of the environment within the designated jurisdiction [8]. Will the establishment of green finance policy effectively reduce air pollution in designated regions? Through what mechanisms does it influence air pollution? These questions are central to evaluating the policy’s effectiveness and guiding the implementation of pilot programs. For China, clarifying the relationship between green finance policy and air pollution is essential to improving urban air quality and advancing sustainable development.

The extant literature on green finance can be divided into two principal areas of investigation. The first of these concerns green finance’s effects on the macroeconomy. The current literature indicates that the advancement of green finance can facilitate the greening of the economy [9], superior economic development [10], and green total factor productivity [11]. Green credit policy can contribute to the green transformation of the economic structure [12]. The second area concerns the impact of green finance on microenterprises. Existing studies have shown that the development of green finance can promote corporate green technology innovation [13]. Green bond issuance can promote corporate green innovation [14]. Green credit policy has the potential to enhance the environmental and social responsibility of corporations [15], promote corporate environmental investment [16], and foster corporate green innovation [17]. Green funds have been shown to enhance firms’ carbon performance, especially in high-emission sectors, with stronger effects when executives possess green awareness and financial expertise [18]. Since the inception of green finance policy, its impact has been the subject of considerable scrutiny.

In examining the factors that contribute to air pollution, existing studies have primarily focused on direct sources and indirect factors. In terms of direct sources, transport emissions, coal combustion industrial emissions, open burning [19], and fossil fuel extraction [20] are all significant factors that directly contribute to air pollution. In terms of indirect factors, firstly, national policies such as the low-carbon city pilot policy [21], environmental protection tax [22], and the “ten articles of the atmosphere” [23] are able to mitigate air pollution. Second, the development of the Internet improves air quality through promoting industrial upgrading and fostering technological innovation [24]. A number of studies have previously examined the factors influencing the improvement in air quality from a variety of perspectives. There is a substantial body of literature examining the potential benefits of green finance policy for improving environmental effects. For instance, some literature indicates that green finance policy has the potential to reduce energy consumption intensity [25] and carbon emissions [6]. It also can enhance the efficiency of the green economy [26]. However, one limitation of the existing literature is that it only analyzes the effect of policy implementation from a single pollutant perspective. There is a lack of examination of the synergistic air pollution management effects of green finance policy. Moreover, it is yet unknown how precisely green finance reform measures air quality impact.

This paper makes the following contributions to the existing literature. First, this study conducts a more granular empirical analysis based on prefecture-level city data, offering sharper resolution than that of prior research that has typically focused on national or provincial levels. To enhance the robustness and precision of environmental measurement, this study constructs a composite air pollution index using the entropy weight method, incorporating both industrial SO₂ emissions and PM_2.5 concentrations. This integrated index allows for a more accurate assessment of policy impacts at the city level. Second, this paper contributes to the mechanism literature by identifying two key transmission pathways through which green finance affects air pollution: resource allocation and green innovation. This channel-based analysis clarifies the internal logic of environmental governance via financial instruments and enriches the theoretical framework. Third, the paper conducts a series of heterogeneity analyses, revealing how the effects of green finance vary across regions with different geographical characteristics, resource endowment, and financial development levels. These findings highlight the spatially differentiated policy impacts and provide valuable implications for targeted environmental governance.

The remaining portions of the study are structured as follows: Section 2 introduces the policy background and research hypothesis. Section 3 presents the research design. Section 4 provides the empirical results. Finally, policy recommendations are provided.

2. Policy Background and Research Hypothesis

2.1. Policy Background

According to the OECD (2020) [27], effective green finance governance requires coordinated action among regulatory institutions, financial markets, and environmental agencies to align investment flows with sustainability goals. Collaborative governance frameworks, such as taxonomies, disclosure mandates, and incentive structures, are critical in directing capital toward green sectors in developing economies.

In 2017, the State Council decided to make Zhejiang, Guangdong, Jiangxi, Guizhou, and Xinjiang the first pilot regions for green finance policy. Among them, Zhejiang is a tertiary-sector-led, urban area; Jiangxi, at mid-industrialization, is an industrial–rural province; Guizhou is a rural–ecological region dominated by primary industry; and Xinjiang is an industrial–rural frontier whose economy is resource-based and secondary-sector-oriented. The pilot districts have carried out a great deal of research in terms of organizational systems, product innovation, and supporting policies.

Regarding the organizational system, each pilot zone has established a Green Finance Leadership Group, which is networked at the provincial, regional, and municipal levels.

In terms of the development of financial products, the pilot zone has promoted the diversification of green financial products, which usually merit more attractive conditions, such as more favorable interest rates, lower handling fees, or higher potential for returns. This diversification strategy mainly encompasses the promotion of green credit, green bonds, green insurance, and green funds. In terms of green credit, the green branch has established a “four priorities” service channel for green credit, providing enterprises with priority approval and more favorable loan interest rates, increasing credit support for green enterprises and at the same time, consolidating credit resources for high-pollution and high-energy-consuming enterprises. In terms of green bonds, the Financial Services Department proactively assists eligible green enterprises by issuing more green corporate bonds, helping them obtain more favorable financing conditions and lower issuance costs. In terms of green insurance, by improving the environmental damage compensation mechanism in the pilot zone, a compulsory environmental pollution liability insurance system will be implemented for highly polluting and energy-consuming enterprises, and premium subsidies may be provided to reduce the insurance costs for enterprises. In terms of the establishment of green funds, each pilot region has set up a guiding fund for the development of green industries, with fiscal resources inclined towards green industries such as energy conservation and environmental protection, aiming to provide investors with opportunities to participate in high-growth green industries with potentially high returns.

