SO₂ Emissions Reduction Effect of China’s Pollution Levy Standard Adjustment: A Short-Term and Long-Term Analysis

Abstract

As one key environmental regulation measure, China’s pollution levy policy has been widely discussed; however, existing research has solely concentrated on the emissions reduction effect of pollution levy policies in the short term but has ignored long-term effectiveness, which may cause research bias. Thus, taking pollution levy standard adjustments as the core independent variable, this study builds firm-level pollution data and proves that the pollution levy policy is effective both in the short and long term. Furthermore, it demonstrates that the emissions reduction effect is achieved mainly by decreasing production and increasing the removal of sulfur dioxide ($S{O}_2$). In addition, it is uncovered that the emissions reduction effect is mainly a result of two processes—the first is the greater use of clean energy and higher energy efficiency that can cause a decrease in the production of $S{O}_2$, and the second is the utilization efficiency of waste gas treatment facilities, which can increase the removal of $S{O}_2$.

1. Introduction

Sustainable development has aroused wide concern, both public and academic. Concerning the progression of industrialization, many Western countries have adopted the “pollute first, control later” strategy, which sacrifices the interests of the next generation and goes against the original intentions of sustainable development. For the excessive pursuit of economic interests, many countries tend to abuse resources and release enormous amounts of pollutants, which have inflicted substantial strain on the environment. The resulting environmental problems not only slow down economic development [1] but also cause huge damage to human health [2,3,4]. This type of development model cannot be perpetuated. Moreover, with its opening up and reform strategy, China has achieved substantial economic development. Meanwhile, extensive development aggravates the consumption of energy resources (as shown in Figure 1) and produces substantial amounts of pollution [5,6,7]. This development pattern of “high-energy consumption, high-polluting, and high-emission” is no longer desirable [8]. How sustainable development can be achieved is, therefore, an urgent issue, especially for developing countries like China. In such a scenario, the Chinese government has implemented a variety of policy tools to retain the balance of economic development and environmental protection [9]. Among these measures, the pollution levy is one crucial step towards sustainable development.

A pollution levy (or pollution tax) comprises pollution control fees levied on enterprises, institutions, and individual industrial and commercial households that directly discharge pollutants into the environment, generally encompassing sewage, waste gas, noise, solid, and hazardous waste discharge costs. This has been demonstrated efficiently in many developed countries [10,11,12]. Although “endogenous enforcement” may be extensively observed in developing countries [13,14,15,16], the benefits are significant considering the different or even the opposite result between China and India [17]. As the greatest developing country, China granted status to this policy early in 1979 and proceeded to put it into effect progressively among different provinces. Moreover, since a significant difference exists in the levy standard, the actual situation of the economy, and the environmental situation, there is a debate about the “lower collection standards”, “local protectionism” [18,19,20], and “lack of national and regional regulation and financial support” [21] when talking about its effectiveness. Despite this, research on its effectiveness has gained consistency [7,22,23,24,25,26]; however, current research is largely concentrated on its short-term influence while study of its long-term influence remains insufficient, especially at the firm level. Compared to short-term effects that can be immediately felt through the decrease in emissions, long-term effects sometimes have a time lag and are hard to record and measure. As environmental pollution issues involve many aspects of economic and social development, the essence is an extensive development model and backward production capacity, which are concentrated in high-energy-consuming and high-polluting industrial structures [8] and coal-based energy consumption [27,28]. These difficulties have strong medium- to long-term features, and addressing them in the short term is difficult. If we do not research the emissions reduction effects of policies in the long term, we cannot guarantee a correct conclusion. This is also the case when we evaluate the emissions reduction effect of pollution levy policies. Moreover, it is logical to question the following: Does this policy have short-term consequences, long-term implications, or a combination of both? In all these cases, what types of aims and measures take effect? To resolve these problems, we select pollution levy standard adjustments as the primary explanatory variable and perform theoretical and empirical analyses in the succeeding sections. Although some articles have analyzed the policy’s effects and put forward suggestions in the short- and long-term [29], they still ignore the policy’s time-varying differences and lack a combination of short-term and long-term considerations. Different from existing studies, we illustrate both the policy’s short- and long-term success, which contributes to the literature concerning the effects of environmental regulations on pollutant emissions. In detail, our article is outlined as follows.

Firstly, we create firm-level panel data by matching industry–enterprise data and pollution data from 2000 to 2013. After that, we integrate temporal trends into Xttobit (random-effects Tobit models) to examine both the short- and long-term effectiveness of the pollution levy policy. Taking the total amount of $S{O}_2$ emissions as the dependent variable, we observe that the increase in pollution levy standards indeed decreased $S{O}_2$ emissions both in the short and long term. This conclusion is robust when controlled by the interaction of specific coefficient factors and nonlinear temporal trends. To evaluate the robustness further, we altered the baseline model using Reghdfe (linear regression absorbing multiple levels of fixed effects), PPML (Poisson pseudo maximum likelihood), Tobit (pooled tobit model), and Pantob (estimation of static panel data censored regression models), respectively; substituted the $S{O}_2$ emissions response variable with $S{O}_2$ emission density; and used subset observations without some specific years and firms. Using all these terms, we achieved the same results.

As for mechanism analyses, we detected both the production adjustment and pollution treatment and observed that firms’ output and production hours did not decrease but increased when facing higher pollution levy norms. This indicates that the drop in $S{O}_2$ emissions is not achieved by decreasing the output when confronted by stricter pressure in the short term but may continue with economic progress. Furthermore, we studied this decline both with terminal and front-end treatment and found that the higher pollution levy norm not only forces firms to upgrade terminal treatment but also causes firms to use more clean energy and improve energy efficiency concerning front-end treatment.

To assess if there is heterogeneity, we divide enterprises into distinct groups according to their characteristics, such as ownership, scale, and yearly energy usage. Firstly, we conclude that SOEs are more likely to produce cleanly, and other enterprises may opt to increase the removal of $S{O}_2$. Then, when it comes to the company scale, SMEs (small and micro-companies) are more likely to eliminate more $S{O}_2$, while LMEs (large and medium enterprises) choose to use less coal and more clean gas. In the end, the “Belows” (firms with an annual energy consumption lower than 10,000 tons of standard coal) used less coal and boosted their output per 10 thousand units of coal. The “Aboves” (enterprises with an annual energy consumption of more than 10,000 tons of standard coal) are seeking to increase the removeal of $S{O}_2$ to decrease consumption and may also aim to enhance the consumption ratio of clean gas with an increase in the consumption of coal.

Overall, this article proves to be a useful supplement to the existing literature in three aspects: (1) it demonstrates the emissions reduction effect of pollution levy policies from a new perspective of time trends and provides a new interpretation of the emissions reduction effect of the pollution levy policy; (2) it illustrates the underlying mechanisms in producing less and removing more $S{O}_2$; and (3) it shows that the greater the use of clean energy and higher energy efficiency can result in a decrease in the production of $S{O}_2$, and the utilization efficiency of waste gas treatment facilities can increase the removal of $S{O}_2$. This work proceeds as follows: Section 2. Institutional Background; Section 3. Hypotheses; Section 4. Data and Model; Section 5. Results; Section 6. Conclusions and Recommendations.

2. Institutional Background

The pollutant levy was introduced to OECD countries in the 1970s. China began to explore the construction of a sewage pricing system in the late 1970s. This was first mentioned in the Environmental Protection Work Report Key Points by the Former State Council Environmental Protection Leading Group in December 1978. The following year, the “Environmental Protection Law of The P.R. China (For Trial Implementation)” stipulated that the discharge of pollutants exceeding national standards would be fined according to their concentration and quantity. Until the end of 1981, 27 provinces gradually carried out experimental work on sewage charges, and corresponding policies were issued continuously. In February 1982, the State Council promulgated the “Interim Measures for the Collection of Sewage Discharge Fees”, in which there were extensive regulations on the purpose, objects, charging criteria, management, and usage of sewage charges, etc. This represented the official establishment of the sewer pricing system in China. Relevant policies were created from time to time thereafter, such as the Water Pollution Prevention and Control Law, the revised Environmental Protection Law, Regulations on Cost Management of State-owned Enterprises, Interim Measures for Paid Use of Special Funds for Pollution Source Control, Financial Management, and Accounting Measures for the Collection of Excessive Pollutant Discharge Fees. At the local level, practically all provinces also formulated implementation measures or rules for the collection of sewage charges following national legislation and local conditions, establishing a complete set of legal frameworks for sewage charge systems throughout the country.

