Re-Examination of the Relationship between Industrial Agglomeration and Haze Pollution: From the Perspective of the Spatial Moderating Effect of Environmental Regulation

Abstract

This paper uses panel data from 284 Chinese cities from 2004 to 2020 and employs a dynamic spatial panel Durbin model to re-examine the relationship between industrial agglomeration, environmental regulation, and haze pollution. It further adopts a dynamic spatial moderation effect model to explore the spatial regulatory mechanism of environmental regulation. The results show that both local and neighboring industrial agglomeration have a significant “inverted U-shaped” relationship with local haze pollution, and the scale cumulative optimization effect can only be effectively played after the industrial agglomeration level of the locality and neighboring areas exceeds the inflection point. Local environmental regulation significantly inhibits haze pollution, while neighboring environmental regulation plays a promoting role. The moderating effect of environmental regulation on the relationship between industrial agglomeration and haze pollution shows spatial heterogeneity in the local and neighboring areas. Local environmental regulation has a “U-shaped” non-linear moderating effect while neighboring environmental regulation has a positive linear moderating effect. Therefore, the government should pay attention to the joint effort and coordinated advancement of industrial agglomeration and environmental regulation to further reduce urban haze pollution and enhance urban air quality.

1. Introduction

The Chinese government has announced Carbon peaking and carbon neutrality goals, which demand the improvement of environmental pollution management and the promotion of economic transformation towards high-quality and sustainable development. Since atmospheric pollutants and greenhouse gases share a “common source and process” relationship, China must undertake haze control while reducing carbon emissions [1,2]. The management of atmospheric pollution centered on PM2.5 is not only an important task in the construction of China’s ecological civilization but also an inevitable requirement for China’s deep participation in global environmental governance. The report of the 20th National Congress of the Communist Party of China clearly pointed out to “Continuously and deeply commit to the battle of defending blue skies, clear waters, and clean soil, enhance the coordinated control of pollutants, and essentially eliminate severe pollution weather”, showing the central government’s determination to strengthen ecological and environmental protection in all aspects, regions, and processes. According to the Ministry of Ecology and Environment of China, in the past 10 years, the number of heavily polluted days in China has decreased by 92%, and the PM2.5 concentration has decreased by 57%, achieving a tenth consecutive decrease. These changes fully illustrate that China’s air quality has improved significantly, and PM2.5 concentrations have gradually decreased. However, China’s air quality is still far from the standard of the World Health Organisation [3]. According to the “2023 World Air Quality Report” from the IQAir website, the air quality of China ranks relatively low globally among 134 countries and regions. The areas in China with higher PM2.5 concentrations are primarily distributed in North China, Central China, and the southwestern basin, where the PM2.5 levels are still 7 to 10 times the guideline value, and in some areas, they even exceed 10 times the guideline. Therefore, in the key period of the “14th Five-Year Plan” for the construction of Chinese-style modernization, how to effectively prevent and control the worsening of haze pollution and improve the ecological environment quality of residents has become an important issue of concern to the government, scholars, as well as the public.

However, haze pollution is not merely a localized form of pollution; it can spread to neighboring areas through natural factors such as atmospheric circulation, as well as economic activities like industrial transfer, agglomeration, and transportation [4,5]. Industrial agglomeration is an industrial development model that can fully utilize economies of scale and scope, and it can effectively promote regional economic growth. The innovation incentive effect of industrial clusters can also promote the transformation and upgrading of industries and optimize resource allocation, making the cultivation and strengthening of industrial clusters a policy choice for many local governments. However, from the perspective of green development, governments must consider environmental factors, especially the issue of environmental carrying capacity, when formulating industrial policies to promote industrial agglomeration [6]. Taking the Beijing-Tianjin-Hebei region and its surrounding areas as an example, the government has implemented strict environmental regulation policies and actively promoted the green and low-carbon transformation of industries. This has led to significant progress in the transformation and upgrading of industries, in addition to a substantial improvement in air quality. However, the region still suffers from a heavy industrial structure and heavy chemical industries surrounding the city. As a matter of fact, this region, which accounts for only 7.2% of China’s total land area, is home to 33% of the country’s flat glass, 39% of electrolytic aluminum, 43% of crude steel, 49% of coke, and 60% of the bulk drug industry. Such a phenomenon naturally induces a series of questions. Is industrial agglomeration one of the deep-rooted factors that exacerbate haze pollution? Can stringent environmental regulation policies effectively regulate the relationship between industrial agglomeration and haze pollution? Furthermore, does this regulatory mechanism have spatial spillover effects? The answers to the above questions could provide guidance on how to use environmental regulations in a targeted way, scientifically formulate industrial planning and layout, guide industrial agglomeration, and strive to achieve sustainable development of regional economy, society, and environment. Such guidance is important not only for China but also for developing countries in general, where there are significant differences in the levels of industrial agglomeration and environmental regulation.

2. Literature Review

Currently, research on the interrelationships among industrial agglomeration, environmental regulation, and haze pollution has been relatively abundant. Firstly, the research on the relationship between industrial agglomeration and environmental regulation has formed the following four typical opinions: First, increasing the intensity of environmental regulation can promote industrial agglomeration [7]; Second, increasing the intensity of environmental regulation will inhibit industrial agglomeration [8,9]; Third, the intensity of environmental regulation has no significant impact on industrial agglomeration [10]; Fourth, there is a non-linear characteristic between the two [11,12]. Zhu et al. (2022) [13] conducted a study on China and found that ex-ante environmental regulation had a negative impact on the level of industrial agglomeration, while in-process and ex-post environmental regulation significantly promoted industrial agglomeration, exhibiting spatial heterogeneity at the regional level. The “pollution haven” hypothesis posits that environmental regulations can lead to the agglomeration of different types of industries in various regions, thereby affecting environmental pollution.

Secondly, the existing literature on the relationship between industrial agglomeration and haze pollution has been relatively scarce, and it often focused more on the relationship between economic growth or industrial agglomeration and environmental pollution, suggesting that there is a positive correlation [14,15], a negative correlation [16,17], or an uncertain relationship [18]. Zhang et al. (2015) [19] believed that the impact of agglomeration on various pollutants varies, and due to the complex sources of haze pollution and its stronger cross-regional diffusion characteristics compared to water pollution, it may not be appropriate to generalize the relationship between industrial agglomeration and environmental pollution and directly apply it to the relationship with haze pollution. In recent years, some scholars have begun to focus on the relationship between industrial agglomeration and haze pollution, but a unified conclusion has not yet been formed. There are mainly the following three perspectives: First, there is a high positive correlation between industrial agglomeration and haze pollution [20,21]; Second, industrial agglomeration, industrial collaborative agglomeration, and specialized agglomeration are conducive to mitigating haze pollution [6,22,23]; Third, there is a threshold effect, a “U”-shaped relationship, or an “N”-shaped relationship between the two [24,25,26]. However, the relationship between industrial agglomeration and environmental pollution is complex and may involve multiple mechanisms of action, such as the presence of several “moderator variables” that affect the relationship between the two. This paper considers environmental regulation to be one of the important “moderator variables.” Previous empirical studies have mostly relied on linear models, neglecting some important moderating variables, which is likely to cause research bias.

Thirdly, there has been no consensus in the academic community regarding the relationship between environmental regulation and haze pollution. The majority of scholars believe that an increase in the level of environmental regulation can suppress haze pollution [27,28]. Some scholars, from the perspectives of the displacement effect and the informal economy, argued that environmental regulation policies cannot reduce haze pollution [29,30]. Some scholars perceive that the relationship between the two is still uncertain and may exhibit non-linear characteristics. Huang (2016) [31] examined the non-linear relationship between environmental regulation and haze pollution from the perspective of the shadow economy. Kuang et al. (2023) [32] observed that the threshold effects of formal and informal environmental regulations on haze pollution exhibit significant regional differences.

