2.1 Institutional background

As a catalyst for industrial development, traditional industrial parks have made important contributions to China's economic development, but they have caused serious pollution to the environment (Lu et al., Reference Lu, Wang and Zhu2019; Wu et al., Reference Wu, Nie, Wang and Li2023). This issue is proposed to be addressed through their transformation into EIPs which follow the green development concept and seek to achieve the win-win balance between economic growth and environmental protection. Enterprises within EIPs are required to achieve green and low-carbon transformation and development, and control pollution emission load within a permissible range. EIPs have gradually established an infinitely reusable, recyclable, and renewable industrial ecosystem, and the air pollutants and emissions in EIPs have been significantly reduced (Chertow, Reference Chertow2000; Qian et al., Reference Qian, Xu, Sun and Zhou2022).

What is more, enterprises in EIPs can reduce haze pollution by applying cleaner production technologies, optimizing energy structure, and taking other emission reduction measures. On the one hand, EIPs provide green technical support to enterprises in the parks by introducing high-level talent and teams. On the other hand, strict environmental regulation implemented in EIPs forces enterprises to increase R&D expenditures for clean energy technologies (Wu et al., Reference Wu, Nie, Wang and Li2023), thus reducing air pollutant emissions. Moreover, EIPs promote green technological progress by encouraging enterprises to optimize energy structure and use non-fossil and renewable energy, which is conducive to reducing haze pollution (Qian et al., Reference Qian, Xu, Sun and Zhou2022). Besides, high technology and low pollutant emissions enterprises attracted by EIPs can also promote the green transformation of traditional enterprises, which is conducive to achieving industrial structure upgrading and emission reduction. Detailed operational aspects of the EIPs are described in the online appendix of this paper.

Figure A1 in the online appendix shows the number of newly added EIPs every year in China up to 2016. During the sample period of this paper, in 2008, the former State Environmental Protection Administration named Suzhou Industrial Park, Suzhou National Hi-Tech District and Tianjin Economic-Technological Development Area for the first time as the National Eco-industrial Demonstration Parks. Since then, the construction of EIPs has been in full swing. By 2016, 48 EIPs had been approved for operating, and 45 EIPs had been approved to start construction.

The geographic distribution of EIPs in China is shown in figure 1. There are 31, 11, and 6 EIPs in eastern, central and western China, respectively. Because the number of EIPs in the eastern region significantly exceeds that of other regions, one may be concerned that the location of EIPs may be endogenous. To alleviate the endogeneity caused by the location of EIPs, we employ the propensity score matching (PSM) method and instrument variable (IV) strategy to strengthen causal inference.

Figure 1.

Distribution of EIPs in China.

2.2 Theoretical analysis

EIP policy aims to promote regional eco-industrial development, and it offers an innovative pathway to realizing energy saving, emission reduction and environmental protection. Under the rigorous and standardized requirements, the EIPs effectively drive technological innovation, industrial structure upgrading, and environmental regulation, which in turn affects haze pollution. The influence mechanisms through which EIPs affect haze pollution are shown in figure A2 in the online appendix.

First, the construction of EIPs reduces haze pollution through technological innovation in two main ways. (1) EIP projects have accelerated technological innovation. According to the National Eco-Industrial Demonstration Park Management Measures, the government must offer abundant financial and fiscal support for EIPs to introduce advanced innovation talents and cleaner technologies, which can help EIPs improve their technological innovation level and reduce air pollutant emissions. Meanwhile, after being certified as EIPs, the parks introduce many favorable policies to encourage the development of high-tech enterprises, thus attracting a tremendous amount of well-established high-tech companies at home and abroad. For example, there are about 500 production-oriented enterprises and most of them are high-polluting enterprises in Nanchang National High-tech Industrial Development Zone (NNHIDZ). With the construction of environmental protection platforms and advanced environmental protection technologies, NNHIDZ significantly improves local air quality. According to the above analysis, we believe that EIPs can reduce haze pollution through promoting technological innovation. In addition, according to the existing literature, location-oriented policies such as EIPs can significantly improve the innovation level. For example, Wu et al. (Reference Wu, Nie, Wang and Li2023) regard EIPs as a quasi-natural experiment and used a difference in differences (DID) model to examine the impact of EIPs on urban green innovation. They found that the establishment of EIPs significantly increases the number of green patent applications.

