Increase in Industrial Sulfur Dioxide Pollution Fee and Polluting Firms’ Green Total Factor Productivity: Evidence from China

Abstract

Green total factor productivity (GTFP) is an important indicator to measure sustainable development, since it considers both the desired and undesired output of the production process. To evaluate whether and how an environmental policy implemented by the central government of China that doubled the emission fee of industrial sulfur dioxide affected polluting firms’ GTFP, and to evaluate the heterogeneity of its effects, the generalized difference-in-difference model was applied to firm-level production and pollution data. There are three main findings. First, this policy significantly increased polluting firms’ GTFP in pilot provinces, and the magnitude of this impact is equivalent to the gap between the sample median and the 85th percentile of the sample GTFP distribution. Second, this positive impact is mainly caused by polluting firms in eastern pilot provinces and by foreign polluting firms. Third, the mechanism analysis shows that polluting firms in eastern pilot provinces significantly enhanced their innovation in green technology and increased their exports, and they increased the installation of pollution-reducing facilities per unit of output value. This paper provides new and insightful policy implications for environmental protection and sustainable development, especially in developing countries.

1. Introduction

The promotion of the transition of economic development to a sustainable path has always been a key concern globally. The balance between economic growth and pollution control is therefore critical. While emission fees, as the most popular form of pollution control policies, have been proven to be effective in reducing emissions, no consensus has yet been reached on its economic effects, especially in developing countries [1,2,3,4,5].

On the one hand, although several studies have found that emission fees would encourage polluting firms to reduce emissions by focusing on clean production and the optimization of energy usage, and can hence increase their economic revenue, this channel does not exist in emerging economies. The implementation of emission fees would increase polluting firms’ production cost and reduce their profit, making firms cut production and adversely affecting their performance [6,7,8].

On the other hand, the possibility of innovation as a response to emission fees may also exist, since the Porter hypothesis states that polluting firms can benefit from properly designed environmental policies because such polices can induce firms to innovate and reduce the cost of production [9]. There are a great number of studies that empirically test the Porter hypothesis. Studies about emission fees in developed countries find that when the policy is properly designed, the negative effect on the production cost can be offset by the positive effect from innovation, which leads to productivity growth [10,11]. Evidence from developing countries such as the BRICS and Mexico also supports this result [12,13]. However, some studies argue that, although emission fees can stimulate innovation, the direction and extent to which the comprehensive effect of regulation and innovation will affect polluting firms are ambiguous [14,15].

A gap in the extant literature about the impact of emission fees at the firm level is the lack of an appropriate and comprehensive measure that can capture the coordinated development of environmental efficiency and economic performance. In addition, these studies do not distinguish between general innovations aiming at improving economic productivity and green innovations aiming at reducing emissions. Hence, whether and how green technology innovation can be polluting firms’ response to emission fees and the resulting influence on their overall performance in pollution control and productivity growth are still ambiguous.

This paper investigates the firm-level impact of an environmental policy in China which raised the emission fee of industrial sulfur dioxide emissions. This policy was implemented in 2007 by the State Council, when it issued the “official guidance on energy conservation and emission abatement”, which raised the national standard of the emission fee of sulfur dioxide of polluting firms from 0.63 to 1.26 CNY per kilogram. (This policy also required pilot provinces to adjust the pollution fee of COD. However, it did not force different provinces to follow the same standard. Hence, this policy is regarded as more of a regulation of the emission of sulfur dioxide by the extant literature such as [16]). According to this guidance, starting from 2007, 14 pilot provinces had to implement this new standard within 3 years. This paper studies the policy’s impact on firm-level green total factor productivity (GTFP), which is an important indicator that takes into account the undesired output (pollutant) together with the desired output [17]. The empirical analysis seeks answers to the following research questions. Did this policy affect polluting firms’ GTFP? Was the impact heterogeneous across different regions and why? What were the influencing channels of this impact?