Finally, some scholars have analyzed the relationship between environmental regulation, industrial agglomeration, and air pollution. Chen et al. (2024) [33] used a static spatial Durbin model to verify the impact of environmental regulation and industrial agglomeration on air pollution. According to their study, local environmental regulation shows an inverted U-shaped relationship with local air pollution, while neighboring environmental regulation shows a U-shaped relationship. They believed that stricter environmental regulation in local cities may lead to a reduction in manufacturing agglomeration and an increase in productive service industry agglomeration, thereby reducing air pollution.

To summarize, conclusions of existing research on the relationships between industrial agglomeration, environmental regulation, and haze pollution exhibit considerable inconsistency due to differences in the selection of research samples, the application of econometric models, the choice of indicators for industrial agglomeration, and the methods of calculation. More importantly, previous studies have often treated environmental regulation as a dependent or control variable to examine its impact on haze or environmental pollution, neglecting the potential moderating role that environmental regulation might play. Therefore, the relationship between industrial agglomeration and haze pollution requires further examination. In this regard, the marginal contributions of this paper are mainly as follows: First, it studies industrial agglomeration, environmental regulation, and haze pollution within the same framework, fully considering the spatial correlation among the three and re-examining their intrinsic relationships using a dynamic spatial Durbin model; Second, it accurately identifies and tests the moderating role of environmental regulation and further extends it to the spatial dimension, exploring the moderating effects of local and neighboring environmental regulations from a more comprehensive perspective, which is a deepening and expansion of previous related research; Third, our choice of measuring industrial agglomeration from the perspective of output density not only takes into account the spatial bias caused by differences in smaller geographical units but also focuses on the load of industrial activities per unit of geographical area, thereby further enriching the methods for measuring industrial agglomeration indicators.

3. Mechanistic Analysis and Research Hypothesis

3.1. The Spatiotemporal Scale Cumulative Optimization Mechanism of Industrial Agglomeration on Haze Pollution

During the initial stage of industrial agglomeration, where scale expansion is predominant, the local economic growth model is primarily based on the traditional extensive growth, neglecting environmental protection and gradually forming a pattern of pollution first and management later [22]. Under the dominance of the extensive growth philosophy, the scale effect of industrial agglomeration outweighs the technological and structural effects. Enterprises focus more on scale expansion, which may trigger cutthroat competition in the production factor market and product market. This can result in the negative environmental externalities of industrial agglomeration. Moreover, due to the strong radiation capacity of the agglomeration area, production is overly concentrated, and the crowding effect is prominent. The exhaust gases and other environmental pollutants produced are all gathered around the agglomeration area, exacerbating local haze pollution [34]. As the quality of industrial agglomeration continues to improve and its structure is continuously optimized, the technological and structural effects gradually surpass the scale effect. At this time, the capacity for pollution control and green total factor productivity are also continuously improving [35,36]. After the economy shifts from high-speed growth to high-quality development, enterprises continuously explore their own transformation and upgrading path to adapt to the new growth model. Especially the investment promotion service management departments of urban industrial agglomeration areas will pay more attention to attracting high-level production factors and raising entry thresholds, promoting industrial agglomeration to develop towards specialized division of labor, technological cooperation, and industrial chain collaboration to meet the requirements of sustainable development [37]. Moreover, as the agglomeration area becomes more mature, the technological cooperation and resource sharing among enterprises becomes even deeper. The environmental protection standards of enterprises and agglomeration areas in the production and operation process will also become the threshold conditions for the agglomeration area to retain and attract higher-level production factors. The sharing of environmental protection technology and facilities will also be strengthened accordingly. All the above factors promote the reduction of environmental pollution levels during the process of industrial agglomeration, tending to maturity and advancement. The emissions of various pollutants and harmful gases continue to decrease, which is conducive to controlling local haze pollution. In addition, due to the spatial attraction and agglomeration effects of industrial agglomeration itself, it has a spatial spillover effect; that is, the level of industrial agglomeration in neighboring areas may also affect the concentration of local haze pollution. This depends on the type of industry that is agglomerated in the adjacent areas as well as the stage of industrial agglomeration development they are in. If it is a high-pollution and high-energy-consumption type of industrial agglomeration, it will exacerbate local haze pollution and vice versa. Based on the above analysis, the theoretical Hypothesis H1 is proposed as follows.

H1: 

There is an “inverted U-shaped” curve relationship between industrial agglomeration and haze pollution, and the industrial agglomeration in neighboring areas will affect local haze pollution.

3.2. The Competitive Promotion and Screening Mechanism of Environmental Regulation on Haze Pollution

As public concern for environmental quality and haze pollution continues to grow, governments are placing greater emphasis on green economic development. To alleviate haze pollution, governments at all levels have continued to carry out air pollution prevention and control actions, formulated various regulatory measures, and encouraged enterprises to transform and upgrade. They have severely cracked down on and shut down highly polluting and energy-intensive enterprises, achieving significant progress in haze control [38]. However, generally speaking, environmental regulation can increase the production costs of a company by requiring additional investment in environmental protection technology and equipment without changing other production and operation conditions. This may reduce profit margins, especially for polluting enterprises with lower productivity levels and poor pollution control capabilities. These enterprises may choose to exit the market due to a lack of funds, technology, and also management capabilities to cope with environmental regulations. Ultimately, the enterprises that survive in the market are those with technological and financial capabilities and relatively lower costs [39]. Once the competitive promotion and screening mechanism of environmental regulation takes effect, it also helps to alleviate haze pollution. In addition, if neighboring areas increase the intensity of environmental regulation, it may impose higher environmental costs and pollution reduction costs on energy-consuming and polluting industries within the region. Some smaller-scale enterprises with weaker profitability may transfer to areas with looser environmental regulations to avoid high environmental costs. As a result, the influx of such enterprises may increase air pollutants in the region of relocation and exacerbate haze pollution, potentially creating an effect of shifting pollution to neighboring regions. Therefore, the theoretical Hypothesis H2 is proposed.

H2: 

There is a negative correlation between environmental regulation and haze pollution, and the level of environmental regulation in neighboring areas will affect local haze pollution.

3.3. The Spatial Moderating Mechanism of Environmental Regulation on Industrial Agglomeration and Haze Pollution

On one hand, industrial agglomeration may directly reduce the intensity of haze through pathways such as the spillover of pollution control and emission reduction technologies, centralized treatment and regulation of pollutants, specialized division of labor, and cost savings in pollution control. On the other hand, the direction and extent of the impact of industrial agglomeration on haze pollution are likely to be influenced by the intensity of environmental regulation. In other words, environmental regulation will serve as a moderating variable in the influence of industrial agglomeration on haze pollution. Since environmental regulation affects the production and investment behaviors of enterprises, it subsequently influences the location choices of enterprises, leading to the agglomeration of different types of industries in different regions [39]. Areas with lower levels of environmental regulation may attract the agglomeration of highly polluting and energy-intensive industries, thereby exacerbating haze pollution, whereas areas with higher levels of regulation focus on developing low-pollution, low-energy-consuming industries, which is conducive to reducing haze pollution [40]. Furthermore, the level of local or neighboring environmental regulation standards and the adoption of different regulatory measures for different industries can lead to the agglomeration of different industries in local or neighboring areas, thereby affecting the local environment or haze pollution. Therefore, the impact of environmental regulation on the degree of industrial agglomeration in a particular area is a dynamically adjusting process and exhibits spatial correlation. Local or neighboring environmental regulation may play a moderating role in the relationship between industrial agglomeration and haze pollution. That is, as the degree of environmental regulation in the local area or neighboring areas and the types of regulated industries vary, the production costs, technological innovation capabilities, and factor allocation of enterprises will change, thereby affecting the smog pollution through the agglomeration of industries with different energy consumption and pollution levels. Based on the discussion above, the theoretical Hypothesis H3 of this paper is proposed.

H3: 

Environmental regulation plays a moderating role in the relationship between industrial agglomeration and haze pollution, and such a moderating effect has a spatial correlation.

