Pollution acceleration before braking: Evidence of environmental deterioration from the anticipated steel restriction policy in China

This research employs China’s steel restriction policy as a backdrop to investigate environmental policies’ unintended and counterproductive effects. Using high-resolution satellite-derived data and panel Difference-in-Differences regression, we found that the air pollution concentration in cities implementing the steel restriction policy is 5.688 μg/m3 higher than in control group cities. Additionally, the growth rate of air pollution in these cities is 6.577% faster. This quantitative evidence substantiates the backfire effect of the anticipation of environmental policy, where the delay between a policy’s announcement and its enforcement leads to a short-term surge in pollution levels. For China and other emerging economies, the development of a thorough and deliberate intergovernmental cooperation strategy is critical when formulating environmental policies. It involves synchronizing the efforts of different government levels in applying pollution controls and diminishing the interval of potential intense pollution in the pre-implementation phase.


Introduction
The steel industry is a vital part of the global economy.According to the World Steel Association, global crude steel production reached 1951 million tons in 2021 (World Steel Association 2022).However, this industrial achievement comes with significant environmental challenges, mainly due to the extensive use of fossil fuels in steel production, which leads to severe pollutant emissions that threaten public health and well-being (Tilt 2013).Consequently, many countries have imposed restrictions or introduced technical regulations to reduce emissions from the steel industry.For example, the EU's Industrial Emissions Directive (IED) sets stringent standards for steel production (European Commission 2010).In contrast, China has implemented strict administrative measures to adjust the steel industry, ensuring it meets the environmental protection goals set by the central government.
Since China's reform and opening up, the steel industry has burgeoned, catapulting the nation to the forefront of global steel production.The World Steel Association reported that in 2021, China was responsible for 1032.8 million tons of crude steel, accounting for approximately 53% of the global production, thus cementing its status as the world's steel powerhouse again (World Steel Association 2022).However, this industrial success has been tinged with environmental challenges, as the steel production process, heavily reliant on fossil fuels, is a significant source of pollutant emissions, casting a pall over the well-being of its citizens.Recognizing these environmental challenges, the Chinese government has implemented regulatory measures to curb pollution and promote sustainable industrial practices.
The Chinese government has made considerable efforts, enacting a series of environmental protection policies.One of the key measures is the Air Pollution Prevention and Control Action Plan (APPCAP), formulated in 2013, which stands as China's first national plan for air pollution prevention.The APPCAP plan focuses on the Beijing-Tianjin-Hebei region, a significant source of pollution activities, particularly from the steel industry, which involves processes such as coking, iron and steelmaking, and rolling (Guo et al 2017, Dong et al 2023, Cheng et al 2024, Li et al 2024).To align with national policy, the Hebei provincial government implemented a localized initiative, the Steel Industry Restructuring Program (HSIRP), to alleviate the tension between steel industry development and urbanization (Hebei Provincial Government 2014).The program proposes to relocate steel production firms from Hebei to coastal regions such as Fujian to improve air quality and encourage sustainable development within the region (Hebei Provincial Development and Reform Commission 2015).With Hebei province at the epicenter of these industrial and environmental dynamics, assessing how these interventions specifically affect this vital area is crucial.
Hebei Province is recognized as a critical hub in China's steel production, contributing markedly to global total steel output.Nevertheless, the industry's growth has led to serious environmental pollution challenges, especially regarding air quality in both Hebei and the wider Beijing-Tianjin-Hebei region.Consequently, the HSIRP, a large-scale steel restriction policy, warrants a comprehensive evaluation due to its significant implications for balancing economic development and environmental conservation.Past research has considered the impact of APPCAP on air quality, but the specific effects of the HSIRP on air pollution reduction remain unexplored (Cai et al 2017, Feng et al 2019, Yu et al 2022, Dong et al 2023, Li et al 2024).This study aims to fill this gap by assessing the impact of HSIRP on PM 2.5 concentrations, thereby exploring the environmental effects of China's large-scale steel restriction policy and the mechanisms through which these effects occur.
This research uses high-resolution satellite imagery to extract air pollution data for selected cities and designates the study area using local Moran's I to identify high-pollution clusters.Furthermore, the Difference-in-Differences (DID) model is employed to analyze the effect of the HSIRP's policy on local air pollution levels.Besides, the study area is further refined using the artificial impervious surface (AIS) layer, which serves as a more accurate indicator of built-up areas impacted by human activity.This method offers a closer approximation of pollution that affects human activities, reducing the dilution effect of pollution levels in areas with little to no human presence, such as forests and bodies of water (Qiang et al 2020(Qiang et al , 2021(Qiang et al , 2023)).
Our anticipated contributions cover some key areas.First, we identify areas of high pollution in China and use them as a research sample.Second, using a DID approach, we hope to outperform traditional atmospheric pollution control plans by evaluating the impact of China's steel restriction policy on air pollution.Furthermore, we incorporate data from AIS to create a pollution sample that more accurately reflects areas of human activity.Finally, we present policy recommendations based on the DID findings.
The structure of this paper is as follows: section 2 provides an overview of the data, including the study area, the variables selected, and their statistical description.Section 3 details the methodology, covering local Moran's I and the DID approach.Section 4 presents the study's findings, and sections 5 and 6 provide discussion and conclusion, respectively.

