Does Air Quality Ecological Compensation Improve Total Factor Energy Efficiency?—A Quasi-Natural Experiment from 282 Cities in China

Abstract

The impact of air-quality ecological compensation on total factor energy efficiency can help a country to achieve high-quality economic development with the goals of peak carbon emissions and carbon neutrality, and to explore a win–win path for the economy and the environment. This study investigates the impact of air-quality ecological compensation (AQEC) on total factor energy efficiency using the time-varying difference-in-difference model with a sample of 282 Chinese cities from 2004 to 2022. The results show AQEC significantly increases total factor energy efficiency by 1.71% in the pilot areas. This result remains robust after undergoing PSM-DID, considering only the first three pilot batches, an instrumental variable approach, the exclusion of other policies interference, and dual machine learning. The mechanisms analysis indicates that AQEC enhances total factor energy efficiency by promoting industrial structure advancement and green technology innovation. Furthermore, it is noteworthy that heterogeneity exists in the effect of AQEC on total factor energy efficiency, particularly in cities with an old industrial base and small cities. Overall, this study refines the causal relationship between air-quality ecological compensation and total factor energy efficiency, providing empirical evidence and policy insights for China and other countries to enhance energy efficiency and promote urban ecological civilization.

1. Introduction

Since the Industrial Revolution, the global population and economic growth have created natural resource shortages and environmental degradation. Climate change has compelled the international community, such as the United States and the European Union, to rethink the current economic growth model by implementing an energy transition towards clean energy [1,2]. Similarly, the Chinese government has placed greater emphasis on clean energy consumption and production in its government performance appraisals to accelerate the transformation from a large energy-importing country to an energy powerhouse. Indeed, China is the world’s second largest economy as well as the largest producer and consumer of energy in the world. Its economic reforms are focused on the energy supply [3]. However, China’s overall coal consumption is still growing. Thus, energy efficiency improvements are needed to overcome the constraints of the energy supply structure and ecological and environmental pressures. There are numerous examples of successful initiatives in such situations. More than forty countries have accomplished amazing feats related to ecosystem service projects under extreme pressures for environmental management [4]. Examples include the cross-border lake mechanism in the Great Lakes between the United States and Canada, cross-regional synergistic management of the Murray–Darling River Basin in Australia, and the Biodiversity Offsetting Pilots program in England [5]. China has also creatively explored major innovative practices such as the air-quality ecological compensation policy, which has greatly stimulated the enthusiasm for intergovernmental environmental governance. Can this policy alleviate energy efficiency in China, which is a pressing need under the pressure to extract more out of the same energy resources and reduce emissions? This is what this study focuses on.

This study’s contributions are four-fold: Firstly, we are the first to consider the direct impact of air-quality ecological compensation on total factor energy efficiency and provide evidence on the relationship between the air-quality ecological compensation mechanism and total factor energy efficiency. Secondly, this research examines the underlying driving mechanism of air-quality ecological compensation on total factor energy efficiency from the viewpoints of green technology innovation and industrial structure advancement. Thirdly, this paper explores the differential effects of air-quality ecological compensation on total factor energy efficiency under different urban industrial characteristics and urban size classes to further improve the related research. Lastly, the findings of this paper are of significant reference value for the government in terms of improving the air-quality ecological compensation policy and enhancing total factor energy efficiency.

The remainder of our paper is divided into seven parts. Section 2 introduces the related literature review. Section 3 presents the theoretical analysis and the study’s theoretical hypotheses. Section 4 describes the model, variables, and data sources used in this study. Section 5 analyzes the empirical findings. Section 6 completes the mechanism discussion and heterogeneity analysis. Section 7 presents the major findings and policy suggestions.

2. Literature Review

The first theme of interest in the literature is the study of ecological compensation policies. Regarding the connotations of ecological compensation, the concept of ecological compensation dates to the 1970s and has fueled public debate about the economic value of ecosystems. Ecological compensation can be regarded as a comprehensive mechanism to coordinate the economic interests between ecological protection beneficiaries and ecological protectants [6]. It can effectively internalize a negative ecological externality and improve environmental equity [7]. Ecological compensation comprehensively accounts for the cost of ecological protection and the opportunity cost of the development and value of ecological services by fully utilizing financial transfer payment. Essentially, ecological compensation is realized through market transactions involving immaterial benefits like money and raw materials. In terms of the types of ecological compensation, compared to the extensive research on forests [8], grasslands [9], watersheds [10], and croplands [11,12], there are fewer results explicitly examining air-related ecological compensation policies because of the transboundary nature of air pollution and the complexity of its sources. Regarding the evaluation of ecological compensation policies and the computation of compensation rates, studies have used the synthetic control method and difference-in-differences method to evaluate the environmental benefits of ecological compensation policies and to quantify the economic and poverty reduction effects. Dai et al. [7] constructed a policy and critique index model based on text mining to examine the implementation benefits of China’s ecological compensation policies which were performing above expectations. Furthermore, the definition of a reasonable range of ecological compensation standards is crucial in ecological compensation projects. Extant measurement methods focus on the ecosystem-service value approach [6,12] and bargaining game analysis model [13].

A second theme of interest in the literature is the impact of environmental policies on energy efficiency. Research on how environmental regulatory policies affect energy efficiency focuses on the “compensation for innovation” and “cost of compliance theory” perspectives. Existing studies point out that the “compliance cost effect” of environmental regulation significantly increases the cost of pollution control and inhibits energy efficiency [14,15]. Nevertheless, with the introduction of Porter’s hypothesis, some scholars have obtained more positive findings. Wang et al. examined the impacts of market-incentive, command-and-control, and voluntary-participation environmental regulations on energy efficiency based on the SYS-GMM model. They discovered that the command-and-control type promotes energy efficiency more than the other two types [16]. Mandal and Madheswaran provided evidence that environmental regulations had a greater effect on energy efficiency in the Indian cement sector [17]. Several academics who have examined the effects of carbon emissions trading policy [18], innovation-driven policy [19], and an environmental protection tax [20] on energy efficiency have reached similar results on “compensation for innovation”.