4. Model Setup and Variable Selection

4.1. Dynamic Spatial Panel Model

Existing studies have indicated that industrial agglomeration [41], environmental regulation [28], and haze pollution [42,43] all exhibit strong spatial correlations. Ignoring the spatial spillover effects of these three variables may lead to biased results. Therefore, this paper incorporates the spatial lag or spatial error terms of the aforementioned three variables into the model, constructing the Spatial Autoregressive Model (SAR), Spatial Error Model (SEM), Spatial Autocorrelation Model (SAC), and Spatial Durbin Model (SDM), respectively. The models were first subjected to Wald and Lratio tests, which rejected the null hypotheses, suggesting that the SAR and SEM models were not suitable. Subsequently, the applicability of the SAC and SDM models was assessed based on Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) results. The findings indicated that the AIC and BIC values for the SDM model were lower than those for the SAC model, thus making the SDM model the most appropriate for the scope of this paper.

Considering the potential time-dependent effects, specifically the temporal lag effects of haze pollution [42], as well as the endogeneity issues that may arise from the bidirectional causal relationships between haze pollution and factors such as industrial agglomeration and environmental regulation, a one-period lagged term of haze pollution is introduced into the equation. The following dynamic spatial panel Durbin model is constructed:

$$\begin{array}{cc}& {\mathrm{p}\mathrm{m}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{p}\mathrm{m}}_{\mathrm{i}\mathrm{t}-1}+{\mathsf{\rho }}_1{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{p}\mathrm{m}}_{\mathrm{j}\mathrm{t}}+{\mathsf{\alpha }}_2{\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\rho }}_2{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{j}\mathrm{t}}{+{\mathsf{\alpha }}_3\mathrm{e}\mathrm{r}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\rho }}_3{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{e}\mathrm{r}}_{\mathrm{j}\mathrm{t}}+\\ & {\mathsf{\alpha }}_4{\mathrm{s}\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\rho }}_4{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·\mathrm{s}{\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{j}\mathrm{t}}+\mathsf{\delta }\sum {\mathrm{X}}_{\mathrm{i}\mathrm{t}}+\mathsf{\lambda }{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{X}}_{\mathrm{j}\mathrm{t}}+{\mathsf{\theta }}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}},\end{array}$$

(1)

where ${\mathrm{p}\mathrm{m}}_{\mathrm{i}\mathrm{t}}$ indicates the level of haze pollution, with $\mathrm{i}$ the index for the city, and $\mathrm{t}$ for the time. ${\mathrm{p}\mathrm{m}}_{\mathrm{i}\mathrm{t}-1}$ denotes the intensity of haze pollution with a one-period lag, ${\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}$ is the measure of industrial agglomeration, ${\mathrm{s}\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}$ is the quadratic term of industrial agglomeration, ${\mathrm{e}\mathrm{r}}_{\mathrm{i}\mathrm{t}}$ is the environmental regulation, ${\mathsf{\alpha }}_{0,1,\dots ,4}$ and $\mathsf{\delta }$ are the coefficients to be estimated, ${\mathsf{\rho }}_{1,\dots ,4}$ and $\mathsf{\lambda }$ are the spatial lag coefficients, ${\mathrm{w}}_{\mathrm{i}\mathrm{j}}$ is the spatial weight matrix, ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ is a set of control variables affecting haze pollution, $\mathsf{\theta }$_it is the time fixed effect, ${\mathsf{\mu }}_{\mathrm{i}\mathrm{t}}$ is the city fixed effect and ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ represents the error term that varies with both city and time. To ensure the robustness of the research conclusions and to systematically examine the spatial correlation characteristics of haze pollution, this paper constructs four types of spatial weight matrices ${\mathrm{W}}_{1,\dots ,4}$ that include geographical and economic factors, based on the approach of Shao et al. (2016) [4]. The four types of spatial weight matrices are defined as follows: (1) ${\mathrm{W}}_1$ is the inverse distance weight matrix, for which its matrix element ${\mathrm{w}}_{\mathrm{i}\mathrm{j}}$ represents the inverse of the nearest highway distance between city $\mathrm{i}$ and city $\mathrm{j}$; (2) ${\mathrm{W}}_2$ represents the matrix of geographic economic distance weight, with ${\mathrm{w}}_{\mathrm{i}\mathrm{j}}$ denoting the product of the reciprocal of the nearest highway distance between city $\mathrm{i}$ and city $\mathrm{j}$ and the proportion of the annual average GDP per capita of city $\mathrm{i}$ to the annual average GDP per capita of all cities; (3) ${\mathrm{W}}_3$ is the geographical distance weight matrix, for which the matrix element ${\mathrm{w}}_{\mathrm{i}\mathrm{j}}$ represents the inverse of the nearest railway distance between city $\mathrm{i}$ and city $\mathrm{j}$; (4) ${\mathrm{W}}_4$ is the adjacent spatial weight matrix, for which the matrix element ${\mathrm{w}}_{\mathrm{i}\mathrm{j}}=1$ if two cities are geographically adjacent, otherwise ${\mathrm{w}}_{\mathrm{i}\mathrm{j}}=0$.

4.2. Dynamic Spatial Moderating Effect Model

To test whether environmental regulation acts as a moderating variable, a spatial moderating effect model is further employed for empirical examination. Considering the time lag of haze pollution, the model (1) is augmented with the interaction term between environmental regulation and industrial agglomeration, as well as its spatial lag term, so as to explore the moderating effect of local and neighboring environmental regulation on the relationship between local industrial agglomeration and haze pollution. The model is given by:

$$\begin{array}{cc}& {\mathrm{p}\mathrm{m}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{p}\mathrm{m}}_{\mathrm{i}\mathrm{t}-1}+{\mathsf{\gamma }}_1{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{p}\mathrm{m}}_{\mathrm{j}\mathrm{t}}+{\mathsf{\beta }}_2{\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\gamma }}_2{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·\\ & {\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{j}\mathrm{t}}{+{{\mathsf{\beta }}_3{\mathrm{s}\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}+\mathsf{\gamma }}_3{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{s}\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{j}\mathrm{t}}+{\mathsf{\beta }}_4\mathrm{e}\mathrm{r}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\gamma }}_4{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{e}\mathrm{r}}_{\mathrm{j}\mathrm{t}}+{\mathsf{\beta }}_5{\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}·\\ & {\mathrm{e}\mathrm{r}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\gamma }}_5{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}·{\mathrm{e}\mathrm{r}}_{\mathrm{i}\mathrm{t}}+\mathsf{\tau }\sum {\mathrm{X}}_{\mathrm{i}\mathrm{t}}+\mathsf{\phi }{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{i}\mathrm{j}}·{\mathrm{X}}_{\mathrm{j}\mathrm{t}}+{\mathsf{\theta }}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}\end{array}$$

(2)

In this model, ${\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}\mathrm{t}}·{\mathrm{e}\mathrm{r}}_{\mathrm{i}\mathrm{t}}$ represents the interaction term between industrial agglomeration and environmental regulation, ${\mathsf{\gamma }}_{1,\dots ,5}$ and $\mathsf{\phi }$ are the spatial lag coefficients, ${\mathsf{\beta }}_{0,\dots ,5}$ and $\mathsf{\tau }$ are the coefficients to be estimated, with the remaining letters retaining their meanings. If the estimated values of coefficients ${\mathsf{\beta }}_5$ and ${\mathsf{\gamma }}_5$ are significant, it implies that environmental regulation serves as a moderating variable between industrial agglomeration and haze pollution.

4.3. Selection of Variables

This study utilizes panel data from 284 Chinese cities spanning the years 2004 to 2020. Missing values are supplemented using mean imputation and smoothing methods. All indicators involving price factors are measured in Chinese Yuan (CNY) and adjusted to real value data based on the price index, taking 2003 as the base year.

Explained Variable: Haze Pollution (pm). Scholars have traditionally used a single or several conventional pollutants, such as SO₂ and CO₂, to represent the degree of haze pollution [44]. However, in recent years, attention has increasingly focused on PM2.5, the primary culprit of haze pollution [45]. This paper measures haze pollution using the annual average of urban PM2.5 levels, with data from the Socioeconomic Data and Applications Center of Columbia University.

