Assessing the abatement potential and cost of Chinese industrial water pollutants

China is suffering escalating negative consequences from environmental pollution, e.g., increasing respiratory diseases and mortality (Zhang et al., 2010) and water-related food safety risk (Jawahar & Ringler, 2009). Water pollution is less visible but has emerged as more threatening over time (World Bank 2007; Wu et al., 2015). Water pollution has a direct and negative impact on public health (more sick days, more cancer, etc.), the ecological system (more pollution in developed areas) and economic development (tourism, fisheries, hunting, etc.) (Stonich, 1998; Hu & Cheng, 2013; Wang & Yang, 2016). Taking the health impact for example, Wang & Yang (2016) reveal that water pollution in China has imposed significantly negative influence on health outcomes. Moreover, they suggest that the common pollutants in industrial wastewater have differential impacts on health outcomes. Dwight et al. (2005) investigate the health cost of four kinds of diseases due to exposure to polluted coastal water. They find that the average economic burden for the residents of Orange County in California will increase by about $44.63 per person. Besides, stated preference analysis (e.g. hedonic method) is also widely adopted to estimate the environmental externalities of urban river pollution or the impacts of pollution (Guignet, 2012; Chen, 2017).

In 2012, the total discharge of wastewater in China reached 68.5 × 10⁹ tons which is, in volume terms, comparable to the annual flow of the Yellow River of 58 × 10⁹ m³ per annum (National Bureau of Statistics, 2014). According to the Ministry of Environmental Protection (MEP) of China, 61.5% of groundwater monitoring sites do not meet the average water quality standard. Among the 968 sections of 423 major rivers and lakes under national monitoring, 63.1% of sections had water quality ratings from Grade I to III, 27.7% from Grade IV to V, and 9.2% Grade V and below¹ (MEP, 2014, 2008). The most important and most thoroughly collected indicators of wastewater in China are the chemical oxygen demand (COD) and ammonia nitrogen (NH₄), which are also the two main objects of government regulation.

To cope with the challenge, the Chinese government has issued and implemented many water policies in the past several decades. The strictest and most effective way to achieve a mitigation target for water pollutants is implementation of a total emission control policy for wastewater. The policy was first introduced in 1988 with the efforts of the Environmental Protection Bureau of China (the former body of MEP) to implement the ‘Interim Measures for the Administration of Water Pollution Discharge Permits’. Until 1996, when the State Council of China promulgated the ‘Decision on Several Issues Concerning Environmental Protection’, the quantity of wastewater was targeted at a certain level and each region was required to meet a basic concentration standard for wastewater. The policy has recently been shifted to control the quantity of specific water pollutants according to the environmental carrying capacity of water.

During the 10th Five-Year-Plan (FYP, 2001–2005), China planned to mitigate specific pollutants and set a non-binding target of a 10% reduction of national SO₂ and COD emissions. However, emissions of SO₂ and soot increased by 8–15% in 2003 and most indexes of environmental protection failed to meet the target. During the 11th FYP (2006–2010), the central government committed to a mandatory target to cut 10% of national SO₂ and COD emissions. This binding goal was also distributed among provinces. At the end of the 11th FYP, the country as a whole accomplished a 12.45% reduction of COD and a 14.29% reduction of SO₂. In the 12th FYP (2011–2015), China has continued a policy of reducing the main pollutants and added two more pollutants, NH₄ and NOx, to the policy. The corresponding national abatement targets for NH₄ and NOx are set at 10%, while for SO₂ and COD the targets are 8% nationwide.

Although the government has made great efforts in response to these urgent practical needs, the subject has attracted less attention from the academic community. Environmental scientists usually focus on the driving forces of water pollutants. Nevertheless, assessment studies of water pollutant reduction potential and costs are rare. Most assessment studies have assessed the potential to reduce the use of energy or reduce emissions of air pollutants (Price et al., 2011; Wei et al., 2015). Some studies estimate the potential water savings for the agricultural sector (Ensink et al., 2004; Hongyun & Liange, 2007; Christian-Smith et al., 2012; Bassi, 2014). Recently, Ng (2015) estimated the cost of seawater flushing for 15 major coastal cities. However, to the best of our knowledge, few studies have investigated the feasible reduction potential and marginal cost to cut industrial water pollution in China. This paper aims to fill this gap.