In terms of supporting policies, the pilot zone has formulated constraints, incentives, and safeguard mechanisms to ensure the smooth implementation of green finance policy. The government of the pilot zone has introduced special assessment measures for the construction of the pilot zone, covering the administrative districts under it and all relevant departments. Rewards and penalties are implemented according to the fulfilment of tasks in each county and district. The People’s Bank of China incorporates the results of green credit performance evaluation of financial institutions into the comprehensive evaluation system of these institutions. Financial institutions will be incentivized to vigorously develop green financial business and curb the inflow of loans from the “two high” industries. On the financial front, the pilot zone has arranged for financial subsidies and risk compensation for green credit and green bonds with corresponding financial funds. In addition, premiums for green insurance are subsidized to fully mobilize financial institutions to develop green finance.

2.2. Research Hypothesis

Green finance policy is a new type of environmental governance instrument that promotes sustainable economic growth by improving the environmental efficiency of financial resource allocation [28]. Unlike traditional financial systems, green finance focuses on directing capital flows toward green industries such as energy conservation, environmental protection, and clean energy [29], thereby reducing overall pollutant emissions and supporting low-carbon development. Green finance supports the implementation and expansion of environmentally friendly projects by offering financial tools such as green credit, green bonds, dedicated funds, insurance services, and fiscal subsidies [30]. These projects include renewable energy facilities and clean technology applications. The use of these financial instruments reduces the cost of capital for enterprises undertaking green transformation. It also helps alleviate financing constraints and promotes the optimization of industrial structures, leading to improvements in both economic and environmental performance. Furthermore, green finance promotes the concept of green investment and green consumption, which enhances public environmental awareness, encourages low-carbon lifestyles, and helps to reduce dependence on energy-intensive and highly polluting sectors [31]. At the institutional level, green finance also serves as a policy tool for strengthening environmental regulation. Financial institutions are increasingly required to integrate environmental criteria into financing decisions, and firms are expected to disclose environmental information and improve their ecological responsibility [32]. In green finance pilot zones, regulatory frameworks are more stringent, credit resources are more carefully evaluated, and capital allocation is guided more strictly in favor of green sectors. At the same time, green financial performance indicators are incorporated into the assessment systems of both financial institutions and local governments, which helps ensure effective policy implementation and enhances accountability for environmental outcomes [33]. These mechanisms collectively contribute to the formation of a comprehensive governance system. Within this system, green finance directly helps reduce air pollution by promoting green investment, limiting financing for heavily polluting industries, and strengthening environmental supervision through coordinated financial and administrative efforts. In light of this, this paper proposes the following hypothesis.

Hypothesis 1.

Green finance policy has a catalytic effect on mitigating air pollution in pilot areas.

Green finance can enhance the efficiency of capital utilization by directing financial resources toward high-efficiency and low-pollution sectors, thereby improving green total factor productivity through more targeted investment [34]. It also reduces the cost of capital for environmentally responsible firms and encourages them to upgrade their production processes and adopt cleaner energy use patterns [35]. One of the primary mechanisms through which green finance contributes to air pollution reduction is the resource allocation effect. In this study, the resource allocation effect is defined as the capacity of green finance to reallocate capital away from fossil fuel-intensive industries and toward renewable, energy-efficient, and low-carbon sectors. This reallocation process gradually transforms the energy consumption structure by increasing the share of clean energy and reducing dependence on high-carbon energy sources. Such structural adjustment lowers overall energy intensity and helps mitigate pollution in the long run [36]. Green financial instruments, particularly green credit and green bonds, play a critical role by facilitating access to financing for green industries while restricting capital availability to high-emission sectors [37]. As green investment expands, the regional energy mix becomes more balanced, sustainable, and environmentally sound [38]. In contrast to the catalytic effects emphasized in Hypothesis 1, which operate through direct incentives and enhanced regulatory enforcement, the resource allocation effect focuses on long-term structural change in energy use. Therefore, green finance policy mitigates air pollution in pilot areas by optimizing the energy consumption structure through improved capital allocation. In light of this, this paper proposes the following hypothesis.

Hypothesis 2.

Green finance policy has a resource allocation effect, mitigating air pollution in pilot areas by optimizing energy consumption structure.