Some of the literature is most relevant to the object of our study. Whether ecological compensation can effectively improve energy efficiency has been examined by scholars. Li and Zipp found that through ecological compensation, the U.S. government effectively stimulates the public to provide Switchgrass and other biofuels as an alternative to fossil fuels, and to improve the level of energy utilization [21]. Everard et al. found that the Converging World Programme, which includes the UK and India, has made cross-regional ecological compensation possible. This model of cooperation optimizes the conventional energy mix by providing assistance for wind turbines in India and further investing surplus funds in tropical dry evergreen forest [22]. Cases from countries like Costa Rica and Mexico confirm that watershed ecological compensation promotes the protection of cloud forests by the energy sector and hydropower companies, which is essential for optimizing hydropower outputs [23]. Ecological compensation like the Conversion of Cropland to Forest Program and the Ecological Welfare Forest Program in China will encourage people to switch from biomass fuel to modern fuel [24]. In addition, a small number of scholars have focused on the policy effects of the implementation of air-quality ecological compensation. Based on a time-varying difference-in-difference model, Gan and Zong found that air-quality ecological compensation has an accelerating effect on the improvement of the atmospheric environment [25]. The “pollution reduction” effect is particularly prominent in the eastern region and resource cities [26]. In summary, most of the existing studies focus on the impacts of cropland ecological compensation, forest ecological compensation and watershed ecological compensation on energy efficiency. While several studies have assessed the effects of the pilot scheme of the airquality ecological compensation policy, they have mainly focused on atmospheric quality. To our knowledge, no study has yet focused on the relationship between air-quality ecological compensation and total factor energy efficiency.

The extant literature has the following research gaps: ① Currently, the research on ecological compensation concentrates on measuring ecological compensation standards, improving the operation mechanism, and its consequences on the economy. However, there are limited assessments of the outcomes of ecological compensation policies. ② Government-led ecological compensation has gained popularity over time. However, most studies on the subject concentrate on natural resources areas such as watersheds, forests, grasslands, and arable land. There are very few theoretical analyses and empirical investigations of air-quality ecological compensation. ③ Numerous policy considerations, such as carbon emissions trading regulations, innovation-driven policy, and environmental protection taxes, have been identified in the literature as having an impact on urban total energy efficiency. Nonetheless, a deficiency persists in terms of the thorough examination of the underlying mechanisms and direct investigation of how an air-quality ecological compensation policy can impact urban total factor energy efficiency.

This study focuses on evaluating the effect of the air-quality ecological compensation policy on urban total factor energy efficiency and its mechanisms of action, aiming to answer the following important concerns: First, is it possible for the air-quality ecological compensation policy to successfully encourage an increase in the total factor energy efficiency of metropolitan areas? Second, how does the air-quality ecological compensation policy affect the total factor energy efficiency of cities? Third, does the influence of air-quality ecological compensation construction on urban total factor energy efficiency vary by region? In order to investigate the efficacy of the air-quality ecological compensation practice, build and enhance the air-quality ecological compensation system, and encourage the transformation of the energy economy, the answers to these questions are extremely valuable.

3. Theoretical Analysis and Hypothesis Development

3.1. Impact of Ecological Compensation on Total Factor Energy Efficiency

In the early stage of Chinese urban development, economic growth directly leds to severe energy consumption [27]. However, energy consumption irreversibly impacts the ecological environment, resulting in air and water pollution in both the local and distant regions without the urban area having to bear the external costs, but also enjoying the same spillovers of ecological positive externalities from other regions. In the long run, there is a free-rider phenomenon in governmental environmental governance, which can diminish the initiative of those taking efforts to mitigate urban air pollution [28]. To achieve sustainable economic development, cities in developed and developing countries have identified green and low-carbon development strategies [1,2]. The Chinese government has adopted the air-quality ecological compensation policy to mitigate the inequity of environmental governance among governments, aiming to internalize the negative externality of air pollution emissions [7] and promote the green and low-carbon transition of cities [29]. Ecological compensation is in turn paid to the provincial government by cities that emit more emissions [30]. Research shows that the policy helps reduce urban haze emissions [31] and improves the air quality of prefectural-level cities subject to the policy. Furthermore, the policy successfully lowers CO₂ emissions [32]. As a crucial instrument to improve the environment, the air-quality ecological compensation policy will change the flow and consumption structure of energy in industry, thus affecting the consumption demand of different energy sources. Thus, Hypothesis 1 is proposed as follows:

Hypothesis 1. 

The air-quality ecological compensation policy has a positive impact on total factor energy efficiency.

3.2. Mechanisms

Firstly, the ecological compensation policy raises the environmental standards of pilot cities [33]. By redistributing the ecological benefits from different administrative agencies and different regions [6], it forces local governments to take effective measures to adjust the industrial structure. On the one hand, in recent years, the central government has heavily pushed the targets of peak carbon and carbon neutrality, which has made the flaws of heavily polluting and energy-consuming industries more visible. Ecological compensation constraints the behavior of heavily polluting and high-emission enterprises and may force high-energy-consuming industries to change their development paradigms [34]. In addition, the implementation of the ecological compensation policy will further accelerate the withdrawal of backward or pollution-intensive industries and hasten the phase-out of the traditional high-pollution development model [35]. Finally, the implementation of the ecological compensation policy has expanded the attention of local governments to environmental protection areas, which has led to the emergence of new economic growth sectors such as related green or high-tech industries. Over time, the tertiary industry has become a new engine in the green and low-carbon transformation of cities. And the optimization and transformation of the industrial structure are often accompanied by the transfer and reallocation of resources and production components, which is crucial for promoting the efficient use of energy.

