Estimating the environmental efficiency and marginal CO2 abatement cost of coal-fired power plants in China

Energy Policy 85 (2015) 347–356

Contents lists available at ScienceDirect

Energy Policy journal homepage: www.elsevier.com/locate/enpol

Estimating the environmental efficiency and marginal CO2 abatement cost of coal-fired power plants in China Limin Du a,n, Jie Mao b a b

China Academy of West Region Development, Zhejiang University, Hangzhou 310058, China School of International Trade and Economics, University of International Business and Economics, Beijing 100029, China

H I G H L I G H T S


A novel plant-level dataset derived from the National Economic Survey in China is used.
There are large opportunities for CO2 emissions reduction in China's coal-fired power plants.
Subsidies can reduce environmental inefficiency but increase shadow price.

art ic l e i nf o

a b s t r a c t

Article history: Received 7 April 2015 Received in revised form 1 June 2015 Accepted 10 June 2015

We estimate the environmental efficiency, reduction potential and marginal abatement cost of carbon dioxide (CO2) emissions from coal-fired power plants in China using a novel plant-level dataset derived from the first and second waves of the National Economic Survey, which were implemented in 2004 and 2008, respectively. The results indicate that there are large opportunities for CO2 emissions reduction in China's coal-fired power plants. Given that all power plants operate fully efficiently, China's CO2 emissions in 2004 and 2008 could have been reduced by 52% and 70%, respectively, accompanied by an expansion in electricity output. In other words, the opportunities for ‘double dividend’ exist. In 2004, the average marginal abatement cost of CO2 emissions for China's power plants was approximately 955 Yuan/ton, whereas in 2008, the cost increased to 1142 Yuan/ton. The empirical analyses show that subsidies from the government can reduce environmental inefficiency, but the subsidies significantly increase the shadow price of the power plants. Older and larger power plants have a lower environmental efficiency and marginal CO2 abatement cost. The ratio of coal consumption negatively affects the environmental efficiencies of power plants. & 2015 Elsevier Ltd. All rights reserved.

JEL classification: Q52 Q54 Q58 Keywords: Coal-fired Power Plants Environmental Efficiency Shadow Price China

1. Introduction China is currently one of the largest greenhouse gas emitters in the world because of its rapid economic growth and soaring energy consumption. According to the IEA (2014), China's total carbon dioxide (CO2) emissions in 2012 was approximately 8.2 billion tons, which accounted for approximately 26% of the world's total CO2 emissions that year. The majority of China's CO2 emissions primarily comes from the power industry. In China, the CO2 emissions from electricity and heat production account for approximately half of the total emissions from combustion in 2012 (IEA, 2014). Although the Chinese Government has encouraged the development of renewable energy (such as hydropower, wind n

Corresponding author. E-mail addresses: [email protected] (L. Du), [email protected] (J. Mao).

http://dx.doi.org/10.1016/j.enpol.2015.06.022 0301-4215/& 2015 Elsevier Ltd. All rights reserved.

power and solar power) over the past decade, at present, fossil fuel-fired power (especially coal-fired power) remains the dominant form of electricity generation (Xie et al., 2012). Therefore, it is particularly important to investigate the environmental efficiency, reduction potential and marginal abatement cost of CO2 emissions from the power sector, for the sake of China's overall greenhouse gas reduction. Previous studies have emphasized the important role of the power sector in environmental protection. The environmental efficiency and marginal abatement cost of various pollutants (such as SO2, CO2 and NOX) for power plants have been widely discussed. Related studies include those of Färe et al. (2005), Lee (2005), Murty et al. (2007), Zhou et al. (2012), and Sueyoshi and Goto (2013), among others. However, only a few researchers have focused on the environmental efficiency and pollution abatement cost of power plants in China, partly because of the lack of highquality plant-level data. Yang and Pollitt (2009) investigated the

348

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

SO2 emission performance of China's power industry based on a dataset of 221 surveyed coal-fired power plants. Wei et al. (2013) estimated the shadow price of CO2 emissions and its determinants, using the data of 124 coal-fired power plants located in Zhejiang Province, China. Their data are derived from the first wave of the National Economic Survey, which was implemented by Zhejiang Province in 2004. Zhang and Choi (2013) analyzed the dynamic changes of the carbon performance of fossil-fuel power plants in China, but their analysis focused on a limited sample of 259 large (capacity greater than 1 GW) state-owned plants. In this paper, we estimate the environmental efficiency, CO2 reduction potential and marginal abatement cost of CO2 emissions for China's coal-fired power plants. We utilize a cross-sectional plant-level dataset derived from the first and second waves of the National Economic Survey of China, which were implemented in 2004 and 2008, respectively. The novelty of our paper is that we have constructed a new dataset by combining the data of the National Economic Survey with those of the Compilation of Statistical Data of the Power Industry. Compared to the previous studies, our dataset consists of a much larger sample and more information on China's coal-fired power plants. Consequently, our empirical results may be more representative and reliable. We also analyze the effects of subsidies on the environmental efficiency and shadow price of the power plants, which are usually ignored in the literature. Our work thus contributes to the literature by offering new evidences and more insights. With regard to the results from the multifold specifications of regression, we find that the environmental efficiency of China's coal-fired power plants has decreased between the years 2004 and 2008. There is ample room for CO2 emissions reduction in China's coal-fired power plants. Specifically, if all of the power plants had operated efficiently (at the frontiers), the CO2 emissions in 2004 and 2008 could have been reduced by 52% and 70%, respectively. The average marginal abatement cost of CO2 emissions for the power plants has increased with time, approximately 955 Yuan/ton in 2004 and 1142 Yuan/ton in 2008, respectively. We also find that subsidies from the government increase the shadow price but significantly reduce the environmental inefficiency. Older and larger power plants are less environmentally efficient, but it is cheaper for them to reduce their CO2 emissions. A higher ratio of coal consumption in a plant results in a lower environmental efficiency. Compared to the central region, the power plants located in the western region have both lower environmental efficiencies and a lower shadow price, whereas those in the eastern region only exhibit a higher environmental efficiency. The remainder of the paper is organized as follows: Section 2 briefly reviews the existing studies and describes the estimation methodology; Section 3 reports the empirical results; Section 4 is conclusions and policy implications.