Our paper focuses on industrial COD and NH₄ emissions, which are controlled by the 11th FYP and the 12th FYP, respectively. Using data covering the 10th, 11th and 12th FYP (2003–2012), we investigate dynamic change and spatial variation in terms of abatement potential and abatement cost, which conveys important policy implications. We employ environmental production theory and a directional distance function (DDF) to model the multi-input and multi-output structure and the simultaneous production of pollutants. This approach has a flexible structure that enables researchers to derive interesting results, such as the marginal cost of abating various pollutants (Färe et al., 2005, 2006). Our results show that 13.18% of national COD and 13.27% of national NH₄ are over-emitted due to inefficiency². For the period 2003–2012, the average cost to abate one additional unit of COD was 710 Yuan/kg and the average cost to abate an additional unit of NH₄ was 7,390 Yuan/kg. We further investigate the determinants of the abatement cost using a fixed-effect panel estimation. Our estimation indicates that the abatement cost is a function of the economic development level, pollution intensity and capital-labor ratio. We find that pollution abatement policies affect abatement costs through various channels. Moreover, our results suggest that the abatement of COD is becoming more costly and difficult, while reduction of NH₄ presents ‘win-win’ opportunities.

The paper is presented as follows. Section 2 introduces the model and presents the empirical specification. Section 3 presents the data and variables. Section 4 reports the main results and Section 5 presents a further discussion. The conclusion and policy implications are presented in Section 6.

Two major industrial water pollutants (COD and NH₄) are considered as byproducts associated with an industrial production process. The inputs (labor, capital stock, coal and petrol) and desirable output data can be collected from the official statistical yearbooks. The National Bureau of Statistics (NBS) has published industrial water pollutant data since the beginning of the 10th Five-Year-Planning (FYP) period; data for COD is available starting in 2001 and data for NH₄ starting in 2003. Our data covers 30 mainland provinces and 10 years (2003–2012). All the data is collected from the China Statistical Year Book (NBS, various years), China Energy Statistical Year Book (NBS, various years), China Water Resources Bulletin (MEP, various years) and China Environment Yearbook (NBS, various years).

In our case, the desirable output (y) is represented by the industrial value added; the capital input (x₁) is measured by the net value of fixed assets; the proxy of labor (x₂) is the annual average number of employees; and the energy input (x₃) is tons of total coal and petroleum products. The two bad outputs are tons of emissions of COD and NH₄ in the industrial wastewater. The monetary value is deflated to the 1998 constant price level. Tibet is excluded due to unavailability of data. The descriptive statistics are summarized in Table 1.

In order to avoid the convergence problem when solving Equations (3) and (5), we normalize the input and output vectors by their mean values (Färe et al., 2005). This indicates that the decision-making unit uses the mean input to produce the average output. Table 2 reports the estimated parameters of Equation (5), which are obtained by solving the linear programing using the General Algebraic Modeling System. There are in total 300 observations between 2003 and 2012 for 30 mainland provinces. However, the environmental production technology and the DDF must fulfill the requirement of null-jointness. It implies that the observation (y,0) is not in P(x) if D(x, y, 0) < 0. As a result, 10 out of 300 observations violate this condition and are dropped. Moreover, eight observations are further dropped because of a zero denominator when estimating the shadow price. Finally, 282 observations are kept for further analysis.

Reading from Table 3, we can find that the mean value of the DDF is 0.106 during the whole period. This indicates that the industrial value-added could be expanded by 0.106*365.8 = 38.77 × 10⁹ Yuan, while COD and NH₄ emissions could be reduced by 0.106*155.1 = 16.44 thousand tons and 0.106*11.7 = 1.24 thousand tons, respectively, for a hypothetical province. The score of β starts increasing from 0.092 in 2003 to 0.139 in 2006, then decreases significantly until 2009; thereafter, it shows a clear upward trend. Because β measures the distance and serves as an inefficiency measurement, we can conclude that the emission performance of industrial COD and NH₄ got worse during 2003–2005, but the turning point appeared in 2006, when the 11th FYP set up the mandatory emission reduction targets. Emission performance improved for the period 2006–2009 but began turning worse in 2010³.

Table 3.

Abatement potential, shadow price and Morishima elasticity of industrial COD and NH₄.