This policy served as a natural experiment for which the generalized difference-in-difference (DID) model was a suitable statistical method. Besides the basic regression which explores the causal relationship between this policy and the GTFP of polluting firms in pilot provinces, the heterogeneity of the impact on polluting firms in different regions and of different ownership was also investigated. Three possible mechanisms, including innovation in green technology, exports, and management costs for installing emission control filters, were considered.

To the best of our knowledge, this is the first paper to study the effect of this policy on firms’ GTFP in China, although the evaluation of the effect of environmental policies on sustainable development in China has caught much attention [18,19,20,21]. This paper contributes to the extant literature in the following aspects. First, previous studies about this policy focused on the provincial level or industry level [16], while this paper provides a novel microangle as well as the influencing channel, which can help the evaluation of this policy at the firm level. Second, previous studies on emission fees focused on whether they can stimulate general innovation, including studies where GTFP is used as the measure for the firm-level impact [22], while this paper uses green technology innovation which is a more accurate indicator of polluting firms’ attempts to reduce emissions. Third, as the largest developing country, China’s experience can provide important and practical implications for the practice of similar policies in other countries.

The rest of this paper is organized as follows. Section 2 introduces the empirical strategy. Section 3 presents the regression results and robustness check. Section 4 explores the possible mechanisms through which this policy affected polluting firms’ GTFP and the heterogeneity of its effects. Section 5 concludes.

2. Empirical Strategy

2.1. Data

The empirical analysis combined two datasets. The first was the annual survey of industrial firms of China (ASIF), which includes the general production and financial information of all state-owned and above-scale Chinese manufacturing firms (non-state-owned firms with a sales scale above CNY 5 million before 2011 and above CNY 10 million after 2011). The information obtained from the ASIF in conducting the empirical analysis included the number of workers, the establishment year of the firm, fixed assets, total assets, gross industrial output, profits, revenue, etc.

The second dataset was the environmental survey and reports (ESR) collected by the bureau of environmental protection of China. The bureau conducts an annual survey of a set of firms the main pollutant emissions of which account for more than 85% of the total annual industrial emissions in their respective counties. These firms were the main source of pollution in their respective regions, and they were more affected by environmental regulations than the other firms because of their large scale of emissions. This dataset reports firm-level information including the amount of pollutant emissions such as nitrogen oxide, sulfur dioxide, soot, and industrial dust emissions, as well as the industry code, location, industrial output, and the quantity and capacity of wastewater treatment equipment. The ESR is considered to be the most comprehensive and reliable source of the microenvironmental data of China and has been used in many studies [23,24,25].

Table 1. Variable definitions from the ESR.

Variable	Definition and Measure
SO2	The amount of SO2 emitted by firm in tons.
COD	The amount of COD emitted by firm in tons.
Soot	The amount of soot emitted by firm in tons.
NH3-N	The amount of NH3-N discharged by firm in tons.
Output	The firm’s annual industrial output in multiples of CNY 10,000.
Industry	The two-digit SIC code of the industry to which the firm belongs.
Province	The province in which the firm is registered.
Equipment	The number of wastewater treatment facilities the firm owns.
Capacity	The wastewater treatment capacity of the firm in tons/day.

provides the representative available information in the ESR [19].

The two datasets were merged from 2005 to 2013 by matching the firm ID and firm name and following the classic approach to cleaning the data [26]. Shanxi and Heilongjiang were excluded from the treatment group (pilot provinces) because these two provinces only increased the emission fee for polluting firms that had not completed the construction of desulfurization facilities or had exceeded the emission standard of sulfur dioxide. Xinjiang was also excluded because it is a minority autonomous region. In the rest of the pilot provinces, Jiangsu implemented the policy in 2007; Anhui, Hebei, Shandong, and Inner Mogolia in 2008; Guangxi, Yunnan, and Shanghai in 2009; and Guangdong, Liaoning, and Tianjin in 2010.