One of the key mechanisms through which green finance policy contributes to air pollution mitigation is the green innovation effect. In this paper, the green innovation effect is defined as the process by which green finance stimulates the research, development, and adoption of environmentally friendly technologies, thereby reducing pollution and improving energy efficiency. The Porter hypothesis suggests that well-designed environmental regulations can stimulate innovation and create win–win situations for the economy and the environment. Green finance policy stimulates green innovation by raising the cost of financing and environmental violations for polluting firms [39]. When firms face tight financing constraints, their capacity to invest in innovation is often limited. Green finance helps ease these constraints, especially for green technology development, by directing capital toward low-carbon and clean energy solutions [40]. Moreover, the participation of financial institutions in green finance enhances market information transparency and reduces information asymmetry. This enables investors to make informed decisions based on credible environmental performance metrics [7]. Technological progress is essential to breaking fossil energy dependence. Green innovation, in particular, plays a vital role in reducing pollution through energy efficiency improvement and the substitution of cleaner alternatives [41]. In the long term, technological innovation drives companies to use more advanced energy technologies to promote energy conservation and emission reduction [42]. Therefore, green finance policy can enhance the green innovation capacity of enterprises served in pilot areas. It promotes renewable energy consumption and reduces pollution intensity. In light of this, this paper proposes the following hypothesis.

Hypothesis 3.

Green finance policy has a green innovation effect, mitigating air pollution in pilot areas via green technological innovations.

In summary, the mechanism of green finance policy for mitigating air pollution is shown in Figure 1.

3. Research Design

3.1. Model

In order to examine the impact of green finance policy on air pollution, this paper sets up the following regression model:

$$AP{I}_{it}={\alpha }_0+{\alpha }_{1}Trea{t}_i\times Tim{e}_t+\beta X_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(1)

where the explanatory variable API_it denotes the air pollution index of the city i in year t; Treat_i is the dummy variable for the city; Time_t is a time dummy variable; X_it is a control variable group; μ_i is the city fixed effect, and λ_t is the year fixed effect. ε_it is the random disturbance term. α₀ denotes the intercept term. This paper focuses on the coefficient α₁ of the interaction term in Model (1), which captures the policy’s impact on air pollution.

3.2. Variable Selection

3.2.1. Explained Variables

At present, the academic community has not yet formed a unified standard for the measurement of air pollution levels. Most scholars use only a single indicator to assess the regional air pollution level, which is not able to comprehensively reflect the real air pollution level in the region. Therefore, this study drew upon [43] and finally selected representative industrial SO₂ air pollutant emissions, as well as PM_2.5 concentrations, in each city. The air pollution index was constructed via the entropy weight method to measure the air pollution level in each city.

The air pollution index (API) in this study is constructed using the entropy weight method. Compared with the analytic hierarchy process (AHP), which assigns weights based on subjective expert judgment, the entropy weight method offers a more objective approach by relying on the statistical variation of each indicator. This helps reduce bias arising from human subjectivity. In contrast to the equal weighting method, which assumes all indicators are equally important, the entropy method captures the relative importance of each pollutant based on the amount of information it contributes to the system. This enhances the methodological transparency and robustness of the index.

The calculation procedure involves four steps: (1) normalization of raw data to ensure comparability between indicators; (2) computation of entropy values to evaluate the dispersion degree of each indicator; (3) determination of entropy-based weights according to the information contribution of each pollutant; and (4) aggregation of weighted indicators into a single composite index. The higher the value of the API, the more severe the level of air pollution in the city.

3.2.2. Independent Variables

Treat_i × Time_t is the core explanatory variable of this paper, which represents the dummy variable of green finance policy, and Treat_i is a dummy variable for prefecture-level city; the value of Treat_i is 1 if a city is the pilot zone; otherwise, it is 0. Time_t is a time dummy variable, with the boundary set in 2017. The value of Time_t is 1 if the time is after the year 2017, and 0 otherwise.

3.2.3. Control Variables

This paper draws on the work of [44,45,46] and selects the following control variables to consider other possible factors that may affect air pollution. (1) Level of industrial structure (IND): the ratio of industrial added value to GDP. (2) Population density (DEN): measured as the logarithm of the number of people per square kilometer in each city. (3) Urbanization level (URB): measured by the ratio of urban population to total population. (4) Level of government intervention (GOV): measured by the ratio of municipal fiscal expenditure to GDP. (5) Level of human capital (HUM): measured as the ratio of the number of students enrolled in general higher education per 10,000 people.

The symbols and calculations for the above variables are shown in

Table 1. Variable description table.

Variable Types	Variables	Description
Explained variables	API	Air pollution index (the entropy method)
	PM2.5	Annual mean PM2.5 concentration (μg/m3)
	SO2	Industrial SO2 emissions per unit of GDP (10,000 tons per 100 million CNY)
Independent variables	Treat × Time	Green finance policy dummy
Control variables	IND	Industrial structure ratio (%)
	DEN	Population density log(People/km2)
	URB	Urbanization level (%)
	GOV	Government intervention (fiscal/GDP) (%)
	HUM	Human capital (higher education ratio) (%)

.

3.3. Data Description

In terms of sample selection, this study constructs a panel dataset covering 276 prefecture-level cities in China from 2011 to 2020. The information on green finance pilot cities is based on the list issued by the State Council in 2017. The PM_2.5 data used in the air pollution indicator were obtained from Washington University in St. Louis (https://sites.wustl.edu/acag/datasets/surface-pm2-5/) (accessed on 1 August 2025), and the annual average PM_2.5 concentrations for each city were extracted using the zonal statistics tool in ArcGIS [47]. Industrial SO₂ emissions per unit of GDP were sourced from the China Urban Statistical Yearbook. Other control variables were also collected from the China Urban Statistical Yearbook and statistical bulletins of prefecture-level cities.