	2003–2012	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012
Abatement potential
Mean of β (Std. Dev)
	0.106 (0.126)	0.092 (0.099)	0.096 (0.111)	0.139 (0.177)	0.119 (0.153)	0.105 (0.134)	0.067 (0.075)	0.066 (0.072)	0.099 (0.085)	0.133 (0.143)	0.150 (0.162)
COD: Average abatement quantity (tons)
	21,240	20,982	23,480	37,531	29,807	21,458	11,915	11,875	18,003	17,278	20,448
COD: National abatement potential (%)
	13.18	11.6	13.3	20.8	16.5	12.9	7.8	8.1	11.9	13.4	16.3
NH4: Average abatement quantity (tons)
	1,661	1,699	2,041	3,571	2,474	1,542	738	702	1,009	1,330	1,536
NH4: National abatement potential (%)
	13.27	12	14	22	17.5	13.6	7.5	7.7	10.4	12.9	15.7
Shadow price (unit: 108Yuan/ton)
COD: Mean of (Std. Dev)
	0.0071 (0.0110)	0.0041 (0.0017)	0.0045 (0.0021)	0.0044 (0.0024)	0.0074 (0.0167)	0.0049 (0.0041)	0.0064 (0.0070)	0.0084 (0.0119)	0.0083 (0.0101)	0.0099 (0.0138)	0.0131 (0.0206)
NH4: Mean of (Std. Dev)
	0.0739 (0.2203)	0.0334 (0.0541)	0.0410 (0.0637)	0.0501 (0.0901)	0.1210 (0.4446)	0.0469 (0.0590)	0.0873 (0.2378)	0.1152 (0.3621)	0.0595 (0.0812)	0.0922 (0.1811)	0.0884 (0.1675)
Morishima elasticity
COD: Mean of (Std. Dev)
	− 0.0899 (0.0845)	−0.0448 (0.0398)	−0.0511 (0.0421)	−0.0665 (0.0650)	−0.0854 (0.0921)	−0.0901 (0.0779)	−0.0949 (0.0888)	−0.1004 (0.1098)	−0.0978 (0.0538)	−0.1365 (0.1003)	−0.1378 (0.0983)
NH4: Mean of (Std. Dev)
	0.0843 (0.2473)	0.0998 (0.2200)	0.0314 (0.0337)	0.0697 (0.1458)	0.0548 (0.1392)	0.0460 (0.0631)	0.0751 (0.1926)	0.1547 (0.5642)	0.1269 (0.2564)	0.1180 (0.2656)	0.0661 (0.1219)

	2003–2012	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012
Abatement potential
Mean of β (Std. Dev)
	0.106 (0.126)	0.092 (0.099)	0.096 (0.111)	0.139 (0.177)	0.119 (0.153)	0.105 (0.134)	0.067 (0.075)	0.066 (0.072)	0.099 (0.085)	0.133 (0.143)	0.150 (0.162)
COD: Average abatement quantity (tons)
	21,240	20,982	23,480	37,531	29,807	21,458	11,915	11,875	18,003	17,278	20,448
COD: National abatement potential (%)
	13.18	11.6	13.3	20.8	16.5	12.9	7.8	8.1	11.9	13.4	16.3
NH4: Average abatement quantity (tons)
	1,661	1,699	2,041	3,571	2,474	1,542	738	702	1,009	1,330	1,536
NH4: National abatement potential (%)
	13.27	12	14	22	17.5	13.6	7.5	7.7	10.4	12.9	15.7
Shadow price (unit: 108Yuan/ton)
COD: Mean of (Std. Dev)
	0.0071 (0.0110)	0.0041 (0.0017)	0.0045 (0.0021)	0.0044 (0.0024)	0.0074 (0.0167)	0.0049 (0.0041)	0.0064 (0.0070)	0.0084 (0.0119)	0.0083 (0.0101)	0.0099 (0.0138)	0.0131 (0.0206)
NH4: Mean of (Std. Dev)
	0.0739 (0.2203)	0.0334 (0.0541)	0.0410 (0.0637)	0.0501 (0.0901)	0.1210 (0.4446)	0.0469 (0.0590)	0.0873 (0.2378)	0.1152 (0.3621)	0.0595 (0.0812)	0.0922 (0.1811)	0.0884 (0.1675)
Morishima elasticity
COD: Mean of (Std. Dev)
	− 0.0899 (0.0845)	−0.0448 (0.0398)	−0.0511 (0.0421)	−0.0665 (0.0650)	−0.0854 (0.0921)	−0.0901 (0.0779)	−0.0949 (0.0888)	−0.1004 (0.1098)	−0.0978 (0.0538)	−0.1365 (0.1003)	−0.1378 (0.0983)
NH4: Mean of (Std. Dev)
	0.0843 (0.2473)	0.0998 (0.2200)	0.0314 (0.0337)	0.0697 (0.1458)	0.0548 (0.1392)	0.0460 (0.0631)	0.0751 (0.1926)	0.1547 (0.5642)	0.1269 (0.2564)	0.1180 (0.2656)	0.0661 (0.1219)