And (2), green technological innovation is conducive to reducing environmental pollution. The specific manifestations are front-end prevention and back-end governance. In terms of front-end prevention, green technological innovation drives the emergence of digital environmental protection projects such as intelligent pollutant monitoring systems and early environment warning systems, which effectively prevent and control pollution sources. Moreover, the continuous development and application of production, energy-saving and environmental protection technology enables enterprises to substantially improve energy efficiency and reduce pollutant emissions (Chen et al., Reference Chen, Paudel and Zheng2022). In terms of back-end governance, green technological innovation promotes the improvement of pollution control facilities and strengthens the reuse of pollutants, thereby improving environmental quality. For example, Tianjin municipality launched a special fund of $100 million to support green development in the Tianjin Economic-Technological Development Area. The fund provided subsidies for appropriate green innovation projects including the comprehensive utilization of resources, the construction of distributed photovoltaic power stations, cleaner production, and the construction and operation of hydrogen refueling stations. This measure effectively incentivizes enterprises in parks to promote technological innovation in the production process, increase capital investment in pollution control and reduce greenhouse gas emissions.Footnote ¹

Second, the construction of EIPs reduces haze pollution through industrial structure upgrading, again in two main ways. (1) EIPs have positively contributed to industrial transformation and upgrading. In EIPs, there are two ways to promote industrial upgrading, namely, supporting the green transformation of energy-intensive industries and developing emerging green industries. The green transformation of traditional high-polluting and energy-intensive industries is a gradual process that is likely to span a decade or more. In this process, the negative impacts of green transformation on haze pollution can come into effect within a short period of time, and they will increase significantly as time goes on (Tong et al., Reference Tong, Wang and Xu2020). What is more, EIPs introduce favorable policies for high-tech industries with green technological innovation, encouraging the green and low-carbon transformation of traditional pollution-intensive industries (Li et al., Reference Li, Pan, Kim, Linn and Chiang2015). Therefore, EIPs can reduce haze pollution within two years through industrial upgrading. (2) The industrial upgrading promotes pollution prevention and control (Grossman and Krueger, Reference Grossman and Krueger1995). The secondary industry is considered as the major contributor to environmental pollution. Therefore, it is bound to reduce pollutant emissions when the proportion of the secondary industry drops and the proportion of the emerging industry rises (Cole et al., Reference Cole, Elliott and Shimamoto2005; Li and Li, Reference Li and Li2022). Moreover, compared with polluting industries, the development model of emerging industries is sustainable and can produce a demonstration effect, forcing the polluting industries to reduce pollutant emissions.

Third, the construction of EIPs reduces haze pollution through environmental regulation for two reasons. (1) The establishment of EIPs can strengthen urban environmental regulations. Compared with traditional industrial parks, EIPs are required to meet many compulsory indicators, including elastic coefficient of comprehensive energy consumption, application ratio of renewable energy and so on. This means that EIPs impose higher requirements for cleaner production and environmental protection. (Zeng et al, Reference Zeng, Cheng, Shi and Luetkenhorst2021; Qian et al., Reference Qian, Xu, Sun and Zhou2022; Wu et al., Reference Wu, Nie, Wang and Li2023). Therefore, EIPs can produce a strong demonstration effect on the local area and improve the level of environmental regulations. (2) The strict environmental regulation is conducive to reducing haze pollution. EIPs have stricter environmental regulations, and set the performance indicators such as industrial technology level, resource and energy utilization efficiency, pollutant emissions, and economic benefits. The enterprises in the parks are required to meet these environmental goals. EIPs also implement compulsory audits of cleaner production, improving clean production levels of enterprises. What is more, in EIPs, there are specific standards for the total discharge quota of key pollutants. In addition, EIPs require enterprises to disclose environmental information, including waste, by-products and residual energy, which may be useful for other enterprises' production. During this process, other enterprises can recycle and reuse the production materials and products as long as possible, thus promoting the establishment of a circular economy model and reducing haze pollution.

Based on the above theoretical analysis, this paper proposes the following hypotheses:

Hypothesis 1 (H1): The construction of EIPs can reduce urban haze pollution.

Hypothesis 2 (H2): The construction of EIPs reduces haze pollution through technological innovation, industrial upgrading, and environmental regulations.