Secondly, the fundamental purpose of air-quality ecological compensation is to “reduce emissions”, which places additional demands on businesses and governments. On the one hand, the local governments with high pollution intensity experience greater financial pressures because they must pay for the excessive pollution. Consequently, through the publication of national economic development reports [36], the government encourages R&D investment and talents to move towards low-carbon and green development. By providing tax breaks and granting R&D subsidies, the government eases the financial pressure faced by innovative activities and guides highly polluting enterprises to develop eco-friendly and energy-efficient technologies and equipment. On the other hand, air-quality ecological compensation raises ecological standards in pilot cities [37]. Economic sectors face greater governance costs and penalty costs as environmental target constraints become tighter. Therefore, economic sectors will aggressively engage in innovation activities to accelerate the transformation of the original polluting production to cleaner production, boost the supply of green technology products and services, and facilitate the low-carbon transition of cities [38]. Numerous studies have demonstrated that green technology innovation, especially independent innovation, is a key driver for promoting clean and low-carbon energy utilization [39,40,41,42]. Therefore, we formulate our second hypothesis as follows:

Hypothesis 2. 

The air-quality ecological compensation policy can improve total factor energy efficiency through industrial structure advancement and green technology innovation.

4. Research Design

4.1. Data

Since 2011, with the joint deployment of the Ministries of Finance and Environmental Protection, Anhui and Zhejiang provinces have carried out China’s first inter-provincial basin ecological compensation mechanism pilot project in the Xin’an River basin. In 2018, China’s National Development and Reform Commission, the National Forestry Administration, and four other departments issued a notice of the Ecological Poverty Alleviation Work Plan to provide more benefits to the poor from ecological protection and the restoration of the environment and achieve a “win-win” situation for poverty alleviation and environmental governance. In 2020, China promulgated the Regulations on Ecological Protection Compensation (Draft for Comments), emphasizing the inclusion of arable land as one of the main types of resources which could receive ecological compensation. China’s ecological compensation policies have broadened in scope, expanded the pool of compensation funds, gradually improved their mechanisms, and now have diverse forms. Financial support from these policies comprehensively covers key areas such as forests, watersheds, and key ecological functional zones.

Similarly, to reduce energy consumption and promote low-carbon transformation, the provinces of Shandong, Hubei, Henan, and Anhui officially launched pilot air-quality ecological compensation policies in 2014, 2015, 2016, and 2018, respectively, including fine and respirable particulate matter in the assessment indicators for ecological compensation. The fund coefficient of ecological compensation is adjusted upward annually, which effectively promotes the internalization of environmental pollution spillovers and can restrain irrational emissions behavior in the region. Given that the air-quality ecological compensation policy pilot has been carried out for nearly a decade, has the policy effectively promoted the improvement of total factor energy efficiency? This is the focus of this study’s investigation.

4.2. Time-Varying Difference-in-Differences Regression Model

To avoid the selectivity bias caused by the non-randomness of policy interventions as much as possible, quasi-experimental techniques such as the matching method, the breakpoint regression method, and DID methods have been applied to policy evaluations. The matching method is a non-experimental method that requires the fulfilment of the conditional independence and common interval hypotheses, as well as the availability of a large amount of data. The breakpoint regression method indicates that there is a continuous variable that determines the probability of an individual receiving a policy intervention on either side of a threshold. The theoretical framework of the DID method is based on “natural experiments”, where causal effects are measured by constructing grouping dummy variables, staging dummy variables, and interaction terms. However, the traditional DID model is only applicable for examining the effectiveness of policies at a single point in time and may be unable to comprehensively identify the effect of air-quality ecological compensation policy, which has been implemented in batches. To better capture the progressive and cumulative impacts of this policy over the long term, this study uses the following time-varying DID model, drawing on Ma et al. and Yang et al. [43,44]:

$$\mathit{Energy}\_{\mathit{efficiency}}_{\mathit{it}}={\beta }_0+{\beta }_1{\mathit{AQEC}}_{\mathrm{it}}+\delta {\mathit{X}}_{\mathit{it}}+{\mu }_i+{\gamma }_t+{\epsilon }_{\mathit{it}},$$

(1)

where i and t denote the city and year, respectively; Energy_efficiency_it is the explanatory variable, representing total factor energy efficiency; $\beta$₀ is a constant term; AQEC_it equals 1 if the city is classified as a pilot city for air-quality ecological compensation in the current or future years, and 0 otherwise; X_it denotes a set of control variables; $\mu$_i and $\gamma$_t denote the city and year fixed effects, respectively; and $\epsilon$_it is a randomized perturbation term. $\beta$₁ is used to measure the causal effect of total factor energy efficiency before and after the policy’s implementation.

4.3. Variable Selection

4.3.1. Explained Variable

Total factor energy efficiency (Energy_efficiency). Total factor energy efficiency refers to the ratio of the optimal amount of energy inputs to the actual amount of inputs required for a given output, according to best production practices, with all factors of production other than energy factor inputs held constant. Studies mainly measure energy efficiency from two perspectives: the single- and total-factor approaches. The measurement process of the single-factor method is simple and direct; specifically, the reciprocal of energy consumption intensity is directly taken as the energy efficiency. However, this method ignores the energy efficiency and industrial technology differences between economies. Therefore, this study uses the input-orientated, scale reward invariant, super-efficient model to measure the total factor energy efficiency of cities. Energy, labor, and capital are input factors; GDP is the desired output; and industrial wastewater, sulfur dioxide, and soot and dust emissions are the non-desired outputs. The total factor energy efficiency is computed as follows:

$$\begin{gathered} \min \mathit{Energy}\_\mathit{efficiency}={\sum }_{i=1}^m\frac{\overline{x}}{x_{ik}}/\frac{1}{r_1+r_2}({\sum }_{s=1}^{r_1}\frac{{\overline{y}}^d}{{\overline{y}}_{sk}^d}+{\sum }_{q=1}^{r_2}\frac{{\overline{y}}^u}{{\overline{y}}_{qk}^u}), \\ \mathrm{s}.\mathrm{t}.\left\{\begin{array}{c}\sum _{j=1,\ne k}^n{x}_{ij}{\lambda }_j\le \overline{x,}\\ \sum _{j=1,\ne k}^n{y}_{sj}^d{\lambda }_j\ge {\overline{y}}^d,\\ \sum _{j=1,\ne k}^n{y}_{qj}^u{\lambda }_j\ge {\overline{y}}^u,\\ x_k\le \overline{x},\\ y_k^d\ge {\overline{y}}^d,\\ y_k^u\le {\overline{y}}^u,\\ {\lambda }_j\ge 0,\\ i=1,2,\dots ,m,\\ j=1,2,\dots ,n,\\ s=1,2,\dots ,r_1,\\ q=1,2,\dots ,r_2,\end{array}\right. \end{gathered}$$