With the adoption of the “Environmental Protection Tax Law of the P.R.C.” on 1 January 2018, the “pollutant discharge fee” came to an end. On 12 April 2018, the Ministry of Ecology and Environment’s “Decision on Abolishing Regulations and Normative Documents on Pollution Charges (Draft)” abolished one regulation and 27 normative documents on pollution charges. Among them, the regulations on the “Inspection Measures for the Collection of Sewage Charges” (formerly Decree No.42 of the State Environmental Protection Administration, promulgated on 23 October 2007) were repealed. At the same time, 27 normative documents, such as the notice on the format of common legal documents for the collection and inspection of unified sewage charges (Environmental Protection Office No.19, published on 25 February 2008), were abolished. This means that the pollution charge system implemented in China for nearly 40 years was withdrawn from the historical stage. Reviewing the history of sewage charges for nearly 40 years, the Comprehensive Work Plan for Energy Conservation and Emission Reduction (called “the plan” hereafter), initially announced in 2007, is a remarkable landmark.

In its early phases, the pollution levy standard was calculated according to the average governance cost method but charged half considering the affordability of firms at that time [30,31]. As time went by, the levy standard was significantly lower than the actual cost of pollution management, and this was not conducive to incentivizing firms to reduce emissions [29]. In addition, due to the lack of tight limits on energy conservation and emission reduction, the effect of policy implementation was not very satisfying. Thus, in the plan that was initially announced in 2007, the Chinese government clearly set up the target tasks and overall criteria for energy conservation and pollution reduction during the 11th Five-Year period. Furthermore, adjustments to pollution levy standards were introduced on the agenda. During the 11th, 12th, and 13th Five-Year periods, similar plans were proposed and named the same. After a thorough analysis, we identified three important features of the plan.

One of the main features is the environmental target accountability system. The plan defined a clear aim for approximately the next five years and would be changed practically every five years. The plan issued in 2007 argued that by 2010, energy consumption per ten thousand CNY of GDP would decrease from 1.22 tons of standard coal in 2005 to less than 1 ton of standard coal—a decrease of about 20%; water consumption per unit of industrial added value would be reduced by 30%; and sulfur dioxide emissions would decrease from 2549 million tons in 2005 to 22.95 million tons. Furthermore, total sulfur dioxide emissions, chemical oxygen demand, energy consumption per unit of GDP, and other environmental indicators were introduced in the performance evaluation of local officials, which is also known as implementation accountability and “one-vote veto” system assessment. On the one hand, due to the strong constraint, the breakdown and assessment of environmental protection constraint indicators for a specific year motivate enterprises and local officials to implement measures immediately to reduce the emissions of the current period. Because this happens immediately, we call this the plan’s “short-term” emissions reduction effects. On the other hand, overall requirements throughout the 11th Five-Year period make enterprises and local officials carry out a prognosis on tougher environmental regulations in the future. As a result, they will build a durable effect work mechanism. Since this will last for the entire duration, we name this the “long-term” impact.

Another noticeable element of the strategy is that the Chinese central government constructed a clear road map for the adjustments of pollution levy standards. In more detail, the pollution levy threshold for $S{O}_2$ must be doubled within three years, increasing from 0.63 CNY per kilogram to 1.26 CNY per kilogram. To achieve this need, provinces in China proceeded to gradually change pollution levy standards (as detailed in

Table 1. Pollution levy standard adjustment action among provinces from 2007 to 2013.

Num.	Prov. Name	Carried
1	Jiangsu	Raised to 1.26 yuan in July 2007
2	Anhui	Raised to 1.26 yuan in three years and 0.84 yuan in January 2008
3	Hebei	Raised to 1.26 yuan in two years and 0.96 yuan in July 2008
4	Shandong	Raised to 1.26 yuan in July 2008
5	Neimenggu	Raised to 1.26 yuan in two years and 0.95 yuan in July 2008
6	Guangxi	Raised to 1.26 yuan in two years and 0.95 yuan in January 2009
7	shanghai	Raised to 1.26 yuan in January 2009
8	Yunnan	Raised to 1.26 yuan in two years and 0.95 yuan in January 2009
9	Guangdong	Raised to 1.26 yuan in April 2010
10	Liaoning	Raised to 1.26 yuan in August 2010
11	Tianjin	Raised to 1.26 yuan in December 2010
12	Xinjiang	Raised to 1.26 yuan in August 2012

). Almost all provinces had fully modified the pollution discharge fee standard by the end of 2015 pursuant to the Notice on Relevant Issues Concerning Adjustment of Sewage Discharge Fee Collection Standards. Thus, this paper uses the adjustment of the pollution levy relative to $S{O}_2$ starting from 2007 as an opportunity to analyze the policy’s effects on $S{O}_2$ emissions reduction and demonstrate the underlying mechanistic approaches. Since $S{O}_2$ was one significant indication until the beginning of the 13th Five-Year period, this study largely focuses on the emission of $S{O}_2$.

Other than that, to ensure that the stated targets are reached, several specific steps were included in the plan, such as eliminating some backward production capacity, placing desulfurization units into operation, and other means. Among these measures, both the overall requirements during the 11th Five-Year period and the breakdown and assessment of environmental protection constraint indicators for a single year are also present. In the following sections, we analyze all these connected measures both in the short and long term.

3. Hypotheses

3.1. Pollution Costs

Due to environmental externalities, the core policy is external cost internalization preventing the “tragedy of the commons” [32,33]. Moreover, one clear mechanism of the pollution levy reform comprises modifications of levy standards. Thus, when pollution costs grow, firms need to pay more for the same amount of $S{O}_2$ emissions. As rational actors, corporations will choose the best options between pollution or environmental treatment. The pollution levy standard is one significant indication of pollution costs. Thus, we set a dummy variable, DID, and its value is equal to the situation where the pollution levy standard is raised, while the others are zero. Thus, this leads to the first assumption:

Hypothesis 1.

An increase in pollution levy standards can cause $S{O}_2$ emissions to fall both in the short term and long term.

3.2. Output Adjustment

There is an argument about the relationship between economic development and environmental protection [34]. We examine whether economic development will be affected by the increase in pollution levy rules.

One crucial feature of the ongoing plan is to hasten the reduction in backward manufacturing capacity. Faced with implementation accountability and the “one-vote veto” system assessment, local governments will promptly extrude the living space of the backward production capacity by shutting down, merging, or implementing production transformations. Facing increasing environmental restrictions, one easy approach for various firms is to decrease production and pollute less, but this is considered a fix for the short term and may not be helpful for the long term. Thus, this comes down to consumption:

Hypothesis 2.

Pollution levy policies can affect $S{O}_2$ emissions via production adjustments in the near future; the higher the pollution levy standard, the lower the production hours and the lower the output, which will decrease $S{O}_2$ emissions.

3.3. Terminal Treatment

Another essential component of the plan is to extend input to building the pollution control project [23]. As these effects occur in the short term, this form of terminal treatment would cut emissions quickly. These measures include increasing the amount of pollution abatement equipment, improving the processing capacity of pollution abatement equipment, improving the utilization rate of pollution abatement equipment, strengthening energy conservation and environmental protection management capacity building, accelerating the improvement of energy conservation and emission reduction laws and regulations systems, improving the punishment standard, and effectively solving the problem of “low illegal cost and high law-abiding cost”. Thus, these assumptions can be described as follows:

Hypothesis 3.

Through terminal treatment, enterprises can boost sulfur dioxide removal, thereby reducing their emissions.

Hypothesis 3a.

By expanding the number of waste gas treatment facilities, enterprises can enhance the removal of $S{O}_2$, thereby minimizing their emissions.

Hypothesis 3b.

By improving the processing capacity of waste gas treatment facilities, enterprises can enhance the removal of $S{O}_2$, thereby minimizing their emissions.

Hypothesis 3c.

By boosting the utilization rate of waste gas treatment facilities, enterprises can enhance the removal of $S{O}_2$, thereby minimizing their emissions.

3.4. Front-End Treatment

Last but not least, the plan takes measures to develop advanced production capacities with low energy consumption and low levels of pollution. For example, it develops renewable energy; promotes the utilization of wasted heat and pressure, especially in key energy-consuming industries; accelerates the saving and replacing of oil; and reduces the use of coal, which can reduce pollution emissions significantly [23,35]. Since more advanced technologies and substantial costs are needed, industrial transformation and upgrading take substantially more time and its consequences will not be substantial in the short term but will be in the long term. Thus, we tested the following assumptions:

Hypothesis 4.

Through front-end treatment, enterprises can limit sulfur dioxide production, thereby reducing their emissions.

Hypothesis 4a.

By employing more clean energy, enterprises can minimize sulfur dioxide production, thereby reducing their emissions.

Hypothesis 4b.

By improving energy efficiency, enterprises can reduce sulfur dioxide production, thereby reducing their emissions.

Figure 2: Organization of the study hypotheses.

4. Data and Model

4.1. Data Description

To examine the influence of the pollution levy policy on pollution emissions, we used environmental data at the firm level—The Green Development Database. This provides information on emissions and the environmental governance of Chinese industrial enterprises from 1998 to 2014, collected from the Annual Environmental Survey of Polluting Firms (AESPFs) of China established by the Ministry of Ecology and Environment (formerly the Ministry of Environmental Protection) in the 1980s. The statistical fields mainly include basic enterprise information; production information; water environment and atmospheric environment, covering resource utilization indicators (industrial water consumption and coal consumption), pollution discharge indicators (industrial wastewater discharge and sulfur dioxide discharge), and pollution governance indicators (number of wastewater treatment facilities and the removal of nitrogen oxides); and dozens of other indicators.