To ensure data quality, regions with a high proportion of missing data, such as Tibet, were excluded from the final sample. For other cities, scattered missing values were supplemented manually by consulting local statistical bulletins. The remaining missing entries were completed using linear interpolation along the original data series.

Table 2. Descriptive statistics.

Variable	Sample Size	Average Value	Standard Deviation	Minimum Value	Maximum Value
API	2760	0.0768	0.0711	0.0006	0.9039
Treat × Time	2760	0.0116	0.1071	0.0000	1.0000
IND	2760	0.4587	0.1087	0.1171	0.8934
DEN	2760	5.7439	0.9422	0.6831	7.8816
URB	2760	0.5626	0.1465	0.0649	1.0000
GOV	2760	0.2012	0.1018	0.0439	0.9155
HUM	2760	0.0198	0.0259	0.0001	0.1938

displays the variables’ descriptive statistics.

4. Empirical Results

4.1. Baseline Regression Results

In this paper, the baseline regression model was regressed as shown in

Table 3. Baseline regression results.

Variables	(1)	(2)
Treati × Timet	−0.054 ***	−0.040 ***
	(−5.02)	(−6.18)
IND		0.075 ***
		(2.97)
DEN		0.093 ***
		(7.97)
URB		0.093 ***
		(2.98)
GOV		0.039 **
		(1.98)
HUM		−0.206
		(−1.56)
Constant	0.077 ***	−0.547 ***
	(123.77)	(−7.88)
City FE	Yes	Yes
Year FE	Yes	Yes
Observations	2760	2760
R-squared	0.814	0.849

Notes: ***, **, * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

. Column (1) shows the regression results using only the explanatory factors, while Column (2) displays the regression results including the control variables. The regression results show that the green finance policy significantly reduces the level of air pollution. Table 3’s Column (1) indicates a significant negative coefficient of Treat_i × Time_t at the 1% level. This suggests that the green finance program is effective in reducing air pollution levels after accounting for individual and time impacts. The negative coefficient of Treat_i × Time_t remains significant at the 1% level after adding control variables. More specifically, when the control variables are considered and not considered, green finance policy reduces air pollution of cities by 4.0% and 5.4%, respectively. As a result, this study concluded that the green finance policy will help to reduce air pollution in the pilot cities, and Hypothesis 1 was confirmed.

4.2. Parallel Trend Test

The application of the DID model requires that the parallel trend assumption is satisfied, which means there should be no systematic difference in air pollution level between pilot and non-pilot cities before the implementation of the green finance policy. To examine the variation in policy effects across different years before and after its implementation, this study takes the year prior to the policy announcement as the baseline period and employs the event study approach to conduct the parallel trend test [48]. Figure 2 presents the test findings. The results show that the coefficient of the effect of the green finance policy on air pollution before the implementation of the policy is not significant, and the original hypothesis that the regression coefficient is 0 is not rejected. The results indicate that there were no significant differences in air pollution trends between the treatment and control groups prior to the implementation of the green finance policy. In 2017–2020, the impact of green finance policy is significantly negative, indicating that green finance policy significantly improves the level of air pollution. Therefore, the parallel trend hypothesis is valid.

4.3. Robustness Test

4.3.1. Placebo Test

Through the use of placebo tests, this study investigates the degree to which random factors and issues with missing variables affect the baseline regression results. Specifically, we employ a randomization approach to randomly assign treatment and control groups, followed by a placebo DID regression to store the estimated coefficients [49]. We conduct 500 rounds of random sampling and plot the kernel density distribution of the estimated coefficients, as illustrated in Figure 3. If the estimated coefficients under the stochastic treatment are distributed around 0, it implies that the spurious policy variables do not have a significant impact on air pollution. The results of Figure 3 show that, in this paper, the spurious regression coefficients are centrally distributed around 0 and away from the baseline regression coefficients (−0.040). This suggests that, to some extent, the baseline regression results are not due to unobserved chance factors.

4.3.2. PSM-DID Model

The DID model is prone to “selectivity bias”. With large sample sizes, individual characteristics of districts in the experimental and control groups can be significantly different, leading to biased estimates. This paper uses propensity score matching (PSM) to correct for sample selection bias by identifying control group cities that share similar characteristics with those in the treatment group [50]. Regarding the matching method, this study selects control variables as matching covariates and applies both kernel matching and radius matching methods to perform the PSM-DID estimation. The regression coefficient remains considerably negative, as seen in

Table 4. PSM-DID test results.

Variables	(1)	(2)
	Kernel Matching	Radius Matching
Treati × Timet	−0.035 ***	−0.036 ***
	(−4.87)	(−5.33)
Constant	−0.569 ***	−0.566 ***
	(−7.99)	(−8.05)
Control	Yes	Yes
City FE	Yes	Yes
Year FE	Yes	Yes
Observations	2157	2307
R-squared	0.854	0.853

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

. This indicates that the inhibitory effect of green finance policy on air pollution is still significant after considering the problems of sample self-selection and sample selection bias, further reinforcing the robustness of the paper’s conclusions.