2.2. Regression Model

The generalized DID model was applied to explore the causal relationship between the increase in the emission fee and the firm-level GTFP, as in [27].

$${GTFP}_{it}={\beta }_0+{\beta }_1\mathrm{S}{DF}_{it}+{\beta }_2{X}_{it}+{\lambda }_i+{\mu }_{\mathrm{t}}+{\epsilon }_{it}$$

(1)

where i and t are indices for the firm and the year; ${\beta }_0$ is the constant; and $G{TFP}_{it}$ is the GTFP of firm i in year t, which is calculated using the data envelopment analysis (DEA) and the Malmquist–Luenberger (ML) index [21]. Labor and capital were used as the input variables in the calculation, where the input of labor was measured by the number of employees at the end of a year, and the input of capital was measured by the total fixed assets deflated using the fixed asset price index. (In the DEA-ML method, the ML index reflects the growth rate of a firm’s GTFP. The GTFP of a firm in 2005 is normalized to be 1, and the GTFP of this firm from 2005 to 2013 is obtained by multiplying its ML index in the corresponding year.) The output variables were categorized into expected output and undesired output, where the expected output was measured by the industrial added value, and the undesired output included sulfur dioxide emissions, chemical oxygen demand, and smoke emissions.

$SD{F}_{it}$ is a dummy variable, which is equal to 1 if the province firm i locates in is a pilot province in year t; otherwise, it equals 0. ${\beta }_1$ captures the average treatment effect of this policy on a firm’s GTFP.

$X$ is a set of control variables at the firm level and the city level, including (1) scale, which is the enterprise scale measured by the total asset of the firm. According to the environmental Kuznets hypothesis, the enterprise scale affects a firm’s output and emissions; (2) age, which is the age of the firm calculated as the current year minus the year of establishment of the firm plus 1. The corporate life cycle theory states that a firm’s age affects its ability to learn and the technology it adopts, which may affect its GTFP; (3) finacons, which is the financing constraints of the firm, calculated as the ratio of corporate interest expenses to fixed assets. A low level of financing costs indicates that it is easier for a firm to obtain funds for its production, operation, technological improvement, and environmental governance; (4) profitlv, which is the firm’s profitability, calculated as the ratio of the total profit of the firm to its sales revenue. Profitability can reflect the ability of a firm to technologically innovate, which can affect the firm’s GTFP.

The region-level control variables include (1) Lngdp, which is the log of the city’s GDP per capita and (2) Lnpop, which is the log of the population size at the end of a year. The agglomeration of economic activities and population can increase the benefit from the sharing of environmental treatment facilities and from economies of scale, which in turn affect firms’ development in the region. ${\mathsf{\beta }}_2$ is a column vector which captures the average influence of each control variable on the dependent variable.

${\mu }_i$ and ${\lambda }_t$ are the firm and year fixed effects. ${\epsilon }_{it}$ is the error term clustered at the firm level.

Table 2. Summary statistics.

	(1)	(2)	(3)	(4)	(5)
Variables	N	Mean	Sd	Min	Max
GTFP	9270	1.004	0.0455	0.768	2.086
DID	9270	0.295	0.456	0	1
scale	9270	11.71	1.412	7.741	17.21
age	9270	15.71	12.21	1	114
finacons	9270	0.082	0.316	0	25.85
profitlv	9270	0.082	0.078	0.001	0.813
lngdp	9270	12.05	1.266	8.744	16.81
lnpop	9079	6.252	0.540	4.380	9.169
Number of id	1012	1012	1012	1012	1012

presents the summary statistics.

Prior to the basic regression, a test of variance on the independent variables was conducted, and the results are reported in

Table 3. Analysis of Variance.

Source	Partial SS	df	MS	F	Prob > F
Model	0.4873604	7	0.06962291	34.39	0.0000
DID	0.02725859	1	0.02725859	13.46	0.0002
scale	0.07721188	1	0.07721188	38.13	0.0000
age	0.00815721	1	0.00815721	4.03	0.0448
finacons	0.02989518	1	0.02989518	14.76	0.0001
profitlv	0.00973627	1	0.00973627	4.81	0.0283
lngdp	0.26840466	1	0.26840466	132.56	0.0000
lnpop	0.00710947	1	0.00710947	3.51	0.0610
Residual	18.366733	9071	0.00202477
Total	18.854093	9078	0.0020769

. Polluting firms’ GTFP was affected by these independent variables.