Other essential data comprise the China Industrial Enterprise Database. This provides data on industrial enterprises above a designated size from 1998 to 2013 (1998–2006 includes all state-owned industrial enterprises and non-state-owned industrial enterprises with an annual sales income of CNY 5 million and above, and 2007–2010 is adjusted relative to the annual main business income. In 2011, an industrial legal enterprise with an income of CNY 5 million or more was adjusted to an industrial legal enterprise with an annual main business income of CNY 20 million or more). The data originate from the industrial survey and statistics carried out by the National Bureau of Statistics following the “Industrial Statistical Reporting System”. Statistical fields contain two categories—basic information about enterprises, and production and financial information for these enterprises. Among these, the production and financial information of enterprises comprises indicators such as industrial operations, production and sales, employee conditions, balance sheets, profit and loss statements, and cash flow statements.

Thanks to the China Microeconomic Data Query System provided by the EPS database, we easily acquired these two matched firm-level data. For the matching of industrial enterprises and green development enterprises, the organization structure code and enterprise name were used. At the same time, a small amount of matching data was supplemented by the relationship among industrial enterprises–custom enterprises, custom enterprises–green development enterprises, etc..According to existing research [36,37], we cleaned the data and obtained an unbalanced panel dataset for the period from 1998 to 2013, encompassing 164,091 firms and 627,165 observations. The main data-cleaning processes were as follows: (1) observations with a total industrial output value less than CNY 5 million were removed; (2) observations with employee numbers less than eight were excluded; (3) observations with total assets valued less than CNY 1 million were removed; (4) observations with a total debt value of below zero were removed; and (5) observations with below zero $S{O}_2$ emissions, $S{O}_2$ removal, or $S{O}_2$ production were deleted. Since industry concordances changed in 2002 and 2011, we followed a previous method [36] to ensure the uniform coding of the data. In addition, we changed aberrant data records as follows: the year of establishment was fixed and the yearly average number of all employees was replaced with the total employees at the end of the year if there were no data. Moreover, the macro-level data originated from the Chinese Research Data Services Platform (CNRDS).

4.2. Model Setting

Because our outcome variables are censored, we adopt Tobit regression rather than OLS regression [38]. We divided the firms into two groups according to whether the provinces they belong to adjusted the pollution levy standard from 1998 to 2013 (as shown in Table 1). The treatment group included firms from Jiangsu, Anhui, Hebei, Shandong, Neimenggu, Guangxi, Shanghai, Yunnan, Guangdong, Liaoning, Tianjin, and Xinjiang; while the others were in the control group.

Since Shanxi and Heilongjiang only altered the standard of sewage discharge fees for units that have not completed the construction of flue gas desulfurization facilities or units that generate sulfur dioxide levels that exceed requirements, these were not considered here [23].

$$\begin{array}{c}{SO}_{2it}={\beta }_0+{\beta }_1\times POS{T}_{it}\times TREA{T}_{it}+{\beta }_2\times TREN{D}_{it}+{\beta }_3\times POS{T}_{it}\times TREA{T}_{it}\times TREN{D}_{it}+{\beta }_{4}TREN{D}_{it}^2\\ +{\beta }_5\times POS{T}_{it}\times TREA{T}_{it}\times TREN{D}_{it}^2+{\beta }_6\times CO{N}_{FIR{M}_{it}}+{\beta }_7\times CO{N}_{PR{V}_{it}}+{\mu }_i+{ϵ}_{it}\end{array}$$

(1)

where $i=1,2,\cdots ,N$; $t=1998,1999,\cdots ,2013$; ${SO}_{2i,t}$ denotes the $S{O}_2$ emissions of a certain firm i in the year t; $POS{T}_{i,t}$ is a dummy variable that equals 0 for all time periods preceding the year in which the policy took effect, otherwise, it equals 1; $TREA{T}_{i,t}$ is also a dummy variable that takes the value of 1 for all enterprises belonging to the provinces that implemented the pollution levy standard adjustment or, otherwise, takes a value of 0; $POS{T}_{i,t}\times TREA{T}_{i,t}$ is a dependent variable that equals 1 if firm i belongs to the treatment group and time is i after the year of the policy’s enforcement; $TREN{D}_{i,t}=1,2,\cdots ,15$, indicates the time trends. Following existing research [39], we used firms’ starting performances and province-specified characteristics as important control variables; $CON\_FIR{M}_{i,t}$ is a series of performance measurements for firm i in the year t, including the firm’s age (AGE_LAG), the square of the firm’s age (AGESQR_LAG), the average annual number of all employees (LABOR_LAG), labor productivity (LDSCL_LAG), capital-labor ratio (LDSCL_LAG), and asset-liability ratio (ZCFZL_LAG). Firm-level control variables take the effects of the firm’s scale, stage of development, and financial performance into consideration. All these firm-specific variables lag by one period considering endogeneity; $CON\_PR{V}_{i,t}$ represents province-predetermined variables of firm i in the year t, including GDP per capita (PRV_GDP), the intensity of industrial sulfur dioxide emission (PRV_SO2), and the proportion of waste gas treatment investment (PRV_ZLTZ) in 2007; ${\mu }_i$ is the fixed consequence of firm i; ${ϵ}_{i,t}$ is the error term that represents all unobserved factors that influence dependent variables; and ${\beta }_1$ captures the policy’s short-term effect on pollutant discharge while ${\beta }_5$ represents the long-term effect.

Let $X_{i,t}=POS{T}_{i,t}\times TREA{T}_{i,t}$, then Equation (1) is represented as follows:

$$\begin{array}{c}{SO}_{2it}={\beta }_0+{\beta }_1\times X_{it}+{\beta }_2\times TREN{D}_{it}+{\beta }_3\times X_{it}\times TREN{D}_{it}+{\beta }_4\times TREN{D}_{it}^2\\ +{\beta }_5\times X_{it}\times TREN{D}_{it}^2+{\beta }_6\times CON\_FIR{M}_{it}+{\beta }_7\times CON\_PR{V}_{it}+{\mu }_i+{ϵ}_{it}\end{array}$$

(2)

To decrease the impact of outliers, all variables used in this study were indented at the 1% level. At the same time, considering the potential impact of price factors [40], we adjusted the gross industrial output values to the constant price gross industrial output values using the GDP deflator index (using 1998 as the base period) [23]. The summary information is presented in

Table 2. Summary statistics.

Var.Name	Obs.	Mean	SD	Min.	Median	Max.
SO2EM	448,522	7.59634	4.71497	0.00000	9.29477	21.50259
SO2EM_DS	448,454	1.73840	1.67255	0.00000	1.44238	10.70678
SO2PD	448,507	7.75523	4.82679	0.00000	9.45728	21.65425
SO2PD_DS	448,439	1.86367	1.76597	0.00000	1.59764	11.70696
SO2RM	448,503	2.14214	4.32498	0.00000	0.00000	21.03954
SO2RM_DS	448,435	0.44757	1.14645	0.00000	0.00000	11.70032
SO2RM_RIO	340,283	0.09240	0.17982	0.00000	0.00000	0.69315
OUTPUT	461,898	8.28493	1.49004	5.67824	8.06396	19.53413
HOURS	261,601	8.22311	1.02189	0.00000	8.43055	12.54018
COAL	376,530	5.24943	4.10207	0.00000	6.55251	16.76766
FUELCOAL	376,530	4.55967	4.01156	0.00000	5.86079	16.76766
RAWCOAL	336,330	1.13802	3.06387	0.00000	0.00000	15.49817
GAS_RIO	74,473	0.15694	0.35787	0.00000	0.00000	1.00000
PERCOAL_OUTPUT	315,887	2.56880	2.51025	0.00001	1.65496	21.18769
PEROUTPUT_COAL	318,795	0.57103	0.77607	0.00000	0.20587	11.36682
EQUIPMENT	376,530	0.79423	0.81736	0.00000	0.69315	11.61848
CAPBILITY	318,958	5.81813	4.83687	0.00000	7.67694	23.61918
PERCAPBILITY_OUTPUT	240,459	7.52156	1.45060	0.03129	7.40792	18.95431
PEREQUIPMENT_OUTPUT	199,301	0.72601	1.18559	0.00000	0.28683	13.15401
X	462,019	0.21864	0.41332	0.00000	0.00000	1.00000
TREND	462,019	4.14191	3.13671	1.00000	3.00000	16.00000
XTREND	462,019	1.09268	2.70168	0.00000	0.00000	16.00000
TRENDSQR	462,019	26.99434	39.87349	1.00000	9.00000	256.00000
XTRENDSQR	462,019	8.49299	29.49315	0.00000	0.00000	256.00000
NUM_LAG	323,003	3.02 × 104	7.98 × 103	1.71 × 104	3.14 × 104	4.19 × 104
AGE_LAG	322,992	2.36163	0.84163	0.00000	2.30259	6.70196
AGESQR_LAG	322,992	6.28565	4.10583	0.00000	5.30190	44.91627
LABOR_LAG	320,259	5.70074	1.11197	2.19722	5.63835	12.20092
LDSCL_LAG	320,257	1.05115	0.77738	0.00064	0.85800	10.61416
ZBLDB_LAG	320,244	5.23232	1.09683	0.29393	5.14444	15.13424
ZCFZL_LAG	298,916	0.55544	0.21614	0.00000	0.56598	3.94376
PRV_GDP	462,019	1.37 × 107	3.11 × 107	7.88 × 103	1.68 × 106	2.54 × 108
PRV_SO2	462,019	181.17165	421.32075	0.39851	19.06713	9.41 × 103
PRV_ZLTZ	462,019	201.49064	453.03440	0.13757	24.39839	3.70 × 103

Data source: edit by authors.

