You are on page 1of 13

Science of the Total Environment 690 (2019) 1190–1202

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

An evaluation of greenhouse gas emission efficiency in China's industry


based on SFA
Jingchao Sun, Tao Du ⁎, Wenqiang Sun, Hongming Na, Jianfei He, Ziyang Qiu, Yuxing Yuan, Yingnan Li
SEP Key Laboratory of Eco-Industry, Northeastern University, Shenyang, Liaoning 110819, People's Republic of China
School of Metallurgy, Northeastern University, Shenyang, Liaoning 110819, People's Republic of China

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• China's GHG efficiency at industry level


is assessed by stochastic frontier analy-
sis.
• At night, net electricity is better than
small-scale on-site coal-fired power
plant.
• Technology advance is better to im-
prove efficiency and mitigate GHG
emissions.
• Energy efficiency and GHG efficiency are
dependent but loosely related indexes.

a r t i c l e i n f o a b s t r a c t

Article history: As one of the largest countries with sound industrial systems in the world, China is a major emitter of greenhouse
Received 21 May 2019 gases. In order to achieve sustainable development, the analysis of greenhouse gas emissions from industry is of
Received in revised form 5 July 2019 great significance. This research evaluates the greenhouse gas emission efficiency (GHG efficiency) at industry
Accepted 6 July 2019
level in 26 sectors of China and analyzes the impacts of its determinants using stochastic frontier approach. Fur-
Available online 8 July 2019
thermore, the correlations for GHG efficiency with its determinants and other proxies are estimated by Kendall's
Editor: Huu Hao Ngo rank analysis. Industry level data from Chinese processing and manufacturing industries spanning over the pe-
riod 2000–2016 are used for this analysis. Results show that there has little potential to improve GHG efficiency,
Keywords: then technology progress is necessary. The GHG efficiency performance responds both to changes in the propor-
GHG efficiency tion of net electricity use and the ratio of electricity price to coal price; meanwhile the technology capacity also
Stochastic frontier analysis has a positive impact. Besides, three main proxies of GHG intensity, energy efficiency and energy intensity
Correlation have their own significance respectively, but could not represent GHG efficiency.
China's emission trading scheme (ETS) © 2019 Elsevier B.V. All rights reserved.

1. Introduction

As the world's largest emitter of greenhouse gases currently, China


has been the main pusher and catalyst for Global Climate Governance
⁎ Corresponding author at: SEP Key Laboratory of Eco-Industry, Northeastern
according to Greenpeace (Greenpeace, 2018). Referring to the
University, Shenyang, Liaoning 110819, People's Republic of China. Greenpeace and Intergovernmental Panel on Climate Change, the total car-
E-mail address: dut@smm.neu.edu.cn (T. Du). bon dioxide emission of China in 2017 is 9232.6 million tons, accounting

https://doi.org/10.1016/j.scitotenv.2019.07.093
0048-9697/© 2019 Elsevier B.V. All rights reserved.
J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202 1191

for about a quarter of global carbon dioxide emission, which leads to the results such as the more molten iron is used, the more carbon dioxide
deterioration of living environment and severe air pollution is emitted in the steel industry (Na et al., 2017). Brännlund et al. ana-
(Greenpeace, 2018; Shen and Wang, 2013). The primary cause of this lyzed carbon intensity performance and the effectiveness of the Swed-
phenomenon is the side effects of rapid China's industrial development ish CO2 tax (Brännlund et al., 2014). It found that Firms' carbon
(Gao et al., 2018). To face a series of environmental impact challenges intensity performance had reactions with the changes of the CO2 tax
(Sun et al., 2019a, 2019b) such as global warming caused by greenhouse and fossil fuel price, but was more sensitive to the tax.
gases, China started Emission Trade Scheme(ETS) in 2013, and it has im- As discussed in the previous studies, there are four main energy effi-
plemented seven pilot carbon markets in various zones, including Bei- ciency evaluation methods. In order to evaluate the efficiency in indus-
jing, Chongqing, Guangzhou, Hebei, Shanghai, Shenzhen and Tianjin to try reasonable, the author illustrates these four methods, including
achieve the promise that China reduces the carbon intensity per unit stochastic frontier analysis, data envelopment analysis, exergy analysis
of GDP by 60–65% by 2030 in the Conference of Parties (Lin and Jia, and benchmark comparison (Li and Tao, 2017). In addition, it is found
2019). Coal is the main fossil resources and energy in China, while it is that the analysis of the relationship between overall energy efficiency
also a greenhouse gas producer. The changes have taken place in the and influencing factors is applicable to econometric models. They also
structure of energy consumption used in manufacturing and processing made a review of energy efficiency estimation using these four different
industries in recent years that are mainly in the proportion of net elec- methods (Li and Tao, 2017). Yang et al. did similar research along with
tricity in entire energy consumption. Of course, the prices of electricity SFA and data environment analysis (DEA) as well, and comparing the
and coal are essential factors affecting greenhouse gas efficiency as well. differences between these two methods (Yang and Lei, 2016).
The global CO2 efficiency was evaluated among 170 countries from The energy efficiency and carbon dioxide emission are evaluated by
1997 to 2007 using a stochastic cost frontier, which indicated that the method of input-output analysis (Lin and Sun, 2010). Cheng et al.
there are great differences in efficiency level among countries and estimated energy efficiency, carbon dioxide emission efficiency and an-
China has the lowest level though CO2 efficiency also improved like alyzed its impacts using DEA (Cheng et al., 2018). As for stochastic fron-
other counties, and the author put forward some political advise at last tier approach, which is adopted in this research, and the different forms
(Herrala and Goel, 2012). Furthermore, it was estimated at province of SFA are used to evaluate energy efficiency in the diverse field. These
level as well. Jiang et al. used stochastic frontier model by means of results show that SFA is a suitable method to estimate energy efficiency
translog production function to evaluate the technical energy efficiency and energy inefficiency, which are so important for policymakers
on the basis of panel data on 29 Chinese provinces from 2003 to 2011 (Filippini and Charles Hunt, 2011; Filippini and Hunt, 2015; Filippini
(Jiang et al., 2016). Then, the author used a spatial Durbin error model et al., 2014).
to judge the major determinants of energy efficiency. In order to achieve The robustness of these two measures of Spearman and Kendall cor-
the sustainable development of industry, measuring the carbon emission relation were researched. The Kendall's rank is better than Spearman's
efficiency from 2009 to 2015 in Anhui province in China with the method rank correlation in the aspects of robust and efficient (Croux and
of data envelopment analysis Malmquist analysis. It shows that the trend Dehon, 2010). That energy and other input prices play in reducing the
of carbon efficiency is rising, and the major factor is technical progress energy intensity of manufacturing industries in Canada were studied.
(Qiu and Yu, 2017). On the city level, the purpose was to evaluate energy The results indicate that both capital and labor are complementary to
efficiency for 285 cities in China during 2008–2012 (Wang et al., 2017a). energy (Gamtessa, 2016).
And the other aim was to reduce carbon emission and improve its energy The main purpose of this research is to evaluate emission reduction
efficiency with the slacks-based measure approach. The results showed potential to increase carbon efficiency for 35 Chinese industries cover-
most of the cities had relatively low energy efficiency. However, it also ing the period 2000–2016 using SFA. Carbon intensity is a normal indi-
means great potentials to improve energy efficiencies. cator but as a measure of carbon efficiency is non-applicable as it could
There are also many studies on industry level. Martínez et al. esti- not reflect real efficiency. In this paper, the estimated efficiency scores
mated the energy efficiency and analyzed the trends in energy con- are to be interpreted as revealing not only the progress of the technol-
sumption and CO2 emissions in Swedish manufacturing industries in ogy of energy saving and emission reduction, the changes of industrial
the period 1993–2008 using production frontier model at two-digit structure and energy structure but also the potential to reduce carbon
level (Pardo Martínez and Silveira, 2012). The results showed that emission while the level of production remains unchanged; in other
many factors like electricity consumption, investments, total factor of words, it is the emission reduction potential of industries. And the
productivity, energy prices and energy and CO2 taxes have negative in- other purposes are to investigate thoroughly and analyze the influenc-
fluences on CO2 consumption. Similarly, Weyman-Jones et al. focused ing factors of carbon efficiency which conditional estimate carbon effi-
on the non-residential electric energy efficiency in the Portuguese econ- ciency of industries, including CHN ETS, the energy structure and the
omy measured by the method of stochastic frontier analysis (SFA) ratio of the price of electricity and coal.
(Weyman-Jones et al., 2016) and using the time-series sample of non- The rest of this paper is organized as follows. Section 2 introduces
residential electrical energy consumption from 1970 to 2014 to analyze the methodology and the applied model. Section 3 describes the data
the influencing factors. It showed that measuring energy efficiency im- and its resource. Section 4 shows the results of the estimation and dis-
provement is very critical for energy policy and energy studies. cussions. Section 5 makes a conclusion.
Moreover, Lundgren et al. estimated energy efficiency of the 14 sec-
tors in Swedish manufacturing using single-stage SFA and analyzed the 2. Materials and methods
influencing factors of energy efficiency such as European Union Emis-
sion Trading System (EU ETS), carbon intensity and so on with a firm In this study, in order to estimate the capability of the maximum
level. Results show that carbon efficiency doesn't equal to energy effi- amount of output that can be produced from a given combination of in-
ciency, and the EU ETS has little effect on energy efficiency (Lundgren puts or the ability of reducing the minimum amount of inputs from a
et al., 2016). The efficiency of carbon trading market was evaluated in given output, the stochastic frontier function approach introduced by
China, and it had weak efficiency but still had the signs of restoring mar- Aigner et al. and Meeusen et al. (Aigner et al., 1977; Meeusen and
ket efficiency as results (Zhao et al., 2017). Recently, Löschel et al. inves- Broeck, 1977) is used.
tigated the impact of EU ETS for the efficiency of economic performance In contrast, the nonparametric efficiency analysis methods such as
on the scale of manufacturing firms in Germany (Löschel et al., 2019). Its data envelopment analysis (DEA) decided by their data without any as-
results suggested that the EU ETS had a significant positive impact on sumptions on the specific statistical distribution of the error terms, but
the regulated firms. Na et al. made a detailed analysis on CO2 of the dif- as for stochastic frontier analysis approach, it has the strength of
ferent process in a specific iron and steel manufacturing route with allowing for random shocks and measurement error, and another
1192 J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202