Key Explanatory Variables: One of the key explanatory variables is industrial agglomeration (agg). Shao et al. (2019) [46] identified the output density as a good indicator to measure the degree of economic or industrial agglomeration in a region, using the ratio of the total non-agricultural output to the total urban administrative area to measure the degree of economic agglomeration. We posit that when measuring industrial agglomeration, emphasis should be placed on the area of industrial and other land uses to accurately reflect the load of industrial activities per unit of geographical area. However, due to data availability, the ratio of the total non-agricultural output (the sum of the added value of the secondary and tertiary industries) to the total urban construction land area is ultimately used to measure industrial agglomeration (agg1). To ensure the robustness of the estimation results, the location quotient is also used to represent the level of industrial agglomeration (agg2), using the formula: ${\mathrm{a}\mathrm{g}\mathrm{g}}_{\mathrm{i}}=({\mathrm{v}\mathrm{a}}_{\mathrm{i}}/{\sum }_{\mathrm{i}=1}^{\mathrm{n}}\mathrm{v}\mathrm{a})/({\mathrm{g}\mathrm{d}\mathrm{p}}_{\mathrm{i}}/{\sum }_{\mathrm{i}=1}^{\mathrm{n}}\mathrm{g}\mathrm{d}\mathrm{p})$, where ${\mathrm{v}\mathrm{a}}_{\mathrm{i}}$ represents the sum of the added value of the secondary and tertiary industries of the $\mathrm{i}$-th city, $\mathrm{v}\mathrm{a}$ represents the total sum of the added value of the secondary and tertiary industries of the sample cities, ${\mathrm{g}\mathrm{d}\mathrm{p}}_{\mathrm{i}}$ represents the GDP of the $\mathrm{i}$-th city, and $\mathrm{g}\mathrm{d}\mathrm{p}$ represents the total GDP of the sample cities. The corresponding quadratic terms for the level of industrial agglomeration are sagg1 and sagg2.

The other key explanatory variable is environmental regulation ($\mathrm{e}\mathrm{r}$). Some scholars have used the measure of operating costs of pollution control facilities [10], while others have measured formal and informal environmental regulation [47]. Drawing on the approach of Zhu et al. (2018) [48], the pollution density method is used to represent the intensity of environmental regulation, which is the average emission intensity of three pollutants: industrial wastewater, industrial sulfur dioxide, and industrial dust, denoted as ${\mathrm{M}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$. The reciprocal value of this value is taken to obtain the intensity of environmental regulation. The calculation formula is as follows: ${\mathrm{R}\mathrm{E}}_{\mathrm{i}\mathrm{t}}=3/{\sum }_{\mathrm{j}=1}^3{\mathrm{M}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$, where ${\mathrm{M}}_{\mathrm{i}\mathrm{j}\mathrm{t}}={\mathrm{P}}_{\mathrm{i}\mathrm{j}\mathrm{t}}/{\mathrm{G}}_{\mathrm{i}\mathrm{t}}$, ${\mathrm{P}}_{\mathrm{i}\mathrm{j}\mathrm{t}}$ is the total emission of the j-th pollutant in city $\mathrm{i}$ in year $\mathrm{t}$; ${\mathrm{G}}_{\mathrm{i}\mathrm{t}}$ is the total industrial output value of city $\mathrm{i}$ in year $\mathrm{t}$; ${\mathrm{R}\mathrm{E}}_{\mathrm{i}\mathrm{t}}$ is the environmental regulation intensity of city $\mathrm{i}$ in year $\mathrm{t}$, the larger value of which indicates stricter the environmental regulation.

Other explanatory variables: There are many factors affecting haze pollution, among which the impact of natural factors such as geography, climate, and meteorology accounts for a large proportion. However, since the impact of these factors is already internalized into the haze pollution data, and the availability of these data is poor and difficult to effectively extract, we focus on the following control variables from an economic development perspective: (1) Population density (den), represented by the number of people per unit area of the city, reflecting the impact of population agglomeration on haze pollution; (2) Per capita income level (lngdp), measured by the natural logarithm of per capita GDP; (3) Industrial structure (ins), described by the proportion of the added value of the secondary industry, including industry and construction, in GDP; (4) R&D investment (t), given by the proportion of expenditure on science and technology in local fiscal budgetary expenditure; (5) Degree of openness to foreign investment (fdi), measured by the actual amount of foreign direct investment utilized by each city, and the average exchange rate of each year is used to convert US dollars into RMB in the calculation; (6) Energy consumption structure (es), represented by the proportion of coal consumption in the total energy consumption; (7) Transportation (veh), represented by the number of motor vehicles in the city. The descriptive statistics of the main variables are shown in

Table 1. Descriptive Statistics of Variables.

Variable	Definition	Observations	Mean	Standard Deviation	Minimum	Median	Maximum
pm	Haze pollution level	4828	59.451	15.274	20.736	61.955	90.856
agg1	Level of industrial agglomeration	4828	14.472	7.521	2.735	13.414	46.438
agg2	Level of industrial agglomeration	4828	1.307	0.166	0.109	1.031	1.809
er	Level of environmental regulation	4828	1.321	1.389	0.049	0.789	8.786
den	Population density	4828	6.421	0.389	5.408	6.547	7.273
lngdp	Per capita income level	4828	10.280	0.676	8.113	10.332	11.768
ins	Industrial structure	4828	0.522	0.110	0.056	0.530	1.021
t	R&D investment	4828	0.076	0.317	0.001	0.013	2.957
fdi	Degree of openness to foreign investment	4828	11.636	31.211	0.002	3.214	273.4
es	Energy consumption structure	4828	0.710	0.176	0.057	0.733	0.992
veh	Level of transportation	4828	69.171	90.071	2.520	39.750	590.900

.

5. Empirical Results and Analysis

5.1. Estimation Results of the Dynamic Spatial Panel Model

Before the model estimation, it is necessary to test for the existence of spatial spillover effects in haze pollution by conducting global and local spatial correlation index tests. The results show that under the four spatial weight matrices, Moran’s I index is greater than 0, and Geary’s C index is less than 1, all of which are significant at the 1% level, indicating that haze pollution exhibits positive spatial correlation characteristics of high-high and low-low type clustering. Then, the residuals based on the Ordinary Least Squares (OLS) estimation results are tested, and the results show that the LMlag, Robust LMlag, LMerr, and Robust LMerr statistics of the four spatial weight matrices are all significant at the 1% level, thereby indicating that haze pollution has significant spatial correlation. Therefore, it is necessary to use the spatial panel model for the empirical study in this paper. Finally, the Hausman test results suggest that the SDM model is suitable for estimation using fixed effects, and further estimation is conducted using the bias-corrected maximum likelihood estimation (MLE) method, with the results shown in

Table 2. Estimation Results of the Dynamic Spatial Durbin Model ¹.