2. Methods 2.1. Literature review The recent development of the environmental production theory and directional distance function makes it possible for researchers to estimate the environmental efficiency and shadow price (marginal abatement cost) of non-marketed pollutants without price and cost information (Färe et al., 1993). Under this estimation framework, the pollutants are usually considered as undesired byproducts. The existing literature on the estimation of the environmental efficiency and shadow price of pollutants can be roughly classified into three groups based on the estimation techniques, i.e., the non-parametric Data Envelopment Analysis (DEA) approach, the parametric Linear Programing (LP) approach,

and the parametric Stochastic Frontier Analysis (SFA) approach (Zhou et al., 2014). The non-parametric DEA approach constructs the production frontier by combining all of the observed inputs and outputs to form a piecewise production boundary. The prominent merit of the DEA approach is that it is not necessary to impose a specific functional form for the underlying technology in advance (Zhang and Choi, 2014). The DEA approach applies to both the Shephard distance function and the directional distance function (Chung et al., 1997; Shephard et al., 1970).1 The related research includes Choi et al. (2012), Kaneko et al. (2010), Lee et al. (2002), Maradan and Vassiliev (2005), and Wei et al. (2012), among others.2 However, the DEA approach does not guarantee the differentiability of the estimated distance functions everywhere. For any inflection point located on the frontier, its slope is not unique. The choice of the slopes for these inflection points by the researcher will affect the values of the shadow prices considerably (Lee et al., 2002). Furthermore, the DEA approach has suffered from many other problems, e.g., the impact of outliers (Vardanyan and Noh, 2006). The parametric LP approach estimates the production frontier by minimizing the sum of the differences between the estimated distance functions of the observed production bundles and that of their projections on the production frontier. Both the Shephard and the directional distance functions can be estimated by this approach. The Shephard distance function is usually parameterized to have a translog functional form, whereas the directional distance function is usually parameterized with a quadratic functional form because of its special properties. The related studies include Coggins and Swinton (1996), Lee and Zhang (2012), Marklund and Samakovlis (2007), Matsushita and Yamane (2012), Rezek and Campbell (2007) and Swinton (2004), among others. The main advantage of the LP approach is that the estimated frontier is differentiable everywhere. It can also perform estimations that take all of the constraints of the environmental production technology into account. The weakness of the LP approach is that it ignores statistical noise. However, this can be remedied by resorting to bootstrap simulations (Simar and Wilson, 2000; Zhang and Choi, 2014; Zhou et al., 2010). The parametric SFA approach constructs the production frontier by econometric estimation, thus it has the merit of taking statistical noise into account. Furthermore, the frontier estimated by the SFA approach is also differentiable everywhere. This approach only applies to the directional distance function but not to the Shephard distance function. The reason is that the SFA approach is essentially based on the translation property, whereas the Shephard distance function does not satisfy this property. Relevant previous studies include Färe et al. (2005), Murty et al. (2007), and Wei et al. (2013), among others. The primary weakness of the SFA approach is that it cannot include the constraints of the environmental production technology in the estimation process. The usual procedure applied in the previous studies is to first run the SFA estimation ignoring the constraints, and then check if the estimated results meet the constraints ex-post. Only those observations meeting the constraints will be kept for further analysis. However, this process may induce estimation bias (Du et al., in press). From the above review of the estimation approaches, we find that the parametric LP approach has specific merits relative to the 1 The Shephard output distance function assumes that the desirable outputs and undesirable outputs only adjust proportionally, whereas the directional output distance function permits an increase of the desirable outputs but a reduction of the undesirable outputs. Actually, the former is a special case of the latter (Chambers et al., 1998). 2 For more detailed review of the DEA approach in energy and environment analysis, please refer to Song et al. (2012) and Zhou et al. (2008).

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

non-parametric DEA approach and the parametric SFA approach. Therefore, we will employ the parametric LP approach to estimate the environmental efficiencies and shadow prices for China's coalfired power plants in this paper. 2.2. Methodology The environmental efficiency and marginal abatement cost of undesirable output can be estimated by using the production theory combined with the directional output distance function. We will only briefly introduce the methodology in this section because it is rather standard in the literature of environmental efficiency and shadow price estimation. Suppose the following production technology: if a producer wants to produce a vector of desirable outputs y = (y1, … , yM ) ∈ R+M by employing a vector of inputs x = (x1, … , xN ) ∈ RN + , it has to produce a y = (y1, … , yM ) ∈ R+M byproduct simultaneously, i.e., a vector of undesirable outputs b = (b1, … , bJ ) ∈ R+J . The production technology can be described by the following output set:

P (x) = {(y , b): x can produce (y , b)}

(1)

In addition to the standard compact and convex assumptions, we assume that the output set satisfies two additional important assumptions. First, the desirable outputs and the undesirable outputs are jointly produced. That is to say, if an output bundle (y, b) ∈ P (x ) and b = 0, then y = 0. This assumption implies that if we want to produce a positive quantity of desirable output, then we must simultaneously have a positive quantity of undesirable output. Second, the desirable outputs and the undesirable outputs are together weakly disposable, i.e., if an output bundle (y, b) ∈ P (x ) and 0 ≤ θ ≤ 1, then the output bundle (θy, θb) ∈ P (x ). The weak disposability assumption requires a proportional reduction of the desirable output and undesirable output. This assumption implies that any reduction of the undesirable output will incur a cost in terms of the desirable output. The production technology can be represented functionally by the directional output distance function, which satisfies the aforementioned properties. The formal definition of the directional output distance function is as follows:

→ Do(x, y , b; gy , − gb) = max {β: (y + βgy , b − βgb) ∈ P (x)}

(2)

349

efficiency of this output bundle will increase by α . The derivation of the shadow price is based on the duality theory, i.e., the duality of the directional output distance function and the revenue function. As long as the market price of the m-th desirable output pm is given, the shadow price of the j-th undesirable output qj can be calculated by the following formula:

⎡ → ⎤ ∂Do(x, y , b; 1, − 1)/∂bj ⎥ , qj = − pm ⎢ → ⎢ ⎥ ⎣ ∂Do(x, y , b; 1, − 1)/∂ym ⎦

→ → Do(x, y + αgy , b − αgb ; gy , − gb) = Do(x, y , b; gy , − gb) − α

(4)

The shadow price reflects the trade-off between the undesirable output and the desirable output on the production boundary of the output set. Specifically, for the case of one desirable output and one undesirable output, the price ratio (  q/p) can be depicted as the slope of the tangent line through the point, which is the projection of the output bundle (y, b) on the production frontier. For empirical estimation of the directional output distance function, both the parametric method and non-parametric method can be used (see the literature review aforementioned). In this paper, we employ the parametric LP approach to estimate the directional output distance function because it has the advantages of differentiability and constraints inclusion. In previous studies, the quadratic form is the most widely used functional form to parameterize the directional output distance function (Chambers et al., 1998; Färe et al., 2005; Murty et al., 2007). The quadratic functional form satisfies the translation property. It is also twice differentiable and flexible. Therefore, we assume a quadratic functional form for the directional output distance function, too. Additionally, we assume that the direction vector (gy , − gb) = (1, − 1). Such an assumption means that we hope to achieve an increase of the desirable output and a reduction of the undesirable output at the same scale simultaneously.3 Specifically, for the case of one desirable output (y), one undesirable output (b) and three inputs (x1, x2, x3), the directional output distance function for the k-th power plant with quadratic functional form can be written as follows:

→ Do(xk , yk , bk ; 1, − 1) 3

where g = (gy , gb) ∈ R+M × R+J is a direction vector that specifies the movement directions of the desirable outputs and undesirable outputs. For any given production technology, the directional output distance function seeks the largest possible increase of the desirable outputs and reduction of the undesirable outputs simultaneously, as long as the production is feasible. The directional output distance function represents the environmental efficiency. A higher value of β implies a lower environmental efficiency. Particularly, if the value of β is equal to zero, then the producer is located on the production frontier and its production is completely efficient. If the value of β is larger than zero, then we can tell that inefficiency exists in the production process. In addition to the properties inherited from the output set P(x), the directional output distance function also exhibits another important property, i.e., the translation property:

j = 1, …, J

=α+

∑ αnx nk + β1yk

+ γ1bk +

n= 1

+

1 γ (bk )2 + 2 2

3

1 2

3

3

∑∑ n = 1 n ′= 1

3

∑ ηnx nkbk + ∑ δnx nkyk n= 1

1 αnn x nk x n k + β2(yk )2 ′ ′ 2

n= 1

+ μyk bk (5)

The deterministic linear programing algorithm proposed by Aigner and Chu (1968) can be used to estimate the parameters of the quadratic directional output distance function represented in Eq. (5). This approach minimizes the sum of the differences between the estimated directional output distance functions of the observed points and that of their corresponding projection points on the production frontier (the points of complete efficiency). Specifically, the linear minimization problem can be represented as follows:

(3)

where α is a scalar. The translation property describes the relationships of the environmental efficiencies of the output bundles along a given direction vector. Specifically, for a given direction vector g = (gy , gb), if the desirable output y is increased by αgy and the undesirable output b is reduced by αgb , then the environmental

3 Shadow price is affected by the choice of direction vector. Essentially, a direction vector represents a particular mapping rule which projects the observations in the sample to the production frontier. Given the market price of good output, the shadow price of bad output for a particular observation is determined by the slope of tangent on the frontier. Therefore, given the direction of good output gy, a larger value of bad output direction gb will lead to a higher estimated shadow price (Vardanyan and Noh, 2006).

350

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

K

min

Table 1 Capacity and output of power plants.

→

∑ (Do(xk , yk , bk ; 1, − 1) − 0) k=1

Year

s. t.

(i) (ii) (iii)

(iv) (v)

→ Do(xk , yk , bk ; 1, − 1) ≥ 0, k = 1, …, K → t t Do(xk , yk , 0; 1, − 1) < 0, k = 1, …, K → ∂Do(xk , yk , bk ; 1, − 1) ≤ 0, k = 1, …, K ∂y → ∂Do(xk , yk , bk ; 1, − 1) ≥ 0, k = 1, …, K ∂b → ∂Do(x¯ , yk , bk ; 1, − 1) ≥ 0, n = 1, 2, 3; k = 1, …, K ∂x n

(vi) β1 − γ1 = − 1,

β2 = γ2 = μ ,

δn − ηn = 0,

(vii) αn, n = αn , n, ′ ′

n, n′ = 1, 2, 3

2004 2008

where

yk⁎

n = 1, 2, 3

(6)

(7)

= (yk + βkgy ) is the maximum attainable level of elec-

3

γ1 + γ2bk + ∑n = 1 ηnx nk + μyk 3

β1 + β2yk + ∑n = 1 δnx nk + μbk

Capacity (kW)

Output (kWh)

Capacity (%)

Output (%)

1.54  108 3.29  108

0.92  1012 1.65  1012

47 55

51 59

Yearbook of 2005 and 2009.

tricity output and bk⁎ = (bk − βkgb) is the minimum CO2 emissions that can be realized for the k-th power plant if its production process was completely efficient. Based on the estimated parameters, the shadow price of the undesirable output for each power plant can be calculated according to Eq. (4). Specifically, the shadow price of the undesirable output for the k-th plant can be written as follows:

qk = − pk

Percentage

Note: The data of installed capacity and electricity output of the country are derived from the China Power

The first K restrictions in (i) guarantee that the production processes of all of the power plants are feasible. The restrictions in (ii) describe the null-jointness property, which implies that, for any desirable output y40, the output bundle (y, 0) is not located within the output set. The restrictions in (iii) and (iv) are monotonicity assumptions imposed on the desirable output and the undesirable output, respectively. These two sets of restrictions are used to guarantee the correct signs of the estimated shadow prices. Typically, an increase in any of the three inputs with the desirable output and undesirable output unchanged signifies a decrease in environmental efficiency. Thus, we impose a set of positive monotonicity constraints (at mean level) on the three inputs listed in (v). Additionally, the translation properties are imposed by the restrictions listed in (vi) and the symmetry property is imposed by the restrictions listed in (vii). The estimates of the directional output distance functions make it possible for us to go further, i.e., to estimate the feasible increase potential of electricity output and reduction potential of CO2 emissions for each power plant relative to the plants producing on the frontier. The calculation formulas are as follows:

Δyk = (yk + βk gy) − yk Δbk = bk − (bk − βk gb)

Quantity

(8)

consumption information of the power plants.5 Consequently, we attain a novel cross-sectional dataset of 1158 observations, i.e., 518 observations for the year 2004 and 640 observations for the year 2008. Table 1 reports the total installed capacity and electricity output of the sample and their corresponding proportions in China. In terms of installed capacity, our sample accounts for approximately 47% and 55% of the total installed thermal capacity of China in 2004 and 2008, respectively. In terms of electricity output, our sample accounts for approximately 51% and 59% of the total thermal electricity output of China in 2004 and 2008, respectively. Thus, our dataset is a good representative of China's coal-fired power industry. In this paper, we focus on the case of two outputs (one desirable output and one undesirable output) and three inputs. The desirable output is electricity output and the undesirable output is CO2 emissions. The three inputs considered are labor, capital and energy. In our dataset, the electricity output is measured in annual kilowatt-hours (kWh), and the labor is measured in annual average employees. The capital is measured in monetary terms. In our dataset, we have information on the average annual net value of fixed assets, which captures the remaining value of fixed assets with depreciations excluded. We use this term to mean the capital input. The data of energy input is not directly available, however we have information on the quantity of coal, oil and gas consumption for each plant. Thus, we are able to derive the total energy input by converting all of those different types of fuels into Standard Coal Equivalent (SCE).6 It is worthy to note that some coal-fired power plants also consume oil or gas, although their primary fuel is coal. There are in total four types of plants in our sample, in terms of fuel usage combinations, i.e., plants that only use coal, plants that use coal and oil, plants that use coal and gas, and plants that use coal, oil and gas. Table 2 reports the distribution of the power plants with different fuel usage combinations. More than 65% of plants use both coal and oil, and approximately 33% of the plants use coal only, but only no more than 2% of the plants use gas. The data of the undesirable output, of the CO2 emissions, are not directly available either. However, we are able to calculate the CO2 emissions from the fuel burned for each plant in each year because we have information on the different types of energy consumption. Following IPCC (2006) and Du et al. (2012), we calculate the CO2 emissions from coal, oil and gas usage for each power plant in each year using the following formula:

2.3. Data Our analyses are based on two datasets. One is derived from the first and second waves of the National Economic Survey collected in 2004 and 2008, respectively.4 The second is from the Compilation of Statistical Data of the Power Industry in the corresponding years. We combine these two datasets based on the firm codes of the plants to derive both the financial information and energy 4 There are three waves of the National Economic Survey that exist (2004, 2008 and 2012, respectively), but we only have access to the data of the first and second waves.