advantage of the SFA approach is that it is useful to analyze the factors of Based on the theoretical framework of Aigner et al. and the ‘true’
and investigate the determinants of producer performance (Cullinane random effects specification of the stochastic production frontier refer-
et al., 2006). ring to Belotti et al. (Aigner et al., 1977; Belotti et al., 2012) is as follows:
There are many SFA models using different assumptions, and the !
ytm  
condition of time invariance is a strong assumption, especially in longer ln t ¼ βm;0 þ βm;L ln Ltm þ βm;K lnK tm þ βm;E −1 lnEtm
Em
panel. With the development of the stochastic frontier production func-
tion, a stochastic frontier analysis model with time-varying inefficiency þ α m ln T þ βm;LK ln Ltm ln K tm þ βm;LE ln Ltm C tm
was built up by Coelli (Battese and Coelli, 1995; Coelli, 1995). It relaxes þ βm;LT ln Ltm ln T þ βm;KE lnK tm ln Etm
this assumption and the modified inefficiency stochastic frontier pro- þ βm;KT lnK tm ln T þ βm;ET ln Etm ln T þ vtm þ utm ð3Þ
duction function is a significant improvement over the former stochas-
where ytm denotes the sell total production value of industry m at year t,
tic frontier functions which do not have the technical inefficiency
L and K represent labor and fixed capital, respectively. T is a time trend
effects. We use the industrial specific distance to the frontier as a mea-
variable in our case according to Greene et al., E stands for the aggregate
sure of the level of energy inefficiency. Green et al. pointed out two
greenhouse gas emission for GHG efficiency or energy use for energy ef-
models to address the issue of unobserved heterogeneity when using
ficiency (Greene and Dan, 2004). In Eq. (3), labor, capital, time trend and
the panel data, the true fixed effects model, and the true random effects
energy use are inputs, and they are treated as independent variables.
model (Greene, 2005). And the true random effects model could con-
We take an assumption of the deterministic frontier to adopt the form
sider unobserved heterogeneity of the two stochastic terms that is for
of a trans-log production function. We use maximum likelihood estima-
time-variant factors. Thus, in order to discuss comprehensive factors
tion to implement the model. The residual two error terms in Eq. (3) are
of the industrial level carbon efficiency in the estimations, the true ran-
a statistical noise term, vtm, and a non-positive random variable
dom effects model is used. The usefulness and advantages of SFA in
depicting inefficiency utm.
Bayesian statistics in over traditional SFA which just uses classical statis-
In the first part, the statistical noise term vtm is assumed to be nor-
tics are proved as well (Carvalho and Marques, 2016).
mally distributed usually that vtm~N(0, σv m2). The second part, utm,
The production process is affected by the compound errors, which
which represents the underlying efficiency and is assumed to have
consists of two different unknown random error terms. One of the
truncated-normal distribution that utm~N(μtm, σtm). In the following
two error components results in the deviation from the production
Eq. (4), utm is denoted as conditional GHG inefficiency, and it is assumed
frontier by statistical objectives themselves and the reasons of the phe-
to be a function of a combination of GHG inefficiency factors. The func-
nomena may be mismanagement, policies or unsatisfactory inputs, they
tion of conditional technical inefficiency is as follows:
all can also be interpreted as the indicators of nonnegative technical in-
efficiency. The other error component is a random variable which is not t
utm ¼ γm;CETS CETStm þ γm;value valuem þ γm;ELE ELEtm þ γmEC EC tm
controlled by themselves, such as weather, statistical error and the dif- þ γ mEC TC tm þ υtm ð4Þ
ferent performances of machinery and employees daily (Kumbhakar
and Lovell, 2000). where CETS denotes the China's National emission trading system,
It is so crucial to estimate the inefficiency and evaluate the drivers of which is a dummy variable taking the value one if industry m joins
inefficiency for policymakers (Battese and Coelli, 1995; Kumbhakar and this scheme. value is annual average price of carbon emissions transac-
Lovell, 2000). These two ways to achieve the estimations of inefficiency tion. ELE is the ratio of purchased electricity to total energy consump-
and inefficiency effects are two stages (Kalirajan, 1981) and single- tion. EC is the ratio of annual average electricity price to annual
stage (Battese and Coelli, 1995). And the single-stage approach is used average coal price. TC is the technical capacity, which indicates the
in the random effects SFA model in this paper in order to estimate the in- total value of out-put per capita. γ are the coefficients of different deter-
efficiency and obtain the conditional determinants of it as it avoids the es- minant variables respectively. υ is the random error term.
timation inconsistency in two-stage approach (Battese and Coelli, 1995). Eq. (4) means that the energy efficiency and GHG efficiency control
Carbon intensity of industry as a relevant criterion is to analyze the variables are displayed in the mean of the exponential distribution of
performance of the climate policy (Brännlund et al., 2014). According utm. υtm is explained as a random error term. Estimated energy efficiency
to the Intergovernmental Panel on Climate Change (IPCC), there are and carbon equivalent efficiency estimates can be translated to efficiency
mt mt
many greenhouse gases as atmospheric polluters such as CO2, N2O, scores and carbon equivalent scores according to EIe, c = exp (−ue, c).
CH4 and especially CO2. Due to the limitation of diverse technology There are three measures that are used to looking into the cross cor-
levels in different industries, the combustion rate is commonly incom- relation between two different values. The Pearson correlation is one of
plete combustion, the average oxidation rate of different fuels is used the most often used in statistics, but Spearman's rank and Kendall's rank
in this paper. We use the emission factors of IPCC to calculate pollutants have more robustness and higher efficiency than Pearson's correlation
emission from industry level in this paper. The function is as follows: (Atkinson et al., 2004). The Kendall's rank measure is more robust and
X more efficient than Spearman's rank correlation, making it the prefera-
EGHG ¼ FQ i  FOij  EF ij  GWPij ð1Þ
ble estimation (Croux and Dehon, 2010). This research will estimate the
where EGHG is greenhouse gas emissions, its unit is carbon dioxide correlation between different values using Kendall's rank measure.
equivalent. FQi is the quantity of fuel i. FOij is oxidization ratio of green- The model approach is based on the assumption of the different in-
house gas j in fueli. EFij is the emission factors of greenhouse gas j in fuel i dustries which have the same characteristic of energy input and sell
. GWP is global warming potential of gas j in fuel i, whose data sources total production value output. In this case, we can get the GHG efficiency
refer to data provided by IPCC. as the index used for measuring technical progress of greenhouse gas
Greenhouse gas intensity is a crucial role in the research field of at- emission reduction and its reduction potential.
mospheric environmental protection, and it is also an indispensable
data basic for policymakers. In this paper, greenhouse gas intensity is 3. Data description
as follows:
The data of this study consist of a panel data which covers 26 Chinese
Em manufacturing sectors above designated size from 2000 to 2016, includ-
cm ¼ GHG ð2Þ
Ym ing Ferrous Metals, Articles, Beverage, Chemical Fiber, Chemical Raw
where cm is greenhouse gas intensity of industry m. Emis greenhouse
GHG Material and Chemical Products, Electronic Equipment, Electrical Ma-
gas emission of industry m, it is calculated by Eq. (1). Ym is gross indus- chinery, Foods, Furniture, General Purpose Machinery, Leather, Measur-
trial output of industry m. ing Instrument, Medicines, Metal Products, Non-metallic Mineral
J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202 1193