. Due to the limit of article length, the variable definitions are presented in Appendix

Table A1. Variable Definitions.

Var.Name	Definition
Core variables
SO2EM	$S{O}_2$ emissions
SO2EM_DS	$S{O}_2$ emissions density
SO2PD	$S{O}_2$ production
SO2PD_DS	$S{O}_2$ production density
SO2RM	$S{O}_2$ removal
SO2RM_DS	$S{O}_2$ removal density
SO2RM_RIO	$S{O}_2$ removal ratio
OUTPUT	total industrial output value
HOURS	normal production time per year
COAL	industrial coal consumption
FUELCOAL	consumption of fuel coal
RAWCOAL	consumption of raw coal
GAS_RIO	proportion of clean gas consumption
PERCOAL_OUTPUT	output value per unit of coal; equal to total industrial output value divided by coal consumption
PEROUTPUT_COAL	coal consumption per unit output value; equal to coal consumption divided by the total industrial output value
EQUIPMENT	number of waste gas treatment facilities
CAPBILITY	capacity of waste gas treatment facilities
PERCAPBILITY_OUTPUT	per unit capacity of waste gas treatment facilities for the output value; equal to output value divided by the capacity of waste gas treatment facilities
PEREQUIPMENT_OUTPUT	per amount of waste gas treatment facilities for the output value; equal to output value divided by the number of waste gas treatment facilities
X	treat # post; equal 1 if firm i belongs to the treatment group and time is i after the year of the policy’s enforcement
TREND	indicates the time
XTREND	X#TREND
TRENDSQR	the square of TREND
XTRENDSQR	X#the square of TREND
Control variables
NUM_LAG	number of firms in the specific year
AGE_LAG	firm age
AGESQR_LAG	square of the firm age
LABOR_LAG	average annual number of all employees
LDSCL_LAG	labor productivity
ZBLDB_LAG	capital–labor ratio
ZCFZL_LAG	asset–liability ratio
PRV_GDP	per capita GDP in 2007
PRV_SO2	$\mathrm{intensity} \mathrm{of} \mathrm{industrial} S{O}_2$ emissions in 2007
PRV_ZLTZ	proportion of waste gas treatment investment in 2007

.

5. Results

5.1. Baseline Regression

Table 3. Baseline regression results.

	(1)	(2)	(3)	(4)
X	−1.32585 ***	−1.47584 ***	−1.40652 ***	−1.55763 ***
	(−27.24732)	(−19.58152)	(−28.78677)	(−20.58012)
TREND	−0.07058 ***	−0.09894 ***	0.05816 ***	0.04850 **
	(−12.22964)	(−11.00406)	(6.33276)	(3.27665)
XTREND	0.28838 ***	0.30559 ***	0.32864 ***	0.34026 ***
	(20.87557)	(15.44569)	(23.47931)	(17.01999)
TRENDSQR	−0.00115 *	0.00466 ***	−0.02547 ***	−0.01780 ***
	(−2.42013)	(7.17899)	(−17.80631)	(−9.33563)
XTRENDSQR	−0.01558 ***	−0.01571 ***	−0.01908 ***	−0.01843 ***
	(−16.70070)	(−12.70507)	(−20.00459)	(−14.66366)
_cons	6.34612 ***	4.30911 ***	6.20204 ***	4.16269 ***
	(271.14494)	(33.92776)	(250.75405)	(32.64066)
FIRM_LEVEL	No	Yes	No	Yes
PROVINCE_LEVEL	No	No	Yes	Yes
N	448,522	285,544	448,522	285,544

t statistics in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001.

reports the estimates of the pollution levy standard adjustment effect on $S{O}_2$ emissions. Column (1) does not introduce any control variables; column (2) introduces a set of control variables at the firm level; column (3) adds a set of basic attributes of provinces; and column (4) adds all the above control variables. In all of the columns, the estimates of ${\beta }_1$ in Equation (2) are all notably negative, which demonstrates that $S{O}_2$ emissions drop in the short term after an increase in the pollution levy. Moreover, the estimates of ${\beta }_5$ in Equation (2) are also strongly negative, which suggests that $S{O}_2$ emissions will also fall in the long term with adjustments in the pollution levy standards. These findings support Hypothesis 1. Furthermore, we can observe that the absolute values of the estimates of ${\beta }_1$ are greater than those of ${\beta }_5$. This reflects that the reform’s short-term influence is substantially greater than its long-term influence. In other words, enterprises are more likely to implement short-term but not long-term steps to reach the emissions reduction goal. This reveals the fact that the reform is more reliant on command–control measures than market means.

5.2. Robustness Analysis

In the earlier analysis, we controlled the underlying factors that would affect the adoption of pollution levy standard adjustment. Next, we changed the model set, replaced the dependent variable, and used selected observations for robustness analysis. Re-examination showed that the results were still robust, as detailed in the following sections.

5.2.1. Change Model

In

Table 4. Robustness analysis results—change model.

	(1)	(2)	(3)	(4)	(5)
X	−1.55763 ***	−0.38182 ***	−0.05508 ***	−1.44792 ***	−0.51925 ***
	(−20.58012)	(−4.86197)	(−5.08941)	(−10.89825)	(−5.05200)
TREND	0.04850 **	−0.08964 ***	−0.01087 ***	0.78918 ***	−0.10526 ***
	(3.27665)	(−6.05681)	(−5.92219)	(28.27472)	(−5.83949)
XTREND	0.34026 ***	0.09228 ***	0.01165 ***	0.32566 ***	0.11014 ***
	(17.01999)	(4.09455)	(4.01420)	(7.79184)	(3.94449)
TRENDSQR	−0.01780 ***	−0.00610 **	−0.00081 ***	−0.09400 ***	−0.00791 ***
	(−9.33563)	(−3.18853)	(−3.38204)	(−24.36129)	(−3.36641)
XTRENDSQR	−0.01843 ***	−0.00626 ***	−0.00074 ***	−0.01788 ***	−0.00711 ***
	(−14.66366)	(−4.10938)	(−3.93217)	(−6.31488)	(−3.86823)
_cons	4.16269 ***	4.64388 ***	1.85344 ***	5.56599 ***
	(32.64066)	(21.47558)	(68.50178)	(27.73420)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes
N	285,544	265,476	229,978	285,544	285,544

t statistics in parentheses; ** p < 0.01, *** p < 0.001.

, column (1) is the baseline model that incorporates all control variables; and in columns (2) to (5), we substitute the baseline model with Reghdfe, PPML, Tobit, and Pantob, respectively. It is evident that both the estimates of ${\beta }_1$ and ${\beta }_5$ are notably negative, which suggests strong robustness.

5.2.2. Change Variable

In columns (1) to (5) in

Table 5. Robustness analysis results—change variable.

	(1)	(2)	(3)	(4)	(5)
X	−0.52054 ***	−0.11482 ***	−0.07622 ***	−0.42172 ***	−0.17266 ***
	(−19.92314)	(−4.20976)	(−4.03907)	(−9.78835)	(−4.36836)
TREND	0.03289 ***	−0.01368 **	−0.00614 *	0.28634 ***	−0.01572 *
	(6.37874)	(−2.63521)	(−2.16872)	(29.78494)	(−2.31094)
XTREND	0.10809 ***	0.02279 **	0.01412 **	0.08704 ***	0.03014 **
	(15.57057)	(2.93702)	(2.76368)	(6.44616)	(2.80375)
TRENDSQR	−0.00687 ***	−0.00273 ***	−0.00172 ***	−0.03325 ***	−0.00376 ***
	(−10.31276)	(−4.18088)	(−4.65216)	(−25.71103)	(−4.33794)
XTRENDSQR	−0.00563 ***	−0.00141 **	−0.00097 **	−0.00505 ***	−0.00181 *
	(−12.84951)	(−2.72262)	(−2.89402)	(−5.53692)	(−2.55598)
_cons	2.92689 ***	2.52002 ***	1.43079 ***	3.69338 ***
	(67.12247)	(34.99733)	(36.38956)	(53.02301)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes
N	285,514	265,443	229,945	285,514	285,514

t statistics in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001.