3

CO2i =

∑ Eif f =1

× CFf × CCf × COFf × 3.67

(9)

where i is the index of the power plant; f is the index of energy 5 The data from the National Economic Survey only provide the financial information of the power plants (such as assets, profit, subsidies, etc.), whereas the data from the Compilation of Statistical Data of the Power Industry provide information of energy consumption and electricity output. 6 Here we use the following transition coefficients: 1 ton of coal¼ 0.714 ton of SCE; 1 ton of oil ¼1.457 ton of SCE; 10 thousand square meters of natural gas¼ 12.143 ton of SCE.

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

Table 2 Fuel usage combinations of power plants. Fuel usage combination

No. of plants

Percentage (%)

Coal Coal þoil Coal þgas Coal þoil þgas Total

381 758 7 12 1158

32.90 65.46 0.60 1.04 100

types, representing coal, oil and gas, respectively; Eif is the consumption quantity of fuel f for plant i; CFf represents the transformation factor of fuel f; CCf is the carbon content of fuel f; and COFf is the carbon oxidation factor of fuel f. The term 3.67 is the transformation coefficient of pure carbon emissions to CO2 emissions. Table 3 reports the summary statistics of the three inputs and two outputs. The table shows that, on average, the two outputs for the year 2008 are much larger than that of the year 2004. For the three inputs, the average inputs of capital and energy in 2008 are larger than that in 2004, but the average labor input in 2008 is lower than that in 2004. This implies an increase in the average scale of the power plants from 2004 to 2008, as well as a substitution of capital input for labor input. From the table, we also can observe that there are substantial variations among power plants for all of the variables.

3. Results and discussion In this section, we first report the estimates of the environmental efficiency, reduction potential and marginal abatement cost of CO2 reductions based on the parametric LP approach. Then, we analyze the impact factors of environmental efficiency and the marginal abatement cost using econometric regressions. 3.1. Environmental efficiency and shadow price We solve the linear programming problem in Eq. (6) twice for the data from 2004 and 2008, respectively. Before the estimation, we apply a mean normalization process to the raw data to manage the problem of convergence, i.e., we divide the input and output variables by their sample means of the corresponding year (Färe et al., 2005). The mean of the normalized variable is equal to 1. The on-grid electricity prices for different power plants in China are usually different. However, we do not have the information for on-grid electricity prices for each plant. Wei et al. (2013) provided the average on-grid price of the fossil-fueled power plants in Zhejiang Province, which is approximately 0.46 Yuan/kWh. In this paper, we follow Wei et al. (2013) to assume that all of the power plants receive a same on-grid electricity price,

351

i.e., 0.46 Yuan/kWh. It is worthy of noting that the on-grid prices for the power plants located in Zhejiang Province are usually slightly higher than that of the plants located in the central and west provinces. The reason is that Zhejiang Province lacks coal resources and has to import coal from the west and central regions. If this is the case, our setting of the on-grid electricity price may lead to an overestimation of the marginal CO2 abatement cost. As we have mentioned above, the main weakness of the LP approach is that it cannot take statistical noise into consideration. Nevertheless, the bootstrapping method pioneered by Efron (1979) provides a possible way to measure statistical precision. We calculate the bootstrapping standard error for parameter θ^ j

B B 1 ∑b = 1 (θ^jb − θ¯j )2 , where θ¯j = B ∑b = 1 θ^jb and B is the bootstrapping repetitions. Table 4 reports the parameter estimates and their corresponding bootstrapping standard errors and normal-based 95% confidence intervals. The bootstrapping procedure is implemented by MATLAB. Table 5 provides the estimation results of the directional output distance functions, the potential of an electricity output increase and the potential of a CO2 emissions reduction. As previously mentioned, the estimated directional output distance function measures the maximum possible augment of the electricity output and reduction of CO2 emissions simultaneously for the power plant. It can serve as an index for measuring the environmental efficiency. A value of zero signifies that the power plant is completely efficient (producing at the frontier), whereas a positive value implies the existence of inefficiency. From Table 5, we can observe that, on average, the power plants' production in 2004 and 2008 was not completely efficient. The average environmental efficiency in 2008 was slightly lower than that in 2004, implying a decrease in relative efficiency. However, the number of frontier plants in 2008 is 13, which is slightly higher than the figure (8 frontier plants) in 2004. The potential increases in electricity outputs relative to the fully efficient plants in 2004 and 2008 are 0.46  1012 kWhs and 1.06  1012 kWhs, respectively, which is approximately 48% and 64% of the total electricity output of the sample in the corresponding year. The potential reductions in CO2 emissions relative to the fully efficient plants in 2004 and 2008 are 4.02  108 ton and 9.36  108 ton, respectively, which is approximately 52% and 70% of the total CO2 emissions of the sample in the corresponding year. These figures reflect the fact that if the technical inefficiencies of the power plants relative to the frontier producers can be eliminated, then the electricity output and the environmental quality of China will be improved greatly; it is a win-win solution. Table 6 provides a summary of the statistics for the shadow price estimates. The shadow price is a type of opportunity cost, i.e., it reflects the cost (measured in electricity output) that the plant has to yield for an additional unit of CO2 emissions reduction,

asseboot (θ^j ) =

1 B−1

Table 3 Statistical summary of inputs and outputs. Year

Variable

Unit

Obs

Mean

Std. dev.

Min

Max

2004

Electricity CO2 emissions Labor Capital Energy

kWh Ton Person 1000 Yuan Ton SCE

518 518 518 518 518

17.83  108 1,486,679 788 700,997 647,006

26.60  108 2,044,886 813 1,238,949 886,966

0.01  108 1650 16 327 714

153.24  108 1.24  107 7453 9,489,340 5,354,045

2008

Electricity CO2 emissions Labor Capital Energy

kWh Ton Person 1000 Yuan Ton SCE

640 640 640 640 640

25.76  108 2,075,765 638 1,380,435 901,304

33.24  108 2,634,511 946 1,793,779 1,143,412

0.02  108 2567 1 295 1112

231.53  108 2.05  107 13,326 1.09  107 8,874,505

352

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

Table 4 Parameter estimates and bootstrapping confidence intervals. Coefficient

Variable

α0 α1 α2 α3 β 1 ¼ γ1  1 α11 α12 ¼α21 α13 ¼ α31 α22 α23 ¼ α32 α33 β11 ¼ μ11 ¼ γ11 δ11 ¼ η11 δ21 ¼ η21 δ31 ¼ η31

2004

Constant L K E y L2 LK LE K2 KE E2 y2 Ly Ky Ey

2008

Estimates

Std. err.

95% Conf. interval

Estimates

Std. err.