Variables	I: Industrial Agglomeration agg1				II: Industrial Agglomeration agg2
	W1	W2	W3	W4	W1	W2	W3	W4
L.pm	0.359 ***	0.398 ***	0.857 ***	0.511 ***	0.837 ***	0.879 ***	0.882 ***	2.119 ***
	(10.78)	(11.92)	(23.67)	(14.19)	(30.08)	(31.93)	(28.81)	(72.02)
agg	2.876 ***	2.895 ***	0.242 ***	1.547 ***	2.725 ***	2.836 ***	3.722 ***	3.559 ***
	(18.14)	(18.27)	(5.59)	(10.72)	(16.13)	(18.04)	(2.74)	(17.22)
sagg	−0.060 ***	−0.061 ***	−0.006 *	−0.042 ***	−1.733 ***	−0.860 ***	−0.985 **	−1.516 ***
	(17.76)	(−17.90)	(−1.75)	(−13.32)	(−17.21)	(−18.86)	(−2.37)	(−18.26)
er	−6.214 ***	−6.308 ***	−0.097 **	−3.023 ***	−2.030 ***	−1.944 ***	−0.049 ***	−1.844 ***
	(−17.67)	(−17.94)	(−2.08)	(−9.64)	(−5.86)	(−5.61)	(−3.15)	(−5.90)
W.agg	4.949 ***	5.456 ***	0.301 ***	1.994 ***	5.445 ***	4.355 ***	5.752**	3.789 ***
	(11.08)	(12.42)	(10.31)	(17.88)	(17.83)	(15.16)	(2.16)	(12.69)
W.sagg	−0.025 ***	−0.038 **	−0.031 **	−0.191 **	−0.289 ***	−0.064 ***	−0.474 **	−0.801 ***
	(−2.71)	(−2.41)	(−2.14)	(−2.43)	(−2.74)	(−4.57)	(−2.05)	(−8.41)
W.er	4.903 ***	2.980 ***	2.384 *	2.572 ***	2.772 ***	2.964 ***	0.028 ***	2.771 ***
	(64.80)	(64.47)	(1.86)	(25.79)	(20.21)	(18.36)	(3.01)	(29.11)
Spatial rho	0.284 ***	0.373 ***	0.701 ***	0.236 ***	2.520 ***	2.513 ***	0.741 ***	0.611 ***
	(3.44)	(4.54)	(3.31)	(6.42)	(3.60)	(3.85)	(3.65)	(7.10)
Control variables	YES	YES	YES	YES	YES	YES	YES	YES
City fixed	YES	YES	YES	YES	YES	YES	YES	YES
Time fixed	YES	YES	YES	YES	YES	YES	YES	YES
R2	0.049	0.046	0.677	0.201	0.002	0.002	0.706	0.004
Log-L	−1190.507	−1190.961	−2247.292	−1332.051	−1259.173	−1253.559	−2487.265	−1448.204

¹ The asterisks (***, **, *) indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses are t-statistics. Log-L represents the Log-likelihood statistic. Due to space limitations, the coefficients of control variables and their spatial lagged terms are not reported and are available upon request.

. The signs of the estimated coefficients under the two measurement methods of industrial agglomeration and the four spatial weight matrices are consistent, with only differences in significance and magnitude. Due to space limitations, the analysis is based on the estimation results of industrial agglomeration agg1 and the spatial weight matrix ${\mathrm{W}}_1$.

First, we analyze the impact of industrial agglomeration on haze pollution and its spatial spillover effects. The estimated coefficients of the first-order term and its exogenous interaction term of industrial agglomeration are significantly positive at the 1% significance level, while the estimated coefficients of the second-order term and its exogenous interaction term are significantly negative, with a significant spatial spillover effect. This indicates that there is an “inverted U-shaped” relationship between the levels of local and neighboring industrial agglomeration and local haze pollution. Following the method of Shao et al. (2019) [46], the inflection point of the first-order term of industrial agglomeration is calculated to be 18.445. Before the inflection point, the focus of the development of the agglomeration area is to expand scale. To attract more enterprises to agglomerate, environmental and industrial management departments of local governments often adopt low-standard threshold policies for corporate pollution emissions. The initial dispersion of industries and mixed spatial distribution is inevitable, and the public facilities and means for pollution control and emission reduction are not fully equipped, and the advantages of joint emission reduction have not been effectively utilized. This brings many loopholes in management, subjectively and objectively, and the lower levels of local and neighboring industrial agglomeration will exacerbate the occurrence of local haze pollution. After crossing the inflection point, as the scale of the agglomeration area, the composition of the industry, and the connections between the industrial chain tend to be stable, and the intrinsic demand for high-level development of the agglomeration area is generated, various public supporting facilities and services for pollution control and emission reduction are increasingly mature. In order to seek increasing returns to scale, the willingness of enterprises to cooperate will also increase accordingly, and the agglomeration area faces the continuous improvement of environmental protection standards, forcing the reform of traditional low-standard management systems and methods. At this stage, the positive learning effect and technological spillover effect of local and neighboring industrial agglomeration are fully exerted, which is conducive to reducing the emission of pollutants and mitigating haze pollution.

Second, we examine the influence of environmental regulation on haze pollution and its spatial spillover effects. On the one hand, the estimated coefficient of environmental regulation is significantly negative, indicating that local environmental regulation can suppress haze pollution. This is the result of the dual choice of enterprises to cope with changes in the external policy environment and the market. A variety of environmental regulation policies and measures can encourage enterprises to accelerate their research and development and use environmentally friendly production technologies to achieve the dual goals of profit acquisition and environmental protection. They also force some enterprises to reselect their locations to avoid environmental constraints, so environmental regulation is a factor in promoting the reduction of haze pollution. On the other hand, the coefficient of the spatial lag term of environmental regulation is significantly positive, indicating that neighboring environmental regulation has a positive impact on local haze pollution. The neighboring areas increase the intensity of environmental regulation, which has the effect of driving out enterprises that are difficult to adapt to this regulatory change. Of course, this kind of migration to avoid environmental regulation faces a trade-off problem between compliance costs and migration costs. Transferring to neighboring areas with relatively loose environmental regulation is undoubtedly a choice to reduce migration costs, but this kind of regional migration will greatly reduce the effectiveness of environmental regulation due to the particularity of haze. For a city, the increase in the intensity of its environmental regulation may actually exacerbate air pollution in neighboring areas and will not significantly weaken local haze pollution.

5.2. Robustness Test

To ensure the reliability of the conclusions, we conduct robustness tests from four main aspects: replacement of the main explanatory variable indicators, establishment of different spatial weight matrices, elimination of outlier tests, and correction of endogeneity issues within the model.

Firstly, we replaced the main explanatory variable indicators. On the one hand, we used the location entropy method to measure industrial agglomeration (agg2) and re-estimated the dynamic SDM model, with the results presented in Table 2. On the other hand, following the method of Chen et al. (2024) [33], we utilize data on industrial sulfur dioxide and industrial dust emissions, reconstruct the environmental regulation intensity indicator (er2) using a linear weighting method, and then re-estimate the model, with the results presented in

Table 3. Robustness check with the substitution of environmental regulation indicators ¹.

Variables	I: Industrial Agglomeration agg1				II: Industrial Agglomeration agg2
	W1	W2	W3	W4	W1	W2	W3	W4
agg	0.762 ***	1.563 ***	2.482 ***	1.274 ***	1.772 **	2.559 ***	1.454 ***	2.514 ***
	(4.08)	(6.45)	(8.26)	(7.73)	(2.28)	(7.01)	(5.03)	(3.03)
sagg	−0.021 ***	−0.083 ***	−0.834 ***	−0.878 ***	−0.026 *	−0.021 **	−0.602 **	−0.163
	(−6.05)	(−9.64)	(−8.40)	(−3.15)	(−1.65)	(−2.33)	(−2.43)	(−1.36)
er2	−1.227 ***	−1.694 **	−2.10 3 **	−4.810 ***	−0.514	−0.442 **	−1.005 ***	−1.269 ***
	(−2.88)	(−2.14)	(−2.32)	(−8.30)	(−1.11)	(−2.18)	(−3.49)	(−3.52)
L.pm	0.813 ***	1.755 ***	1.596 ***	2.784 ***	0.534 ***	0.322 ***	0.414 ***	0.370 ***
	(8.70)	(3.96)	(4.77)	(6.96)	(6.78)	(5.34)	(5.28)	(5.16)
W.agg	2.290 ***	4.369 ***	1.650 ***	1.491 ***	2.697 **	3.705 ***	3.702 ***	5.230 **
	(3.50)	(6.54)	(9.65)	(5.76)	(2.32)	(5.04)	(4.86)	(2.17)
W.sagg	−4.439 ***	−5.823 ***	−3.094 ***	−6.006 ***	−1.888 ***	−2.159 **	−4.012 ***	−2.847 ***
	(−6.47)	(−5.04)	(−3.01)	(−4.08)	(−3.65)	(−2.06)	(−3.78)	(−3.22)
W.er	6.886 ***	2.597 ***	6.585 ***	7.805 ***	2.895 ***	0.351 ***	2.199 ***	0.406 **
	(7.10)	(4.50)	(4.11)	(8.81)	(3.08)	(5.56)	(4.01)	(2.12)
Spatial rho	0.738 ***	0.519 ***	1.870 ***	1.292 ***	0.029 ***	0.508 ***	0.182 ***	0.365 ***
	(7.10)	(7.15)	(4.74)	(7.97)	(3.17)	(10.52)	(7.96)	(6.46)
Control variables	YES	YES	YES	YES	YES	YES	YES	YES
City effect	YES	YES	YES	YES	YES	YES	YES	YES
Time effect	YES	YES	YES	YES	YES	YES	YES	YES
R2	0.182	0.049	0.147	0.303	0.378	0.344	0.533	0.641
Log-L	−2949.271	−2289.801	−2439.906	−2860.406	−2489.344	−2873.834	−2519.556	−2611.042