Table 1 example, the transport equipment was divided into three sectors after
Descriptive statistics of variables in the stochastic production frontier functions. Mean 2012: automobile, railway-ship-aerospace and other transportation
values 2000–2016.
equipment manufacturing, which are merged in this study. Besides,
Variable/unit Minimum Maximum Mean Standard the transaction prices of variables come from the China's Carbon Price
deviation Survey (Slater et al., 2018), the China Coal Resources Network (Sxcoal.
Output/billion yuan 5.8763 10.8906 8.8437 1.0727 com., 2019) and the National Power Price Regulatory Bulletin
Labor/person 2.9237 6.8499 5.2130 0.8820 (Administration, 2017). The detailed descriptive statistics are shown
Capital stock/billion yuan 5.0947 10.5680 8.1205 1.0916
in Table 1.
Emissions/t CO2e 5.5471 12.5115 8.5573 1.6436
Energy/kg standard coal (kgce) 4.1326 11.6204 7.2219 1.7317 As an output variable for stochastic production frontier model, the value
of production measured in monetary values reflects 2000-year price de-
noted in RMB, the values of other variables are the same as it is. The annu-
Products, Paper, Special Purpose Machinery, Textile, Textile Wearing ally average number of employees in industry indicates the use of labor
Apparel, Tobacco, Transport Equipment, Non-ferrous Metals, Printing, from China Labor Statistics Yearbook (BPESC, 2001–2017). Capital stocks
Agricultural Products, Petroleum, Wood. These data come from the Sta- are calculated by the perpetual inventory method using gross investment
tistics of China (NBSC, 2019) and China Statistical Yearbook (NBSC, data denoted in RMB with reference to China Statistical Yearbook (NBSC,
2001–2017a). Because of the change of the statistical caliber in 2003 2001–2017a). The aggregated energy use is computed based on the elec-
and 2012, more detailed classifications of sectors were collected. For tricity and fuel use is measured in kilogram standard coal (kgce) according
to the Statistics of China (NBSC, 2019) and the China Energy Statistical Year-
Table 2
book (NBSC, 2001–2017b). We calculate the CO2 equivalent emissions from
Descriptive statistics of variables in the inefficiency function Mean values 2000–2016
(standard deviation in parenthesis).
the aggregate fuel use and the purchase of electricity referring to the Na-
tional Power Price Regulatory Bulletin (Administration, 2017), the conversion
NO. Sector Electricity GHG Technical Energy coefficients refer to the Intergovernmental Panel on Climate Change (Shen
proportion intensity capacity intensity
and Wang, 2013) and China Industrial Statistical Yearbook (NBSC,
1 Ferrous metals 0.0919 7.2500 0.7002 8.5371 2001–2017c). The CO2 emissions are measured in t CO2 equivalent (t CO2e).
(0.0109) (2.5598) (0.1830) (0.5819)
The variables in the inefficiency function are given in Eq. (4), includ-
2 Articles 0.4786 0.9974 0.2217 3.0413
(0.0639) (0.4449) (0.0292) (0.3905) ing ETS (CHN ETS dummy variable), electricity proportion (purchasing
3 Beverage 0.1571 0.8000 0.1994 3.5209 electricity/total energy consumption), the GHG intensity (emission/
(0.0384) (0.3862) (0.1210) (0.2330) sales), ratio of net electricity price to coal Price (value of a unit standard
4 Chemical fiber 0.2961 2.2003 0.1113 5.6275 coal), technical capacity (sales/labor). Descriptive statistics for electric-
(0.1075) (1.6780) (0.0232) (0.4870)
5 Chemical raw material and 0.1373 3.9167 0.8352 6.5391
ity proportion, GHG intensity, ratios of net electricity price to coal
chemical products (0.2526) (1.7690) (0.1234) (0.3531) price and technical capacity are presented in Table 2.
6 Electronic equipment 0.6517 0.1457 0.8725 4.2091 Fig. 1 illustrates the price ratio of net electricity to coal in Chinese in-
(0.1414) (0.0233) (0.1826) (0.3990) dustry from 2000 to 2016. The trend of the price ratio is stable during
7 Electrical machinery 0.5119 0.2091 0.7092 4.2877
the first four years of the 2000s. In the followed ten years the price
(0.0949) (0.0573) (0.1528) (0.7252)
8 Foods 0.1763 0.9147 0.2305 5.4086 ratio fell. However, in 2015 it was back to 2006 level. Then in 2016 the
(0.0173) (0.4742) (0.0370) (0.3155) ratio fell to approximately the minimum point that is at 2012 level.
9 Furniture 0.4279 0.2870 0.0901 1.3657 In Fig. 2, the greenhouse gas (GHG) intensity and energy intensity in
(0.1263) (0.1941) (0.0207) (0.4428) all 26 sectors during 2000–2016 are presented, calculated as ten thou-
10 General purpose 0.3601 0.5162 0.5167 6.8318
machinery (0.0696) (0.2416) (0.1158) (0.8635)
sand tons CO2e/billion yuan output and ten million kgce/billion yuan
11 Leather 0.4510 0.2355 0.2090 2.3977 output respectively. The general trend of GHG intensity and energy in-
(0.1306) (0.0711) (0.0216) (0.7440) tensity is to fall, and the downward trend is more significant from
12 Measuring instrument and 0.5759 0.2112 0.1038 2.4345 2000 to 2011, while the gentle trend after 2012 is shown. As mentioned
machinery (0.1190) (0.0901) (0.0221) (0.3215)
before, GHG intensity is normal deemed as a proxy of greenhouse gas
13 Medicines 0.2564 0.6923 0.2076 4.6511
(0.0389) (0.3244) (0.0311) (0.5399) emissions. That the efficiency of greenhouse gas emission has negative
14 Metal products 0.5948 0.7353 0.3984 3.6280 correlation with its intensity is expected. Results of that will be shown
(0.1160) (0.2927) (0.0732) (0.6413) in the next section.
15 Non-metallic mineral 0.1185 5.9252 0.6986 7.5512
products (0.0237) (2.9974) (0.1201) (0.6246)
Notes: Ferrous Metals represents Manufacture and Processing of Ferrous Metals; Articles
16 Paper 0.1713 2.7279 0.2311 5.2720
represents Manufacture of Articles for Culture, Education and Sport Activity; Beverage rep-
(0.0109) (0.8445) (0.0425) (0.9936)
resents Manufacture of Beverage; Chemical Fiber represents Manufacture of Chemical Fi-
17 Special purpose 0.3504 0.4819 0.3182 3.6742
ber; Chemical Raw Material and Chemical Products represents Manufacture of Chemical
machinery (0.1161) (0.2849) (0.0490) (0.5872)
Raw Material and Chemical Products; Electronic Equipment represents Manufacture of
18 Textile 0.3823 1.0124 0.7722 5.6229
Communication Equipment, Computer and Other Electronic Equipment; Electrical Ma-
(0.0858) (0.2406) (0.1885) (0.2815)
chinery represents Manufacture of Electrical Machinery and Equipment; Foods represents
19 Textile wearing apparel 0.3956 0.2680 0.3093 4.1429
Manufacture of Foods; Furniture represents Manufacture of Furniture; General Purpose
(0.0896) (0.0728) (0.0256) (0.2644)
Machinery represents Manufacture of General Purpose Machinery; Leather represents
20 Tobacco 0.3541 0.2554 0.0568 1.8994
Manufacture of Leather, Fur, Feather and Its Products; Measuring Instrument and Machin-
(0.1412) (0.2023) (0.0161) (0.3186)
ery represents Manufacture of Measuring Instrument and Machinery for Cultural Activity
21 Transport equipment 0.3841 0.3285 0.7140 4.9810
and Office Work; Medicines represents Manufacture of Medicines; Metal Products repre-
(0.1081) (0.1884) (0.1243) (0.2747)
sents Manufacture of Metal Products; Non-metallic Mineral Products represents Manufac-
22 Non-ferrous metals 0.4237 2.6161 0.4406 6.8333
ture of Non-metallic Mineral Products; Paper represents Manufacture of Paper and Paper
(0.4877) (0.9636) (0.1547) (0.7266)
Products; Special Purpose Machinery represents Manufacture of Special Purpose Machin-
23 Printing 0.5511 0.4685 0.0780 1.9302
ery; Textile represents Manufacture of Textile; Textile Wearing Apparel represents Manu-
(0.1086) (0.2058) (0.0052) (0.3288)
facture of Textile Wearing Apparel, Footwear and Caps; Tobacco represents Manufacture
24 Agricultural products 0.2284 0.6151 0.8185 6.9439
of Tobacco; Transport Equipment represents Manufacture of Transport Equipment; Non-
(0.0565) (0.4289) (0.1657) (1.8136)
ferrous Metals represents Manufacture and Processing of Non-ferrous Metals; Printing
25 Petroleum 0.0092 12.6854 0.3579 7.5337
represents Printing, Reproduction of Recording Media; Agricultural Products represents
(0.0016) (2.7439) (0.0726) (0.3263)
Processing of Food from Agricultural Products; Petroleum represents Processing of Petro-
26 Wood 0.3371 0.8298 0.1749 4.4822
leum, Coking, Processing of Nucleus Fuel; Wood represents Processing of Timbers, Manu-
(0.1203) (0.4188) (0.0462) (0.2195)
facture of Wood, Bamboo, Rattan, Palm and Straw Products.
1194 J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202