95% Conf. interval

 0.0005 0.0011 0.0288 0.0325  0.3407  0.0009  0.0041 0.0107  0.0114  0.0133  0.0299  0.0253  0.0004 0.0199 0.0000

0.0018 0.0091 0.0140 0.0655 0.0371 0.0114 0.0175 0.0146 0.0170 0.0240 0.0696 0.0129 0.0002 0.0101 0.0004

 0.0040  0.0168 0.0013  0.0960  0.4134  0.0232  0.0385  0.0179  0.0448  0.0603  0.1663  0.0507  0.0009 0.0002  0.0008

0.0002 0.0021  0.0156  0.0570  0.2857  0.0007 0.0009  0.0024 0.0056 0.0091 0.0502  0.0301 0.0000 0.0000 0.0000

0.0007 0.0024 0.0114 0.0484 0.0323 0.0037 0.0122 0.0263 0.0149 0.0177 0.0565 0.0153 0.0020 0.0113 0.0000

 0.0011  0.0026  0.0379  0.1519  0.3491  0.0080  0.0231  0.0538  0.0237  0.0257  0.0605  0.0601  0.0039  0.0221 0.0000

0.0030 0.0190 0.0563 0.1610  0.2679 0.0214 0.0302 0.0393 0.0219 0.0337 0.1065 0.0000 0.0001 0.0396 0.0008

0.0015 0.0068 0.0068 0.0379  0.2223 0.0067 0.0249 0.0491 0.0348 0.0438 0.1609  0.0001 0.0039 0.0221 0.0000

Note: Bootstrapping with 1000 repetitions and normal-based 95% confidence intervals are reported. Table 5 Estimates of environmental efficiency. Year

Obs. Mean Std. dev. Min Max No. of frontier plants

2004 518 2008 640

0.25 0.29

0.27 0.37

0 0

1.84 8 3.61 13

Potential electricity increase (kWh)

Potential CO2 reduction (ton)

0.46  1012 1.06  1012

4.02  108 9.36  108

assuming the plant is producing on the production frontier. The shadow price is also considered to be the marginal abatement cost (Coggins and Swinton, 1996; Wei et al., 2013). Table 6 shows that the average marginal CO2 abatement cost was 955 Yuan/ton in 2004 and 1142 Yuan/ton in 2008. The shadow prices in both years exhibit a wide variation, ranging from 325 Yuan/ton to 1557 Yuan/ ton in 2004 and from 98 Yuan/ton to 1427 Yuan/ton in 2008. It is worth noting that the average shadow price has increased with time, which implies that China has to sacrifice more to achieve its greenhouse gas reduction target over time. Fig. 1 plots the kernel densities of the shadow price estimates for the year 2004 and 2008. From Fig. 1, we can observe that the peak of the kernel density curve for 2008 moves to the right side of that for 2004, indicating that the number of the power plants with higher shadow price increased over time. Specifically, most of the shadow price estimates were distributed around 1000 Yuan/ ton in 2004, while they move to 1400 Yuan/ton in 2008. It is interesting to compare our estimated shadow prices with those of the previous studies. The comparison results are reported in Table 7. From the table, we can observe that the estimates of the previous studies lie within a wide range. Different approaches employed in these studies are one of the major reasons for disparities in the results. Additionally, the choice of different dataset and sample period may also affect the results. Among these previous studies, Wei et al. (2013) is directly comparable with this study since both of the papers investigate the power plants in China and employ a similar approach. They Table 6 Estimates of shadow price. Year

Obs.

Mean

Std. dev.

Min

Max

2004 2008

518 640

955 1142

169 296

325 98

1557 1427

Fig. 1. Kernel density of the shadow price.

report a mean shadow price of 2059.8 Yuan/ton using the LP approach, which is a bit higher than our results. It is reasonable since they focus on the power plants located in Zhejiang Province while our sample covers the whole country. As we know, the development of China is unbalanced among regions. The energy efficiency of power plants located in Zhejiang Province may be much higher than those located in other regions, since Zhejiang Province is one of the most developed regions in China. It is also interesting to compare our estimated shadow price with the market price of carbon emissions trading. China has launched seven pilot regional carbon trading markets since 2013 in Beijing, Shanghai, Guangdong, Tianjin, Shenzhen, Hubei and Chongqing. We find that our estimated CO2 shadow prices exceed the prices reported by these pilot carbon trading markets.7 The possible reason is that the shadow price represents the opportunity cost, i.e. the value of electricity output that must be forgone to achieve the reduction at the margin, whereas the market price reflects the supply and demand of permits. Accordingly, the market price does not necessarily reflect all the abatement costs 7 Currently, the market prices of the seven pilot regional carbon trading markets are relatively low. The price for the Beijing market on May 26, 2015 is only about 48 Yuan/ton and the prices for the other pilot markets are even lower. More price information of the pilot carbon markets in China is available at: http://www. tanjiaoyi.com/.

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

Table 7 Comparison of shadow price estimates. Studies

Method

Sample

Wang et al. (2011) DDF/DEA 30 Provinces, 2007 Choi et al. (2012) SBM/DEA 30 Provinces, 2001–2010 Lee and Zhang (2012) Yuan et al. (2012) Wei et al. (2013)

SDF/LP

30 Industries, 2009

DDF/DEA

24 Industries, 2004 and 2008 124 Power plants, 2004

DDF/LP DDF/LM Zhang and Xie (in DDF/DEA press) This paper DDF/LP

Electronic information industry, 1980–2012 518 Power plants, 2004 640 Power plants, 2008

Shadow Price 475.3 Yuan/ton 41.2–46.9 Yuan/ ton 19.7 Yuan/ton 200–120,300 Yuan/ton 2059.8 Yuan/ton 612.6 Yuan/ton 115 Yuan/ton 955 Yuan/ton 1142 Yuan/ton

Notes: SDF, DDF, LP, LM, SBM, DEA denote Shephard Distance Function, Directional Distance Function, Linear Programming, Maximum Likelihood, Slacks-based Measurement, Data Envelopment Analysis, respectively. Choi et al. (2012) and Lee and Zhang (2012) report their results in dollars. We transform them to RMB by the exchange rate 1 dollar ¼6.3 RMB Yuan.

(Smith, et al., 1998; Wei, et al., 2013). Shadow price could be considered as the upper-bound of CO2 reduction cost. 3.2. Impact factors of the shadow price and environmental efficiency To investigate the potential impact factors of the shadow price and environmental efficiency, we consider the following crosssectional regression model:

qi = α + βXi + εi

(10)

where qi represents the shadow price of CO2 emissions or the environmental efficiency, taking the logarithmic form for the shadow price; Xi is the impact factor; εi is the error term; and α and β are parameters that have to be estimated. Specifically, we consider the following impact factors: Ownership (denoted by SOE). The existing literature show that state-owned power plants exhibit a lower production efficiency relative to their non-state-owned counterparts (Du et al., 2013; Gao and Van Biesebroeck, 2014). Our dataset includes information on the ownership structure for each plant. Specifically, the ownerships of the plants are classified into 10 types in our dataset, i.e., state-owned, collectively owned, privately owned, Hong Kong/ Macao/Taiwan-owned, foreign-owned, joint-venture, limited liability company, limited stock company, joint-equity cooperative company, and others. We use a dummy SOE to indicate the ownership of the plant. If the plant is state-owned, the dummy variable SOE is equal to 1, otherwise it is equal to 0. Scale (denoted by SCALE). It has been found that a scale economy exists in the power generation industry, i.e., a larger power station is typically associated with a higher energy efficiency and a lower carbon intensity (Christensen and Greene, 1976). To capture the scale effect, we include the total assets of the power plants in our regressions. We use the variable SCALE to denote the scale of the plant, and take the logarithmic form for the variable. Age (denoted by AGE). A new power generation plant typically has a higher environmental efficiency than an old one. China's power generation industry has experienced rapid technological progress during the past decades, i.e., the newly built power plants typically have more advanced technology relative to the old plants (Du et al., 2009). Our dataset has information on the establishment time of each plant. It is easy to derive the age of each plant by subtracting the establishment year of the plant from the survey year. We use the variable AGE to denote the age of the plant, and take the logarithmic form for the variable. Energy consumption structure (denoted by ratio_coal). The