As shown in Table 4, we compare our results with related studies. The results vary greatly and depend on the sample data, pollutant, model and estimation strategy. For example, Wang & Lall (2002) set up a cost function in trans-log form and estimate the marginal abatement cost of water pollution based on plant-level data from 1,500 Chinese industrial firms in 1994. They find that the marginal abatement cost of COD discharge for large plants is about 582.3 Yuan per ton, and that it is much more expensive for medium-sized and small firms to reduce COD; their costs are 775.6 Yuan per ton and 1,652.3 Yuan per ton, respectively. Similarly, Dasgupta et al. (2001) also employ a trans-log specification to accommodate the cost function for 260 Chinese plants. Their econometric estimation suggests that the charge for COD emissions should be set at 20 $/ton when the abatement rate is 93%. Differing from the econometric approach, Wang et al. (2015) use the nonparametric Data Envelopment Analysis (DEA) method to evaluate the shadow price of COD and NH₄-N for China's industrial sectors. They find that the average abatement costs for COD in 2009 and 2010 reach 69.26 Yuan/kg and 65.14 Yuan/kg, respectively. Compared to COD, the shadow price for NH₄-N is much higher. It reaches 1,646.22 Yuan/kg in 2009 and 1,446.43 Yuan/kg in 2010.

Beyond these attempts, two studies are close to our methods. Kumar & Managi (2011) use a parametric DDF to model the shadow price. Using data for 92 Indian water-polluting firms, they find the cost to cut one additional ton of COD emission ranged from 20,000 to 170,000 Rupees. Another similar study is by Tang et al. (2016). They also employ the parametric DDF to examine the marginal abatement cost of COD, total nitrogen (TN) and total phosphorus (TP) emissions for the Chinese agricultural sector. Their results indicate the average shadow prices of COD, TN and TP from 2001 to 2010 are 8,266 Yuan/ton, 25,560 Yuan/ton and 10,160 Yuan/ton, respectively. In the present study, we adopt a method similar to that of Tang et al. (2016) but focus on the industrial sector, which has higher productivity and greater opportunity cost compared with the agricultural sector.

Besides the abatement potential and abatement cost, we also explore the complex substitution nexus between desirable and undesirable outputs. The Morishima elasticity of substitution in Equation (7) measures how the relative price ratio responds to the change in intensity between pollutants and industrial value-added. A negative sign of Morishima elasticity reflects a trade-off relationship. By contrast, a positive coefficient suggests ‘win-win’ opportunities because the reduction of the pollutant is associated with the expansion of industrial value-added (Kumar & Managi, 2011).

The Morishima elasticity of COD also exhibits an upward trend in terms of its absolute value, which suggests that it is getting more difficult and expensive to further reduce COD emissions. However, the Morishima elasticity of NH₄ does not show a clear trend and, instead, fluctuates over time.

The great provincial disparity and time-varying trends of abatement costs further motivate us to explore their driving forces. In the light of previous literature, we account for the following possible factors.

• 1. Pollutant intensity. As our correlation tests show, pollutant intensity is strongly correlated with abatement cost, which is in line with the theory of the marginal abatement cost curve (Murty et al., 2007). A higher value of COD intensity means a ‘higher pollution level’ for one unit of economic output. Consequently, it is much easier and cheaper to reduce one unit of additional emission. Thus, we expect a negative nexus between the pollutant intensity and its abatement cost. The COD intensity is measured with the COD emission over industrial value added. Similarly, the NH₄ intensity is expressed as the NH₄ emission per unit of industrial value added.

• 2. Economic development level. In order to control pollution, a decision-maker has to either switch their inputs to abatement activities, or reduce their output to reduce the undesirable by-products. In both cases, the pollution abatement activities will move the potential output frontier to a lower level. Thus, the abatement cost measures the opportunity cost when controlling pollution. One would expect that the developed provinces would suffer greater potential economic loss, and thus experience high abatement costs. In our case, the gross domestic product (GDP) per capita (RGDP) is adopted to indicate the economic development level. We also introduce its squared term to test the nonlinearity effect. In accordance with the Environmental Kuznets Curve hypothesis, the abatement cost may first experience a decline and then increase, which in turn shapes a U curve. In that case, the coefficient of RGDP is expected to be negative while the squared term's coefficient is positive.