Fig. 1. The Price Ratio of Net Electricity to Coal in Chinese processing and manufacturing
industry 2000–2016.

In Fig. 3, the energy consumption in all 26 sectors during 2000–2016


are presented. The general trend of total energy consumption is to rise,
and the downward trend is more significant in recent year, while the Fig. 3. Energy consumption in Chinese processing and manufacturing industry
gentle trend after 2013 is shown. As mentioned before, coal is normal 2000–2016.
deemed as the main energy during the whole time in China. Crude oil,
coke and electricity are also in the important positions and showed in and the gross sales output growth rate slowed as the changes of eco-
descending order by the amount of energy. nomic environment. The week negative sign for value which means
The annual mean transaction value of ETS is adopted from 2013 to that the week correlation between value and GHG inefficiency. ELE has
2016. It is hard to see the influence of ETS pilots to GHG efficiency negative effects with GHG inefficiency, and EC has opposite effects. In
from the original data. We will discuss specific correlation through de- general, adding proportion of electric energy in whole energy and
tail calculation in next section. adjusting the radio of electricity to coal price would reduce GHG ineffi-
ciency. The impact of technical capacity has the similar correlation with
4. Results and discussions GHG inefficiency totally like ELE, the increasing of GHG efficiency along
with the progressing of technical capacity as well. The detail analysis for
In this section, the results of GHG efficiency from estimating stochas- each sector is presented later. The last two columns show the variance
tic production frontier conditioned on various factors as mentioned parameters of σu and σv that are statistically significant.
above are presented and analyzed. There are two parts to explain the re- The descriptive statistics for the whole underlying GHG efficiency of
sults. In the first part, the estimated time-variant GHG efficiency score all sectors in China from the econometric estimation in Table 4, showing
and the influencing factors' correlation with it will be shown in that the average value efficiency is estimated to be about 89.55%, the
Section 4.1. Second, comparing these proxies and the results of GHG ef- median efficiency score is 92.80% and there are dispersions near the
ficiency, GHG intensity, energy efficiency and energy intensity will be mean efficiency score.
displayed in Section 4.2. It presents the average GHG efficiency scores for every industry in
three periods of the estimation, in the whole period and its minimum

4.1. GHG efficiency scores and the correlation with its determinants
Table 3
The estimation results for stochastic production frontier model, Estimation results for the GHG efficiency.
Eqs. (3) and (4), are given in Table 3, presenting the estimated coeffi-
Parameters Coefficients Standard error P-value
cient of and the impacts of the determinants of GHG inefficiency in Chi-
Ln E_GHG −0.5856 0.0752 0.000
nese manufacturing and processing industries that cover 26 sectors. The
Ln L 1.2590 0.9479 0.000
positive sign for ETS which means that ETS may has no use to promote Ln K −0.4507 0.8819 0.000
GHG efficiency rising, but the growth of energy use and greenhouse T 0.1551 0.0164 0.000
gas emission slowed down and declined during ETS implementation, Ln E_GHG Ln E_GHG 0.0586 0.0091 0.000
Ln L Ln L 0.0660 0.0085 0.000
Ln K Ln K 0.3229 0.0048 0.000
TT −0.0018 0.0006 0.001
Ln E_GHG Ln L 0.0826 – –
Ln E_GHG Ln K −0.2387 0.0095 0.000
Ln L Ln K −0.3488 – –
T Ln E_GHG 0.0164 0.0026 0.000
T Ln L 0.0272 0.0034 0.000
T Ln K −0.0419 0.0034 0.000

Impacts of inefficiency determinants


ETS 5.1595 1.1788 0.000
value −0.1324 0.0375 0.000
ELE −3.8929 0.9554 0.000
EC 3.6203 0.8077 0.000
capacity −1.6864 0.9114 0.064

Variance parameters
E(σu) 0.1562
Fig. 2. GHG intensity and energy intensity in Chinese processing and manufacturing
σV 0.0888 0.0073 0.000
industry 2000–2016.
J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202 1195

Table 4 that have plausible signs from the economic perspective in Table 3. The
Descriptive statistics of overall GHG efficiency scores. positive sign for time means that output per unit of GHG (CO2e) increas-
Min Max Mean Median St. dev. ing overtime in most sectors. Regarding the efficiency determinant equa-
0.4991 0.9958 0.8955 0.928 0.0931
tion, the results of estimations reveal that the impacts of CHN ETS and
transaction value have a neutral effect on GHG efficiency in 2013–2016
for most sectors, the reasons for this phenomenon may be that the period
and maximum in Table 5. From the perspective of time dimension, the
of ETS from starting is not for long enough, and the coverage is too-
overall trend of GHG efficiency in China's manufacturing industry is up-
narrow for the entire China's industries, such as Zhao et al. found that
ward, but more obvious fluctuations are appeared for some Individual
the efficiency of carbon trading market had weak efficiency in China
manufacturing industries in the period of 2012–2016. Meanwhile, the
(Zhao et al., 2017). The factors of electricity proportion and capacity
GHG efficiency scores in different industries vary remarkably from
have positive effects on GHG efficiency. However, there is a negative ef-
Table 5. Most industries are at high levels of GHG efficiency in this
fect of the impact of Electricity to Coal for vast majority sectors in China.
study. The industry with the highest GHG efficiency is the electronic
Again, there are more detailed analysis results of statistics as shown
equipment, whose GHG efficiency has achieved more than 90% in the
in Table 6. Coal is the primary fossil fuel consumed by China's industry,
whole period. Other industries such as the articles, the electrical ma-
and electricity to coal is a significant index on the performance of GHG
chinery, the measuring instrument and machinery, the textile and so
efficiency. The high correlation between GHG efficiency and the impacts
on also have higher levels of GHG efficiency relatively, whose mean
of electricity proportion and electricity to coal are revealed. The positive
GHG efficiency has also reached 90%. The relatively lower levels of
signs for electricity proportion in most sectors, adding the proportion of
GHG efficiency are estimated in the petroleum, the wood and the non-
electric energy in whole energy would increase GHG efficiencys. For ex-
metallic mineral products, but their mean GHG efficiency has got
ample, the correlation of electricity to coal of petroleum is 0 because the
more than 84% similarly. Nevertheless, the ferrous metals sector has
main fossil fuel of petroleum is fossil oil having no correlation with coal.
the lowest GHG efficiency, that gets 83.74% as well. All sectors have
less underlying GHG efficiency in general description.
Table 6
For the logarithmic form of variables in the stochastic production fron- Descriptive statistics of the cross-correlation between Sector's GHG efficiency and the im-
tier model, the estimated coefficients can be interpreted as elasticities pacts of its determinants.