, the regression results of Xttobit, Reghdfe, PPML, Tobit, and Pantob are provided, respectively. Different from Table 4, we replaced the outcome variable SO2EM with SO2EM_DS. The estimates of ${\beta }_1$ and ${\beta }_5$ are all significantly negative as well, which provides support for baseline regression.

5.2.3. Subset Observations

In

Table 6. Robustness analysis results—subset observations.

	(1)	(2)
	SO2EMxttobit	SO2EM_DSxttobit
X	−0.47518 *	−0.31286 ***
	(−2.23343)	(−3.91760)
TREND	0.33547 ***	0.14514 ***
	(9.75899)	(10.86792)
XTREND	0.18264 *	0.10883 ***
	(2.28549)	(3.57671)
TRENDSQR	−0.02970 ***	−0.01467 ***
	(−5.78776)	(−7.32094)
XTRENDSQR	−0.01807 *	−0.01012 ***
	(−2.57530)	(−3.75263)
_cons	9.94969 ***	5.35771 ***
	(47.13434)	(68.24668)
FIRM_LEVEL	Yes	Yes
PROVINCE_LEVEL	Yes	Yes
N	65,134	65,133

t statistics in parentheses; * p < 0.05, *** p < 0.001.

, we retained the dataset from 2000 to 2011; removed the year 2010 because of poor data quality; only retained manufacturing industries with a two-digit industry code between 13 and 43; removed firms that only existed before 2007 or after 2011; removed firms that always had no or zero $S{O}_2$ emissions; and removed firms with no or zero waste gas treatment facilities [41]. Columns (1) and (2) list the Xttobit regression results of SO2EM and SO2EM_DS, respectively. Notably, we obtained consistent empirical results that show the estimates of ${\beta }_1$ and ${\beta }_5$ are all significantly negative.

5.2.4. Transfer of Pollution across Regions

We modified different model sets and conducted a series of robust analyses to evaluate the emissions reduction effect of pollution levy standard adjustment; however, this approach cannot exclude the transfer of pollution across regions. Indeed, it should be pointed out that China’s environmental regulation causes pollution to be transferred nearby [42]. When local governments in the control group raise the pollution levy standard, high-polluting industries transfer pollution to regions in the contrast group. As a result, $S{O}_2$ pollution in the control group is increased but is decreased in the contrast group.

To investigate this phenomenon, we chose the six major high-polluting industries as a research sample [23]. Theoretically, if the transfer of pollutants occurs, this would first be reflected in the number of enterprises and outputs in this area and then affect the amount of pollution emissions. Therefore, we focused on the six major high-polluting industries in the observations, took the number of enterprises and outputs as the dependent variables, respectively, and conducted a regression analysis to test the existence of pollution transfer [23]. The results are listed in

Table 7. (a) Placebo test-transfer of pollutants. (b) Placebo test-transfer of pollutants (continued).

(a)
	OUTPUT
	(1)	(2)	(3)	(4)	(5)
	xtobit_base	xtobitt_firm	xtobit_prv	xtobit_all	pantob
X	0.33991 ***	0.04201	0.36855 ***	0.05206	0.03175
	(5.65345)	(1.05947)	(6.10189)	(1.30050)	(0.56388)
TREND	0.07823 ***	0.04751 ***	0.02462	0.02426	0.02344 *
	(9.66621)	(6.14674)	(1.90113)	(1.92988)	(2.01689)
XTREND	−0.07767 ***	−0.04761 ***	−0.09235 ***	−0.05297 ***	−0.04692 ***
	(−4.64608)	(−3.35459)	(−5.43663)	(−3.66768)	(−4.39288)
TRENDSQR	−0.00004	−0.00312 ***	0.00875 ***	0.00061	0.00082
	(−0.06656)	(−4.99722)	(4.79125)	(0.34157)	(0.47775)
XTRENDSQR	0.00391 ***	0.00409 ***	0.00537 ***	0.00470 ***	0.00418 ***
	(3.50965)	(4.00008)	(4.68699)	(4.47252)	(5.48025)
_cons	8.53710 ***	0.87783 ***	8.62257 ***	0.90974 ***
	(306.94171)	(17.84720)	(269.43964)	(17.82721)
FIRM_LEVEL	No	Yes	No	Yes	Yes
PROVINCE_LEVEL	No	No	Yes	Yes	Yes
N	23,396	21,254	23,396	21,254	21,254
(b)
	NUM_LAG
	(6)	(7)	(8)	(9)	(10)
	xtobit_base	xtobitt_firm	xtobit_prv	xtobit_all	pantob
X	510.37767 ***	934.03031 ***	341.78895 ***	778.28691 ***	1.45 × 103 ***
	(8.95829)	(16.65826)	(6.00497)	(13.88609)	(9.85380)
TREND	75.93481 ***	20.73803 *	309.64986 ***	232.18968 ***	93.60098
	(8.68337)	(2.28264)	(22.69846)	(16.65618)	(1.38149)
XTREND	199.67490 ***	127.29312 ***	266.78541 ***	193.41690 ***	81.45678
	(11.66637)	(7.26981)	(15.61191)	(11.03806)	(0.92415)
TRENDSQR	8.45307 ***	10.17337 ***	−29.09016 ***	−23.53005 ***	−18.85652 *
	(12.19969)	(14.16342)	(−15.14818)	(−12.14059)	(−2.08358)
XTRENDSQR	−16.25384 ***	−12.14881 ***	−22.53687 ***	−18.51915 ***	−14.24187 *
	(−13.86414)	(−10.05288)	(−19.11733)	(−15.18159)	(−2.20119)
_cons	2.22 × 103 ***	1.04 × 103 ***	1.85 × 103 ***	772.23824 ***
	(90.51050)	(13.22718)	(63.14343)	(9.69179)
FIRM_LEVEL	No	Yes	No	Yes	Yes
PROVINCE_LEVEL	No	No	Yes	Yes	Yes
N	23,398	21,256	23,398	21,256	21,256

t statistics in parentheses; * p < 0.05, *** p < 0.001.

, in which the dependent variable of columns (1) to (5) is the number of enterprises in the six major high-polluting industries, and that of columns (6) to (10) is the output of these industries. To control the fixed effects, we used the Pantob model in columns (5) and (10). We found that in the case of output, the coefficients ${\beta }_1$and ${\beta }_5$ were both significant, as shown in columns (1) to (5). From this, we concluded that economic development and emissions reduction by the six major high-polluting industries were realized. For the number of enterprises, the coefficient ${\beta }_1$ was significantly positive and the coefficient ${\beta }_5$ was significantly negative. This indicated that the number of enterprises in the six major high-polluting industries would increase in the short term but decrease in the long term. This reflects the fact that local governments temporarily allow market participants to enter the high-polluting industries in the early stage, accounting for current economic development, but impose strict control later. As a result, the goal of reducing pollution emissions can be realized without excessive disturbance to economic development. Based on the result of an increase in output and a decrease in the number of enterprises, we conclude that the policy can play an important role in optimizing inventory and controlling increment.

5.2.5. Emissions Reduction Effect on $\mathrm{N}{\mathrm{O}}_{\mathrm{X}}$

Up to this point, we investigated the pollution emissions reduction effect of this policy on $S{O}_2$. One fact is that this policy involves controls on various kinds of pollutants other than $S{O}_2$, and given the existence of co-benefits [43,44], we conducted further analysis using another kind of important pollutant, $N{O}_X$, as the dependent variable. In doing so, we aimed to see whether the adjustment of pollutant levy standards would have different effects on different kinds of pollutants.

The results for $N{O}_X$ are listed in

Table 8. Placebo test—$N{O}_X$EM.

	(1)	(2)	(3)	(4)
X	−0.08719	−0.35203 ***	−0.11113	−0.35226 ***
	(−1.44815)	(−3.65063)	(−1.83884)	(−3.64435)
TREND	0.38606 ***	0.34210 ***	0.46353 ***	0.43531 ***
	(34.58324)	(19.16237)	(28.39036)	(15.94271)
XTREND	−0.04351 *	−0.00520	−0.03746 *	−0.00588
	(−2.32247)	(−0.18956)	(−1.98364)	(−0.21360)
TRENDSQR	−0.01212 ***	−0.01405 ***	−0.02534 ***	−0.02651 ***
	(−14.46449)	(−12.13754)	(−10.90893)	(−8.29976)
XTRENDSQR	0.00279 *	0.00187	0.00210	0.00171
	(2.13404)	(1.06634)	(1.58673)	(0.97079)
_cons	2.90147 ***	−0.28567	2.82610 ***	−0.34253
	(77.67217)	(−1.51943)	(72.03797)	(−1.80994)
FIRM_LEVEL	No	Yes	No	Yes
PROVINCE_LEVEL	No	No	Yes	Yes
N	276,350	174,637	276,350	174,637

t statistics in parentheses; * p < 0.05, *** p < 0.001.