• 3. The capital intensity. The capital-labor ratio is a good indicator to measure the capital intensity. It can be used to proxy whether the industrial sector is capital-intensive or labor-intensive. Capital is a complement to intermediate materials and energy in the short run (Griffin & Gregory, 1976). The province with greater capital usage per unit of labor is expected to demand more intermediate inputs and energy resources for one unit of economic output, which leads to worse environmental quality (Cole et al., 2008). We expect a positive sign for capital intensity.

• 4. Policy dummy. Environmental policy is regarded as the effective way to attain environmental goals, like the cap-and-trade system (Du et al., 2016). Three important public policies were adopted during the period of our sample. In 2006, the government set a mandatory reduction target for COD emission. In 2011, a mandatory target for NH₄ abatement was introduced and implemented. The tightening regulation of COD and NH₄ emissions may have led to the rise of abatement costs. Moreover, the government implemented an economic stimulus program in 2009 to boost growth in the context of the global economic recession. All three policies also provide us with a chance to examine their possible impacts on abatement costs. Three dummies (Dummy2006, Dummy2009 and Dummy2011) are introduced to examine the policy effects in 2006, 2009 and 2011. The variable Dummy2006 equals 1 for year 2006–2012; otherwise, it equals zero. The same setting is used for Dummy2009 and Dummy2011.

In addition, we introduce cross-terms between year dummies and control variables to explore the mechanisms. We assume that the policy interventions may affect the abatement cost through the change of pollution intensity, the level of economic activity and capital intensity. The statistical descriptions of all the regression variables are summarized in Table 5.

Table 5.

Statistic description of regression variables.

Variables	Unit	Mean	Std. Dev.	Min	Max
ln	104Yuan/ton	3.879	0.715	1.872	6.851
ln	104Yuan/ton	5.363	1.548	0.262	10.110
Ln(CODinten)	ton/108 Yuan	3.993	1.201	0.573	7.727
Ln(NH4inten)	ton/108 Yuan	1.314	1.228	−1.949	4.580
RGDP	104Yuan	0.823	0.435	0.241	2.449
K/L	104Yuan/person	18.906	8.816	7.479	51.808
Dummy2006	–	0.702	0.458	0	1
Dummy2009	–	0.387	0.488	0	1
Dummy2011	–	0.184	0.388	0	1

Variables	Unit	Mean	Std. Dev.	Min	Max
ln	104Yuan/ton	3.879	0.715	1.872	6.851
ln	104Yuan/ton	5.363	1.548	0.262	10.110
Ln(CODinten)	ton/108 Yuan	3.993	1.201	0.573	7.727
Ln(NH4inten)	ton/108 Yuan	1.314	1.228	−1.949	4.580
RGDP	104Yuan	0.823	0.435	0.241	2.449
K/L	104Yuan/person	18.906	8.816	7.479	51.808
Dummy2006	–	0.702	0.458	0	1
Dummy2009	–	0.387	0.488	0	1
Dummy2011	–	0.184	0.388	0	1

In Equation (8), a one-year lagged term for abatement cost is included to reduce the endogeneity problem resulting from the omitted variables. is the error term.

The regression results for COD and NH₄ are listed in Tables 6 and 7, respectively. We first regress against all control variables and then add the policy dummy and its interaction terms. The Hausman test favors the fixed effect estimation. All continuous variables are expressed in log forms to avoid the heterogeneity problem.

Table 6.

Regression result on the determinants of COD's shadow price.