3. Regression Results

3.1. Basic Regression

Table 4. Basic regression.

	(1)	(2)
Variables	All	All
SDF	0.0047 *	0.0059 **
	(1.90)	(2.20)
scale		0.0041
		(1.20)
age		0.0002
		(1.36)
finacons		0.0048 **
		(2.17)
profitlv		−0.0165 *
		(−1.69)
lngdp		0.0075 ***
		(6.58)
lnpop		0.0007
		(0.79)
Constant	1.0000 ***	0.8625 ***
	(1030.06)	(19.33)
Firm FE	NO	YES
Year FE	NO	YES
Observations	9270	9079
R-squared	0.007	0.025

Robust t-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

reports the results of the basic regression model (1). The first column reports the result when no control variable is included, and the second column reports the result when all control variables are included.

The results indicate that, compared with polluting firms in the control group, this policy increased the GTFP of firms in pilot provinces by 0.0059 units at the significance level of 5%. (The absolute value of the change in GTFP is usually small. For example, [23] studies the effect of an increase in the strength of environmental regulation on polluting firms’ GTFP. The magnitude of this effect was 0.0049.) The magnitude of this positive impact is equal to the gap between the sample median (which equals 1.0001) and the 85th percentile of the sample distribution (which equals 1.0046), which is considerably large.

The results in column (2) also indicate that firms with more financial resources have higher GTFP, because they can adjust their production process and attempt to innovate at a lower cost. Firms with higher profits have lower GTFP on average, possibly because many of these firms are heavily or moderately polluting firms. Firms located in cities with a higher level of GDP per capita have a higher level of GTFP, which is also consistent with our conjecture.

3.2. Parallel Trend Test

To guarantee the robustness of the estimates of model (1), the sample has to satisfy the parallel trend assumption [28]: before the increase in the emission fee, the GTFP of firms in the provinces of the treatment group and the control group should have similar time trends. Following the classic approach, the following regression model was constructed as in [27]:

$$GTF{P}_{it}={\displaystyle \sum _{k=-5,k\ne -1}^6{\beta }_k{D}_{t_{i0}+k}}+\phi X_{it}+{\lambda }_i+{\mu }_t+{\epsilon }_{it}$$

(2)

where $t_{i0}$ is the year in which the province firm i located in became a pilot province; $D_{t_{i0}+k}$ is a dummy variable indicating whether $t-t_{i0}=k,k=-5,\dots ,5$ except for $k=-1$. The other variables are consistent with the definitions in model (1). When $k<0,{\beta }_k$, it tests the parallel trend assumption; when $k\ge 0,{\beta }_k$, it captures the dynamic effects of this policy. The estimates are plotted in Figure 1.

As shown in Figure 1, polluting firms in the treatment group and the control group followed similar time trends at least five years before the province which the firms located in became a pilot province. The figure also shows that after a pilot province increased the emission fee, the GTFP of polluting firms in this province significantly increased. It is noticeable that the positive impact remained at a relatively high level six years after this policy. Hence, this policy also had a long-term effect in improving firms’ GTFP.

3.3. Propensity Score Matching (PSM)

In order to verify the robustness of the result in column (2) of Table 4, and to mitigate concerns about sample selection in the DID model which may cause provinces with a higher potential for GTFP improvement to be selected as pilot provinces of this policy, the “propensity index matching” (PSM) method proposed by Heckman et al. (1997) [29] for calibration was applied.

A control group was constructed based on the observable control variables which exhibited a similar pattern to that of the treatment group before the treatment, and thereby generated a more reliable causal reference. In particular, the nearest neighbor matching method was used: for each polluting firm in the treatment group, four polluting firms in the control group were selected so that the difference between the propensity scores of the firm in the treatment group and the average of the four firms is the smallest. Most of the samples in the treatment group and the new control group were also in the common value range.