Table 5 Sector Electricity Electricity ETS Transaction Technical


Descriptive statistics of Sector's GHG efficiency scores. proportion to Coal value capacity

Ferrous metals 0.3529⁎⁎ −0.3088⁎ −0.4281⁎⁎ −0.3603⁎ 0.6324⁎⁎⁎


Sector 2000–2005 2006–2011 2012–2016 Min Max Mean
Articles 0.5588⁎⁎ −0.3677⁎⁎ −0.2616 −0.2702 0.1471
(Mean) (Mean) (Mean)
Beverage 0.6912⁎⁎⁎ −0.6618⁎⁎⁎ 0.2616 0.2927 0.5588⁎⁎⁎
Ferrous metals 0.7587 0.9555 0.7902 0.5863 0.9757 0.8374 Chemical fiber 0.5000⁎⁎ −0.5000⁎⁎ −0.2140 −0.1576 0.4412⁎⁎
Articles 0.9152 0.9519 0.9240 0.8635 0.9830 0.9308 Chemical raw 0.6118⁎⁎⁎ −0.6765⁎⁎⁎ −0.0238 0.0225 0.8823⁎⁎⁎
Beverage 0.7757 0.9014 0.9231 0.7008 0.9676 0.8634 material and
Chemical fiber 0.7854 0.9450 0.8684 0.6590 0.9718 0.8662 chemical
Chemical raw 0.8074 0.9523 0.9159 0.7310 0.9787 0.8905 products
material and Electronic 0.5588⁎⁎⁎ −0.6471⁎⁎⁎ −0.0476 0.0000 0.5000⁎⁎⁎
chemical equipment
products Electrical 0.5294⁎⁎⁎ −0.5882⁎⁎⁎ −0.0951 −0.0676 0.8529⁎⁎⁎
Electronic 0.9420 0.9876 0.9731 0.9055 0.9958 0.9672 machinery
equipment Foods 0.4265⁎⁎ −0.7353⁎⁎⁎ 0.1189 0.1576 0.6618⁎⁎⁎
Electrical 0.8693 0.9815 0.9501 0.8025 0.9901 0.9327 Furniture 0.5588⁎⁎⁎ −0.7059⁎⁎⁎ 0.0951 0.1351 0.5588⁎⁎⁎
machinery General 0.3971⁎⁎ −0.6471⁎⁎⁎ −0.0951 −0.0450 0.8382⁎⁎⁎
Foods 0.7767 0.9165 0.9189 0.7576 0.9627 0.8679 purpose
Furniture 0.8638 0.9427 0.9399 0.8251 0.9762 0.9140 machinery
General 0.7600 0.9646 0.9225 0.6582 0.9779 0.8800 Leather 0.0294 −0.1177 −0.2854 −0.2252 −0.0294
purpose Measuring 0.3668⁎⁎ −0.4853⁎⁎⁎ −0.1903 −0.1351 0.4706⁎⁎⁎
machinery instrument
Leather 0.9387 0.9409 0.9287 0.8785 0.9856 0.9365 and
Measuring 0.9023 0.9673 0.9357 0.8490 0.9796 0.9350 machinery
instrument Medicines 0.3971⁎⁎ −0.5441⁎⁎⁎ 0.1427 0.1802 0.5⁎⁎⁎
and Metal products 0.6176⁎⁎⁎ −0.7059⁎⁎⁎ 0.7059 0.1351 0.7647⁎⁎⁎
machinery non-metallic 0.6324⁎⁎⁎ −0.7059⁎⁎⁎ 0.2616 0.3153 0.7206⁎⁎⁎
Medicines 0.8348 0.9143 0.9217 0.7750 0.9733 0.8884 mineral
Metal products 0.8567 0.9725 0.9551 0.6955 0.9905 0.9265 products
Non-metallic 0.7177 0.9134 0.9466 0.6936 0.9762 0.8541 Paper −0.2206 0.1471 −0.5946⁎⁎⁎ −0.5179⁎⁎ 0.5735⁎⁎⁎
mineral Special 0.6176⁎⁎⁎ −0.7794⁎⁎⁎ 0.1903 0.2252 0.8382⁎⁎⁎
products purpose
Paper 0.8982 0.9320 0.8530 0.7831 0.9470 0.8968 machinery
Special 0.6721 0.9370 0.9453 0.5965 0.9813 0.8459 Textile 0.5147⁎⁎⁎ −0.4265⁎⁎ −0.3092 −0.2252 0.2059
purpose Textile wearing 0.1765 −0.1912 −0.0714 −0.0225 −0.0147
machinery apparel
Textile 0.9386 0.9726 0.8971 0.7825 0.9849 0.9384 Tobacco 0.2941⁎ −0.3382⁎ 0.3805⁎ 0.4279⁎⁎ −0.25
Textile wearing 0.9234 0.9337 0.9347 0.8732 0.9800 0.9304 Transport 0.5882⁎⁎⁎ −0.7941⁎⁎⁎ 0.2378 0.2477 0.8235⁎⁎⁎
apparel equipment
Tobacco 0.9035 0.8946 0.9501 0.8304 0.9738 0.9141 Non-ferrous 0.4706⁎⁎⁎ −0.5147⁎⁎⁎ −0.1189 −0.0676 0.6912⁎⁎⁎
Transport 0.7643 0.9513 0.9621 0.6573 0.9907 0.8885 metals
equipment Printing 0.4706⁎⁎⁎ −0.75⁎⁎⁎ 0.3805⁎ 0.3828⁎ 0.2794
Non-ferrous 0.7246 0.9725 0.9130 0.4991 0.9845 0.8675 Agricultural 0.6176⁎⁎⁎ −0.6471⁎⁎⁎ 0.0000 0.0676 0.8529⁎⁎⁎
metals products
Printing 0.8592 0.9412 0.9682 0.8318 0.9848 0.9202 Petroleum −0.3529⁎⁎ 0.0000 −0.5470⁎⁎⁎ −0.4729⁎⁎ 0.5735⁎⁎⁎
Agricultural 0.8060 0.9599 0.9395 0.7342 0.9855 0.8996 Wood 0.5735⁎⁎⁎ −0.6324⁎⁎⁎ 0.3805⁎ 0.3828⁎ 0.4706⁎⁎⁎
products
⁎ Statistical significance at 10%.
Petroleum 0.8622 0.9204 0.7218 0.5154 0.9466 0.8414
⁎⁎ Statistical significance at 5%.
Wood 0.7155 0.8900 0.9597 0.6521 0.9797 0.8489
⁎⁎⁎ Statistical significance at 1%.
1196 J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202

But for electricity to coal, it has negative effects on the contrary with Table 7
electricity proportion that means adjusting the ratio of electricity to Estimation results for the energy efficiency.