Variables	ln (log COD's shadow price)
Lag 1 term of ln	0.642*** (0.054)	0.586*** (0.057)	0.632*** (0.059)	0.592*** (0.060)
ln(CODinten)	−0.166*** (0.043)	−0.002 (0.066)	−0.151*** (0.052)	−0.001 (0.067)
RGDP	−1.498* (0.774)	−1.417* (0.793)	−1.544** (0.779)	−1.497* (0.790)
RGDP2	0.529** (0.243)	0.606** (0.250)	0.522** (0.253)	0.583** (0.256)
K/L	0.017*** (0.005)	0.021*** (0.008)	0.026*** (0.007)	0.025*** (0.009)
Dummy2006		1.011*** (0.370)		1.232*** (0.400)
Dummy2006×ln(CODinten)		−0.177*** (0.060)		−0.229*** (0.066)
Dummy2006×RGDP		−0.150 (0.143)		−0.273* (0.155)
Dummy2006×(K/L)		−0.004 (0.006)		0.001 (0.007)
Dummy2009			−0.029 (0.259)	−0.352 (0.281)
Dummy2009×ln(CODinten)			0.040 (0.048)	0.105** (0.053)
Dummy2009×RGDP			0.105 (0.130)	0.188 (0.136)
Dummy2009×(K/L)			−0.010** (0.005)	−0.011** (0.005)
CONS.	2.557*** (0.591)	1.780*** (0.646)	2.407*** (0.616)	1.773*** (0.647)
Obs.	245	245	245	245
R2	0.639	0.659	0.647	0.671
AIC	−26.408	−32.855	−24.371	−33.068

Variables	ln (log COD's shadow price)
Lag 1 term of ln	0.642*** (0.054)	0.586*** (0.057)	0.632*** (0.059)	0.592*** (0.060)
ln(CODinten)	−0.166*** (0.043)	−0.002 (0.066)	−0.151*** (0.052)	−0.001 (0.067)
RGDP	−1.498* (0.774)	−1.417* (0.793)	−1.544** (0.779)	−1.497* (0.790)
RGDP2	0.529** (0.243)	0.606** (0.250)	0.522** (0.253)	0.583** (0.256)
K/L	0.017*** (0.005)	0.021*** (0.008)	0.026*** (0.007)	0.025*** (0.009)
Dummy2006		1.011*** (0.370)		1.232*** (0.400)
Dummy2006×ln(CODinten)		−0.177*** (0.060)		−0.229*** (0.066)
Dummy2006×RGDP		−0.150 (0.143)		−0.273* (0.155)
Dummy2006×(K/L)		−0.004 (0.006)		0.001 (0.007)
Dummy2009			−0.029 (0.259)	−0.352 (0.281)
Dummy2009×ln(CODinten)			0.040 (0.048)	0.105** (0.053)
Dummy2009×RGDP			0.105 (0.130)	0.188 (0.136)
Dummy2009×(K/L)			−0.010** (0.005)	−0.011** (0.005)
CONS.	2.557*** (0.591)	1.780*** (0.646)	2.407*** (0.616)	1.773*** (0.647)
Obs.	245	245	245	245
R2	0.639	0.659	0.647	0.671
AIC	−26.408	−32.855	−24.371	−33.068

Standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01. AIC refers to the Akaike Information Criterion which is a measure of the relative quality of statistical models for a given set of data.

Table 7.

Regression result on the determinants of NH₄'s shadow price.

Variables	ln (log NH4's shadow price)
Lag 1 term of ln	0.337*** (0.053)	0.318*** (0.053)	0.328*** (0.052)	0.315*** (0.053)
ln(NH4inten)	−0.168*** (0.055)	−0.148** (0.061)	−0.124* (0.069)	−0.121* (0.070)
RGDP	0.530 (1.234)	0.909 (1.354)	1.070 (1.268)	1.398 (1.384)
RGDP2	−0.174 (0.402)	−0.215 (0.422)	−0.239 (0.430)	−0.273 (0.444)
K/L	−0.001 (0.009)	−0.002 (0.010)	0.016 (0.013)	0.015 (0.014)
Dummy2011		0.044 (0.325)		−0.041 (0.380)
Dummy2011×ln(NH4inten)		−0.292** (0.120)		−0.245* (0.145)
Dummy2006×RGDP		−0.299 (0.261)		−0.289 (0.322)
Dummy2011×(K/L)		0.018* (0.011)		0.019 (0.011)
Dummy2009			0.142 (0.292)	0.055 (0.342)
Dummy2009×ln(NH4inten)			−0.182** (0.085)	−0.106 (0.109)
Dummy2009×RGDP			−0.078 (0.219)	0.076 (0.272)
Dummy2009×(K/L)			−0.004 (0.009)	−0.008 (0.009)
CONS.	3.580*** (0.712)	3.382*** (0.799)	2.915*** (0.758)	2.752*** (0.843)
Obs.	245	245	245	245
R2	0.309	0.329	0.335	0.347
AIC	261.200	261.738	259.654	263.141