3.1 Model

Based on the panel data of 276 prefecture-level cities in China from 2007 to 2018, this paper regards the establishment of EIPs as a quasi-natural experiment. Because the start-up time of EIPs is not the same, with reference to Bertrand et al. (Reference Bertrand, Duflo and Mullainathan2004), a DID model is constructed as follows:

(1)\begin{equation}\ln PM{2.5_{it}} = {\alpha _0} + {\alpha _1}EI{P_{it}} + \sum \gamma _{j}\ X_{it} + {\mu _i} + {\nu _t} + {\varepsilon _{it}}, \end{equation}

where i and t denote city and year, respectively. The dependent variable is the natural logarithm of annual PM2.5. EIP_it is a dummy variable, and if city i started the construction of the EIP in year t, then EIP_it equals 1 since year t, and otherwise it equals 0. ${\alpha _1}\;$ is the coefficient that we focus on to test whether EIPs have a reduction effect on annual PM2.5 concentrations. ${X_{it}}$ is a set of control variables. ${\mu _i}$ and ${\nu _t}$ refer to the city and year fixed effects respectively. ${\varepsilon _{it}}$ is the error term.

3.2 Variables and data sources

1. (1) Dependent variable. The dependent variable is urban haze pollution levels measured as PM2.5 concentrations. PM2.5, as one of the primary air pollutants, is believed to reflect the overall level of haze pollution effectively (Burkhardt et al., Reference Burkhardt, Bayham, Wilson, Carter, Jesse, O'Dell, Ford, Fischer and Pierce2019; Heyes and Zhu, Reference Heyes and Zhu2019; Chen et al., Reference Chen, Chen, Xie, Mueller and Davis2023b). The existing literature mainly obtains the annual average PM2.5 concentration data from Columbia University (Lin et al., Reference Lin, Fu, Jiang, Hu, Dong, Huang and Zhao2014; Fu et al., Reference Fu, Jiang, Lin, Liu and Wang2015). Instead of this data set, the data used in this paper come from the Atmospheric Composition Analysis Group at Dalhousie University in Halifax. ArcGIS software is used to extract and transform the data from the global data set published by this institution, and then we obtain the urban annual average PM2.5 concentration for China. Although the data from both Columbia University and Dalhousie University are collected with ground-based tracking stations and National Aeronautics and Space Administration satellites, the data set used in this paper has the following advantages. First, the data set published by Dalhousie University is available for a much longer time period than data released by Columbia University. Second, the data set in the paper achieves higher accuracy. The reason is that, before 2014, the data set produced by Columbia University only contains three-year average PM2.5 concentrations rather than annual average observations.

2. (2) Independent variable. The independent variable is EIP policy, an indicator to reveal whether the EIPs have been approved for establishment in the city. EIP equals 1 if the city was approved for the park construction, or one of the industrial parks in the city is approved for an EIP officially; EIP equals 0 otherwise. The list of EIPs comes from the website of the Ministry of Ecology and Environment of the People's Republic of China. It is worth noting that both the operational parks and the parks under construction are included in the treatment group. In further analysis, we investigate heterogeneity in treatment effects across the two groups of parks. Because our sample period is from 2007 to 2018, we exclude the cities whose establishment year of EIP is 2001 and 2003. Furthermore, some cities have more than one EIP which were established in different years; in this case, we take the year when the first EIP was established as the year that the EIP policy came into effect. Finally, there are a total of 276 cities, including 48 in the treatment group and 228 in the control group.

3. (3) Control variables. To ensure that the results are not driven by omitted variables, we control some factors that may influence haze pollution and use lag-one period as controls. Specifically, the control variables are defined as follows. The level of economic development, ED, is measured by GDP per capita. Population size, PS, is measured as the total population of the city at the end of the year. Industrial structure, IS, is measured as the proportion of the output value of the tertiary industry to GDP. The level of secondary industry development, SD, is measured as the proportion of the output value of the secondary industry to GDP. The level of infrastructure development, ID, is defined as the length of road construction. The level of opening-up to the outside world, OP, is measured by the ratio of FDI and GDP.Footnote ²

3.3 Descriptive statistics

The sample of this paper consists of 2007–2018 data on 276 prefecture-level cities in China. The descriptive statistics of each variable are presented in table 1.

Table 1.