Study period
The timeframe for this study is specifically delineated from 2008 to 2016, grounded on several pivotal considerations.The initial rationale is the Chinese government's announcement of the 'Four Trillion Investment Plan' in 2008, which significantly propelled the nation's industrial development while simultaneously exacerbating environmental pollution.Moreover, beginning in April 2017, the Ministry of Environmental Protection initiated an extensive enforcement campaign, assigning 5600 officials to a year of intensified supervision across the Beijing-Tianjin-Hebei region and surrounding areas.This campaign's objective was to enhance the enforcement of environmental regulations and policy adherence, potentially impacting the results of research studies such as ours (Yu et al 2022).Therefore, the study period concludes in 2016 to effectively isolate the effects of policy changes from the research findings.

Study area
Air pollution within China is marked by enduring spatial clusters.Local Moran's I is applied to define the study area (see section 3.1).Two significant pollution agglomeration clusters are identified in China: (1) The North China Plain and the Yangtze River Delta, and (2) the South Xinjiang region (figure 1).It is important to recognize that the pollution agglomerations in South Xinjiang are influenced by natural environmental factors, specifically the Gobi deserts.In consideration of the scope of this research and to guarantee that the control group in the DID analysis (see section 3.2) shares consistent pollution characteristics with the treatment group, the study is confined to the pollution agglomeration in the North China Plain and the Yangtze River Delta.
Within the pollution agglomeration of the North China Plain and the Yangtze River Delta, regions (including prefecture-level cities, county-level cities, and counties) that have the high-high PM 2.5 concentrations from 2008 to 2016 are classified as the treatment group based on the Local Moran's I value.Eight cities with notably high air pollution concentrations are identified as part of this group: Baoding, Langfang, Bazhou, Shijiazhuang, Tangshan, Wu'an, Qinhuangdao, and Xingtai.The other cities within these plains are the control group (figure 1).This classification is essential for conducting the DID regression analysis.