Variables	ln (log NH4's shadow price)
Lag 1 term of ln	0.337*** (0.053)	0.318*** (0.053)	0.328*** (0.052)	0.315*** (0.053)
ln(NH4inten)	−0.168*** (0.055)	−0.148** (0.061)	−0.124* (0.069)	−0.121* (0.070)
RGDP	0.530 (1.234)	0.909 (1.354)	1.070 (1.268)	1.398 (1.384)
RGDP2	−0.174 (0.402)	−0.215 (0.422)	−0.239 (0.430)	−0.273 (0.444)
K/L	−0.001 (0.009)	−0.002 (0.010)	0.016 (0.013)	0.015 (0.014)
Dummy2011		0.044 (0.325)		−0.041 (0.380)
Dummy2011×ln(NH4inten)		−0.292** (0.120)		−0.245* (0.145)
Dummy2006×RGDP		−0.299 (0.261)		−0.289 (0.322)
Dummy2011×(K/L)		0.018* (0.011)		0.019 (0.011)
Dummy2009			0.142 (0.292)	0.055 (0.342)
Dummy2009×ln(NH4inten)			−0.182** (0.085)	−0.106 (0.109)
Dummy2009×RGDP			−0.078 (0.219)	0.076 (0.272)
Dummy2009×(K/L)			−0.004 (0.009)	−0.008 (0.009)
CONS.	3.580*** (0.712)	3.382*** (0.799)	2.915*** (0.758)	2.752*** (0.843)
Obs.	245	245	245	245
R2	0.309	0.329	0.335	0.347
AIC	261.200	261.738	259.654	263.141

Standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01.

In Table 6, the coefficient of the one-year lagged term of ln is significantly positive. It favors the ‘path dependence’ hypothesis that the present abatement cost is heavily determined by the historical abatement cost. The COD intensity is found to be significantly negative, associated with the abatement cost of COD. It suggests that the abatement cost for a ‘dirty’ province (which emits more COD for one unit of industrial output) will be much cheaper. In contrast, the ‘clean’ province that has lower COD intensity is associated with higher abatement cost. In general, a 1% decrease of COD intensity will lead to a 0.15–0.17% increase in the abatement cost.

The negative sign of the RGDP and the positive sign of its squared term show a U-shaped relationship between the economic development level and COD's abatement cost. This seems consistent with the Environmental Kuznets Curve. The turning point is about 13,400 Yuan. This means that the abatement cost for the poor provinces whose RGDP is less than 13,400 Yuan will decline along with the abatement activities. On the contrary, further abatement activities will become more costly for the developed provinces, whose per capita GDP exceeds the turning point.

We also notice that the capital-labor ratio is an important factor. The significant positive coefficient is consistent with our expectation. It indicates that those provinces whose industrial composition is capital-intensive will have higher abatement costs. In general, a 1% increase of capital intensity will lead to a 0.017–0.026% increase in abatement cost.

The Dummy2006 in Table 6 shows a significantly positive impact on the abatement cost. It shows that the COD mandatory abatement policy implemented in 2006 does increase the abatement cost. Among its interaction terms, only Dummy2006×ln(CODinten) is remarkable, while the other two are not significant. This suggests that COD regulatory policy generates an indirect impact on the abatement cost through the change of COD intensity. In other words, the marginal effect of the policy in the second column in Table 6 can be expressed as d(lnq₁)/d(dummy2006) = 1.011 - 0.177* ln(CODinten). We can conclude that the COD mandatory abatement program that started in 2006 has two effects. On the one hand, it directly pushes up the abatement cost; on the other hand, it reduces the abatement cost through the change of COD intensity.

The direct impact of the economic stimulus program (Dummy2009) on the abatement cost is not significant, while it creates an indirect effect through the change of capital-labor ratio. Given its negative coefficient for the interaction term Dummy2009× (K/L), it lowers the abatement cost after 2009 and the abatement cost becomes much lower for capital-intensive provinces.