Descriptive statistics

4.1 Parallel trend test

The prerequisite for the DID is to meet the parallel trend test, that is, the haze pollution levels in the treatment group and the control group should have the same changing trend before the implementation of EIP policy. Referring to Beck et al. (Reference Beck, Levine and Levkov2010), this paper tests the parallel trend based on equation (2),

(2)\begin{equation}\ln PM{2.5_{it}} = {\beta _0} + {\beta _k}\mathop \sum \limits_{k ={-} 7}^7 EIP_{it}^k + \sum \gamma _j\ X_{it} + {\mu _i} + {\nu _t} + {\varepsilon _{it}},\end{equation}

where $\ln PM{2.5_{it}}$ represents the haze pollution level of city i in year t. k denotes the k-th year of the establishment of the EIPs. $EIP_{it}^k$ represents a series of dummy variables for the relative time of establishment of EIPs (7 to 2 years before the EIPs were established, in the year when the EIPs are established, and 1 to 7 years after the EIPs are established). In particular, we define $EIP_{it}^{ - 7}$ as equal to 1 if the establishment year is for 7 or more years before the establishment of the EIP, otherwise 0. $EIP_{it}^7$ is equal to 1 if the observation is for seven or more years after the establishment of the EIP, otherwise 0. The coefficient ${\beta _k}\;$captures the differential effect of the construction of EIPs on urban haze pollution. Other parameters and variables are defined in accordance with model (1).

The test results are presented in figure 2. It shows no significant difference between the haze pollution levels of the treatment group and the control group before the construction of EIPs, which provides evidence for satisfying the parallel trend assumption and the effectiveness of the DID model. The effect of haze pollution reduction follows parallel trends until one year after the establishment of EIPs (time ‘1’ on the horizontal axis); thus, there is a lagged effect of EIPs on air quality improvement. The reason is that the traditional business philosophy that prioritizes production over environmental protection is still rooted at the beginning of policy implementation. In addition, it takes time for EIPs to affect urban haze pollution. In the second year after their establishment, the EIPs exert the effect of decreasing haze pollution levels gradually and the effect remains stable until the fourth year. In the following years, the policy effect gradually strengthens, illustrating that the effect of EIPs on haze pollution in the long run is still significant.

Figure 2.

Parallel test.

Notes: All effects are relative to the one year before the policy went into effect.

4.2 Baseline results

After confirming that the parallel trend hypothesis is satisfied, the paper adopts the DID approach to evaluate the effect of the establishment of EIPs on haze pollution based on the two-way fixed effects model. Table 2 presents the results. We control the city and time fixed effects in all columns. Column (1) does not include any control variables and the estimated coefficient is significantly negative at the 1 per cent level. Columns (2)–(7) gradually cover the full set of control variables. Almost all of the estimates are still significant at the 1 per cent level, indicating that the construction of EIPs can effectively reduce haze pollution and improve urban air quality. The main reason is that the EIPs, which are designed with the principle of reducing, reusing and recycling, improve the urban air quality.

Table 2

The effect of establishing EIPs on haze pollution

Notes: The level of economic development (ED), measured by GDP per capita (logarithm). Population size (PS), measured as the total population of the city at the end of the year (logarithm). Industrial structure (IS), measured by the proportion of the output value of the tertiary industry to GDP. The level of secondary industry development (SD), measured by the proportion of the output value of the secondary industry to GDP. The level of infrastructure development (ID), defined as the length of road construction (logarithm). The level of opening-up to the outside world (OP), measured by the ratio of FDI and GDP. All control variables are lagged by one period. Standard errors in parentheses.

4.3 Instrumental variable

Another potential pitfall is the non-randomness of the distribution of EIPs across cities. We tested for the common trends in the DID estimation to rule out the possibility that confounding events might bias our estimates, and this section provides additional support using an IV approach. We use predicted EIPs as the IV for real establishment of EIPs. Specifically, first, following the forecasting method proposed by Lipscomb et al. (Reference Lipscomb, Mobarak and Barham2013), we predicted the likelihood of the placement of EIPs based on predetermined local geographic attributes over each of the five-year periods (see online appendix table A1). Second, referring to Lu et al. (Reference Lu, Sun and Wu2023a), we further adopted a jackknife method and estimated the probability of having an EIP. This ‘leave-out'estimate partly alleviates the concern about a weak instrument problem as it is less likely to violate the exclusion restriction criterion. Then, we ranked the cities based on the estimated probability, as was done by Duflo and Pande (Reference Duflo and Pande2007), and generated a 0/1 variable on the predicted EIP, for the top N_j cities if the province launched N_j EIPs during that period. The predicted EIP serves as an IV for the actual EIP. Table 3 shows the results of IV estimation based on the two-stage least square (2SLS) strategy. Column (1) shows the first-stage result, which indicates the IV is a very strong positive predictor of establishing EIPs with a Cragg-Donald Wald F-statistic larger than 10, eliminating the problem of weak IVs. The second-stage result is shown in column (2); although the absolute estimation is much larger than those in the traditional DID estimation, the 2SLS estimate confirms the negative relationship of EIPs and haze pollution.