Air pollution
In literature addressing air pollution in China, the China Air Quality Index (AQI) and satellite-derived raster data are commonly leveraged as sources of air quality.AQI based on ground stations combines the monitored control concentrations into a conceptual value and grades them to indicate the air quality status (Yang et al 2019).However, this calculation method has many drawbacks, such as it fails to reflect the extreme fluctuations over time, the upper limit of the score is too low, and the exceedance of pollution standards cannot be measured.More critically, the technique has a limited coverage and time horizon.In 2012, China announced the environmental air quality standard (GB3095-2012), and 74 cities were identified as the first batch of cities to implement the standard, which was then gradually expanded.Therefore, the AQI data source lacks coverage of data before 2012 and observations of small cities and counties.
The satellite raster data, by contrast, has multitemporal and high-coverage properties that can be adapted to our study area.In terms of algorithm, the uniform global algorithm makes the monitoring results easier to compare.The Surface PM 2.5 V5.GL.02 dataset combines the GEOS chemical transport model with a series of AOD datasets (MISR, SeaWiFS, NASA, and MODIS) and uses Geographically Weighted Regression (GWR) to extract the mean PM 2.5 values in each study area.Regional ground-based total and compositional mass observations are used to correct the GWR.
Using satellite raster data for urban pollution measurement requires carefully selecting city boundaries.Considering the inconsistency of the official urban areas published by the Chinese government and the problem that uninhabited areas within administrative regions may reduce the true pollution levels, we introduce built-up areas to make the calculated pollution concentrations more accurate.Several studies have pointed out that current global .Therefore, we can use artificial surface data in land cover as a proxy for urban built-up areas, which helps us circumvent the impact of non-anthropogenic areas on urban pollution calculations.In this study, we use the GAIA dataset from Gong et al (2020).This dataset can provide data about 30 m of land cover and observations about the artificial surface.We extract the artificial ground surface based on this dataset for the Surface PM 2.5 V5.GL. 02 dataset for convergence.

Treatment variable
At the end of 2012, China convened the Central Economic Work Conference, the highest-level meeting on economic policy (State Council PRC 2012).Unlike previous meetings, this one frequently mentioned terms like economic overheating and overcapacity.Following the conference, the National Development and Reform Commission introduced a series of policies aimed at eliminating excess capacity in the steel industry.This set the tone for subsequent years, focusing on eliminating outdated capacity and controlling pollution, with the central government incorporating these targets into local government evaluations and official promotions (Chen et al 2021).Consequently, many provinces, led by the government, began actively implementing industrial restriction measures.Figure 2 provides a timeline chart to deconstruct central and Hebei provincial government policy milestones.By analyzing the timeline of policy proposals and implementations, we infer that the policy was planned in 2013.However, due to China's unique approval system, there is nearly a one-year window between planning and specific approvals.Therefore, 2013 is set as our base year, but unlike most DID models, we focus more on the changes around the policy approval in 2014.

Control variables
In this paper, we would like to investigate whether HSIRP affects the PM 2.5 concentration in the treated cities.Hence, the control group will contain the basic economic characteristics of cities.All the value of the control variables is either in ratio or natural log format.The control variable group contains six factors: ln(GDP), ln(pop), ln(area), fiscal_ratio, sec_ind, and tert_ind.
These control variables are closely linked to air pollution, and managing them is essential to isolate factors that might affect the estimation of policy impacts.First, ln(GDP) highlights the connection between economic development and air pollution.Economic growth often coincides with industrialization and urbanization, leading to increased energy consumption and emissions (Grossman and Krueger 1995).Thus, GDP growth may be associated with higher levels of air pollution, especially in developing countries.Second, ln(pop) indicates the relationship between population density and air pollution.An increase in urban population can lead to traffic congestion and higher energy consumption, escalating air pollution (Chen et al 2020).Third, ln(area) reflects the relationship between urban area size and air pollution.Expanding of urban built-up areas can result in a more dispersed distribution of pollution sources, affecting the diffusion and dilution of pollutants (Qiang et al 2021).Fourth, fiscal_ratio represents the proportion of fiscal expenditure to GDP.Higher fiscal revenue may imply that the government has more resources