353

CO2 emissions from the combustion of different fuels largely differ (Auffhammer and Carson, 2008; Du et al., 2012). Coal emits much more CO2 than that is emitted by oil and natural gas despite burning the same quantity of energy (Zhang, 2000). Our dataset provides information on the quantities of coal, oil and natural gas consumption for each plant. Thus, we are able to calculate the ratio of coal consumption for each plant by converting all types of energy consumption into SCE. We use raio_coal to denote the energy consumption structure of the power plants. Subsidy (denoted by per_subsidy). Energy subsidies from governments are somewhat universal in China, which may affect the efficiency and productivity of the power plants (Lin and Jiang, 2011). Specifically, China's coal-fired power plants receive two types of government subsidies. The first type is designed to reduce SO2 emissions. Any plant equipped with a SO2 scrubber receives a subsidy of 1.5 cents per kilowatt-hour to remedy their costs. Scrubbers consume energy and, therefore, increase CO2 emissions (Xu, 2011). The second type (typically implemented by local governments) is designed to reduce power shortages during peak periods. This type of subsidy may encourage the units to operate at excess capacity, consequently reducing the energy and carbon efficiencies of subsidized power plants. We include subsidy information in monetary terms. We divide the subsidy of each plant by its electricity output to derive the value of the subsidy per kWh, which is denoted by the variable per_subsidy. Location (denoted by EAST and WEST). The regional development of China is unbalanced. Consequently, power plants located in different regions may have different productivities. Researchers usually divide mainland China into three regions: the eastern region, western region and central region.8 The eastern region is more developed and richer than the other two regions but lacks natural energy resources. The western region is the least developed region, but it is rich in natural energy resources, especially coal. Two dummies, EAST and WEST, are created to capture the impact of location. Time trend (denoted by time). The power plant's production technology may be improved over time. We create a dummy variable, time, to capture any possible technological progress and other time trends. The time trend is equal to 1 for the data from 2008 and 0 for those from 2004. Table 8 provides a summary of the key variables. There are wide variations in the scale and age of the power plants, as evidenced by the high standard deviations. The average subsidy per kWh is approximately 0.01 Yuan, with a standard deviation of 0.04 Yuan/kWh. Some plants receive no subsidies but others receive a subsidy much higher than average. The average ratio of coal consumption is quite high and the variation is not large. Table 9 includes the estimation results. Model 1 is the regression of the shadow price on the independent variables and Model 2 is the regression of the directional output distance function (the environmental efficiency) on the independent variables. To handle the possible problem of heteroscedasticity for the OLS estimation based on cross-sectional data, we adjust the variance-covariance matrix using the Huber/White/sandwich robust estimator. We also check the multi-collinearity problems in the regressions using the Variance Inflation Factor (VIF) tests. The results show that all of the figures of the VIFs are lower than 1.5, implying that the multicollinearity problems in our regressions are not severe. The result of Model 1 in Table 9 shows that the marginal CO2 8 The eastern region of China includes 11 provinces: Beijing, Tianjin, Hebei, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, Hainan and Liaoning. The western region of China includes 12 provinces: Inner Mongolia, Guangxi, Sichuan, Chongqing, Guizhou, Yunnan, Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang and Tibet. The central region of China includes 8 provinces: Shanxi, Anhui, Jiangxi, Henan, Hubei, Hunan, Heilongjiang and Jilin.

354

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

Table 8 Summary statistics of variables. Variable

Unit

SCALE

1000 Yuan AGE Year per_subsidy Yuan/ kWh ratio_coal – SOE – time – EAST – WEST –

Obs.

Mean

1158 1,574,480

Std. dev.

Min

Max

2,264,342

2856

1.61  107

1158 1158

15.39 0.01

15.49 0.04

1158 1158 1158 1158 1158

0.99 0.26 0.55 0.57 0.17

0.03 0.44 0.50 0.49 0.37

1 0

135 0.87

0.46 0 0 0 0

1 1 1 1 1

abatement cost of the state-owned power plants is not significantly different from that of the non-state-owned power plants. The effect of subsidies on the shadow price is positive at the 1% significance level. Specifically, if the subsidy increases by 1 Yuan/kWh, the marginal CO2 abatement cost increases by 57% on average. This effect may be due to the sulfur scrubber subsidy or the output shortage subsidy. However, we are inclined to attribute this effect to the output shortage subsidy because the sulfur scrubber subsidy began in 2007, and only 60% of the coal-fired power plants were installed with SO2 scrubbers by the end of 2008. The result is consistent with our expectation. The output shortage subsidy may induce the power plants to operate at excess capacity and, consequently, make it more difficult to reduce CO2 emissions. The result also shows that older power plants have a lower marginal CO2 abatement cost. On average, a 1% increase in age will result in a 4.1% decrease in the shadow price, and the coefficient is significant at the 1% level. This is reasonable because the gap between an old plant and the frontier plants is usually larger relative to that for a new plant. The coefficient of the variable SCALE is negative at the 1% significance level. This implies that scale expansion can help the power plants reduce CO2 emissions at a lower cost. However, even at the 10% significance level, the effect of the energy consumption structure of the power plants is not significant. The reason may be that oil and gas consumption only accounts for a very small share of the total energy consumption for Table 9 Estimation results. Dependent variable

Model 1 ln_shadowprice

Model 2 ddf

SOE

 0.005 (  0.26) 0.570nnn (3.07)  0.041nnn (  5.50)  0.114nnn (  20.06)  0.156 (  1.22) 0.248nnn (19.18)  0.022 (  1.51)  0.056nn (  2.54) 8.573nnn (53.74) 1158 0.456

 0.023 (  1.24)  0.484nnn (  3.90) 0.050nnn (6.13) 0.129nnn (22.16) 0.188nn (2.56)  0.061nnn (  4.66)  0.063nnn (  4.51) 0.123nnn (3.85)  1.673nnn (  12.81) 1158 0.468

per_subsidy ln_AGE ln_SCALE ratio_coal time EAST WEST Constant N Adj. R2

Note: t statistics in parentheses. np o0.1. nn

p o0.05. p o0.01.

nnn

the power plants. The coefficient of the time trend is positive and significant at the 1% level, which implies an increase in the shadow price over time. On average, the marginal CO2 abatement cost of 2008 is 24% higher than that of 2004. The dummies for location show that the power plants located in the western region have a lower average shadow price relative to that of the benchmark group (the central region), however those located in the eastern region do not exhibit a significant difference to the benchmark group. The result of Model 2 in Table 9 shows that the environmental efficiencies of the state-owned and non-state-owned power plants are not significantly different. A higher subsidy is helpful for improving the environmental efficiency of the power plants, whereas age and scale are harmful to the environmental efficiency of the power plants. A plant with a higher ratio of coal consumption presents a significantly lower environmental efficiency. The coefficient of the time trend is negative, implying that the environmental efficiency of the power plants (on average) has improved with the time. The result is significant at the 1% level. The coefficients of the location dummies show that, relative to the baseline group (the central region), the environmental efficiency of the eastern region is higher, whereas that of the western region is lower. This is consistent with our expectation.