(2)

where, the production system is assumed to have n decision-making units, each of which has m inputs, $r_1$ desired outputs, and $r_2$ non-desired outputs. x, $y^d$, and $y^u$ are the elements in the corresponding input, desired output, and non-desired output matrices, respectively. λ is the proportion of combinations of n decision units in the reconstruction of a valid combination of decision units with respect to the selected decision unit. It is not only the weight vector of the jth decision-making unit, but also the weight coefficient of the output and input indices. The subscript k denotes the evaluated decision unit. $x_{ij}$ is the i-th input of the j-th prefecture-level city, and $y_{sj}$ is the s-th output of the j-th prefecture-level city.

4.3.2. Explanatory Variable

The air-quality ecological compensation policy (AQEC). The policy covers the whole province and has the same effectiveness and constraints on all cities within the province. Hence, prefecture-level cities within the provinces of Henan, Hubei, Anhui, and Shandong, where the air-quality ecological compensation policy is implemented, are selected as the treatment group. This results in 61 cities. Meanwhile, the cities in the other provinces not subject to policy are selected as the control group; this group has 221 cities.

4.3.3. Control Variable

Following Wen et al. [45] and Wang et al. [46], this study incorporates the following control variables: ① The number of urban employees (Lnstaff), measured by the average number of urban workers. Labor and energy are substitutes for each other in the production process of products. Consequently, a higher proportion of employees will reduce energy inputs, and thus, improve energy efficiency. ② The level of economic development (Lnpgdp), measured by per capita GDP. A high level of economic development correlates with increased social productivity and resource allocation efficiency; consequently, energy efficiency will be high in more economically developed cities. ③ The wage level (Lnwage), measured by the average wage of employees. Usually, the wage level is positively correlated with total factor energy efficiency. ④ Innovation capacity (Lninvent), measured by the number of invention patents registered by city. ⑤ Population size (Lnpeople), measured by the total population of the city.

4.4. Data Sources

After excluding Laiwu City, Chaohu City, and other newly established prefecture-level cities due to administrative division adjustments, the balanced panel data of 282 prefecture-level cities in China from 2004 to 2022 are selected. The Chinese city-level data are obtained from the China Urban Statistical Yearbook, the China Regional Economic Statistical Yearbook, and the China Energy Statistical Yearbook. The numbers of invention and green invention patent grants are from the Chinese Research Data Services Platform.

Table 1. Descriptive statistical characteristics of variables.

Variable	Variable Definition	Number of Cities	Obs	Mean	Std. Dev.	Min	Max
Energy_efficiency	Pilot cities for air-quality ecological compensation policy	61	1159	0.4993	0.1553	0.1702	1.2188
	Non-air-quality ecological compensation policy pilot cities	221	4199	0.4772	0.1788	0.0880	1.5943
	First batch of pilot cities for air-quality ecological compensation policy	16	304	0.5264	0.1473	0.2003	1.0531
	Second batch of pilot cities for air-quality ecological compensation policy	12	228	0.5151	0.1787	0.1809	1.0795
	Third batch of pilot cities for air-quality ecological compensation policy	17	323	0.4911	0.1575	0.2063	1.2188
	Fourth batch of pilot cities for air-quality ecological compensation policy	16	304	0.4692	0.1350	0.1702	0.8211
	Full sample		5358	0.4820	0.1742	0.0880	1.5943
AQEC	Ecological compensation pilot policy (1 for carrying out ecological compensation pilot, 0 otherwise)		5358	0.0819	0.2743	0.0000	1.0000
Lnstaff	Urban incumbency (natural logarithm of the average number of urban workers on the job)		5358	3.4357	0.8314	0.3075	6.6490
Lnpgdp	Level of economic development (natural logarithm of GDP per capita)		5358	10.4070	0.8081	4.5951	12.5130
Lnwage	Wage level (natural logarithm of the average wage of employees)		5358	10.5991	0.6426	8.7335	12.2922
Lninvent	R&D innovation capacity (natural logarithm of the number of patents granted for inventions)		5358	4.3341	2.0305	0.0000	11.2797
Lnpeople	Population size (natural logarithm of total urban population)		5358	5.8693	0.6976	2.8190	8.1366

shows the descriptive statistics of the main variables. The mean total factor energy efficiency in the treatment group exceeds that in the control group. In particular, the first batch of pilot cities show stronger effects. Compared to the control group cities, the differences in total factor energy efficiency between the treatment group cities are less prominent. This suggests that the policy mitigates the free-rider phenomenon and enhances the initiative of cities in terms of air pollution management.

5. Empirical Analysis

5.1. Main Regression

In the benchmark regression, this study uses a time-varying DID model to examine the policy’s effectiveness in improving total factor energy efficiency. The results are shown in

Table 2. Reference regression results.