Table 3

IV estimation

Notes: Standard errors in parentheses.

5.1 Robustness test

5.1.1 PSM-DID

Cities with EIPs tend to invest more in environmental protection and the public may have stronger environmental awareness to embrace a better ecological environment. For this reason, referring to Heckman et al. (Reference Heckman, Ichimura and Todd1998), this paper uses the PSM method to match the treatment group and the control group cities within a common value range, and then estimate the model with the DID method. Our matching approach begins with selecting matching variables. The variables include the level of economic development, population size, industrial structure, level of secondary industry development, level of infrastructure development, and level of opening-up to the outside world. Next, we use the logit model parameter estimation to obtain the propensity score. Finally, the observations sharing a similar propensity score are matched. This paper adopts three approaches to find a matched control city for each of the treatment cities, namely, nearest neighbor matching, radius matching and kernel matching. We then perform DID estimation after matching.Footnote ³ The results are reported in table 4, and indicate that EIPs have significantly improved urban air quality.

Table 4

PSM-DID estimation results

Notes: Standard errors in parentheses.

5.2 Heterogeneous analysis

5.3 Mechanism analysis

In this section, we examine intrinsic influence mechanisms through which the EIP affects urban haze pollution. We check the connection between EIP and green technological innovation (GTI), industrial structure upgrading (ISU) and environmental regulation (ER). The specific regression model is as follows:

(3)\begin{equation}{M_{it}} = {\lambda _0} + {\lambda _1}EI{P_{it}} + \sum \theta _j\ X_{it} + {\mu _i} + {\nu _t} + {\varpi _{it}},\end{equation}

where M represents the mechanism variable The other variables are consistent with equation (1). First, we use the number of patents of the city as a measure of technology innovation. Second, referring to Gan et al. (Reference Gan, Zheng and Yu2011), we use two measures to calculate the industrial structure upgrading: rationalization of industrial structure (RIS) and advancement of industrial structure (AIS). We obtain industrial structure using equations (4) and (5) where ${Y_{i,t}}/{Y_t}$ is a ratio with ${Y_{i,t}}$ being output of industrial sector i and ${Y_t}$ being total output. ${L_{i,t}}/{L_t}$ refers to the ratio of employment in the industrial sectors i to all sectors. Third, referring to Dong et al. (Reference Dong, He, Wang and Wang2020) and Hou et al. (Reference Hou, Yu and Fei2023), this paper measures the intensity of environmental regulations by calculating the composite pollution index of gaseous emissions, including industrial wastewater, soot, and sulfur dioxide emissions. Specifically, we first calculate the emission intensity of each pollutant and normalize it. Then, we use the adjustment weight, the ratio of each pollutant emission intensity to its average emission intensity, to reflect the difference in pollution among the samples. Last, we calculate the environmental regulation intensity of each city in the panel data by summing up the product of adjustment weight and each pollutant emission intensity.

(4)\begin{gather}RI{S_{i,t}} = 1 - \frac{1}{3}\mathop \sum \limits_{i = 1}^3 \left|{\frac{{{Y_{i,t}}}}{{{Y_t}}}} \right. - \left. {\frac{{{L_{i,t}}}}{{{L_t}}}} \right|\end{gather}

(5)\begin{gather}AI{S_{i,t}} = \frac{{{Y_{3,t}}}}{{{Y_{2,t}}}}\end{gather}

The regression results are reported in table 5. The result in column (1) shows the coefficient of EIP is positive and significant at the 5 per cent level, which indicates that the establishment of EIPs can improve urban green technological innovation. Columns (2)–(3) show that the regression coefficient of EIP is significantly positive, implying that the establishment of EIPs can also promote industrial structure upgrading. From the result in column (4) of table 5, we find that the construction of EIPs has a significant positive effect on improving the level of urban environmental regulation. In sum, the above results confirm that the implementation of EIP policy is driving air quality improvement by promoting technological innovation, upgrading industrial structure and strengthening urban environmental regulation.

Table 5

Influence mechanisms

Notes: GTI represents green technology innovation, measured by the number of green patents of the city. AIS represents advancement of industrial structure. RIS represents rationalization of industrial structure. ER represents environmental regulation. Standard errors in parentheses.