¹ The asterisks (***, **, *) indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses are t-statistics. Log-L represents the Log-likelihood statistic. Due to space limitations, the coefficients of control variables and their spatial lagged terms are not reported and are available upon request.

. Both tables show that the main conclusions remain largely unchanged, thus indicating the robustness of the conclusions.

Second, we construct different spatial weight matrices. When examining the relationship between industrial agglomeration, environmental regulation, and haze pollution, as well as the spatial spillover effects, four types of spatial weight matrices that include both geographical and economic factors were set up. The estimation results are shown in Table 2. The coefficients and significance levels of the key variables have changed, but the signs have remained unchanged, demonstrating robustness.

Third, we eliminate the outliers in the sample and re-estimate the model. To exclude the potential impact of outliers on the results, we have performed a 1% tail-trimming treatment on all continuous variables. Data below the 1st percentile and above the 99th percentile were replaced with the 1st and 99th percentiles, respectively. The estimation results are shown in

Table 4. Robustness Test After Excluding Outliers ¹.

Variables	I: Industrial Agglomeration agg1				II: Industrial Agglomeration agg2
	W1	W2	W3	W4	W1	W2	W3	W4
agg	1.536 ***	1.535 ***	1.067 ***	0.442 ***	2.619 **	2.976 **	2.746 ***	2.384 ***
	(7.15)	(7.17)	(5.20)	(2.60)	(2.20)	(2.22)	(2.73)	(2.80)
sagg	−0.016 ***	−0.015 ***	−0.016 ***	−0.006	−1.039 **	−1.238 ***	−1.341 **	−1.902
	(−3.24)	(−3.20)	(−3.61)	(−0.28)	(−2.17)	(−2.63)	(−2.40)	(−1.48)
er	−0.449 ***	−0.467 **	−0.332 **	−0.394 *	−0.018 ***	−0.017 **	−0.851 ***	−0.455
	(−3.05)	(−1.99)	(−2.03)	(−1.92)	(−3.03)	(−2.16)	(−3.46)	(−1.00)
L.pm	0.376 ***	0.396 ***	0.558 ***	0.460 ***	0.336 ***	0.535 ***	0.547 ***	0.551 ***
	(8.25)	(8.33)	(9.75)	(8.73)	(43.24)	(43.21)	(44.38)	(46.12)
W.agg	4.825 ***	4.910 ***	2.817 ***	1.127 ***	5.514	4.898 **	5.665 ***	6.082 **
	(8.34)	(8.73)	(11.07)	(9.43)	(0.34)	(2.30)	(6.05)	(2.17)
W.sagg	−4.405 ***	−3.603 ***	−3.973 ***	−4.481 ***	−2.776 ***	−3.176 **	−2.847 ***	−2.240 **
	(−7.82)	(−4.13)	(−7.17)	(−8.48)	(−5.02)	(−1.99)	(−6.32)	(−2.03)
W.er	5.083 **	4.271 **	4.741	0.974 *	5.535 ***	8.369 ***	3.814 ***	1.271 **
	(2.36)	(2.01)	(1.21)	(1.80)	(8.96)	(8.36)	(7.60)	(2.01)
Spatial rho	0.668 ***	0.670 ***	1.046 ***	0.622 ***	0.747 ***	0.758 ***	1.355 ***	0.761 ***
	(8.42)	(8.50)	(4.75)	(9.86)	(11.88)	(12.53)	(7.30)	(24.88)
Control variables	YES	YES	YES	YES	YES	YES	YES	YES
City effect	YES	YES	YES	YES	YES	YES	YES	YES
Time effect	YES	YES	YES	YES	YES	YES	YES	YES
R2	0.053	0.053	0.083	0.369	0.045	0.045	0.003	0.214
Log-L	−1485.991	−1484.792	−1472.298	−1382.901	−1555.930	−1559.603	−1555.583	−1471.368

¹ The asterisks (***, **, *) indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses are t-statistics. Log-L represents the Log-likelihood statistic. Due to space limitations, the coefficients of control variables and their spatial lagged terms are not reported and are available upon request.

. The signs of the coefficients for all variables have not changed, supporting the conclusions of the previous text.

Last, we have thoroughly addressed the issue of endogeneity present in the model. The potential endogeneity issues in this paper may stem from the omission of major explanatory variables and the possible reverse causality between industrial agglomeration, environmental regulation, and haze pollution. Moreover, Equation (1) includes the temporal and spatial lagged terms of haze pollution. Elhorst (2003) [49] suggests that for spatial panel econometric models, even without introducing external instrumental variables, the System Generalized Method of Moments (SGMM) can spontaneously select appropriate instrumental variables from the temporal trend changes of the variables. Therefore, we chose the SGMM method to estimate the dynamic spatial SDM model using the lag one period of the respective variable as an additional instrumental variable, and the results are shown in

Table 5. Estimation Results of Dynamic Spatial Panel Durbin Model (SGMM) ¹.

Variables	I: Industrial Agglomeration agg1				II: Industrial Agglomeration agg2
	sgmm-W1	sgmm-W2	sgmm-W3	sgmm-W4	sgmm-W1	sgmm-W2	sgmm-W3	sgmm-W4
L.pm	0.151 ***	0.349 ***	0.157 ***	0.101 ***	0.137 ***	0.194 ***	0.035 ***	0.119 ***
	(3.78)	(2.92)	(3.67)	(4.19)	(3.08)	(3.93)	(2.81)	(3.02)
agg	1.961 ***	2.005 ***	1.758 ***	1.862 ***	3.642 ***	3.735 ***	3.599 ***	3.496 ***
	(10.30)	(11.27)	(9.13)	(9.65)	(3.89)	(3.94)	(3.34)	(3.53)
sagg	−0.024 ***	−0.031 ***	−0.019 ***	−0.022 ***	−6.624 ***	−6.665 ***	−4.910 **	−5.513 ***
	(−5.34)	(−5.35)	(−4.25)	(−4.77)	(−3.21)	(−2.86)	(−2.37)	(−3.26)
er	−0.412 ***	−0.421 ***	−0.576 **	−0.602 ***	−0.926 ***	−0.944 ***	−0.149 **	−1.131 **
	(−7.67)	(−6.94)	(−2.28)	(−2.64)	(−2.86)	(−2.61)	(−2.15)	(−2.20)
W.agg	0.671 **	0.629 ***	0.821 **	0.795 ***	4.504 ***	4.305 ***	4.752 ***	3.002 ***
	(2.08)	(3.42)	(2.31)	(3.88)	(7.83)	(5.16)	(2.86)	(2.69)
W.sagg	−3.112 ***	−3.838 ***	−2.917 ***	−3.550 ***	−3.092 **	−0.933 ***	−4.144 ***	−2.003 **
	(−4.68)	(−6.96)	(−7.85)	(−3.57)	(−2.10)	(−5.80)	(−4.22)	(−2.24)
W.er	3.435 ***	3.297 ***	2.237 ***	1.572 ***	2.763 ***	2.603 ***	2.028 **	2.851 ***
	(4.80)	(4.47)	(3.86)	(5.79)	(3.21)	(3.36)	(2.01)	(3.11)
Control variables	YES	YES	YES	YES	YES	YES	YES	YES
Wald[P]	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
AR(1)[P]	0.010	0.011	0.000	0.002	0.010	0.000	0.020	0.012
AR(2)[P]	0.381	0.552	0.253	0.201	0.212	0.471	0.394	0.365
Sargan[P]	0.456	0.454	0.293	0.356	0.179	0.178	0.152	0.115

¹ The asterisks (***, **) indicate statistical significance at the 1%, 5% levels, respectively. Values in parentheses are t-statistics. Due to space limitations, the coefficients of control variables and their spatial lagged terms are not reported and are available upon request.