Model (1) was regressed on the PSM sample, and the results are presented in

Table 5. PSM regression.

	(1)	(7)
Variables	FE	FE
SDF	0.0047 *	0.0060 **
	(1.86)	(2.22)
scale		0.0043
		(1.25)
age		0.0002
		(1.33)
finacons		0.0120
		(1.52)
profitlv		−0.0159
		(−1.63)
lngdp		0.0071 ***
		(5.60)
lnpop		0.0008
		(0.83)
Constant	1.0000 ***	0.8633 ***
	(1025.23)	(19.35)
Firm FE	YES	YES
Year FE	YES	YES
Observations	9067	9067
R-squared	0.007	0.025

Robust t-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. The estimated coefficient of the key policy variable SDF is the same in magnitude as its counterpart in Table 4 and is also significant at the level of 5%. Hence, the result in Table 4 is robust.

3.4. Placebo Test

As well as this policy, other policies, events, or random factors may also have affected polluting firms’ GTFP, which may bias the empirical results. In order to investigate the extent to which the results in Table 4 are affected by such factors, a placebo test was conducted [20]. Given that from 2007 to 2010 different pilot provinces implemented this policy in different years, the following procedure was adopted. A year t’ from 2006 to 2009 was randomly selected as the starting year of this policy, and 1, 4, 3, and 3 provinces were also randomly selected as the pseudotreatment group of the policy which implemented the policy in t’ till t’ + 3. This is because from 2007 to 2010 1, 4, 3, and 3 of the 11 pilot provinces implemented this policy. The generalized DID model (1) was then applied to estimate the effect of this policy on this pseudotreatment group. To guarantee the reliability of the placebo test, this procedure was repeated 500 times.

Because the data in this test are randomly generated, the estimated coefficients should be close to zero and insignificant if the result in Table 4 is robust. The distribution of the estimated coefficients is plotted in Figure 2. The red dots represent the estimated coefficients in the 500 regressions, and the vertical red line is the coefficient estimate of the benchmark regression in Table 4. It can be seen that the estimated coefficients concentrate around 0: the mean is −0.00009, and the standard deviation is 0.00247. Hence, the randomly constructed policy shock and the treatment group have no effect on polluting firms’ GTFP. It is also worth noting that the estimated coefficients in Table 4 only intersect with a sufficiently small part of the distribution. To sum up, the significant and positive effect of this policy on polluting firms’ GTFP is not caused by unobserved factors.

4. Heterogeneity

Polluting firms in different pilot provinces may face different strengths of the enforcement of the policy. This is probably due to the highly decentralized political and fiscal system of China and the execution of environmental regulation policies usually being relegated to local governments. Hence, there is some misalignment between the provincial governments’ incentive and that of the central government. In order to stabilize economic growth and employment, some provinces may reduce the strictness of supervision to secretly allow polluting firms to emit above the reported level. They may even subsidize the emission fee to make up the firms’ loss (see https://www.jiemian.com/article/968676.html, accessed on 6 June 2023).

To check whether regional heterogeneity existed, model (1) was regressed on the sub-samples of the eastern and non-eastern provinces of China, and the result is reported in

Table 6. Regional heterogeneity.

	(1)	(2)
Variables	Eastern	Non-Eastern
SDF	0.0063 **	0.0063
	(2.31)	(0.87)
Control	YES	YES
Firm FE	YES	YES
Year FE	YES	YES
Observations	6568	2511
R-squared	0.543	0.345

Robust t-statistics in parentheses; ** p < 0.05.

. The central, western, and northeastern provinces were categorized together as non-eastern provinces for two reasons. First, the sample size of polluting firms in the central, western, and northeastern provinces was relatively small compared with that of the eastern provinces. Second, other than the eastern provinces which have benign market conditions, good infrastructure, a relatively rich stock of human capital, and easy access to the exporting sector, the three other regions on average do not differ significantly in these aspects.