coal price to decline could improve GHG efficiency. The causes leading Parameters Coefficients Standard error P-value
to this result may be the difference in generating efficiency between Ln E_EN 1.4847 0.0883 0.000
smaller-capacity generating set and larger-capacity generating set, Ln L −0.7438 – –
which have higher power generation efficiency. In order to reduce en- Ln K 2.899415 – –
ergy costs, the manufacturing and processing industries in China prefer T 0.0211 0.0364 0.562
Ln E_EN Ln E_EN −0.0506 0.0099 0.000
to build their own captive power plant to satisfy self-demand for elec-
Ln L Ln L −0.1164 0.0127 0.000
tricity, but the captive power plant has lower power generation effi- Ln K Ln K −0.2648 0.0082 0.000
ciency that means producing the same amount of electricity requires TT −0.0019 0.0008 0.022
more energy and more greenhouses gas emissions, it is not recom- Ln E_EN Ln L −0.1239 0.0205 0.000
mended for GHG efficiency. Power grid, waste heat power generation, Ln E_EN Ln K −0.0675 – –
Ln L Ln K 0.3933 – –
renewable energy power generation and on-site coal-fired power T Ln E_EN 0.0347 0.0018 0.000
plants are the main sources of power energy for manufacturing and pro- T Ln L −0.0191 0.0049 0.000
cessing industries, moreover, waste heat power generation and new en- T Ln K −0.0046 0.0068 0.500
ergy power generation have no excess greenhouse gas produced. In fact,
Variance parameters
according to the relevant investigation, it is more appropriate to choose σu 1.2099 0.5609 0.031
to use net electricity when the power load is low at night. Furthermore, σV 0.1401 0.0155 0.000
industry black carbon is the second strongest contribution to current λ 8.6342 0.5613 0.000
global warming that mainly comes from fuel incomplete combustion
in mini-plant. The volume of black carbon can be reduced by the The correlations of that of 26 industry sectors are presented in Table 9,
above-mentioned methods (Ramanathan and Carmichael, 2008; detailed see Fig. A2 of appendix.
Surabi et al., 2002). The impact of technical capacity has the similar ef- The negative correlations between GHG efficiency and GHG inten-
fects with electricity proportion, which has the positive signs in most sity are expected as the previous researches. Meanwhile the closer the
sectors, while neutral in some, indicating a progressing of technical ca- score of the correlation is to −1, the more consistent the results are
pacity along with an increase of the GHG efficiency. For example, there with previous studies. However, the average score of the correlations
is a strong positive correlation for GHG efficiency with the impacts of that are above statistical significance at 10% is 0.5853, which means me-
electricity proportion and technical capacity, but negative for electricity dium correlation between them. So, the GHG efficiency cannot be the
to coal in non-metallic mineral products. So, the industry of non- proxy that represents GHG intensity completely, but the GHG intensity
metallic mineral products should be encouraged to adopt waste heat could be one of the measurements of GHG efficiency. Furthermore, the
power generation and renewable energy power generation firstly, positive correlations between GHG efficiency and energy efficiency are
while the inadequate demand for energy should use purchasing elec- expected, but the correlation score of some sectors like Special Purpose
tricity rather than commission small-scale coal-fired power stations, Machinery is negative and some sectors have no relationship with GHG
improve production technology to increase labor productivity for efficiency, indicating that energy efficiency cannot be used to stand for
higher GHG efficiency, and in terms of price regulation, the price ratio GHG efficiency despite having a strong correlation in a minority of sec-
of electricity to coal should be reduced. tors. There are many ways to reduce greenhouse gases, one of the most
In general, there is less potential for GHG emission reduction from important and popular ways is to improve energy efficiency (Sun et al.,
Table 5, illustrating the booming emission reduction technologies re- 2018; Sun et al., 2017), the other one of that is to reduce the greenhouse
cently have not been widely applied in China's industries. From a com- gas emission directly by emission reduction technology like CO2 recov-
mon point of view, GHG efficiency is often assumed to be synonymous ery and utilization technology (Wang et al., 2017b). But the energy
with GHG intensity. And energy efficiency is always assumed to take
Table 8
place of GHG efficiency to be studied. In the next section, we investigate
Descriptive statistics of energy efficiency.
the difference among these four proxies of GHG efficiency, GHG inten-
sity, energy efficiency and energy intensity. The following analysis pro- Sector Min Max Mean Median
vides enough evidence to support if this assumption of these four Ferrous metals 0.6292 0.9605 0.8168 0.8324
proxies mentioned-above is valid. Articles 0.6566 0.9359 0.8621 0.8871
Beverage 0.6671 0.9321 0.8429 0.8719
Chemical fiber 0.6294 0.9309 0.8133 0.9058
Chemical raw material and chemical products 0.7563 0.9360 0.8704 0.8915
4.2. The correlations between GHG efficiency and these indexes of GHG in- Electronic equipment 0.6386 0.9318 0.8400 0.8599
tensity, energy efficiency and energy intensity Electrical machinery 0.7019 0.9518 0.8464 0.8707
Foods 0.7045 0.9469 0.8334 0.8255
Furniture 0.5857 0.9340 0.8244 0.8838
The estimation results of energy efficiency using the similar model,
General purpose machinery 0.3095 0.9580 0.7465 0.8606
Eq. (3), are given in Table 7, that presents the estimated coefficient of Leather 0.4598 0.9325 0.7931 0.8835
energy efficiency. The last two columns show the variance parameters Measuring instrument and machinery 0.7680 0.9335 0.8679 0.8589
of σu and σv are statistically significant. The sign of λ indicates that en- Medicines 0.4623 0.9414 0.8005 0.8664
ergy inefficiency is demonstrated in this model. Metal products 0.7416 0.9562 0.8501 0.8512
Non-metallic mineral products 0.5431 0.9506 0.8438 0.8734
The descriptive statistics for the whole underlying GHG efficiency of Paper 0.3125 0.9631 0.7505 0.8565
all sectors in China from the econometric estimation are listed in Special purpose machinery 0.5932 0.9449 0.8406 0.8610
Table 8, showing that the average value efficiency is estimated to be Textile 0.8531 0.9310 0.8776 0.8762
about 89.55%, the median efficiency score is 92.80% and there are dis- Textile wearing apparel 0.6831 0.9360 0.8488 0.8788
Tobacco 0.4110 0.9461 0.7782 0.8835
persions near the mean efficiency score. The more detailed descriptions
Transport equipment 0.6488 0.9591 0.8130 0.8558
are given in Fig. A1 of appendix for GHG efficiency, GHG intensity, en- Non-ferrous metals 0.6378 0.9508 0.8560 0.8688
ergy efficiency and energy intensity. Printing 0.6215 0.9488 0.7993 0.8265
In this section, the main purpose is that comparing the time-variant Agricultural products 0.6964 0.9415 0.8454 0.8424
industry level GHG efficiency scores with calculated indexes of GHG in- Petroleum 0.7734 0.9295 0.8684 0.8720
Wood 0.7015 0.9581 0.8468 0.8523
tensity, energy efficiency and energy intensity by Kendall's correlation.
J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202 1197