Local Moran'I
The majority of research that analyzes spatial autocorrelation uses Moran's I statistic.The approach may be categorized into two distinct types: global Moran's I and local Moran's I. Local Moran's I is a method that offers insights into spatial autocorrelation through a visual depiction, in contrast to global Moran's I, which assesses the overall distribution.In other words, it defines the similarity between a spatial observation and its neighboring observations and offers individual statistics for each separate observation.For observation i, the equation representing the local Moran's I statistic is as follows: where n represents the number of observations.The attribute values of observations i and j are denoted as x i and x j , respectively.x represents the mean attribute value of all observations.ω ij is the element of the spatial weight matrix that represents the spatial relationship between observations i and j.Local Moran's I statistic calculates the weighted covariance between the value of observation i and its neighboring observations, with the denominator being the sum of squared differences between the attribute values of all observations and their mean.After normalization, the range of values lies between −1 and 1. Local Moran's I can assist in identifying spatial clustering patterns and outliers.The clusters can be categorized into four types based on the z-scores: statistically significant high-high clusters with highvalue observations are surrounded by other highvalue observations; statistically significant low-low clusters with low-value observations are surrounded by other low-value observations; and statistically significant high-low clusters with low-value observations are surrounded by high-value observations.The high-high and low-low clusters facilitate the discernment of hot spots and cold spots in spatial observations.Conversely, the implementation of high-low and low-high clusters assists in identifying spatial outliers.
In this study, we conducted a local Moran's I analysis on annual mean air pollution values from 2004 to 2018 and identified cities that consistently belonged to high-high clusters over the years.We believe that these communities share a high degree of spatial similarity and can be included in the research framework as a high-pollution clustering entity.

DID regression
The DID model is a widely employed method for environmental and public policy analyses, designed to address issues stemming from causality and endogeneity in exogenous policies.In this research, the DID model is employed to evaluate the effects of the HSIRP on PM 2.5 concentrations, accounting for both time-fixed and city-fixed effects.McConnell (2023) pointed out that in the DID model, taking the logarithm of the dependent variable versus not taking the logarithm has different implications.Without taking the logarithm, the coefficient of the treatment effect in the DID model represents the actual unit difference between the treatment group and the control group, i.e. the difference in PM 2.5 concentration (µg/m 3 ).When taking the logarithm, the coefficient of the treatment effect represents the difference in growth rates between the treatment group and the control group.Therefore, the baseline DID model can be represented as: where i denotes the city and t denotes the year.PM 2.5 represents PM 2.5 concentration, measured in µg/m 3 .ln(PM 2.5 ) represents the logarithm of air pollution concentration.β 0 is the intercept.The did is treatment effect.If a city is among those implementing the steelfree policy during the policy implementation period, its value is 1; otherwise, it is 0. β 1 is our DID estimator, for the models without taking the logarithm of PM 2.5 , the policy effect is β 1 .In the models taking the logarithm of PM 2.5 , the policy effect is exp (β 1 ) − 1. X represents the vector of explanatory variables, and is its coefficient vector.is the city-fixed effect, is the time-fixed effect, and is the error term.

Parallel trend test
Performing a parallel trend test is crucial to ensure the validity of our analysis using the DID regression model with doubly fixed effects.This test examines whether the treatment and control groups exhibited similar trends in PM 2.5 before the implementation of the policy, which is a key assumption for applying the DID regression model.The results of the parallel trend test show that, for both PM 2.5 and ln(PM 2.5 ), the treatment and control groups tend to be parallel before the policy implementation, thereby meeting the necessary conditions for conducting DID regression analysis (figures 3 and 4).The policy had an effect when it was implemented in 2013, peaked in 2014, and quickly dissipated after implementation.Additionally, to facilitate visualization, we plotted the time trends of PM 2.5 and ln(PM 2.5 ) for the experimental and control groups.The trend changes in the treatment group can be clearly observed (figures 5 and 6).

DID regression results
As the DID regression model's fundamental assumptions have been satisfied, we run the DID regression analysis.The control variables are systematically incorporated into the analysis to show the changes in the variable coefficients (table 2).All models have a variance inflation factor (VIF) of less than 10, indicating the absence of multicollinearity problems.As shown in table 2, models 1 and 2 present the results for PM 2.5 and ln(PM 2.5 ) without control variables, respectively, while models 3 and 4 present the results with all control variables included.All regression methods employ time-fixed and city-fixed effects.Regardless of whether control variables are included, the coefficients for the interaction term are positive.Focusing on models 3 and 4, which include control variables, the results indicate that when a city implements HSIRP at a 1% significance level, the air pollution concentration in cities that implemented the steel restriction policy is 5.688 µg/m 3 higher than in the control group cities.Additionally, the growth rate of air pollution is 6.577% faster.