. For the coefficient X, no matter the base-line mode in column (1) or the extant model in columns (2) to (4), ${\beta }_1$ is negative but only significant in columns (2) and (4). This reflects the fact that, in the short term, the policy indeed would have an effect on $N{O}_X$ emissions reduction, but the effect is not as significant as that on $S{O}_2$. For the coefficient XTRENDSQR, ${\beta }_5$ is positive in columns (1) to (4) but only significant in column (1). This means that in the long-term, an emissions reduction effect of this policy on $N{O}_X$would not be expected. Combining the short- and long-term analyses, we can conclude that the policy mainly focuses on the control of $S{O}_2$ in the observed period and may not sufficiently support reductions in other kinds of pollutants. Focusing on the core pollutants, the policy realized the goal of $S{O}_2$ emissions reduction and the Chinese central government subsequently dropped $S{O}_2$from the pollutant control measures during the 14th Five-Year period (2021–2025). In contrast, the regulations on $N{O}_X$ emissions in China have lagged, resulting in higher costs for $N{O}_X$ emissions mitigation [29,45].

5.3. Heterogeneity Analysis

Through our analyses, we confirmed the policy’s emissions reduction effect on $S{O}_2$. However, due to the “endogenous enforcement” [13,14,15] and “local protectionism” [18,19,20] of the pollution levy, institutional, central, provincial, and other enterprises may be faced with different degrees of environmental protection requirements. Thus, we divided enterprises into different categories according to their features to examine if there is heterogeneity.

5.3.1. Firm Ownership

We divided firms into SOEs (state-owned enterprises) and other types to investigate whether there is a difference in ownership. From columns (1) to (6) in

Table 9. Heterogeneity analysis by SOE.

	(1)	(2)	(3)	(4)	(5)	(6)
	SO2PDbySOE = Others	SO2PDbySOE = SOE	SO2PD_DSbySOE = Others	SO2PD_DSbySOE = SOE	SO2RMbySOE = Others	SO2RMbySOE = SOE
X	−1.30690 ***	−0.87047	−0.40619 ***	−0.00506	3.38667 ***	3.25808 *
	(−16.68975)	(−1.89706)	(−14.68838)	(−0.02951)	(10.24409)	(2.17316)
TREND	0.14967 ***	−0.28386 ***	0.07322 ***	−0.09733 ***	0.99477 ***	−0.09313
	(9.12276)	(−7.24111)	(12.53774)	(−6.99514)	(13.44324)	(−0.64926)
XTREND	0.27160 ***	0.15965	0.07933 ***	0.00855	−0.83271 ***	−0.49342
	(13.02888)	(1.49578)	(10.71799)	(0.21885)	(−9.02371)	(−1.35229)
TRENDSQR	−0.02432 ***	−0.00223	−0.00964 ***	0.00052	−0.05061 ***	0.00830
	(−11.59562)	(−0.42431)	(−12.88403)	(0.28048)	(−5.35892)	(0.42978)
XTRENDSQR	−0.01481 ***	−0.00886	−0.00421 ***	−0.00095	0.05689 ***	0.02304
	(−11.22640)	(−1.45581)	(−8.96296)	(−0.43088)	(9.58816)	(1.07348)
_cons	4.47219 ***	1.82220 ***	2.97088 ***	2.06426 ***	−20.70810 ***	−33.05475 ***
	(32.80229)	(4.18539)	(62.30987)	(12.49041)	(−41.00973)	(−23.60216)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
N	252,632	32,901	252,602	32,901	252,628	32,901

t statistics in parentheses; * p < 0.05, *** p < 0.001.

, the estimates of ${\beta }_1$ and ${\beta }_5$ are not significantly positive for SOEs, whereas they are significantly positive for other enterprise types. This indicates that SOEs have a higher probability of clean production, and other enterprises may choose to increase the removal of $S{O}_2$.

5.3.2. Firm Scale

From columns (1) to (6) in

Table 10. (a) Heterogeneity analysis by LME. (b) Heterogeneity analysis by LME (continued).

(a)
	(1)	(2)	(3)	(4)	(5)	(6)
	SO2EMbyLME = Others	SO2EMbyLME = LME	SO2EM_DSbyLME = Others	SO2EM_DSbyLME = LME	SO2PDbyLME = Others	SO2PDbyLME = LME
X	−1.50549 ***	−1.27231 ***	−0.52863 ***	−0.53565 ***	−1.40972 ***	−1.11904 **
	(−19.06111)	(−3.29594)	(−18.53510)	(−4.45420)	(−17.63092)	(−2.84390)
TREND	0.18991 ***	−0.10521 ***	0.08098 ***	−0.03189 ***	0.19938 ***	−0.08790 **
	(10.45238)	(−3.80522)	(12.07513)	(−3.75490)	(10.86967)	(−3.13015)
XTREND	0.29123 ***	0.37837 ***	0.09742 ***	0.14796 ***	0.27047 ***	0.35351 **
	(13.56924)	(3.39259)	(12.37998)	(4.26978)	(12.46996)	(3.11280)
TRENDSQR	−0.02908 ***	0.00235	−0.01121 ***	0.00036	−0.02925 ***	0.00225
	(−12.18112)	(0.70305)	(−12.66393)	(0.35177)	(−12.13954)	(0.66391)
XTRENDSQR	−0.01565 ***	−0.02397 **	−0.00494 ***	−0.00950 ***	−0.01467 ***	−0.02295 **
	(−11.44033)	(−3.25177)	(−9.79262)	(−4.15409)	(−10.62373)	(−3.06038)
_cons	4.40125 ***	3.99897 ***	3.01532 ***	2.60075 ***	4.33904 ***	3.77084 ***
	(31.98816)	(11.18522)	(62.14533)	(23.43609)	(31.08197)	(10.33999)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
N	215,553	69,991	215,523	69,991	215,542	69,991
(b)
	(7)	(8)	(9)	(10)	(11)	(12)
	SO2RMbyLME = Others	SO2RMbyLME = LME	COALbyLME = Others	COALbyLME = LME	GAS_RIObyLME = Others	GAS_RIObyLME = LME
X	3.54852 ***	1.04161	−0.67368 ***	−1.08029 **	0.09553	0.64422 ***
	(10.18279)	(0.84183)	(−5.16220)	(−2.90263)	(1.21681)	(4.76567)
TREND	1.01507 ***	0.64411 ***	0.12844 ***	−0.08137 **	−0.07788 ***	0.00477
	(11.46322)	(6.50057)	(7.49344)	(−3.27238)	(−5.07933)	(0.36158)
XTREND	−0.83733 ***	0.08135	0.10888 **	0.29252 **	0.00222	−0.17134 ***
	(−8.34896)	(0.22092)	(2.84767)	(2.77679)	(0.09122)	(−4.23744)
TRENDSQR	−0.05000 ***	−0.04479 ***	−0.00787 ***	0.00544	0.01106 ***	0.00443 **
	(−4.27466)	(−3.71892)	(−3.41029)	(1.81125)	(5.13315)	(3.06524)
XTRENDSQR	0.05825 ***	−0.01507	−0.00635 *	−0.01862 **	0.00097	0.01118 ***
	(8.96538)	(−0.60644)	(−2.49327)	(−2.71303)	(0.60080)	(4.12293)
_cons	−21.42484 ***	−23.89686 ***	4.17103 ***	2.08363 ***	−3.89540 ***	−3.35235 ***
	(−40.54460)	(−21.60095)	(25.61156)	(6.02808)	(−42.77362)	(−25.96545)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
N	215,538	69,991	155,975	69,991	36,760	19,238

t statistics in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001.

, the estimates of ${\beta }_1$ and ${\beta }_5$ are significantly more negative for SMEs (small and micro-enterprises) than those for LMEs (large and medium enterprises); in columns (7) to (8), these trends are reversed. This indicates that large and medium firms are more likely to produce cleanly, and SMEs are more likely to eliminate more $S{O}_2$. Furthermore, in columns (9) to (10), the estimates of ${\beta }_1$ and ${\beta }_5$ are more significantly negative for LMEs than for SMEs. However, in columns (11) to (12), the estimates of ${\beta }_1$ and ${\beta }_5$ are more significantly positive for LMEs than for SMEs. LMEs prefer to use less coal and more clean gas.

5.3.3. Coal Consumption

We observed that the plan sets clear goals for firms with an annual energy consumption of more than 10,000 tons of standard coal. To test if there is a difference in coal consumption, we first divided firms into two groups according to whether their annual energy consumption exceeded 10,000 tons of standard coal. Then, we replaced the output variable with SO2PD, SO2PD_DS, SO2RM, COAL, GAS_RIO, and PERCOAL_OUTPUT in the baseline model. From the results provided in

Table 11. (a) Heterogeneity analysis by ALLCOAL. (b) Heterogeneity analysis by ALLCOAL (continued).