4. Conclusions and policy implications 4.1. Conclusions In this paper, we estimate the environmental efficiency, reduction potential and marginal abatement cost of CO2 emissions for China's coal-fired power plants. To that end, we make full use of two datasets. One is derived from the first and second waves of the National Economic Survey implemented in 2004 and 2008, respectively, and the other is from the Compilation of Statistical Data of the Power Industry in the corresponding years. Compared to the data used in the previous studies, our newly constructed dataset covers a much wider range of China's coal-fired power plants and provides more information. Thus, our estimation results are more reliable and may be more conclusive. We further analyze the effect of subsidies on the environmental efficiency and shadow price of the power plants, which have typically been ignored in the literature. The results indicate that there are great opportunities for CO2 emissions reduction in China's coal-fired power plants. If all of the power plants were operating at their frontiers, then the CO2 emissions in 2004 and 2008 could have been reduced by 52% and 70%, respectively. The average marginal abatement cost of CO2 emissions for the power plants is approximately 955 Yuan/ton in 2004 and 1142 Yuan/ton in 2008. We also find that the subsidies from the government increase the shadow price but significantly reduce the environmental inefficiency of the power plants. The older and larger power plants have a lower environmental efficiency but can reduce the CO2 emissions at a lower cost. The ratio of coal consumption has a negative effect on the environmental efficiencies of the power plants. Compared to the central region, the western region has both a lower environmental efficiency and a lower shadow price, whereas the eastern region only exhibits a relatively higher environmental efficiency. It is worth to note that, in this paper, we simply assume that all the plants share a common production frontier, while possible plant heterogeneities are neglected. Actually, different types of firms may perform differently (O'Donnell et al., 2008). One possible way to consider plant heterogeneity is to implement a metafrontier analysis, which classifies the plants into different groups according to their characteristics first and then estimates the

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

group-specific production frontier and overall meta-frontier (Battese et al., 2004; Oh, 2010). It is a possible direction for future research. 4.2. Policy implications The Chinese government has committed to reducing its carbon intensity by 40–45% by the year 2020 compared with the 2005 level and to achieving a peaking of CO2 emissions by approximately 2030. To achieve these reduction goals, it is essential to scientifically formulate a low carbon policy for the power sector because it accounts for approximately half of China's total CO2 emissions. Thus, our findings have important policy implications for China's greenhouse gas reductions. First, the environmental performance of China's coal-fired power plants substantially differs. This implies that there are large opportunities for further CO2 emissions reductions and simultaneous electricity output expansions for coal-fired power plants that are not completely efficient. That is to say, ‘double dividend’ opportunities do exist. What policy-makers must do is to provide various incentives that encourage power plants to improve their technical efficiencies. Second, substantial variations in the marginal CO2 abatement cost exist among power plants, which implies that different plants should undertake different CO2 abatement tasks from the perspective of social cost minimization. However, cost is only one element that warrants consideration. Fairness is another dimension that the government must consider. The government should take balance between these two dimensions when CO2 abatement tasks are allocated to the power plants by administrative orders. Third, the market-oriented emissions trading system is beneficial to the power plants' carbon reduction. Currently, aside the existing seven pilot regional carbon trading markets, the Chinese government is planning to construct a nationwide carbon trading market. The coal-fired power generation plants will certainly be included as the primary participants of the carbon trading market because of their huge CO2 emissions. The shadow price estimation in this paper can then provide the pending carbon trading market with a benchmark for the initial market price setting. Finally, the regression results show that replacing the technological obsolescent old units with more advanced equipment is an efficient way to improve the environmental performance of the power plants. Environmental subsidies are a useful policy tool for accelerating this technological replacement. However, we should also concern that long-term subsidies may reduce the productive efficiency of the plants because of the problem of soft constraints. Therefore, a subsidy policy should be combined with corresponding competition policies, such as implementing wholesale electricity competition.

Acknowledgments We truthfully acknowledge having received financial support from the National Natural Science Foundation of China (Nos. 71303213 and 71433002), the Ministry of Education of China (No. 13YJA790015), the Zhejiang Provincial Natural Science Foundation of China (No. LQ12G03013), and the Fundamental Research Funds for the Central Universities in China.

References Aigner, D.J., Chu, S.F., 1968. On estimating the industry production function. The. Am. Econ. Rev. 58, 826–839. Auffhammer, M., Carson, R.T., 2008. Forecasting the path of China's CO2 emissions