Variable	(1)	(2)	(3)	(4)	(5)	(6)
AQEC	0.0349 *** (0.0063)	0.0350 *** (0.0063)	0.0249 *** (0.0060)	0.0256 *** (0.0060)	0.0213 *** (0.0061)	0.0171 *** (0.0060)
Lnstaff		−0.0049 (0.0067)	−0.0221 *** (0.0065)	−0.0264 *** (0.0067)	−0.0359 *** (0.0069)	−0.0600 *** (0.0071)
Lnpgdp			0.1372 *** (0.0067)	0.1438 *** (0.0073)	0.1414 *** (0.0073)	0.1528 *** (0.0072)
Lnwage				−0.0331 ** (0.0147)	−0.0325 ** (0.0147)	−0.0408 *** (0.0145)
Lninvent					0.0144 *** (0.0025)	0.0093 *** (0.0025)
Lnpeople						0.2338 *** (0.0196)
City effect	Yes	Yes	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes	Yes	Yes
Cons	0.2561 *** (0.0055)	0.2718 *** (0.0223)	−0.9465 *** (0.0629)	−0.6795 *** (0.1344)	−0.6692 *** (0.1340)	−1.9626 *** (0.1711)
N	5358	5358	5358	5358	5358	5358
R2	0.5275	0.5276	0.5642	0.5646	0.5674	0.5792

Note: ***, **, and * indicate significant at the 1%, 5%, and 10% levels, respectively.

. Year and city fixed effects are controlled for to avoid the problem of omitted variable bias to some extent. Column (1) shows that when controlling only for the city and year fixed effects, and with no control variables, the coefficient AQEC is positive at the 1% significance level. The coefficient is still positive at the 1% significance level after successively incorporating the control variables in columns (6), indicating that the policy’s implementation promotes the improvement in urban total factor energy efficiency. The treatment group sample outperforms the control group sample in terms of total factor energy efficiency during the sample window period, with an average improvement of 1.71 percent, according to the regression coefficient of the core explanatory variable, which is around 0.0171. The impact of using resources, taxes, and land reclamation deposits as compulsory ecological compensation is not pronounced compared to air-quality ecological compensation, resulting in only a 0.7 percent increase in energy efficiency.

5.2. Parallel Trend and Dynamic Test

The parallel trends assumption must hold to employ the DID approach. This study follows Jacobson et al. [47] to test the parallel trends assumption using the event study method and whether the policy has a lagged effect. The model is shown below.

$$\mathit{Energy}\_{\mathit{efficiency}}_{\mathit{it}}={\beta }_0+{\beta }_1{\left({AQEC}_{it}\right)}^{-3}+{\beta }_2{\left({AQEC}_{it}\right)}^{-2}+\dots +{\beta }_8{\left({AQEC}_{it}\right)}^4 +\delta X_{\mathit{it}}+{\mu }_i+{\gamma }_{\mathrm{t}}+{\epsilon }_{\mathit{it}},$$

(3)

where ${\left({AQEC}_{it}\right)}^{\pm n}$ denotes the effect of implementing the air-quality ecological compensation policy. $\beta$_k is the policy’s average treatment effect on the total factor energy efficiency. Other variables are consistent with those in the benchmark model. The results of the parallel trend test are shown in Figure 1. Before the policy’s implementation, no significant differences are observed in the development trend of total factor energy efficiency between the treatment and control groups. Thus, the parallel trends assumption holds. The results imply that in the first year after the policy’s implementation, the coefficient of ${\left({AQEC}_{it}\right)}^1$ is significantly positive. The policy’s effect on total factor energy efficiency appears only after its implementation in pilot cities; no time lag is observed.

5.3. Placebo Test

To mitigate the influence of potential random events on the selection of the treatment group, this study uses a placebo test. The principle of the placebo test is to randomize the dummy treatment and control groups several times in the full sample, and continuously repeat the baseline regression to test the robustness of the original findings. The sampling is repeated 500 times among all 282 sample cities. Each sample randomizes 61 prefecture-level cities as the dummy treatment group, and the remaining 221 prefecture-level cities as the dummy control group, resulting in 500 estimated coefficients of AQEC. The kernel density plot shows that the sampled estimated coefficients show a standard normal distribution clustered around zero. The mean value of the AQEC coefficient is 0.0003628, which is close to zero and much smaller than the baseline regression coefficient of 0.0171 (shown by the vertical line in Figure 2). Thus, the effect of the air-quality ecological compensation policy on the improvement of total factor energy efficiency is not causally related to potentially unknown events.

5.4. Robustness Tests

5.4.1. Propensity Score Matching Difference-in-Differences

In the time-varying DID estimation using a large sample, distinguishing the individual differences between the pilot and non-pilot cities prior to the policy’s implementation is difficult. As such, it may be difficult to rule out that some unobserved characteristic influenced the selection of pilot cities. To mitigate selection bias, this study uses the propensity score matching method to create matched treatment and control groups that are indistinguishable from one another before treatment. Then, the DID estimation is performed on these matched groups. Again, in the second column of

Table 3. Various robustness treatments.

Variable	PSM-DID	Consider the First Three Batches of Pilot Cities	First-Stage Regression (AQEC)	Second-Stage Regression (Energy_efficiency)
AQEC	0.0167 *** (0.0059)	0.0214 *** (0.0066)		0.4814 ** (0.2332)
Greenbelt			−0.0040 *** (0.0013)
Controls	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes
Cons	−1.9146 *** (0.1785)	−1.9789 *** (0.1705)	−1.9732 ** (0.8661)	−1.8354 ** (0.7237)
N	5255	5358	5358	5358
R2	0.5899	0.5794	0.4485	0.3883
RKF test			13.93

Note: ***, **, and * indicate significant at the 1%, 5%, and 10% levels, respectively.

, the results of the PSM-DID model show that the air-quality ecological compensation policy still significantly promotes urban total factor energy efficiency improvements. Thus, our conclusion remains robust after controlling for selectivity bias.

5.4.2. Considering Only the First Three Batches of Pilot Cities

The policy was implemented in the fourth batch of pilot cities in 2018, resulting in a relatively short study period for this batch. This may mean that certain effects, which may take longer to observe, cannot be identified. To mitigate this concern here, this study considers only the first three batches of pilot cities. The results in column (3) of Table 3 show that the coefficients are still significantly positive, and the policy significantly improves urban total factor energy efficiency in pilot cities relative to non-pilot cities.