Table 9 5. Conclusions
Correlation between GHG efficiency and these indexes of GHG intensity, energy intensity
and energy efficiency.
One of the main purposes of this research is to evaluate the GHG ef-
Sector GHG Energy Energy ficiency in Chinese manufacturing and processing industries that cover
intensity intensity efficiency 26 sectors during 2000–2016. GHG efficiency is defined as to explain
Ferrous metals −0.5735⁎⁎ −0.5735⁎⁎⁎ 0.4265⁎⁎ that the room could be improved on the production frontier of industry
Articles −0.1618 −0.2059 0.4265⁎⁎ that can also be interpreted as in the best practice technology. Further-
Beverage −0.6618⁎⁎⁎ −0.6471⁎⁎⁎ −0.2647
more, GHG efficiency is evaluated by stochastic production frontier
Chemical fiber −0.6177⁎⁎⁎ −0.6177⁎⁎⁎ 0.5882⁎⁎⁎
Chemical raw material and chemical −0.6324⁎⁎⁎ −0.6618⁎⁎⁎ 0.5882⁎⁎⁎ analysis and estimated by applying Greene's ‘true’ random effect
products model. The results show that there is little space to improve GHG effi-
Electronic equipment −0.07353 −0.1618 −0.6324⁎⁎⁎ ciency in fixed technology. This reflects that the new energy saving
Electrical machinery −0.6324⁎⁎⁎ −0.6471⁎⁎⁎
0.3529⁎⁎ and emission reduction technology should make new breakthroughs
Foods −0.6177⁎⁎⁎ −0.6471⁎⁎⁎
−0.2206
Furniture −0.6765⁎⁎⁎ 0.6618⁎⁎⁎
0.0147
and be applied in Chinses industry.
General purpose machinery −0.4706⁎⁎⁎ 0.8088⁎⁎⁎
−0.4706⁎⁎⁎ In addition, the results of impacts of GHG efficiency determinants in-
Leather 0.0147 −0.0441 0.1912 dicate three policy advises for GHG efficiency. First, using more grid elec-
Measuring instrument and machinery −0.3824⁎⁎ −0.3971⁎⁎
−0.0147 tricity instead of small-scale on-site coal-fired power plant in most
Medicines −0.5147⁎⁎⁎ −0.5147⁎⁎⁎
−0.0588
sectors in Chinese industry when it is impossible to generate electricity
Metal products −0.6471⁎⁎⁎ −0.6618⁎⁎⁎
0.4265⁎⁎
Non-metallic mineral products −0.6029⁎⁎⁎ −0.6177⁎⁎⁎
−0.3971⁎⁎ from waste heat or renewable energy sources. Then the lower the ratio
Paper −0.0735 −0.0735 0.1912 of the electricity purchase price to coal purchase price appropriately,
Special purpose machinery −0.6177⁎⁎⁎ −0.6029⁎⁎⁎ −0.4265⁎⁎ the higher the GHG efficiency is. Finally, enhancing the application of
Textile −0.4559⁎⁎ −0.4706⁎⁎⁎ −0.0147 some advanced technologies in industry can improve the GHG efficiency.
Textile wearing apparel −0.1029 −0.1324 0.0441
The impact of the China's Emission Trading System (CHN ETS) has a
Tobacco −0.2647 −0.2647 0.3235⁎
Transport equipment −0.6618⁎⁎⁎ −0.6618⁎⁎⁎ 0.8088⁎⁎⁎ moderate effect on GHG efficiency, while the results are unbelievable in
Non-ferrous metals −0.7794⁎⁎⁎ −0.8088⁎⁎⁎ 0.0441 most sectors. Also, the carbon trading value has a moderate effect in mi-
Printing −0.6324⁎⁎⁎ −0.6471⁎⁎⁎ 0.7794⁎⁎⁎ nority industry sectors.
Agricultural products −0.6471⁎⁎⁎ −0.6471⁎⁎⁎ 0.3971⁎⁎
The other purpose is to investigate the correlation between GHG ef-
Petroleum −0.3235⁎ −0.2941⁎ 0.2941⁎
Wood −0.5588⁎⁎⁎ −0.5882⁎⁎⁎ −0.1029
ficiency and three indexes of GHG intensity, energy intensity and energy
efficiency, which are common used in the sphere of learning and of pub-
⁎ Statistical significance at 10%.
⁎⁎ Statistical significance at 5%. lic. For example, there is a widespread assumption in energy statistic
⁎⁎⁎ Statistical significance at 1%. and econometrics that energy intensity and energy efficiency are equiv-
alent measures of energy performance of economies. GHG intensity, en-
ergy intensity and energy efficiency are dependent indexes for measure
efficiency cannot be suitable as the proxy to measure the greenhouse GHG and energy performance, but they cannot represent each other. For
gas emission with the progress of emission reduction technology. More- instance, the proxy of energy efficiency could not be used as a measure
over, that the different fuel has different compositions is another essen- of GHG efficiency. The same is true on other proxies.
tial reason, which is demonstrated by the correlation between GHG Finally, there is still a lot to be done in order to achieve the goal that
efficiency and energy intensity; in other words, the structures of energy China reduces the carbon intensity per unit of GDP by 60–65% by 2030
use have a significant impact on GHG emission. The negative correla- and how to pursue climate and policy to stimulate industry to invest
tions between GHG efficiency and energy intensity are expected, the av- in energy saving and emission reduction technology and ultimately
erage score of the correlations that are above statistical significance at lower carbon emission. Now the implementation time and coverage of
10% is 0.6641, which means medium correlation. As for the other unbe- the China's Emission Trading System (CHN ETS) are insufficient, and
lievable correlations which are under statistical significance at 10% may its effect is not obvious for GHG efficiency. For a more comprehensive
have other possible reasons like insufficient statistical data or others, the understanding of the GHG efficiency and effect of CHN ETS, it should
exact reasons are not clear. be left for the future when more data is available to us.
To sum up, GHG efficiency scores show the emission reduction po-
tential which means that there is room to improvement in energy-
saving and emission-reduction in Chinese industries, and the correla- Acknowledgements
tions between GHG efficiency and three indexes of GHG intensity, en-
ergy intensity and energy efficiency illustrate that there is medium The authors are grateful to the financial support provided by Na-
and strong correlation in most sectors. However, they cannot represent tional Key Research and Development Program of China
each other due to the individual but different impacts, like the previous (2017YFB0304000 & 2017YFB0304001).
study considered that energy intensity can only provide indirect and de-
layed evidence of technological and engineering energy efficiency Appendix A
(Proskuryakova and Kovalev, 2015). The impacts of GHG efficiency de-
terminants, including electricity proportion, electricity to coal, ETS, The detailed descriptions of GHG efficiency, GHG intensity, energy
transaction value and technical capacity are analyzed above. The CHN efficiency and energy intensity are depicted in Fig. A1. The blue lines
ETS and its transaction value are not statistically significant in most sec- represent energy efficiency. The red lines represent energy intensity.
tors. The analysis suffers to some degree from the CHN ETS trading pe- The green lines represent GHG efficiency. The black lines represent
riod (2013–2016) admittedly, which launched seven pilots in China GHG intensity.
from 2013, and now it is ready to turn to the whole country. Although The detailed descriptions of the correlations between GHG efficiency
we would like to get more accurate data from the monitoring system, and three determinants of ELE, EC and TC are depicted in Fig. A2. The red
they are not developed fully now. The volume of black carbon should scatter points represent the correlation for GHG efficiency with the price
be considered in further research because of the limitations of current ratios of electricity to coal (EC). The green scatter points represent the
technology. In any case, it is meaningful since the results discussed are correlation for GHG efficiency with the proportion of net electricity in
so crucial for some rules of GHG efficiency, and they are useful for policy the total energy used (ELE). The black scatter points represent the cor-
formulation. relation for GHG efficiency with technical capacity (TC).
1198 J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202
J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202 1199

Fig. A1. the proxies of GHG efficiency, GHG intensity, Energy efficiency and Energy intensity in 26 sectors of China's industry, 2000–2016.
1200 J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202
J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202 1201

Fig. A2. Correlations between GHG efficiency and three determinants of ELE, EC and TC.