Similarly, we report the regression result for NH₄'s abatement cost in Table 7. Because the regulation of NH₄ emissions started in 2011, data is collected for only two years (2011 and 2012). We make the following observations. First, the coefficient of ln(NH₄inten) is significantly negative, which is consistent with the COD case. This indicates that the province with higher NH₄ intensity is associated with a lower abatement cost to cut NH₄ emissions. Second, the economic development level and the capital-labor ratio have no significant effect on the NH₄ abatement cost. Third, the NH₄ emission reduction policy launched in 2011 does not affect the abatement cost directly. Instead, it has two indirect effects: it reduces the abatement cost through changing NH₄ intensity while increasing the abatement cost by changing capital intensity. Moreover, the impact of the economic stimulus program launched in 2009 is slight and insignificant.

Using a parametric DDF approach, this paper simultaneously models the joint production of industrial value added and pollutant emissions. The abatement potential, abatement cost and Morishima elasticity of substitution for industrial COD and NH₄ are examined for the Chinese industrial sector during 2003–2012. The results suggest that the national industrial COD and NH₄ emissions can be further reduced by 13.18% and 13.27%, respectively, if inefficiency can be eliminated. This abatement activity is associated with opportunity costs, which should be taken into account. The average marginal cost to cut one unit of industrial COD is 710 Yuan/kg, and the average marginal cost to cut one unit of industrial NH₄ is 7,390 Yuan/kg. The abatement potential and cost exhibit similar trends during 2003–2012, during which we observe significant shocks in 2006 and 2010. Our results also demonstrate great regional disparity. The developed provinces are associated with expensive abatement costs while the developing regions are associated with greater abatement potential and lower costs.

The great heterogeneity among the provinces implies that it is not cost-effective to distribute the abatement target uniformly among different regions. In the developed provinces, it is much more expensive to mitigate COD and NH₄ than in the developing provinces. Thus, it would reduce the overall abatement cost to allocate more burdens to the developing provinces, where ‘win-win’ opportunities exist. There may be concern that this will lead to inequity for poor areas. However, a tradable permit market can deal with both efficiency and equity issues (Moore, 2015), i.e., the developed provinces can buy the abatement quotas from the developing regions, which will happen if and only if the trading price is lower than the abatement cost in rich areas and is higher than the abatement cost in poor areas. This trade will reduce the total transaction costs and achieve the overall emission-abatement target. Moreover, it also will improve efficiency in the developing areas, while trading will provide revenue or investment capital to the poorer areas. In short, the diversified abatement potential and abatement costs among provinces urgently call for the establishment of a national Emissions Trading System, which can assist China in reaching its environmental goals in a cost-effective way while balancing its regional development.

In addition, we notice a remarkable increase in abatement cost in 2006 and a significant drop in 2010. We infer that the binding targets for COD and NH₄ abatement at the beginning of the 11th FYP created a strong incentive for local decision-makers. This policy shock led to vast investment in pollution abatement. Consequently, we observe a sharp rise in marginal cost. In contrast, the massive economic stimulus program during the global financial crisis in 2009 may have allowed inefficient and pollution-intensive firms to survive or even expand. Another explanation is that the global crisis shifted the government's focus from reducing pollution to boosting the economy. In order to confirm our assumption, we ran a regression against the abatement cost of COD and NH₄. The empirical results suggest that ‘the dirtier the emission, the cheaper the abatement cost’ for both water pollutants. The results also demonstrate a U-shaped relationship between economic level and abatement cost of COD (which does not hold for NH₄). More importantly, we find that the mandatory reduction target for COD emissions in 2006 significantly affected the COD abatement cost through decreasing the COD intensity in production. The economic stimulus plan in 2009 affected the abatement costs through the change of pollution intensity and capital intensity. Moreover, we observe that the mandatory abatement for NH₄ emissions in 2011 affected the abatement cost through the decline of the NH₄ intensity in production. However, the 2009 economic stimulus plan had an insignificant effect on the cost of abatement.

Last but not least, a more comprehensive framework concerning the externality, the ecosystem, and the public health impacts of water pollution should be scrutinized based on the economic impact assessment. The other valuation methods, like the cost of illness approach or human capital approach, could also be conducted for a small river basin or a specific economic region using the firm or resident data. The information regarding abatement cost in most regulations and policy-making process is not the issue because most costs cannot be properly captured. In the present paper, the abatement cost of industrial water pollutants we assessed is focused on the opportunity cost, rather than the overall negative externality cost. Besides, a fair national emissions trading system for water pollutants especially on the allocation of water quotas and water environmental capacity is left for further studies.

This research was financially supported by the National Social Science Foundation of China (No. 12AJY003). We appreciate the reviewer's meaningful comments and suggestions. The usual disclaimer applies.