355

using province-level information. J. Environ. Econ. Manag. 55, 229–247. Battese, G.E., Rao, D.P., O'Donnell, C.J., 2004. A metafrontier production function for estimation of technical efficiencies and technology gaps for firms operating under different technologies. J. Prod. Anal. 21, 91–103. Chambers, R.G., Chung, Y., Färe, R., 1998. Profit, directional distance functions, and Nerlovian efficiency. J. Optim. Theory Appl. 98, 351–364. Choi, Y., Zhang, N., Zhou, P., 2012. Efficiency and abatement costs of energy-related CO2 emissions in China: a slacks-based efficiency measure. Appl. Energy 98, 198–208. Christensen, L.R., Greene, W.H., 1976. Economies of scale in U.S. electric power generation. J. Polit. Econ. 84, 655–676. Chung, Y.H., Färe, R., Grosskopf, S., 1997. Productivity and undesirable outputs: a directional distance function approach. J. Environ. Manag. 51, 229–240. Coggins, J.S., Swinton, J.R., 1996. The price of pollution: a dual approach to valuing SO2 allowances. J. Environ. Econ. Manag. 30, 58–72. Du, L., Hanley, A., Wei, C., 2015. Marginal abatement costs of carbon dioxide emissions in China: a parametric analysis. Environ. Resour. Econ. 61, 191–216. Du, L., He, Y., Yan, J., 2013. The effects of electricity reforms on productivity and efficiency of China's fossil-fired power plants: an empirical analysis. Energy Econ. 40, 804–812. Du, L., Mao, J., Shi, J., 2009. Assessing the impact of regulatory reforms on China's electricity generation industry. Energy Policy 37, 712–720. Du, L.M., Wei, C., Cai, S.H., 2012. Economic development and carbon dioxide emissions in China: provincial panel data analysis. China Econ. Rev. 23, 371–384. Efron, B., 1979. Bootstrap methods: another look at the jackknife. Ann. Stat. 7, 1–26. Färe, R., Grosskopf, S., Lovell, C.A.K., Yaisawarng, S., 1993. Derivation of shadow prices for undesirable outputs: a distance function approach. Rev. Econ. Stat. 75, 374–380. Färe, R., Grosskopf, S., Noh, D.W., Weber, W., 2005. Characteristics of a polluting technology: theory and practice. J. Econom. 126, 469–492. Gao, H., Van Biesebroeck, J., 2014. Effects of deregulation and vertical unbundling on the performance of China's electricity generation sector. J. Ind. Econ. 62, 41–76. IEA, 2014. World Energy Outlook 2014, London. IPCC, 2006. 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Institutefor Global Environmental Strategies, Hayama, Kanagawa, Japan. Kaneko, S., Fujii, H., Sawazu, N., Fujikura, R., 2010. Financial allocation strategy for the regional pollution abatement cost of reducing sulfur dioxide emissions in the thermal power sector in China. Energy Policy 38, 2131–2141. Lee, J.-D., Park, J.-B., Kim, T.-Y., 2002. Estimation of the shadow prices of pollutants with production/environment inefficiency taken into account: a nonparametric directional distance function approach. J. Environ. Manag. 64, 365–375. Lee, M., 2005. The shadow price of substitutable sulfur in the US electric power plant: A distance function approach. J. Environ. Manag. 77, 104–110. Lee, M., Zhang, N., 2012. Technical efficiency, shadow price of carbon dioxide emissions, and substitutability for energy in the Chinese manufacturing industries. Energy Econ. 34, 1492–1497. Lin, B., Jiang, Z., 2011. Estimates of energy subsidies in China and impact of energy subsidy reform. Energy Econ. 33, 273–283. Maradan, D., Vassiliev, A., 2005. Marginal costs of carbon dioxide abatement: empirical evidence from cross-country analysis. Rev. Suisse Econ. Stat. 141, 377. Marklund, P.-O., Samakovlis, E., 2007. What is driving the EU burden-sharing agreement: efficiency or equity? J. Environ. Manag. 85, 317–329. Matsushita, K., Yamane, F., 2012. Pollution from the electric power sector in Japan and efficient pollution reduction. Energy Econ. 34, 1124–1130. Murty, M.N., Kumar, S., Dhavala, K.K., 2007. Measuring environmental efficiency of industry: a case study of thermal power generation in India. Environ. Resour. Econ. 38, 31–50. O, Donnell, C., Rao, D.S.P., Battese, G., 2008. Metafrontier frameworks for the study of firm-level efficiencies and technology ratios. Empir. Econ. 34, 231–255. Oh, D.-h, 2010. A metafrontier approach for measuring an environmentally sensitive productivity growth index. Energy Econ. 32, 146–157. Rezek, J.P., Campbell, R.C., 2007. Cost estimates for multiple pollutants: a maximum entropy approach. Energy Econ. 29, 503–519. Shephard, R., Gale, D., Kuhn, H., 1970. Theory of Cost and Production Functions. Princeton University Press, Princeton. Simar, L., Wilson, P.W., 2000. A general methodology for bootstrapping in nonparametric frontier models. J. Appl. Stat. 27, 779–802. Smith, A.E., Platt, J., Ellerman, A.D., 1998. The costs of reducing utility SO2 emissions —not as low as you might think. Public Utilities Fortnight. Song, M., An, Q., Zhang, W., Wang, Z., Wu, J., 2012. Environmental efficiency evaluation based on data envelopment analysis: a review. Renew. Sustain. Energy Rev. 16, 4465–4469. Sueyoshi, T., Goto, M., 2013. A comparative study among fossil fuel power plants in PJM and California ISO by DEA environmental assessment. Energy Econ. 40, 130–145. Swinton, J., 2004. Phase I completed: an empirical assessment of the 1990 CAAA. Environ. Resour. Econ. 27, 227–246. Vardanyan, M., Noh, D.-W., 2006. Approximating pollution abatement costs via alternative specifications of a multi-output production technology: a case of the US electric utility industry. J. Environ. Manag. 80, 177–190. Wang, Q., Cui, Q., Zhou, D., Wang, S., 2011. Marginal abatement costs of carbon dioxide in China: a nonparametric analysis. Energy Procedia 5, 2316–2320. Wei, C., Löschel, A., Liu, B., 2013. An empirical analysis of the CO2 shadow price in Chinese thermal power enterprises. Energy Econ. 40, 22–31.

356

L. Du, J. Mao / Energy Policy 85 (2015) 347–356

Wei, C., Ni, J.L., Du, L.M., 2012. Regional allocation of carbon dioxide abatement in China. China Econ. Rev. 23, 552–565. Xie, B.-C., Fan, Y., Qu, Q.-Q., 2012. Does generation form influence environmental efficiency performance? An analysis of China's power system. Appl. Energy 96, 261–271. Xu, Y., 2011. improvements in the operation of SO2 scrubbers in China's coal power plants. Envion. Sci. Technol. 45, 380–385. Yang, H., Pollitt, M., 2009. Incorporating both undesirable outputs and uncontrollable variables into DEA: the performance of Chinese coal-fired power plants. Eur. J. Oper. Res. 197, 1095–1105. Yuan, P., Liang, W., Cheng, S., 2012. The margin abatement costs of CO2 in Chinese industrial sectors. Energy Procedia 14, 1792–1797. Zhang, N., Choi, Y., 2013. Total-factor carbon emission performance of fossil fuel power plants in China: a metafrontier non-radial Malmquist index analysis. Energy Econ. 40, 549–559. Zhang, N., Choi, Y., 2014. A note on the evolution of directional distance function and its development in energy and environmental studies 1997–2013. Renew.

Sustain. Energy Rev. 33, 50–59. Zhang, N., Xie, H., 2015. Toward green IT: modeling sustainable production characteristics for Chinese electronic information industry, 1980–2012. Technol. Forecast. Soc. Change 96, 62–70. Zhang, Z., 2000. Decoupling China's carbon emissions increase from economic growth: an economic analysis and policy implications. World Dev. 28, 739–752. Zhou, P., Ang, B.W., Han, J.Y., 2010. Total factor carbon emission performance: a Malmquist index analysis. Energy Econ. 32, 194–201. Zhou, P., Ang, B.W., Poh, K.L., 2008. A survey of data envelopment analysis in energy and environmental studies. Eur. J. Oper. Res. 189, 1–18. Zhou, P., Ang, B.W., Wang, H., 2012. Energy and CO2 emission performance in electricity generation: a non-radial directional distance function approach. Eur. J. Oper. Res. 221, 625–635. Zhou, P., Zhou, X., Fan, L.W., 2014. On estimating shadow prices of undesirable outputs with efficiency models: a literature review. Appl. Energy 130, 799–806.