The results show that this policy significantly increased the GTFP of polluting firms in eastern provinces but had no effect on the GTFP of polluting firms in non-eastern provinces. The heterogeneity indicates that polluting firms in eastern pilot provinces may have responded to this policy either by increasing their desired output to fix the undesired output or by managing to reduce their emissions to fix the level of desired output. In the mechanism analysis, the channels through which this positive effect may have been caused will be explored in more detail.

The performance of polluting firms in non-eastern pilot provinces may have been caused by reducing production, whereby both the desired and undesired output decrease. It may also have been caused by secret subsidies for the increased pollution fee or arrears in payment (see https://www.dcement.com/article/200611/39637.html, accessed on 6 June 2023), which would have caused the policy to have no effect on firms’ production decisions.

The next possible heterogeneity to be explored was across different ownership types of polluting firms. Firms with different ownership types have different bargaining powers when facing environmental regulation. State-owned firms usually have connections with the local government, hence can receive preferential treatment in external financing, property rights protection, tax rebates, and market opportunities [18]. These differences can induce firms to respond differently to this policy, which may affect their GTFP. To check this possibility, model (1) was regressed on the sub-samples of state-owned firms, private firms, and foreign firms. The results are reported in

Table 7. Ownership heterogeneity.

	(1)	(2)	(3)
Variables	State-Owned	Private	Foreign
SDF	−0.0019	0.0034	0.0090 *
	(−1.14)	(1.02)	(1.72)
Control	YES	YES	YES
Firm FE	YES	YES	YES
Year FE	YES	YES	YES
Observations	687	5389	1428
R-squared	0.798	0.026	0.016

t-statistics in parentheses; * p < 0.1.

.

The results indicate that this policy significantly increased the GTFP of foreign firms. This is possibly due to the technological advantage of foreign firms in innovation and in the flexibility of production adjustment. In contrast, the advantage of state-owned firms in their connections with local governments may have weakened their incentive for innovation.

5. Mechanism

In this section, the possible channels through which an increase in the emission fee can improve firm-level GTFP are explored.

The first test checked whether this policy affected polluting firms’ green technology innovation. This kind of innovation can increase a polluting firm’s GTFP by decreasing the level of emissions per unit of output.

The extant literature on firm-level innovation usually uses the number of applications for patents or the number of patents granted to proxy a firm’s innovation [30]. To proxy a polluting firm’s performance in green technology innovation, the number of applications for green patents (awarded to green technology innovation) in a given year of a polluting firm was used as the dependent variable in model (1) to test whether innovation is a channel.

Table 8. Innovation.

	(1)	(2)	(3)	(4)
Variables	All	Eastern	Non-Eastern	Intersection
SDF	−0.0286	0.0529	−0.2711	−0.1610 **
	(−0.49)	(1.58)	(−1.62)	(−2.32)
SDFxEast				0.1656 ***
				(3.72)
Control	YES	YES	YES	YES
Firm FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	9079	6568	2511	9079
R-squared	0.645	0.388	0.687	0.645

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

reports the results. Columns (1) to (3) report the results on the full sample and the two sub-samples of eastern and non-eastern provinces. The results of column (4) add an intersection term between the policy and a dummy variable $Eas{t}_i$ indicating whether firm i belongs to an eastern pilot province to further explore whether heterogeneity existed in polluting firms’ innovation across regions.

The results indicate that, compared with polluting firms in the control group in the same region, this policy had a non-significant but positive effect on polluting firms in eastern pilot provinces and a non-significant but negative effect on polluting firms in non-eastern provinces. Both effects are close to the significance level of 10%. The results in column (4) indicate that, compared with polluting firms in the control group, polluting firms in non-eastern pilot provinces significantly reduced innovation, while polluting firms in eastern pilot provinces significantly increased innovation. This pattern is consistent with the fact that eastern provinces have better market conditions, a richer stock of human capital, and easier access to external financing. All these advantages can promote firm-level innovation.

The next possible channel to be explored was whether polluting firms also attempted to reduce their emissions per unit of output by installing emission-reducing facilities.