Robustness test 4.3.1. Changes in sample selection
We try different sample selections to see whether they will significantly change the DID regression results (table 3).In table 3, models 5 and 6 present the DID regression results with the entire sample as the data input.This is the baseline model for comparison.In northern China, centralized heating relies on thermal power, significantly increasing air pollution (Xiao et al 2015).This may obscure the policy effect of HSIRP.Therefore, in models 7 and 8, we focus on those cities with the central heating supply in winter to see whether the policy effect of HSIRP is apparent.Beijing and Tianjin are municipalities included in the sample, while the pollution management and control requirements for these municipalities may differ from those of other cities.Thus, in models 9 and 10, Beijing and Tianjin are also removed from the sample, apart from removing those cities without the central heating supply in winter.According to the findings, the interaction terms' coefficients in those models all point in the same direction.The results indicate statistical significance at the 1% level, as shown in table 3, which evidences the robustness of our DID regression results.

Changes in geographic units
In general, heavy-polluting industries are more prevalent in non-urban areas.Hence, counties may exhibit higher pollution than their corresponding municipalities or county-level cities.Perhaps the policy effect of HSIRP impact of policies is more pronounced when counties rather than county-level cities are chosen as geographical units in the DID regression models.Therefore, we extract PM 2.5 concentrations at the county level.In addition, we use the counties of the cities in the treatment group as the treatment group and the counties of the cities in the control group as the control group.Table 4 presents the associated results.In table 4, the DID regression models 11 and 12 are based on county-level city data, while models 13 and 14 are based on county data.We found that when using county data, the pollution level in the experimental group is 8.042 µg/m 3 higher than in  the control group, and the growth rate is 9.407% faster.This indicates that the pollution surge effect of this policy is more pronounced in county areas.Overall, there are no contradications in the results between the models in different geographic units, which further ensures the robustness of our statistical findings.

Placebo tests
To ensure the reliability of the above DID regression result, a random placebo test is conducted by selecting treatment and control cities.We run the placebo tests in the following order: the entire sample of municipalities and county-level cities, the sample with the cities with heating supply in winter excluded, the  sample with the cities with heating supply in winter, and the two municipalities (Beijing and Tianjin) excluded, and the entire sample of counties.We repeat random sampling 500 times in all the placebo tests, and the results for PM 2.5 and ln(PM 2.5 ) are shown in figures 7 and 8, respectively.Most estim-ations have p-values more than 0.1 and are significantly off from the actual estimates produced by the DID regression models.This lends credence to the robustness of the DID regression results and suggests that our estimations were not derived by coincidence.

Discussion
Our results indicate that the HSIRP may have caused a short-term burst of air pollution; however, the long-term reduction in pollution due to this policy is not clearly evident.First, the DID analysis shows that cities subject to HSIRP experienced a 6.577% increase in PM 2.5 levels just before the policy's implementation-a statistically significant finding at the 1% level.It suggests that steel firms' production activities intensified during the policymaking period, possibly because these firms saw it as a 'last chance' or a window of opportunity to maximize short-term profits as a buffer against anticipated future production cost increases before the policies were implemented.This phenomenon can be summarized as 'anticipated policy backfire.'Second, our DID analysis findings suggest that the reduction in PM 2.5 levels in cities subjected to HSIRP after the policy implementation is not statistically significant.This implies that while air pollution decreased after HSIRP was implemented, the reduction may be attributable to other environmental policies or activities rather than HSIRP.Our findings do not support the notion that HSIRP has contributed to reducing air pollution.Instead, HSIRP may have triggered a short-term increase in air pollution before it was fully implemented.
This study focuses on the impacts of HSIRP from the perspective of environmental protection.It is important to note that the primary purpose of HSIRP and related policies may not be to reduce air pollution but to decrease the overcapacity of the steel industry and upgrade the industrial structure (State Council PRC 2013).However, given that the steel industry is a significant source of air pollution and carbon emissions in China, it can play a crucial role in environmental protection (Ministry of Ecology and Environment PRC 2019).While HSIRP could be considered an alternative strategy for environmental protection, our findings from an environmental policymaking perspective suggest that merely reducing the steel industry's capacity may not effectively reduce air pollution.Instead, it can stimulate the market to take preemptive actions, such as increasing production for instant profits before restrictions are fully implemented.Additionally, there should be compound policies that target the different characteristics of steel firms, such as their development level and geographic concentration.