5.4.3. Instrumental Variable Approach

The policy’s impact on urban energy efficiency may be influenced by endogeneity problems, resulting in inconsistent and biased estimation results. To accurately identify the policy’s net effect, an instrumental variable is used with the help of the two-stage least square (2SLS) method. The exogenous instrumental variable selected is urban green land rate for the following reasons: Firstly, the urban green land rate is likely to be highly correlated with urban thermal environment, carbon storage, and atmospheric pollution. Therefore, urban green land rate meets the correlation conditions for instrumental variables. Secondly, urban green land belongs to the natural environment and should be exogenous to the growth of urban total factor energy efficiency. Its impact on energy efficiency can only be realized through the air-quality ecological compensation policy. Specifically, a unique influence path exists: “the lower the urban green land rate → the greater the concentration of air pollution → the greater the demand for the policy → the greater the intensity of environmental regulations → the more likely it is to implement the policy → total factor energy efficiency enhancement”. The results are shown in column (4) and (5) of Table 3. In the first-stage results, the coefficient of urban green land rate is significantly negative. In the second-stage regression, the coefficient of AQEC is significant, and the direction of ecological compensation on energy efficiency is consistent with that of the baseline regression. Thus, the main results hold even after accounting for the endogeneity issues brought on by selection bias, reverse causation, and missing variables.

5.4.4. Exclusion of Disruptive Policies

Considering that many similar or related policies between cities are implemented simultaneously or may have spillover effects, the influence of other policies on the main effects should be considered. This study considers other contemporaneous policies such as the emissions trading pilot policy [48], the new energy demonstration city pilot policy [49], the low-carbon city pilot policy [50], the carbon emissions trading pilot policy [51], and the environmental information disclosure pilot policy [52]. Specifically, the five types of related policies are set as new treatment variables in a DID setup. They are included in the baseline regression as interaction terms of the policy and policy time variables, respectively, in an attempt to strip out the effects of the contemporaneous policies on total factor energy efficiency.

Table 4. Exclusion of disruptive policies.

Variable	(1)	(2)	(3)	(4)	(5)	(6)
AQEC	0.0158 *** (0.0060)	0.0166 *** (0.0060)	0.0180 *** (0.0060)	0.0151 ** (0.0060)	0.0167 *** (0.0060)	0.0148 ** (0.0060)
Emissions trading pilot policy	0.0280 *** (0.0060)					0.0209 *** (0.0061)
New energy demonstration city pilot		0.0149 ** (0.0060)				0.0109* (0.0060)
Low-carbon city pilot policy			0.0299 *** (0.0047)			0.0263 *** (0.0049)
Carbon emissions trading pilot policy				0.0235 *** (0.0075)		0.0161 ** (0.0078)
Environmental information disclosure					0.0457 *** (0.0062)	0.0410 *** (0.0062)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes	Yes	Yes
Cons	−1.9797 *** (0.1708)	−1.9421 *** (0.1712)	−1.9516 *** (0.1705)	−1.8983 *** (0.1722)	−2.1047 *** (0.1713)	−2.0343 *** (0.1718)
N	5358	5358	5358	5358	5358	5358
R2	0.5810	0.5797	0.5825	0.5800	0.5837	0.5884

Note: ***, **, and * indicate significant at the 1%, 5%, and 10% levels, respectively.

reports the results after excluding policy interference. Columns (1) through (5) report the results obtained after separately including the treatment variable DID of other policies in the baseline regression. Column (6) reports the result obtained after simultaneously including the treatment variable DID of other policies in the baseline regression. The magnitude of the coefficient on AQEC and the direction of its impact remain unchanged after excluding potential shocks from disruptive policies.

5.4.5. Extended Analysis for Machine Learning

Although traditional causal inference models are mostly used in studies to assess policy effects, many limitations are present in the application of these models. For example, the SCM can construct a virtual control group that meets the parallel trend, but it requires that the disposal group does not have an “extreme value” characteristic. Further, it is only applicable to the case of “one-to-many”. Meanwhile, PSM has great subjectivity in the selection of matching variables. To make up for the shortcomings of traditional models, many scholars are considering machine learning applications in causal inference, such as dual machine learning [53,54,55]. This study can overcome multicollinearity problems and avoid the problem of the curse of dimensionality brought about by the redundancy of control variables by using the dual machine learning model. The sample segmentation ratios are 1:3, 1:4, 1:5, and 1:6, and the lasso machine learning algorithm was used to predict the primary and auxiliary regressions. The results are shown in columns (1) to (4) of

Table 5. Extended analysis results for machine learning.

Variables	(1)	(2)	(3)	(4)
	Energy_efficiency	Energy_efficiency	Energy_efficiency	Energy_efficiency
	Kfolds = 4	Kfolds = 5	Kfolds = 6	Kfolds = 7
AQEC	0.0192 *** (0.0055)	0.0200 *** (0.0055)	0.0180 *** (0.0055)	0.0205 *** (0.0053)
Controls	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes
Cons	−0.0002 (0.0020)	0.0001 (0.0020)	0.0000 (0.0020)	0.0000 (0.0020)
N	5358	5358	5358	5358

Note: ***, **, and * indicate significant at the 1%, 5%, and 10% levels, respectively.

, respectively. The coefficient of the core explanatory variable is positive at the 1% significance level. Clearly, the use of the dual machine learning model does not change the main conclusion.

5.5. Additional Analysis

The heterogeneity among cities by economic output and resource allocation efficiency may lead to significant differences in the policy’s impact on total factor energy efficiency and the relationship may be non-linear. With reference to Tilov et al. [56], this study uses a panel quantile regression model to estimate the marginal effect of air-quality ecological compensation under various total factor energy efficiencies. The quantile regression results are shown in Figure 3. The estimated coefficients of air-quality ecological compensation overall first increase and then decrease, with the increase in total factor energy efficiency, which has obvious structural characteristics. This may be due to the fact that cities in the lower quartile face more serious environmental problems and need to increase their energy efficiency more urgently. At the same time, the positive effect of air-quality ecological compensation on energy efficiency starts to decrease in cities in the higher quartiles, which are more mature in terms of technology and production structure.