4.3.3. Double Machine Learning (DML)

Even when the functional form of covariates is unknown, Double Machine Learning (DML) can still provide unbiased estimates of treatment effects [51]. First, this study employs a baseline random forest model with a 4-fold sample-splitting scheme, and the results are reported in Column (1) of

Table 5. Double machine learning test results.

Variable	(1)	(2)	(3)
	Random Forest	Gradient Boosting	LassoCV
Treati × Timet	−0.081 ***	−0.065 ***	−0.039 ***
	(−4.09)	(−2.68)	(−6.22)
Constant	0.0014 **	0.0003	0.0001
	(2.55)	(0.38)	(0.21)
Control	Yes	Yes	Yes
City FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	2760	2760	2760

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

. Second, to test the robustness of the findings via different machine learning algorithms, the baseline random forest is replaced with Gradient Boosting and LassoCV, with the corresponding results presented in Columns (2) and (3). Across all specifications, the estimation results remain consistent, indicating that the use of DML does not alter the main conclusions of the baseline regression.

4.3.4. Removing Environmental Policy Interference

Considering that other environmental policies may affect air pollution, this paper reviews the major environmental policies implemented in China during the sample period. Specifically, the analysis focuses on the Environmental Protection Tax, the Carbon Emission Trading Scheme, and the Energy Rights Trading Policy. As these policies may affect urban air quality, they could potentially interfere with the estimation results. In order to exclude the interference of other environmental policies, the pilot regions of these policies are excluded and regressed, respectively. As shown in

Table 6. Robustness test excluding other policy effects.

Variable	(1)	(2)	(3)
	Environmental Protection Tax	Carbon Emission Trading Scheme	Energy Rights Trading Policy
Treati × Timet	−0.040 ***	−0.046 ***	−0.035 ***
	(−3.96)	(−6.39)	(−3.94)
Constant	−0.624 ***	−0.550 ***	−0.593 ***
	(−7.51)	(−7.94)	(−7.27)
Control	Yes	Yes	Yes
City FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	1500	2410	2210
R-squared	0.856	0.850	0.850

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

, the regression coefficients remain significantly negative, indicating that the main findings are still robust, even after accounting for the potential interference of other environmental policies.

4.3.5. Other Robustness Tests

To enhance the robustness and validity of the findings, this paper conducts four additional robustness checks: (1) Addition of control variables. To address the potential influence of omitted variables on the baseline regression results, this study incorporates additional control variables into the model. Specifically, energy intensity, foreign direct investment, and the level of science and technology are included. As shown in Column (1) of

Table 7. Other robustness tests.

Variable	(1)	(2)	(3)	(4)
Treati × Timet	−0.042 ***		−0.040 ***	−0.049 ***
	(−6.67)		(−6.18)	(−6.29)
L1. Treati × Timet		−0.026 ***
		(-3.64)
Constant	−0.533 ***	−0.522 ***	−0.546 ***	−0.330 ***
	(−7.68)	(−6.79)	(−7.87)	(−5.68)
Control	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	2760	2484	2720	2760
R-squared	0.851	0.843	0.849	0.860

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

, the coefficient for the green finance policy remains significantly negative, thereby reinforcing the robustness of the conclusions. (2) Lagging of core explanatory variable. To account for the potential lagged effects of policy implementation, the Treat_i × Time_t variable is lagged by one period. The regression results, reported in Column (2) of Table 7, indicate that the coefficient of the lagged Treat_i × Time_t variable remains significantly negative. This finding suggests that the delayed effect of the policy does not alter the robustness of the main conclusion. (3) Exclusion of municipalities. Since municipalities such as Beijing, Shanghai, Tianjin, and Chongqing differ greatly from other cities in terms of administrative level and economic development, they may affect the robustness of the results. Therefore, these four cities are excluded, and the regression is re-estimated. As shown in Column (3) of Table 7, the coefficient remains significantly positive, indicating that the main conclusion still holds without the inclusion of these municipalities. (4) Winsorization. All continuous variables are winsorized at the 1% level to reduce the influence of extreme values. As shown in Column (4) of Table 7, the policy coefficient remains significantly negative at the 1% level, indicating that the main result is robust.

4.4. Mechanism Test

The baseline regression results presented in the previous section show that the establishment of green finance policy significantly reduces air pollution levels. Building on this finding, this section explores the underlying mechanisms through which green financial reforms contribute to air pollution mitigation. Based on the proposed research hypotheses, two primary channels are examined: the resource allocation effect and the green innovation effect.

There remains considerable debate over the appropriate methodology for testing mediating mechanisms. The traditional three-step mediation analysis is often subject to endogeneity concerns. Therefore, this paper adopts a simplified approach, relying solely on the second step of the conventional mediation test [52]. The relationship between the mediating variables and air pollution is analyzed primarily through the existing literature and logical inference. The corresponding empirical model is presented as follows:

$$Mediat{e}_{it}={\alpha }_0+{\alpha }_{1}Trea{t}_i\times Tim{e}_t+\beta X_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(2)

4.4.1. Resource Allocation Channel

The optimization of the energy consumption structure serves as a primary pathway through which the resource allocation effect of green finance policy is realized. Given that coal accounts for a large share of total energy consumption in prefecture-level cities, the ratio of coal consumption to total energy consumption is used to measure the energy consumption structure and then to verify whether the green finance policy can reduce air pollution by optimizing the energy consumption structure.