Conclusion
Our research uncovers a frequently neglected dimension of environmental policy enactment, spotlighting the potential negative consequences arising from a delay between the announcement and the actual enforcement of a policy.We refer to this as the 'anticipated policy backfire,' which emphasizes the critical importance of strategic foresight in policy development, especially in contexts like China, where economic growth and environmental conservation have yet to find a harmonious balance.It is essential that the timing of policy pronouncements and their enforcement be thoughtfully calibrated to mitigate adverse outcomes and discourage opportunistic behavior.
The study underscores the temporal delays and objective discrepancies between central and local governments in pollution policy implementation.Thus, devising a strategy that harmonizes the conflicting priorities of both levels of governance to attenuate the detrimental effects of such policies should take precedence in upcoming discussions.The rapid and effective deployment of future policies is essential, curtailing the interval between their formulation and enactment, as procrastination in policy implementation can lead to unintended environmental detriments.Additionally, the central government must contemplate the motivations of local officials and craft apt incentives to ensure that national policies are effectively realized at the local administrative tiers.A comprehensive strategy that integrates monitoring, enforcement, education, and awareness campaigns is imperative for successfully attaining environmental protection and sustainable development objectives.
Our study reveals significant implications for policy and practice.Policymakers must consider the anticipated behavior of industries before introducing new regulations and designing policies to minimize the window of opportunity for such behavior.It can be achieved by shortening the time frame between policy announcement and implementation or by implementing transitional measures to prevent a short-term spike in pollution.Additionally, coordinating incentive mechanisms across different government levels is crucial.The central government should establish mechanisms to ensure local officials are motivated to achieve environmental goals, possibly by integrating environmental indicators into their performance evaluations.Methodologically, our study underscores the effectiveness of DID analysis in identifying the unintended consequences of policy interventions.This approach can applied to other areas where policy delays might lead to similar anticipatory effects.Future research should utilize more detailed data and explore additional econometric techniques to validate and expand our findings.Theoretically, we contribute to the literature on environmental policy and governance by introducing the concept of 'anticipated policy backfire.'This concept highlights the complexity of policy implementation within multi-level governance structures and the importance of strategic foresight in policy design, suggesting the need for more nuanced theories to consider the dynamic responses of stakeholders to policy announcements.
There are also limitations in our study that future research will need to address.It focuses on specific policies within particular regions, which may limit the generalizability of the results.Comparative studies across different regions and policies could provide a broader understanding of the anticipated effects of environmental regulations.Additionally, exploring the role of public awareness and media coverage in anticipatory behavior could offer a more comprehensive understanding of the relevant mechanisms.Our study highlights the importance of timely and well-coordinated policy implementation, the coordination of incentives across governance levels, and the potential unintended consequences of delayed policy execution.Addressing these challenges is crucial for achieving environmental protection and sustainable development goals.

Figure 1 .
Figure 1.Map of the study area.

Figure 2 .
Figure 2. Timeline of the policies implemented by the central and Hebei provincial governments.

Table 1 .
Descriptive statistics of variables employed in this study.Zhu et al 2019).Therefore, a greater proportion of the secondary industry in GDP may lead to more severe air pollution.Lastly, tert_ind reflects the relationship between the proportion of the tertiary industry and air pollution.The tertiary industry (such as services) generally produces less pollution than the secondary industry (Wei et al 2018).Hence, a higher share of the GDP of the tertiary industry could help reduce air pollution.Data are sourced from China City Statistical Yearbooks.Table1presents the descriptive statistics for the dependent, independent, and control variables.

Table 3 .
DID regression results with the changes in sample selection.

Table 4 .
DID regression results with the changes in geographical units.* * p < 0.01.Numbers in parentheses present t statistics from the cluster.