References Bureau of Population and Employment Statistics of China (BPESC), 2001-2017. China
Labor Statistics Yearbook. China Statistical Publishing House, Beijing.
Carvalho, P., Marques, R.C., 2016. Estimating size and scope economies in the Portuguese
Administration NE. National Power Price Regulatory Bulletin 2016. National Energy Ad-
water sector using the Bayesian stochastic frontier analysis. Sci. Total Environ. 544,
ministration, Beijing, 2017, pp. 1–10.
574–586.
Aigner D, Lovell CAK, Schmidt P. Formulation and estimation of stochastic frontier pro-
Cheng, Y., Lv, K., Wang, J., Xu, H., 2018. Energy efficiency, carbon dioxide emission effi-
duction function models. J. Econ. 1977; 6: 21–37.
ciency, and related abatement costs in regional China: a synthesis of input–output
Atkinson, A.C., Riani, M., Cerioli, A., 2004. Multivariate Data and the Forward Search.
analysis and DEA. Energy Efficiency 12, 861–877.
Battese, G.E., Coelli, T.J., 1995. A model for technical inefficiency effects in a stochastic
Coelli, T., 1995. Estimators and hypothesis tests for a stochastic frontier function: a Monte
frontier production function for panel data. Empir. Econ. 20, 325–332.
Carlo analysis. J. Prod. Anal. 6, 247–268.
Belotti, F., Daidone, S., Ilardi, G., Atella, V., 2012. Stochastic frontier analysis using stata.
Croux, C., Dehon, C., 2010. Influence functions of the Spearman and Kendall correlation
SSRN Electron. J. 10, 1–48.
measures. Statistical Methods & Applications 19, 497–515.
Brännlund, R., Lundgren, T., Marklund, P.-O., 2014. Carbon intensity in production and the
Cullinane, Kevin, Wang, TengFei, Song, DongWook, et al., 2006. The technical efficiency of
effects of climate policy—evidence from Swedish industry. Energy Policy 67,
container ports: comparing data envelopment analysis and stochastic frontier analy-
844–857.
sis. Transportation Research Part A Policy & Practice 40, 354–374.
1202 J. Sun et al. / Science of the Total Environment 690 (2019) 1190–1202

Filippini, M., Charles Hunt, L., 2011. Energy Demand and Energy Efficiency in the OECD National Bureau of Statistics of China (NBSC), 2001-2017c. China Industrial Statistical
Countries: A Stochastic Demand Frontier Approach. vol Volume 32. Yearbook. China Statistical Publishing House, Beijing.
Filippini, M., Hunt, L.C., 2015. Measurement of energy efficiency based on economic foun- National Bureau of Statistics of China (NBSC), 2019. National data. Available on. http://
dations. Energy Econ. 52, S5–S16. data.stats.gov.cn/easyquery.htm?cn=C01.
Filippini, M., Hunt, L.C., Zorić, J., 2014. Impact of energy policy instruments on the esti- Pardo Martínez, C.I., Silveira, S., 2012. Energy efficiency and CO2 emissions in Swedish
mated level of underlying energy efficiency in the EU residential sector. Energy Policy manufacturing industries. Energy Efficiency 6, 117–133.
69, 73–81. Proskuryakova, L., Kovalev, A., 2015. Measuring energy efficiency: is energy intensity a
Gamtessa, S.F., 2016. The relationship between input prices and energy intensity in Cana- good evidence base? Appl. Energy 138, 450–459.
dian manufacturing industries. Can. Public Policy 42, 1–9. Qiu, S.-B., Yu, D.-H., 2017. Measurement and evaluation of the carbon emission efficiency
Gao, J., Kovats, S., Vardoulakis, S., Wilkinson, P., Woodward, A., Li, J., et al., 2018. Public of manufacturing industry in Anhui Province. Journal of Interdisciplinary Mathemat-
health co-benefits of greenhouse gas emissions reduction: a systematic review. Sci. ics 20, 1521–1526.
Total Environ. 627, 388–402. Ramanathan V, Carmichael G. Global and regional climate changes due to black carbon.
Greene, William, 2005. Reconsidering heterogeneity in panel data estimators of the sto- Nat. Geosci. 2008; 36: págs. 335–358.
chastic frontier model. J. Econ. 126, 269–303. Shen, Y., Wang, G., 2013. Key Findings and Assessment Results of IPCC WGI Fifth Assess-
Greene, W.H., Dan, S., 2004. Profitability and efficiency in the U.S. life insurance industry. ment Report. vol. 35.
J. Prod. Anal. 21, 229–247. Slater, H.D.B.D., Wang, S., Qian, G.Q., 2018. China's carbon price survey 2018. China Carbon
Greenpeace, 2018. BP Statistical Review of Word Energy 2018. BP Statistical Review of Forum 1–51 Beijing.
Word Energy, London. pp. 1–56. Sun, W., Wang, Y., Zhang, F., Zhao, Y., 2017. Dynamic allocation of surplus by-product gas
Herrala, R., Goel, R.K., 2012. Global CO2 efficiency: country-wise estimates using a sto- in a steel plant by dynamic programming with a reduced state space algorithm. Eng.
chastic cost frontier. Energy Policy 45, 762–770. Optim. 50, 1578–1592.
Jiang, L., Folmer, H., Ji, M., Tang, J., 2016. Energy efficiency in the Chinese provinces: a fixed Sun, W.-q., Yue, X.-y., Wang, Y.-h., Cai, J.-j., 2018. Energy and exergy recovery from ex-
effects stochastic frontier spatial Durbin error panel analysis. Ann. Reg. Sci. 58, haust hot water using organic Rankine cycle and a retrofitted configuration. J. Cent.
301–319. South Univ. 25, 1464–1474.
Kalirajan, K., 1981. The economic efficiency of farmers growing high-yielding, irrigated Sun, W., Zhou, Y., Lv, J., Wu, J., 2019a. Assessment of multi-air emissions: case of particu-
rice in India. Am. J. Agric. Econ. 63, 566–570. late matter (dust), SO2, NOx and CO2 from iron and steel industry of China. J. Clean.
Kumbhakar, S.C., Lovell, C.A.K., 2000. Stochastic Frontier Analysis: The Shadow Price Ap- Prod. 232, 350–358.
proach to the Estimation and Decomposition of Economic Efficiency. Sun, W., Xu, X., Lv, Z., Mao, H., Wu, J., 2019b. Environmental impact assessment of waste-
Li, M.-J., Tao, W.-Q., 2017. Review of methodologies and polices for evaluation of energy water discharge with multi-pollutants from iron and steel industry. J. Environ.
efficiency in high energy-consuming industry. Appl. Energy 187, 203–215. Manag. 245, 210–215.
Lin, B., Jia, Z., 2019. What are the main factors affecting carbon price in emission trading Surabi, M., James, H., Larissa, N., Yunfeng, L., 2002. Climate effects of black carbon aerosols
scheme? A case study in China. Sci. Total Environ. 654, 525–534. in China and India. Science 297, 2250–2253.
Lin, B., Sun, C., 2010. Evaluating carbon dioxide emissions in international trade of China. Sxcoal.com, 2019. Coal Price Index.
Energy Policy 38, 613–621. Wang, J., Lv, K., Bian, Y., Cheng, Y., 2017a. Energy efficiency and marginal carbon dioxide
Löschel, A., Lutz, B.J., Managi, S., 2019. The impacts of the EU ETS on efficiency and eco- emission abatement cost in urban China. Energy Policy 105, 246–255.
nomic performance – an empirical analyses for German manufacturing firms. Resour. Wang, Y., Du, T., Song, Y., Che, S., Fang, X., Zhou, L., 2017b. Amine-functionalized mesopo-
Energy Econ. 56, 71–95. rous ZSM-5 zeolite adsorbents for carbon dioxide capture. Solid State Sci. 73, 27–35.
Lundgren, T., Marklund, P.-O., Zhang, S., 2016. Industrial energy demand and energy effi- Weyman-Jones, T., Boucinha, J.M., Inácio, C.F., Martins, P.M., 2016. Measuring non-
ciency – evidence from Sweden. Resour. Energy Econ. 43, 130–152. residential electric energy efficiency in the Portuguese economy. Energy Procedia
Meeusen, W., Broeck, J.V.D., 1977. Efficiency estimation from Cobb-Douglas production 106, 2–13.
functions with composed error. Int. Econ. Rev. 18, 435–444. Yang, K., Lei, Y., 2016. The carbon dioxide marginal abatement cost calculation of Chinese
Na, H.-m., C-k, Gao, M-y, Tian, Qi, Z.-q., Ye, Z., 2017. MFA-based analysis of CO2 emissions provinces based on stochastic frontier analysis. Nat. Hazards 85, 505–521.
from typical industry in urban — as a case of steel industry. Ecol. Model. 365, 45–54. Zhao, X.-g., Wu, L., Li, A., 2017. Research on the efficiency of carbon trading market in
National Bureau of Statistics of China (NBSC), 2001-2017a. China Statistical Yearbook. China. Renew. Sust. Energ. Rev. 79, 1–8.
Available on. http://www.stats.gov.cn/tjsj/ndsj/.
National Bureau of Statistics of China (NBSC), 2001-2017b. China Energy Statistical Year-
book. China Statistical Publishing House, Beijing.

You might also like