(a)
	(1)	(2)	(3)	(4)	(5)	(6)
	SO2PDbyALLCOAL = below	SO2PDbyALLCOAL = above	SO2PD_DSbyALLCOAL = below	SO2PD_DSbyALLCOAL = above	SO2RMbyALLCOAL = below	SO2RMbyALLCOAL = above
X	−0.84109 ***	−1.10630 ***	−0.30428 ***	−0.21800 ***	0.85498	4.31420 ***
	(−5.62937)	(−11.07771)	(−6.23379)	(−5.40929)	(1.54873)	(9.54160)
TREND	0.08868 ***	0.39833 ***	0.04423 ***	0.21618 ***	0.55466 ***	1.66840 ***
	(4.21348)	(18.65513)	(6.37433)	(23.32111)	(6.64406)	(15.22645)
XTREND	0.23645 ***	0.13056 ***	0.07894 ***	−0.02239 *	−0.14877	−0.88240 ***
	(5.26964)	(5.26237)	(5.35588)	(−2.14838)	(−0.86875)	(−7.24982)
TRENDSQR	−0.01453 ***	−0.02515 ***	−0.00543 ***	−0.01592 ***	−0.01520	−0.11301 ***
	(−5.59143)	(−9.15796)	(−6.32402)	(−13.29160)	(−1.47289)	(−7.99090)
XTRENDSQR	−0.01572 ***	−0.00757 ***	−0.00502 ***	0.00165 **	0.01380	0.05370 ***
	(−5.18470)	(−5.16365)	(−5.02588)	(2.62773)	(1.17379)	(7.17201)
_cons	8.80580 ***	6.87752 ***	4.67230 ***	4.00518 ***	−12.98488 ***	−27.45026 ***
	(45.23261)	(40.88899)	(73.58914)	(59.90456)	(−20.01411)	(−39.12274)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
N	165,636	119,897	165,634	119,869	165,636	119,893
(b)
	(7)	(8)	(9)	(10)	(11)	(12)
	COALbyALLCOAL = below	COALbyALLCOAL = above	GAS_RIObyALLCOAL = below	GAS_RIObyALLCOAL = above	PERCOAL_OUTPUTbyALLCOAL = below	PERCOAL_OUTPUTbyALLCOAL = above
X	−0.72016 ***	1.28170 ***	0.12446	0.32600 **	0.17811 ***	−0.01131
	(−5.40228)	(5.64122)	(1.71017)	(2.79187)	(3.49655)	(−0.31418)
TREND	0.07870 ***	0.09172 ***	−0.03730 **	0.01231	−0.05241 ***	−0.01461 ***
	(4.36597)	(5.44829)	(−3.27617)	(0.81635)	(−7.00639)	(−3.65387)
XTREND	0.16938 ***	−0.39167 ***	−0.01019	−0.09211 **	−0.04458 **	0.01094
	(4.28435)	(−6.57800)	(−0.45941)	(−2.75483)	(−2.86922)	(1.09271)
TRENDSQR	−0.00978 ***	0.00231	0.00823 ***	0.00327	0.00504 ***	0.00169 ***
	(−4.40074)	(1.06538)	(5.69136)	(1.82626)	(5.44494)	(3.59267)
XTRENDSQR	−0.01023 ***	0.02471 ***	0.00178	0.00708 **	0.00258 *	−0.00088
	(−3.83940)	(6.77410)	(1.21002)	(3.24438)	(2.43888)	(−1.37841)
_cons	7.00927 ***	2.02536 ***	−3.83492 ***	−3.53708 ***	−0.80382 ***	−1.21005 ***
	(40.06828)	(11.08971)	(−44.68790)	(−31.43109)	(−11.87251)	(−35.54343)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
N	165,636	60,330	34,705	21,293	164,319	32,767

t statistics in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001.

, the estimates of ${\beta }_1$ and ${\beta }_5$ in columns (1) to (4) are more strongly negative for the “Aboves” (firms with an annual energy consumption exceeding 10,000 tons of standard coal) than they are for the “Belows” (firms with an annual energy consumption of less than 10,000 tons of standard coal). Moreover, for the “Belows”, the estimates of ${\beta }_1$ and ${\beta }_5$ are significantly negative in column (7) and significantly positive in column (11). This reflects the fact that the “Belows” used less coal and increased production per 10,000 units of coal. In columns (5), (6), (9), and (10), the estimates of ${\beta }_1$ and ${\beta }_5$ for the “Aboves” are more significantly positive than they are for the “Belows”. This indicates that the “Aboves” chose to remove more $S{O}_2$ and enhanced the consumption ratio of clean gas.

5.4. Mechanism Analysis

As detailed above, there is indeed a reduction in $S{O}_2$ emissions after the adjustment of pollution levy standards. As a result, it is natural to consider what caused this drop. Thus, we next assessed the potential mechanism relative to the direction of production adjustments and pollution treatment.

5.4.1. Output Adjustment

We used OUTPUT and HOURS to replace the experimental variable in the baseline model.

Table 12. Mechanism analysis—output adjustment.

	(1)	(2)
	OUTPUT	HOURS
X	−0.72016 ***	1.28170 ***
	(−5.40228)	(5.64122)
TREND	0.07870 ***	0.09172 ***
	(4.36597)	(5.44829)
XTREND	0.16938 ***	−0.39167 ***
	(4.28435)	(−6.57800)
TRENDSQR	−0.00978 ***	0.00231
	(−4.40074)	(1.06538)
XTRENDSQR	−0.01023 ***	0.02471 ***
	(−3.83940)	(6.77410)
_cons	7.00927 ***	2.02536 ***
	(40.06828)	(11.08971)
FIRM_LEVEL	Yes	Yes
PROVINCE_LEVEL	Yes	Yes
N	296,114	164,673

t statistics in parentheses; *** p < 0.001.

lists the regression results. It is evident that the estimates of ${\beta }_1$ and ${\beta }_5$ are notably positive in both columns (1) and (2). This indicates that in confronting tougher environmental regulations, corporations do not opt to decrease production to meet the objective of $S{O}_2$ reductions. That is to say, this reform can achieve the goal of $S{O}_2$ reduction without inflicting damage on economic progress.

5.4.2. Terminal Treatment

In columns (1), (2), and (3) of

Table 13. Mechanism analysis—terminal treatment.

	(1)	(2)	(3)
	SO2RM	SO2RM_DS	SO2RM_RIO
X	2.92034 ***	0.64801 ***	0.18965 ***
	(9.24865)	(8.38643)	(16.10303)
TREND	0.73279 ***	0.17603 ***	0.02169 ***
	(11.24047)	(11.03747)	(8.82725)
XTREND	−0.68178 ***	−0.15402 ***	−0.03995 ***
	(−7.82952)	(−7.20959)	(−12.26609)
TRENDSQR	−0.03038 ***	−0.00805 ***	−0.00022
	(−3.61920)	(−3.91700)	(−0.70216)
XTRENDSQR	0.04708 ***	0.01025 ***	0.00228 ***
	(8.46987)	(7.49946)	(10.96490)
_cons	−21.86345 ***	−4.86893 ***	−0.70524 ***
	(−46.36140)	(−42.33558)	(−41.59938)
FIRM_LEVEL	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes
N	285,529	285,499	220,981

t statistics in parentheses; *** p < 0.001.

, we replaced the measured variable with SO2RM, SO2RM_DS, and SO2RM_RIO, separately. It is evident that the estimates of ${\beta }_1$ and ${\beta }_5$ are significantly positive, which supports the observation that enterprises indeed chose to enhance the removal of $S{O}_2$ with tighter environmental regulation. To further test the terminal treatment’s impact, we substituted the explained variable with EQUIPMENT, CAPBILITY, PERCAPBILITY_OUTPUT, and PEREQUIPMENT_OUTPUT, as shown in

Table 14. Mechanism analysis—impact of terminal treatment.

	(1)	(2)	(3)	(4)
	EQUIPMENT	CAPBILITY	PERCAPBILITY_OUTPUT	PEREQUIPMENT_OUTPU
X	−0.13090 ***	−0.18804	0.28082 ***	0.03406
	(−4.76641)	(−1.04464)	(8.37477)	(0.87399)
TREND	0.01012 **	0.32804 ***	−0.01537 ***	−0.04099 ***
	(2.90580)	(12.79134)	(−3.35022)	(−6.43508)
XTREND	0.05647 ***	0.21376 ***	−0.07049 ***	−0.01760
	(6.86470)	(3.94149)	(−6.81604)	(−1.40899)
TRENDSQR	−0.00211 ***	−0.03634 ***	0.00266 ***	0.00404 ***
	(−4.61068)	(−11.53199)	(4.37391)	(5.01368)
XTRENDSQR	−0.00406 ***	−0.01767 ***	0.00399 ***	0.00114
	(−7.32079)	(−4.82791)	(5.63163)	(1.31284)
_cons	−0.86840 ***	−1.68298 ***	2.29455 ***	−0.42740 ***
	(−27.80002)	(−7.78101)	(64.50032)	(−10.26046)
FIRM_LEVEL	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes
N	225,966	198,501	149,660	129,281

t statistics in parentheses; ** p < 0.01, *** p < 0.001.