6. Mechanism Analysis and Heterogeneity Analysis

6.1. Mechanism Analysis

Based on the literature review, this study posits that the policy enhances energy efficiency by promoting industrial structure advancement and green technological innovation. The industrial structure advanced index and natural logarithm of the number of green invention patents granted are selected as the proxies of these two mechanisms, respectively. The following model is constructed following Yang et al. [57]:

$${\mathit{Moderator}}_{\mathit{it}}={\beta }_0+{\beta }_1{\mathit{AQEC}}_{\mathit{it}}+{\beta }_2{X}_{\mathit{it}}+{\mu }_{\mathrm{i}}+{\gamma }_{\mathrm{t}}+{\epsilon }_{\mathit{it}},$$

(4)

$$\mathit{Energy}\_{\mathit{efficiency}}_{\mathit{it}}={\beta }_0+{\beta }_1{\mathit{AQEC}}_{\mathit{it}}\times {\mathit{Moderator}}_{\mathit{it}}+{\beta }_2{\mathit{X}}_{\mathit{it}}+{\mu }_{\mathrm{i}}+{\gamma }_{\mathrm{t}}+{\epsilon }_{\mathit{it}},$$

(5)

where Energy_efficiency_it is the total factor energy efficiency calculated using the Super-SBM model, and Moderator is the moderating variable of ecological compensation affecting energy efficiency. This study focuses on the direction, coefficient magnitude, and significance of the interaction term AQEC×Moderator.

6.1.1. Industrial Structure Advancement

The index of industrial structure advancement is constructed to measure the city’s industrial structural advancement. The formula is: IndustStru = ∑Indust_i × i (1 ≤ i ≤ 3), where IndustStru indicates an advanced industrial structure, and Indust_i denotes the proportion of the output value of the i industry to the total output value. The mechanism analysis results are reported in

Table 6. Mechanism regression.

	(1) Industrial Structure Advancement	(2) Total Factor Energy Efficiency	(3) Green Technology Innovation	(4) Total Factor Energy Efficiency
AQEC	0.0387 *** (0.0030)		0.1578 *** (0.0339)
AQEC × Moderator		0.0081 *** (0.0025)		0.0078 *** (0.0016)
Controls	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes
Cons	1.7001 *** (0.0865)	−1.9561 *** (0.1711)	−1.1372 (0.9680)	−1.9287 *** (0.1708)
N	5358	5358	5358	5358
R2	0.7355	0.5794	0.8224	0.5805

Note: ***, **, and * indicate significant at the 1%, 5%, and 10% levels, respectively.

show that the regression coefficient of the interaction term is 0.0081 at the 1% level. Thus, urban industrial structure advancement significantly moderates the relationship between total factor energy efficiency and air-quality ecological compensation policy. Under the effect of air-quality ecological compensation policy, the industrial structure gradually changes from traditional and resource-intensive industries to tertiary and technology-intensive industries. Production factors reorganization accelerates the withdrawal of pollution-intensive enterprises, and transformation and upgrading of regional industrial structure, thereby achieving rational resource allocation. The optimization and restructuring of the industrial structure are important prerequisites for cities to change the traditional extensive management mode and realize the improvement of total factor energy efficiency.

6.1.2. Green Technology Innovation

Next, this study tests whether the policy is conducive to promoting green technological innovation, and thus, energy efficiency. Specifically, a green technological innovation index is constructed using the logarithm of one plus the number of patents granted for green inventions. The regression results show that the policy’s interaction term with green technological innovation has a significant positive effect on total factor energy efficiency. Thus, green technological innovation acts as a major mechanism of the air-quality ecological compensation policy acting on energy efficiency. Improving the green technology innovation capacity contributes to the transformation of the urban economic development model, reduces resource consumption, realizes green and sustainable development, and further improves urban total factor energy efficiency.

6.2. Heterogeneity Analysis

6.2.1. Heterogeneity of Urban Industrial Characterization

Using the National Old Industrial Base Adjustment and Transformation Plan (2013–2022) published by the National Development and Reform Commission, we divide the sample into old industrial bases (OIBs) and non-old industrial bases (NOIBs). The results in

Table 7. Heterogeneity analysis.

	OIB	NOIB	Scities	Mcities	Lcities
AQEC	0.0227 *** (0.0069)	0.0094 (0.0082)	0.0196 ** (0.0078)	0.0017 (0.0105)	−0.0058 (0.0212)
Controls	Yes	Yes	Yes	Yes	Yes
City effect	Yes	Yes	Yes	Yes	Yes
Year effect	Yes	Yes	Yes	Yes	Yes
Cons	−2.3459 *** (0.2014)	−1.5926*** (0.2473)	−1.9584 *** (0.1940)	−0.8280 * (0.4366)	−0.9972 (0.8683)
N	1786	3572	3762	1330	266
R2	0.7550	0.5425	0.5681	0.6152	0.7893

Note: ***, **, and * indicate significant at the 1%, 5%, and 10% levels, respectively.

and Figure 4 show that the pilot of the air-quality ecological compensation policy causes an approximate 2.27 percent increase in the old industrial bases’ total factor energy efficiency, while the effect is not significant for non-old industrial bases. This may be explained by the fact that the old industrial bases are rich in industrial resources, and heavy industry and secondary industry are developing more rapidly. However, old industrial bases have also suffered from the resource curse and Dutch disease. Furthermore, the market development level and technological innovation in the old industrial bases is insufficient. Consequently, the air-quality ecological compensation policy is more likely to inject new momentum and vitality into the green and low-carbon transformation of old industrial bases and improve the total factor energy efficiency. In contrast, non-old industrial bases have a more developed economy and an intensive development model. This provides fertile ground for encouraging cleaner production technologies enterprises and discouraging low energy efficiency enterprises. Therefore, the original energy efficiency of non-old industrial bases is relatively high. The policy does not significantly affect energy efficiency in non-old industrial bases in the short term.