In Column (1) of

Table 8. Mechanism test.

Variable	(1)	(2)	(3)	(4)	(5)	(6)
	ECS	API	GTI1	API	GTI2	API
Treati × Timet	−0.055 ***		0.718 ***		0.721 ***
	(−4.89)		(4.04)		(3.68)
ECS		0.017 **
		(1.99)
GTI1				−0.001 **
				(−2.01)
GTI2						−0.005 ***
						(5.62)
Constant	0.622 ***	−0.568 ***	0.209	−0.557 ***	2.023 ***	−0.546 ***
	(10.65)	(−7.87)	(0.29)	(−7.68)	(2.68)	(−7.52)
Control	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	2760	2760	2760	2760	2760	2760
R-squared	0.765	0.847	0.815	0.847	0.807	0.849

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

, the coefficient of Treat_i × Time_t is significantly negative, indicating that the green finance policy improves the energy consumption structure. In Column (2), the coefficient of ECS is significantly positive, suggesting that a higher coal share is associated with worse air pollution. According to the neoclassical growth model, efficient allocation of production factors is essential to sustaining economic growth. However, when market imperfections allow environmentally harmful industries to access capital without accounting for externalities, resources are misallocated, leading to both lower productivity and higher pollution [53]. Green finance policy addresses this by redistributing financial resources toward low-carbon sectors, thus improving capital allocation efficiency while mitigating environmental degradation [54]. The green finance policy creates financing advantages for enterprises characterized by low energy consumption and low emissions, while imposing financing constraints on those exhibiting high energy consumption and high emissions [55]. This differentiation in financial access incentivizes firms to adjust their energy consumption structures. Such enterprise-level transformations can subsequently drive broader shifts in the energy consumption structure at the municipal level, thereby contributing to the reduction of air pollution. Prior studies have shown that the structure of energy consumption is a crucial factor affecting pollution emissions [56,57]. These findings suggest that green finance policy mitigates air pollution by optimizing the energy consumption structure. Therefore, Hypothesis 2 is supported.

4.4.2. Green Innovation Channel

This study uses the quantity of green inventions per 10,000 people (GTI1) and green utility model patent per 10,000 people (GTI2) to quantify green innovation. Columns (3) and (5) of Table 8 report the results of the mechanism analyses for green invention patents and green utility model patents, respectively. The findings demonstrate that green finance policy can considerably raise the quantity of patent applications for green inventions and green utility models. Columns (4) and (6) of Table 8 show that green technological innovation is significantly and negatively related to air pollution.

Green finance policy steers resource allocation towards the green innovation sector by subsidizing its lending rates. The financing advantages of green finance can contribute to the rapid development of green technology innovation, which can lead to the reduction of air pollution [58]. From a theoretical perspective, green finance acts as an effective environmental regulation tool that internalizes environmental externalities by reducing the financing costs of green sectors and increasing the credit constraints for polluting enterprises [59]. This selective financial support mechanism improves the risk–return profile of green R&D projects, thereby stimulating innovation in clean technologies. Moreover, green finance promotes long-term investment and reduces short-termism in enterprise behavior, which is crucial for fostering green invention activities [60]. These mechanisms align with Porter’s hypothesis, which asserts that well-designed environmental policies can enhance both innovation and environmental performance.

Additionally, green innovation promotes the diffusion of clean technologies across firms and regions, creating positive spillovers that help reduce pollution. As more firms adopt energy-saving practices through imitation and supply chain integration, the economy gradually shifts toward low-emission production modes [61]. This transition encourages firms to upgrade outdated equipment, enhancing efficiency and lowering emissions of PM2.5, NOx, and SO₂. Regions with stronger innovation capacity and R&D support show greater air quality improvements, underscoring the vital role of green technology in environmental governance [62].

The empirical results reveal that green finance policy exerts a stronger impact on green utility model patents than on green invention patents. This disparity may be attributed to the relatively lower quantity, higher application thresholds, and greater technological complexity associated with green invention patents. Consequently, the effectiveness of green finance in promoting such high-level innovations appears more limited. Nonetheless, the overall findings suggest that green finance policies facilitate the mitigation of air pollution by fostering green technological innovation. Thus, Hypothesis 3 is supported.

4.5. Heterogeneity Analysis

4.5.1. Heterogeneity Test Based on Geographical Characteristics

Due to differences in economic development and imbalanced infrastructure across regions, the implementation and impact of the policy may vary significantly. Therefore, this study further analyzes heterogeneity from the perspective of regional differences by dividing the sample into three main categories: Eastern, Central, and Western, according to geographical location.

Table 9. Heterogeneity analysis in different regions.