According to previous studies, costs related to a firm’s effort of emission abatement are categorized under “management costs”, including the purchase of facilities to reduce pollution [31]. To proxy for the firm’s expenditure on such facilities per unit of output value, the management cost per unit of output value of a firm was used to replace the dependent variable in model (1). Column (1) of

Table 9. Management cost per unit of output value.

	(1)	(2)	(3)
Variables	All	Eastern	Non-Eastern
SDF	0.0018	0.0055 ***	−0.0044
	(1.06)	(3.49)	(−0.86)
Control	YES	YES	YES
Firm FE	YES	YES	YES
Year FE	YES	YES	YES
Observations	9079	6568	2511
R-squared	0.637	0.677	0.588

Robust t-statistics in parentheses; *** p < 0.01.

reports the results of the full sample, and columns (2) and (3) report this regression on the sub-sample of firms in the eastern provinces and in other provinces.

Obviously, the management cost per unit of output value of polluting firms in eastern pilot provinces significantly increased. However, the management cost per unit of output value of polluting firms in non-eastern pilot provinces was not affected. Hence, polluting firms in eastern pilot provinces increased their expenditure on direct pollution control, which would have led to the increase in their GTFP, ceteris paribus.

Finally, the possible channel of exports was explored. The extant literature has long found that exporting is an effective way to improve firms’ productivity because of the benefits from possible externalities of competition in global markets [32,33]. Previous studies also found that firms that export have on average a lower emission intensity [34]. In the case of Chinese polluting firms, exporting to developed countries with generally more strict environmental regulations imposed on products may promote Chinese exporting firms to devote more effort to emission abatement. The results are presented in

Table 10. Exports.

	(1)	(2)	(3)
Variables	All	Eastern	Non-Eastern
SDF	0.1472 **	0.3071 ***	−0.4376 ***
	(2.19)	(4.18)	(−2.77)
Control	YES	YES	YES
Firm FE	YES	YES	YES
Year FE	YES	YES	YES
Observations	3676	2959	717
R-squared	0.844	0.853	0.839

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

.

Obviously, this policy increased the exports of polluting firms in pilot provinces. However, this result was mainly caused by polluting firms in eastern pilot provinces. In sharp contrast, polluting firms in non-eastern pilot provinces decreased exports.

From the mechanism analysis, it can be concluded that the increase in the emission fee of sulfur dioxide increased the GTFP of polluting firms in eastern pilot provinces through three channels: an increase in the innovation activities for developing green technology, an increase in the installation of emission-reducing facilities, and an increase in exports. However, this policy did not have any effect on polluting firms in non-eastern pilot provinces through any of these three channels, which may have led to the insignificance of the policy’s impact on their GTFP in Table 6.

6. Concluding Remarks

In this paper, the firm-level panel data in the ASIF and ESR from 2005 to 2013 were used to investigate the impact of the increase in the emission fee of sulfur dioxide. The results indicate that, first, this policy significantly increased polluting firms’ GTFP in pilot provinces. Second, the impact was heterogeneous across regions. In particular, the GTFP of polluting firms in eastern pilot provinces significantly increased, while polluting firms in non-eastern pilot provinces were not affected. Third, the mechanism analysis further shows that the positive firm-level impact on eastern pilot provinces worked through three channels: green technology innovation, installation of emission-reducing facilities, and exports. These results enrich the literature on the evaluation of environmental policies, especially in developing countries at the firm level.

The results also provide three policy implications. First, central or local governments can subsidize polluting firms’ green technology innovation to strengthen their incentive. Because R&D investment is risky, it may not be the priority of polluting firms located in regions with relatively poor conditions for innovation. Considering the regional heterogeneity, the central government can also promote green technology transfer across regions. Second, while the importance of increasing exports has long been realized, our results show that exporting can also help increase polluting firms’ effectiveness in improving firm-level GTFP. Therefore, there is also the benefit of promoting exports to the environment. Third, the central government may increase the level of supervision by installing monitoring equipment that is directly managed by the Ministry of Ecology and Environment to mitigate the misalignment between local governments’ incentive and that of the central government.