6.2.2. Heterogeneity of City Size

To analyze the heterogeneous effects by different city size classes, this study classifies the sample into small cities (Scities), medium cities (Mcities), and large cities (Lcities) based on the Circular on Classification Standards of City Size issued by the State Council of China in 2014. The results in Table 7 and Figure 5 show that air-quality ecological compensation led to a 1.96 percent increase in total factor energy efficiency in small cities. Regression coefficients in the medium-city subsample and large-city subsample are not significant. Medium and large cities are usually the centers and forerunners of national or regional economic development strategies, exhibiting factor agglomeration effects and scale agglomeration effects. These cities exhibit high economies of scale, are in the leading stages of industrialization, have relatively mature factor markets, and have strong innovation and entrepreneurial activities. As such, their urban total factor energy efficiency tends to be relatively high, and further improvements faces greater bottlenecks and constraints. In contrast, small cities have a less balanced economic structure, are structurally overly dependent on primary and secondary industries, and do not have a unique advantage in high-tech industries. In addition, most smaller cities do not have a strict and unified system of environmental regulations, leading to high industrial pollution intensity and poor energy use efficiency. Therefore, in small cities, air-quality ecological compensation has a greater potential to improve total factor energy efficiency.

7. Conclusions and Policy Recommendations

7.1. Conclusions

The implementation of air-quality ecological compensation has been an important part of China’s efforts in realizing the United Nations 2030 Agenda for Sustainable Development. Using the time-varying DID model, this study utilizes balanced panel data of 282 Chinese cities from 2004 to 2022 to examine the impact of the air-quality ecological compensation on urban total factor energy efficiency and its mechanism of action. The conclusions are described below: Firstly, the implementation of air-quality ecological compensation significantly improves total factor energy efficiency. Specifically, air-quality ecological compensation results in a 1.71 percent increase in total factor energy efficiency for the pilot cities. The policy limits crude economic growth and excessive emissions by heavy polluting enterprises, which may improve energy use efficiency. The benchmark regression results remain valid even under several robustness tests such as PSM-DID, considering the first three pilot city batches, instrumental variables approach, excluding disruptive policies, and dual machine learning. Secondly, mechanism analysis shows that the policy improves total factor energy efficiency through the structural and technological effects. The policy accelerates urban industrial structure advancement, and promotes the transformation and upgrading, or withdrawal of pollution-intensive enterprises. Meanwhile, the policy increases the intensity of urban green technology innovation, which helps improve cleaner production techniques and technologies. Lastly, air-quality ecological compensation greatly improves the total factor energy efficiency of old industrial base cities and small cities, and has a statistically insignificant impact on non-old industrial base cities, large cities, and medium cities. The government needs to accurately grasp the characteristics of the economic development of cities, and flexibly adjust the support for the construction of air-quality ecological compensation policies according to local conditions.

Some empirical results are consistent with Gan et al. [25] and Lin et al. [30], who indirectly found that the “punitive” approach of air-quality ecological compensation restricts the rough economic growth and forces energy-consuming industries to improve energy use efficiency. Specifically, Gan et al. found that a 1% reduction in energy efficiency as a result of the implementation of an air-quality ecological compensation policy would reduce concentrations of PM_2.5, PM₁₀, SO₂, and NO₂ by 1.037 $\mathsf{\mu }$g/m³, 2.192 $\mathsf{\mu }$g/m³, 1.028$\mathsf{\mu }$g/m³, and 0.699$\mathsf{\mu }$g/m³. However, they use energy intensity to measure energy use efficiency based on a single-factor perspective, neglecting the influence of other input factors in favor of concentrating on the relationship between economic output and energy. In contrast, from the total factor perspective, we incorporate indicators like labor, capital, energy, GDP, and pollutant emissions into the same framework, and use the SBM model to quantify total factor energy efficiency. In addition, the conclusions of Gan et al. and Lin et al. do not take into account the mechanism of action by which air-quality ecological compensation influences energy use efficiency. We fill these gaps and reveal the promotion effect and influence path of air-quality ecological compensation on total factor energy efficiency.

7.2. Policy Recommendations

Expanding the scope of the pilot scheme: The frequent transregional air pollution occurrences have become a pain point that cannot be ignored for high-quality economic development. To build a regional and even national-level ecological compensation mechanism is to put the main responsibility of environmental protection governance into practice and to coordinate the promotion of green and low-carbon development. Furthermore, policymakers should gradually cultivate a market-based air-quality ecological compensation trading system and try to use innovative compensation methods, such as technological compensation and industrial compensation, to enhance the policy impacts.

Setting appropriate assessment cycles for air-quality ecological compensation: On the one hand, the assessment cycle of ecological compensation is lengthy but the assessment frequency is low. This grossly disregards the fact that the concentration of pollutants in the region will decrease and renders the policy’s implementation a formality. On the other hand, short assessment cycles and frequent assessments increase the implementation costs and put local governments under more strain. Thus, it is necessary to set up an appropriate assessment cycle.

Expanding access to financing for ecological compensation. The essence of the air-quality ecological compensation policy is the within-province inter-municipal horizontal financial transfer payment. This increases the financial pressure on some cities to a certain extent. As the capital coefficient of air-quality ecological compensation increases annually, this imbalance will also expand. To this end, local governments can properly absorb the capital of financial institutions and enterprises to build a diversified financing mechanism for social and market compensation.

7.3. Limitations

This study has some limitations. Firstly, the explanatory power of the study’s findings may be relatively weak in other regions. The empirical evidence and policy recommendations in this study are presented based on the Chinese region, which lacks direct applicability to developed or other developing countries. Therefore, future research could expand and validate our findings by using samples from other countries or regions. Secondly, it is difficult to completely eliminate interference from related events. Although this study considers the impacts of policies like the new energy demonstration city pilot policy and the low-carbon city pilot policy, competing explanations of other policies cannot be completely ruled out. Future research could take other policies into account to further explore the net effect of air-quality ecological compensation. Lastly, the exploration of the mechanisms in this study may still be insufficiently comprehensive. Besides industrial structure advancement and green technology innovation, there may be other potential mechanisms through which air-quality ecological compensation affects total factor energy efficiency. Deepening the research on mechanisms is encouraged and necessary.