Variable	(1)	(2)	(3)
	Eastern Region	Central Region	Western Region
Treati × Timet	−0.025 ***	−0.019 ***	−0.052 ***
	(−6.42)	(−2.90)	(−3.43)
Constant	0.050	−0.131 *	−0.655 ***
	(0.614)	(−1.94)	(−6.90)
Control	Yes	Yes	Yes
City FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	1000	960	800
R-squared	0.880	0.833	0.863
Chow test	8.46 ***

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

presents the regression results of the heterogeneity analysis based on geographical location. The significance of coefficient differences across subgroups in the heterogeneity analysis is assessed using the results of the Chow test. It is evident that the heterogeneity results all pass the Chow test, indicating that the estimated coefficients across groups are comparable. As shown in Table 9, the coefficients of Treat_i × Time_t for the Eastern, the Central, and the Western regions are all clearly negative. However, the absolute value of the coefficient is larger in the Western region. This suggests that the Western region is more affected by the impact of green finance policy on reducing air pollution than are the Central and Eastern regions.

The reasons may include the dominance of high-polluting and resource-intensive industries in the Western region, which face stronger financing constraints under green finance policies. These constraints push firms toward cleaner technologies and greener production modes. Furthermore, due to previously limited regulatory capacity and weaker environmental governance, the Western region has more room for improvement, making the effects of green finance more pronounced. Economically underdeveloped areas also tend to exhibit higher marginal returns for policy implementation, and local governments may show greater responsiveness to national green directives. These findings imply that regional differences in industrial structure, regulatory capacity, and policy motivation jointly shape the heterogeneous effects of green finance.

This result corresponds to the findings of Qiang et al. [63], who observed a more pronounced reduction in haze pollution in Central–Western cities. Overall, these results underscore that regional developmental context, industrial makeup, and institutional capacity significantly contribute to the effectiveness of green finance on environmental outcomes.

4.5.2. Heterogeneity Test Based on Resource Endowment

Based on the National Sustainable Development Plan for Resource-Based Cities (2013–2020), the sample was divided into resource-based and non-resource-based cities.

Table 10. Heterogeneity analysis based on level of resource endowment.

Variable	(1)	(2)
	Resource-Based Cities	Non-Resource-Based Cities
Treati × Timet	−0.054 ***	−0.031 ***
	(−3.46)	(−9.79)
Constant	−0.573 ***	−0.439 ***
	(−5.20)	(−3.97)
Control	Yes	Yes
City FE	Yes	Yes
Year FE	Yes	Yes
Observations	1100	1660
R-squared	0.832	0.890
Chow test	11.30 ***

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

presents the regression results of the heterogeneity analysis based on resource endowment. The significance of coefficient differences across subgroups in the heterogeneity analysis is assessed using the results of the Chow test. It is evident that the heterogeneity results all pass the Chow test, indicating that the estimated coefficients across groups are comparable. As shown in Columns (1) and (2) of Table 10, the implementation of green finance policy can significantly reduce air pollution in both resource-based and non-resource-based cities.

However, the estimated effect is notably larger in resource-based cities, suggesting that such cities benefit more from green finance interventions. This may be due to their substantial reliance on natural resource industries, which face heightened pollution risks and are consequently more sensitive to financing constraints imposed by green finance. Resource-based cities often possess greater potential for clean-industry development and possess stronger incentives for transformation under national green development goals. The policy thus accelerates their structural shift toward low-carbon industries and improves air quality more effectively. These findings align with the empirical evidence of Xu et al. [64], who find that green finance drives stronger energy structure transition in resource-based cities.

4.5.3. Heterogeneity Test Based on Level of Financial Development

The impact of green financial policies on reducing pollution may vary depending on the degree of regional financial growth. This study builds an index of the degree of financial development using the deposit and loan balances of financial institutions as a percentage of GDP. Based on the median value, the sample is classified as having a high financial development level or a low financial development level.

The regression findings for the groups with high and low financial development are displayed in

Table 11. Heterogeneity analysis based on different levels of financial development.

Variable	(1)	(2)
	High Financial Development	Low Financial Development
Treati × Timet	−0.036 ***	−0.018
	(−6.67)	(−1.57)
Constant	−0.380 ***	−1.068 ***
	(−4.11)	(−2.98)
Control	Yes	Yes
City FE	Yes	Yes
Year FE	Yes	Yes
Observations	1356	1373
R-squared	0.854	0.885
Chow test	3.11 ***

Notes: ***, **, and * denote 1%, 5%, and 10% significance levels, respectively; robust standard errors are used; values in brackets below the coefficients are t-statistics.

’s Columns (1) and (2), respectively. The significance of coefficient differences across subgroups in the heterogeneity analysis is assessed using the results of the Chow test. It is evident that the heterogeneity results all pass the Chow test, indicating that the estimated coefficients across groups are comparable. The findings demonstrate that, for both groups, Treat_i × Time_t display considerably negative regression coefficients. However, for the group with the highest level of financial development, the absolute value of the coefficient is higher and more significant. This implies that in locations with greater degrees of financial growth, green finance policies have a more pronounced moderating effect on air pollution. The most likely explanation is that regions with more advanced financial systems tend to possess more efficient resource allocation mechanisms, more mature financial infrastructure, and better regulatory capacity. These factors enhance the transmission of green finance instruments, such as green credit and green bonds, thereby amplifying their environmental benefits [65]. Moreover, financial deepening often fosters technological diffusion and capital mobility, which can accelerate the adoption of green technologies in polluting industries.