. It can be observed that both the Sargan test and the Arellano-Bond test (AR(1) and AR(2)) meet the requirements of SGMM, indicating that the selected instrumental variables are reasonable and effective. A comparison reveals that the variables are in good agreement with the estimation results in Table 2, demonstrating the robustness of the estimation results. In addition, we have also used different GMM methods to estimate the aforementioned models, which leads to consistent estimation results. The results based on the difference in the GMM method are not reported due to space limitations.

6. Examination of Spatial Moderating Effect Mechanism

Before estimating the dynamic spatial moderating effect model, the following two tests are required. First, the mediating effect test. Referring to the mediating effect model and test steps of Shao et al. (2019) [46], after a three-step test, the P-value of the Sobel test is 0.541, indicating that there is no mediating effect. Therefore, the level of environmental regulation is not a mediating variable, which shows that theoretical hypothesis H3 is reasonable. Second, the test of the linear relationship assumption of the interaction term. Scatter plots and Lowess fitting lines are made for the three variables of haze pollution, environmental regulation, and industrial agglomeration, and it is concluded that the data has non-linear marginal effects. Then, the box estimation method is used for testing, and the P-value of the Wald test is 0.000, which rejects the null hypothesis and indicates that there is a non-linear impact. Further testing is performed using the kernel estimation method, and the results show that within the sample period, the moderating effect of environmental regulation on the relationship between industrial agglomeration and haze pollution is non-linear. Therefore, the interaction term and spatial lag term of the quadratic term of environmental regulation and industrial agglomeration are added to the model (2), thus forming the final dynamic spatial moderating effect model.

Table 6. Estimation Results of the Spatial Moderating Effect of Environmental Regulation ¹.

Variables	Two-Way Fixed Effects Model		Dynamic Spatial Moderating Effect Model
	OLS-1	OLS-2	W1	W2	W3	W4
L.pm	0.836 ***	0.841 ***	1.006 ***	1.024 ***	0.946 ***	0.809 ***
	(22.51)	(22.32)	(33.55)	(34.20)	(28.72)	(24.26)
agg1	0.417 **	0.152 **	1.863 ***	1.910 ***	0.195 *	1.208 ***
	(2.05)	(2.21)	(27.72)	(27.94)	(1.76)	(21.91)
sagg1	−0.008 *	−0.017 **	−0.067 ***	−0.069 ***	−0.008 *	−0.047 ***
	(−1.69)	(−2.24)	(−13.83)	(−13.96)	(−1.73)	(−8.41)
er	−0.182 **	−1.343 **	−1.943 ***	−2.929 ***	−6.225 ***	−7.163 ***
	(−2.25)	(−1.99)	(−7.61)	(−4.39)	(−5.19)	(−6.47)
c.agg1#c.er		−0.234 **
		(−2.53)
c.sagg1#c.er		0.007 **
		(2.16)
agg1_er			−7.109 ***	−7.182 ***	−0.386 ***	−5.488 ***
			(−6.91)	(−7.60)	(−3.74)	(−7.06)
sagg1_er			0.118 **	0.119 **	0.004 **	0.090 **
			(2.38)	(2.45)	(2.08)	(2.10)
W.agg1_er			2.273 ***	2.235 ***	0.343 **	0.793 ***
			(14.39)	(14.29)	(2.21)	(16.02)
W.sagg1_er			0.029 ***	0.027 ***	0.019 *	0.029 ***
			(5.95)	(5.70)	(1.76)	(8.02)
Spatial rho			0.346 ***	0.337 ***	0.509 **	0.356 ***
			(4.30)	(4.21)	(2.55)	(9.84)
Control variables	YES	YES	YES	YES	YES	YES
City fixed	YES	YES	YES	YES	YES	YES
Time fixed	YES	YES	YES	YES	YES	YES
R2	0.853	0.859	0.090	0.089	0.658	0.097
Log-L			−1293.367	−4102.88	−1730.997	−2940.518

¹ The results for the dynamic SDM and the two-way fixed effects model corresponding to the four spatial weight matrices of industrial agglomeration agg2 are estimated but not reported due to space limitations and are available upon request. The asterisks (***, **, *) indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Values in parentheses are t-statistics. Log-L represents the Log-likelihood statistic. Due to space limitations, the coefficients of control variables and their spatial lagged terms are not reported and are available upon request.

presents the OLS regression results with and without considering the moderating effect to compare the magnitude of the moderating effect. The results show that the coefficients and significance levels of the main effect have changed when the moderating effect is considered, indicating the existence of the moderating role of environmental regulation. Furthermore, the dynamic spatial panel moderating effect model is used for estimation, and the results are illustrated in Table 6. In the following, we focus on the analysis of results under the ${\mathrm{W}}_1$ matrix.

First, we analyze the “U-shaped” moderating effect of local environmental regulation. On the one hand, the coefficient of the first-order term of industrial agglomeration, which was 2.876 without considering the moderating effect, changed to 1.863, and the interaction term coefficient was −7.109, significant at the 1% level. This indicates that in this stage, the promoting effect of industrial agglomeration on haze pollution is somewhat alleviated as the strictness of environmental regulation increases, and industrial agglomeration still plays a role in promoting growth. When industrial agglomeration is not yet mature, the preference for the scale of agglomeration will become a certain degree of resistance to environmental regulation. The standards and enforcement measures of environmental regulation are not strict and perfect enough, making it difficult to exert the screening and elimination effect on polluting enterprises. However, as the scale of agglomeration continues to expand and environmental issues become increasingly severe, the standards and enforcement of environmental regulation also become increasingly strict, thereby reducing the negative externalities of the environment brought by agglomeration, thus alleviating the promoting effect of industrial agglomeration on haze pollution. On the other hand, the coefficient of the second-order term of industrial agglomeration, which was −0.060 without considering the moderating effect, changed to −0.067, and the interaction term coefficient was 0.118, with a reduced level of significance compared to the first stage. The inhibitory effect of industrial agglomeration on haze pollution is enhanced with the adjustment of environmental regulation, but its moderating effect is declining; that is, the dependence on environmental regulation is reduced. When industrial agglomeration is relatively mature, the agglomeration effect is greater than the congestion effect, and the positive externalities of agglomeration on the environment gradually emerge. The concepts and practices of circular economy and green development are widely accepted and applied. The abilities of learning, communication, and collaborative production within enterprises, between enterprises, within agglomeration areas, and between agglomeration areas are effectively improved. In other words, it can achieve the effect of reducing haze pollution through its own agglomeration, and the dependence on external forces, especially the dependence on government environmental regulation, has decreased.