. In columns (1) and (2) of Table 14, the estimates of ${\beta }_1$ and ${\beta }_5$ are significantly negative, whereas they are positive in columns (3) and (4). This suggests that the improvement in $S{O}_2$ removal comes from increasing the utilization rate of waste gas treatment facilities and not increasing equipment input. Thus, we can say that the tighter environmental regulations resulted in $S{O}_2$ emission reductions without imposing new obligations on enterprises.

At this point in the article, we have established that the adjustments in pollution levy standards can drive firms to utilize cleaner energy and enhance energy efficiency and the utilization rate of waste gas treatment facilities. All these options assist organizations in achieving the $S{O}_2$ emissions reduction objective both in the short and long term.

5.4.3. Front-End Treatment

In

Table 15. Mechanism analysis—front-end treatment.

	(1)	(2)
	SO2PD	SO2PD_DS
X	−1.44881 ***	−0.44889 ***
	(−18.88599)	(−16.53138)
TREND	0.05724 ***	0.03762 ***
	(3.82427)	(7.07694)
XTREND	0.31561 ***	0.09360 ***
	(15.59723)	(13.03499)
TRENDSQR	−0.01743 ***	−0.00677 ***
	(−9.04489)	(−9.86807)
XTRENDSQR	−0.01725 ***	−0.00501 ***
	(−13.56352)	(−11.08126)
_cons	4.10340 ***	2.82209 ***
	(31.69798)	(61.93446)
FIRM_LEVEL	Yes	Yes
PROVINCE_LEVEL	Yes	Yes
N	285,533	285,503

t statistics in parentheses; *** p < 0.001.

, except for the predicted variable that is replaced with SO2PD and SO2PD_DS, the others are the same as those in the baseline model. The estimates of ${\beta }_1$ and ${\beta }_5$ are both significantly negative. This supports the observation that firms apply methods that decrease $S{O}_2$ production during the process. Furthermore, in

Table 16. Mechanism analysis—impact of front-end treatment.

	(1)	(2)	(3)	(4)	(5)	(6)
	COAL	FUELCOAL	RAWCOAL	GAS_RIO	PERCOAL_OUTPUT	PEROUTPUT_COAL
X	−0.71220 ***	−0.58536 ***	−2.22438 ***	0.20693 **	0.24294 ***	−0.09303 ***
	(−5.85586)	(−3.92972)	(−4.34332)	(3.24544)	(5.18354)	(−6.42284)
TREND	0.01401	−0.07597 ***	0.06921	−0.01580	−0.04103 ***	0.01698 ***
	(1.02605)	(−4.27784)	(1.40322)	(−1.71930)	(−6.33171)	(8.45551)
XTREND	0.14891 ***	0.14476 ***	0.36936 *	−0.03899 *	−0.05750 ***	0.01699 ***
	(4.25321)	(3.30122)	(2.52181)	(−2.06574)	(−4.11609)	(3.93398)
TRENDSQR	−0.00046	0.00888 ***	−0.01619 *	0.00621 ***	0.00422 ***	−0.00119 ***
	(−0.25875)	(3.78932)	(−2.50054)	(5.56312)	(5.34442)	(−4.86914)
XTRENDSQR	−0.00933 ***	−0.00859 **	−0.02326 *	0.00339 **	0.00343 ***	−0.00094 **
	(−4.03866)	(−2.92880)	(−2.43748)	(2.73005)	(3.66109)	(−3.23931)
_cons	3.30951 ***	1.97185 ***	−20.25369 ***	−3.62325 ***	0.60919 ***	0.96011 ***
	(23.20864)	(11.54370)	(−48.60406)	(−52.51816)	(10.15025)	(51.72378)
FIRM_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
PROVINCE_LEVEL	Yes	Yes	Yes	Yes	Yes	Yes
N	225,966	225,966	225,966	55,998	197,086	198,401

t statistics in parentheses; * p < 0.05, ** p < 0.01, *** p < 0.001.

, we replaced the outcome variables in the baseline model with COAL, FUELCOAL, RAWCOAL, GAS_RIO, PERCOAL_OUTPUT, and PEROUTPUT_COAL. It is evident that the estimates of ${\beta }_1$ and ${\beta }_5$ are notably negative from columns (1) to (3), but the reverse is observed in column (4). That is to say, increased clean energy use and decreased coal use take effect. In addition, the estimates of ${\beta }_1$ and ${\beta }_5$ are notably positive in column (5) and negative in column (6), which supports the observation that an improvement in energy efficiency exists.

6. Conclusions and Recommendations

6.1. Conclusions

Focusing on pollution levy standard adjustment as an important environmental instrument, this study provides proof of the win–win outcome of pollution emissions reduction and economic development. In this study, we constructed firm-level panel data by matching industry–enterprise and pollution data from 2000 to 2013 and used reduced form regression with time trends added to investigate the emissions reduction effect of China’s pollution levy standard adjustment both in the short and long term. Our key conclusions are as follows:

Firstly, we discovered that there was a decrease in $S{O}_2$ emissions after the pollution levy standard adjustments. Furthermore, the reduction was significant both in the short and long term, but the short-term effect was significantly greater than the long-term effect. This illustrates that the pollution levy policy makes sense in practice, and the emissions reduction effect is mostly short term.

Next, we proved that $S{O}_2$ reduction mostly comes from two sides—the front-end treatment and the terminal treatment. Furthermore, the effect of $S{O}_2$ reduction mostly results from the culmination of clean energy, process improvement, and an enhancement in energy efficiency. Moreover, we point out that increasing pollution levy norms does not cause a decrease in output, which may be concluded from the increase in the output of per unit pollution emissions. This shows that China realized economic development while protecting the environment.

In addition, there is heterogeneity among different types of firms, as follows: (1) SOEs are more likely to produce cleanly and other enterprises may choose to increase the elimination of $S{O}_2$; (2) LMEs prefer clean manufacturing by utilizing less coal and more clean gas, while SMEs are likely to remove more $S{O}_2$; and (3) they may also try to increase the consumption ratio of clean gas with increasing coal consumption, and firms classified as “Belows” would use less coal and increase their output per 10 thousand units of coal.

6.2. Recommendations

Based on our findings, we conclude that more flexible and stricter environmental policies are needed to accelerate China’s sustainable transformation considering the existence of firm heterogeneity and differences in regional development [43,46].

Firstly, the short-term response and long-term development should be balanced, and a balance between high-quality economic development and emissions reduction should be created. Faced with serious environmental pressure, the Chinese government must set rigorous emissions reduction objectives and conduct a series of practical actions. All these measures, with evident features of campaign-style enforcement, indeed encourage local governments and enterprises to reduce pollution [47,48]. However, if we are only concerned with short-term indicators, market entities tend to be short-sighted and lack motivation for long-term governance. When confronted with multiple mandatory goals that cannot be integrated into the short term, such as economy and environment, local officials will naturally pay close attention to the more favorable one, namely economic development [49,50]. There is little question that disregarding long-term development will not be favorable with respect to truly tackling environmental pollution concerns. Thus, releasing a combination of policy tools focused on both short- and long-term effects is recommended. With respect to this, local governments are expected to dedicate more resources to environmental governance in the long run. Enterprises will also have increased endogenous motivation with respect to being environmentally friendly during the entire production process.

Second, precise strategies should be executed without resorting to the “one size fits all” model, and issues should be addressed in accordance with “pilot before promotion”, which is verified to be successful in practice [48]. In actuality, there are considerable discrepancies in economic development levels and environmental carrying capacity among eastern and western provinces, and different types of firms have major differences in scale and production processes. If we do not consider these aspects in reality, policies will have difficulty obtaining optimal results. Thus, leaving room for policies is vital. Under the unified goal, each province is entitled to vary the implementation time and intensity of policies based on its development status. Different types of firms are also encouraged to investigate green development paths that are fit for themselves according to their ownership, scale, energy consumption characteristics, etc.

To conclude, we want to point out that this study still has some limitations. After finding the emissions reduction effect of China’s pollution levy standard reform, this study attempted to explain this phenomenon based on firms’ increased usage of clean energy, process improvement, and improvement in energy efficiency. However, the factors that make organizations realize the culmination of renewable energy, process improvement, and energy efficiency improvements are still understudied. Further investigations could be performed with the availability of a better database, and we plan to investigate the kinds of specific factors that may lead organizations to choose different types of pollution control measures. After this, we could put forward further recommendations. Furthermore, since climatological $S{O}_2$ exhibits pronounced seasonal and regional variations [51], it is necessary to further analyze the spatial and temporal changes in $S{O}_2$ regimes.