5. Conclusions and Policy Recommendations

The establishment of green finance reform and innovation pilot zones is a pilot policy proposed in response to China’s guidance to develop green finance for economic transformation and development. This study integrates green finance policy and air pollution into a unified analytical framework and employs a difference-in-differences (DID) model to empirically assess the policy’s impact and mechanisms. Furthermore, it investigates the heterogeneity of policy effects across regions with different geographical characteristics, levels of financial development, and environmental regulatory capacity.

The main findings of this study are as follows: (1) Baseline regression results show that the implementation of green finance policy has significantly reduced air pollution in pilot cities. This conclusion remains robust after parallel trend testing, placebo tests, and multiple robustness checks. (2) Mechanism analysis reveals that green finance policy alleviates air pollution, primarily through improved resource allocation and enhanced green innovation. Specifically, the policy is negatively associated with the proportion of fossil energy consumption and positively correlated with the level of green technological innovation. (3) Heterogeneity analysis confirms that the policy’s effects vary significantly across regions. The air quality improvements are more pronounced in Western regions, resource-based cities, and financially developed cities. These findings suggest that green finance policy can effectively complement local environmental regulatory efforts and enhance environmental governance effectiveness.

The findings of this study yield several important policy implications for promoting sustainable development through green finance. First, policymakers should consider expanding the coverage of green finance reform and innovation pilot zones, particularly in regions with high pollution intensity and strong industrial bases. At the same time, the successful experiences from existing pilot areas should be systematically summarized and promoted nationwide. To enhance the attractiveness of green investment, targeted incentives such as tax exemptions, interest subsidies, and risk compensation mechanisms should be strengthened to include private capital. Moreover, establishing a dynamic evaluation and feedback system for green finance performance will help ensure continuous policy adjustment and effectiveness.

Second, efforts should be made to optimize the resource allocation function of green finance. Financial instruments should be designed to direct funds toward low-carbon industries and environmentally friendly sectors. Governments should enhance both supply-side and demand-side support, e.g., on the one hand, by encouraging financial institutions to develop green credit, bonds, insurance, and funds; on the other hand, by mobilizing private and public sector participation in green investment. In particular, policies should facilitate financing for renewable energy and clean technology deployment. Simultaneously, R&D support for green technologies should be increased. Establishing green innovation funds and offering preferential innovation loans can stimulate enterprise-led technological advancement and reinforce the role of innovation in pollution reduction.

Third, a differentiated policy strategy should be adopted to reflect regional heterogeneity. In Western and resource-based cities, green finance support should be intensified to facilitate industrial restructuring and ecological restoration. In regions with more mature financial systems, policy should focus on enhancing the depth, diversity, and market orientation of green financial products. Cross-departmental and multi-level policy coordination mechanisms should be strengthened to maximize synergy. Finally, local governments should be empowered to play a more active role in implementation through strengthened environmental supervision, adequate fiscal support, and regulatory safeguards. Ensuring local accountability is key to the effective execution of green finance policies and the realization of environmental goals.

In addition to its implications for China, this study offers valuable insights for other developing countries that are actively exploring green finance mechanisms to address environmental challenges. Countries such as India, Brazil, and South Africa have initiated various green finance strategies—ranging from sovereign green bonds and sustainable banking frameworks to tax incentives for clean technology investments. However, common challenges persist, including weak financial infrastructure, insufficient regulatory coordination, and fragmented policy implementation. By unpacking how green finance can alleviate air pollution through resource allocation and green innovation, this study contributes to a broader understanding of how market-based instruments can support sustainable development in emerging economies.

Furthermore, international comparative studies have illustrated that the effectiveness of green finance instruments depends significantly on local institutional and financial contexts. For instance, India’s green bond market has played a vital role in expanding renewable energy capacity under strong central oversight, while Brazil’s national development bank (BNDES) has embedded environmental screening criteria into its credit practices. These cross-country experiences demonstrate the importance of aligning financial reforms with environmental governance structures. Future research can extend this paper’s framework to cross-national empirical studies, which will help evaluate the transferability and contextual adaptability of China’s green finance policies to other parts of the developing world. In conclusion, this study offers valuable insights for both the Chinese government and other developing countries in advancing green finance to achieve sustainable development.

Despite the robustness of the empirical results, this study is subject to several limitations. First, constrained by data availability, the empirical analysis is conducted at the city level, which limits the investigation of firm-level behavioral responses to green finance policy. Second, the air pollution index employed in this study is constructed based on two representative pollutants, which may not fully capture the multidimensional nature of urban air quality. Third, the analysis focuses on the direct treatment effects of the policy, without explicitly considering potential spatial spillover effects across regions. Future research could address these limitations in multiple ways. With the availability of firm-level data, micro-level analyses could be undertaken to better understand how enterprises adjust their investment behavior, innovation activities, and emission control strategies in response to green financial instruments. Additionally, future studies may employ spatial econometric techniques to examine the diffusion mechanisms and spatial externalities of green finance policies. These extensions would help to provide a more comprehensive understanding of the role of green finance in promoting collaborative and regionally coordinated environmental governance.