Second, we analyze the linear moderating effect of neighboring environmental regulation. The estimated coefficient of the spatial lag of the first-order interaction term is 2.273, which is significant at the 1% level. That is, in the extensive development stage of local industrial agglomeration, the environmental regulation of neighboring areas positively regulates the relationship between industrial agglomeration and haze pollution. During this period, the stricter the environmental regulation in neighboring areas, the more it promotes the agglomeration of polluting industries in the local area, especially the highly polluting industries following the spatial proximity transfer model [44], thereby exacerbating pollution. This effect is more likely to occur in relatively underdeveloped local areas with loose environmental regulation and developed neighboring areas with strict environmental regulation. The coefficient of the spatial lag of the second-order interaction term is 0.029, which is significant at the 1% level. After the local industrial agglomeration is relatively mature, the environmental regulation in neighboring areas positively regulates the relationship between industrial agglomeration and haze pollution. The possible reason is that the strictness of environmental regulation in neighboring areas and the advanced development of local industrial agglomeration forces polluting enterprises to migrate to more remote areas or transform and upgrade in other ways to seek their own development, which strengthens the effect of industrial agglomeration in suppressing haze pollution and makes a greater improvement in haze pollution in the local area and even the whole region. The above conclusions justify Hypothesis H3.

7. Conclusions and Policy Recommendations

Based on a detailed analysis of the potential theoretical mechanisms between industrial agglomeration, environmental regulation, and haze pollution, combined with a robust empirical test of the theoretical hypotheses using a dynamic spatial panel model and a dynamic spatial moderating effect model with a sample of 284 cities in China, we arrive at the three conclusions as follows:

First, the spatial spillover effect of industrial agglomeration is significant, with both local and neighboring industrial agglomeration levels showing an “inverted U-shaped” relationship with local haze pollution. When the local industrial agglomeration level is below the inflection point value of 18.445, the congestion effect of industrial agglomeration is greater than the agglomeration effect, leading to negative environmental externalities and exacerbating haze pollution. After surpassing the inflection point value, the agglomeration effect exceeds the congestion effect, bringing positive environmental externalities that are conducive to alleviating haze pollution. Low levels of industrial agglomeration in neighboring areas negatively impact surrounding areas through spatial spillover effects, while high levels of agglomeration help alleviate haze pollution in surrounding areas.

Second, environmental regulation has a significant spatial spillover effect. Specifically, local environmental regulation suppresses local haze pollution, while the effect of the neighboring environmental regulation is the opposite. High levels of environmental regulation can not only prompt enterprises to improve production technology and reduce pollutant emissions but also drive low-quality polluting enterprises to transfer or exit the market through cost effects, thereby reducing local haze pollution. High levels of environmental regulation in neighboring areas can encourage the migration of pollution-intensive enterprises to local areas, thereby exacerbating pollution.

Third, environmental regulation acts as a moderating variable between industrial agglomeration and haze pollution. Local environmental regulation acts as a “U-shaped” non-linear moderator, while neighboring environmental regulation acts as a positive linear moderator. As the stage of local industrial agglomeration development varies, the moderating effect of environmental regulation also differs, while the moderating effect of neighboring environmental regulation shows consistency.

The above research conclusions have important policy implications for governments at all levels in the new era to better utilize environmental regulation tools for precise haze control and to fully leverage the scale effects and spillover effects of industrial agglomeration to promote healthy and sustainable development of industries and the economy. The specific policy recommendations are as follows:

First, the government should vigorously promote the upgrade and transformation of industrial clusters and formulate differentiated portfolio policies focusing on the development of industrial agglomeration to play the positive environmental externalities and spatial spillover effects of industrial agglomeration. For regions that have not yet crossed the industrial agglomeration inflection point, such as Xingtai, Baoding, Yangquan, Jining, Hebi, and other key cities, it is necessary to improve their level of industrial agglomeration while fully utilizing the income level, R&D investment, and openness to foreign trade to reduce the impact. For regions that have crossed the inflection point, it would be beneficial to encourage industrial and technological innovation, focus on attracting high-level production factors, and guide the agglomeration and development of high-value-added industries such as high-end R&D and design.

Subsequently, the government should fully stimulate the competitive screening function of environmental regulation and accurately employ the dynamic spatial adjustment laws of environmental regulation in the local area and neighboring areas. When the development of industrial agglomeration is immature, it is advantageous to exploit the negative regulatory effect of local environmental regulation, supplemented by strict regulatory standards, to force the transformation of polluting enterprises. Meanwhile, considerable attention should also be delivered to the positive environmental regulation adjustment effect of neighboring areas, and it would be appropriate to raise the entry threshold to limit their entry into the local area when neighboring polluting enterprises transfer or agglomerate to the local area. In contrast, at the advanced development stage of industrial agglomeration, the positive regulatory effect of local environmental regulation gradually weakens. In such a situation, the intensity and means of environmental regulation can be relatively flexible, with adequate reliance on the market and even more on the advanced development path of industrial agglomeration to reduce haze pollution. To further improve the environment of the entire region and reduce air pollution, the government can make full use of the positive regulatory effect of neighboring environmental regulation, actively cooperate with neighboring regions for multiple aspects, and establish a relatively high level of environmental regulation system to promote the transformation and upgrading or outward transfer of polluting enterprises in both places.

8. Discussion

We compare the conclusions of this study with the results in the existing literature using similar methodologies. On the one hand, they share two main similarities. First, it is confirmed that industrial agglomeration has a significant impact on haze pollution, and there is a non-linear relationship between the two, in agreement with the studies of Yang et al. (2018) [24], Zhang et al. (2022) [25], and Tan et al. (2022) [26]. Second, it is confirmed that environmental regulation has a significant impact on haze pollution, and this impact has a spatial spillover effect, which is consistent with the studies of Li et al. (2020) [27], Sadat et al. (2022) [28], and Kuang et al. (2023) [32]. On the other hand, the differences come from the four aspects. First, this paper reveals that the non-linear relationship between local or neighboring industrial agglomeration and haze pollution is an “inverted U-shape”, rather than a “U-shape” or “N-shape” relationship [24,25,26]. Second, this paper finds that environmental regulation affects the spatial heterogeneity of haze pollution; that is, local environmental regulation significantly inhibits haze pollution while neighboring environmental regulation plays a promoting role. Third, this paper accurately identifies and tests the moderating role of environmental regulation and finds that the moderating effect of environmental regulation on the relationship between industrial agglomeration and haze pollution shows spatial heterogeneity in both local and neighboring areas. This research perspective and conclusion provide new insights into understanding the spatial effects of environmental regulation. Fourth, existing studies have mostly employed static spatial panel models for empirical testing [24,25,33], whereas we have utilized a dynamic spatial panel Durbin model and a dynamic spatial moderation effect model. These methodologies can capture the dynamic relationships and spatial dependencies among industrial agglomeration, environmental regulation, and haze pollution more accurately, thereby enabling a deeper understanding of the complex relationships among these variables. Besides, due to its unique economic development mode and national conditions, China shows specificities in various aspects, including distinctive industrial agglomeration characteristics, variability in the enforcement of environmental regulations, imbalances in regional development, and a unique energy structure and consumption pattern.

Despite the insightful conclusions obtained, the following limitations still exist, and future research efforts could be focused on achieving further improvement. First, the limitation of sample selection. We exploited the panel data of prefecture-level cities in China as the research sample without considering other countries or regions, and the conclusions obtained may be applicable only to China. However, haze pollution is not only a concern for China but also an environmental issue that other countries are committed to addressing. Therefore, future studies can expand the scope of samples by collecting relevant data from other typically developed or developing countries and conducting empirical examinations to compare with the findings of this study so as to obtain richer policy recommendations. Second, limitations of the data used. We used annual average PM2.5 values to characterize urban haze pollution levels. However, annual data cannot reflect daily and seasonal variations in haze pollution. Therefore, future research can invest more efforts in collecting daily, monthly, and quarterly data, which in turn can help us identify more accurately the heterogeneous effects of industrial agglomeration and environmental regulation on haze pollution. Third, there is a limitation in controlling variable selection. In this paper, based on the consideration of data accessibility, the control variables are mainly selected from the perspective of economic development in conjunction with the research object. However, due to the complexity of the causes, haze pollution is also affected by natural factors such as geography, climate, and meteorology, in addition to economic factors. In order to further explore other influential factors, a more complete econometric model could be constructed, which includes factors such as geography and climate as control variables.