Can China's Energy Intensity Constraint Policy Promote Total Factor Energy Efficiency? Evidence from the Industrial Sector.
The world has witnessed the miracle of China's economic development over the past four decades. However, the extensive growth mode associated with high investment, high energy consumption, and high pollution has not been fundamentally changed. As such, energy constraints have gradually appeared as a backdrop to China's rapid economic growth. Especially given the acceleration of industrialization and urbanization in recent years, China's economic development is increasingly faced with growing pressure to save energy and reduce emissions. The energy bottleneck has become a significant problem that threatens to the sustainable development of China's economy. It is predicted that, from 2010 to 2030, the demand for new energy in China will account for 33% of total global demand. These demand pressures will be accompanied by a rising dependency on oil from external sources. Without any control, by 2030, China's oil consumption will surpass 800 million tons per annum, and approximately 75% of this oil will be imported. (1) At the same time, various environmental problems will continue to accompany China's rapid economic development, and these problems (such as heavy haze pollution, excessive carbon dioxide emissions, and frequent extreme weather phenomena) are becoming increasingly serious.
To achieve the green transformation of China's economic development, the Chinese government should improve energy restriction and strengthen environmental governance as soon as possible. Improving energy efficiency is generally regarded as the key to resolving the abovemen-tioned problems. Indeed, such thinking has been introduced into the medium- and long-term planning of China's national economic and social development. In particular, at the beginning of 2006, the Chinese government launched the 11th "Five-Year Plan" (FYP). For the first time, a mandatory energy-conservation target was added into the FYP, i.e., the energy consumption per unit of gross domestic product (GDP) should decline by 20% from the 2005 level. In other words, by 2010, the energy consumption per GDP ([10.sup.4] RMB) was expected to drop to approximately 0.98 tons of coal equivalent (tce), from 1.22 tce level in 2005. This was also the first time an energy-saving target was introduced into the local government's performance evaluation systems.
China's central and local governments then implemented a series of relevant measures to execute the new energy intensity constraint policy (EICP). For instance, in June 2007, the Chinese State Council issued a circular entitled the Comprehensive Work Plan for Energy Saving and Emission Reduction. This document claims that local governments should fully realize the importance and urgency of saving energy and reducing emissions. Close attention should be paid to the government's responsibilities and its law-enforcing supervision of energy saving and emission reduction policies and regulations. In short, China's government should fully implement and oversee key energy-saving projects. In October 2007, the Standing Committee of the Chinese People's Congress validated the Energy Conservation Law of the People's Republic of China. This law makes it clear that improving energy efficiency should be the basic purpose of the law's implementation. In August 2008, the Chinese National Development and Reform Commission issued a notice regarding the implementation of the Energy Conservation Law of the People's Republic of China. This notice proposes strengthening energy-conservation management in key projects, enterprises, and areas through related economic policies. In May 2009, the Chinese Ministry of Finance and National Development and Reform Commission issued a circular entitled Energy-saving Product Projects for Benefiting the People. This publication suggests speeding up the popularization of low-energy-use products and improving the energy efficiency of high-energy-use products through financial subsidies.
To respond to such institutional arrangements, local governments began to implement a series of related auxiliary energy-saving policies and measures. These steps included, but are not limited to the following: supporting key energy saving and emission reduction projects, widely distributing efficient energy-saving products and new energy-saving technologies, improving energy-saving management capabilities, implementing tax policies to encourage energy savings and emission reductions, reducing corporate income tax for energy-saving projects, strengthening financial services for projects that could show elements of energy saving and environmental protection, encouraging and guiding financial institutions to increase and improve the availability of credit support for energy-conservation and emission-reduction projects, and preferentially providing direct financing services to specific and eligible energy saving and emission reduction projects.
In terms of the background just described, we raise the following questions: What is the actual effect of these policies and measures that focus on energy saving and emission reduction? Has the EICP achieved any quantifiable success in improving energy efficiency? Unfortunately, existing studies have paid little attention to these issues. To answer these questions, we first observe the trend of the energy intensity of China's industrial sector in recent years in Figure 1.
The reason we choose the industrial sector as our research object is due to the following considerations. Since the Industrial Revolution in the mid-19th century, the rapid development of the industrial sector has led to huge demand for the use of fossil fuels. This has caused a corresponding sharp rise in greenhouse gas emissions, including carbon emissions. Industrial sector is the largest energy consumer in the entire world. The energy consumption of industrial sector alone accounts for more than 40% of the world's total energy consumption. Going even further, the industrial sector's carbon emissions account for as much as 61% of the world's total carbon emissions (IEA, 2009). China is also still in an accelerated industrialization process, which is mainly characterized by rapid energy consumption. Reducing the rigid and intense demand for energy in China's industrial sector is going to be difficult in the short term. According to data from the China Energy Statistics Yearbook, the proportion of China's industrial energy consumption has remained at approximately 70% of the country's total energy consumption over nearly two decades. As a result, the industrial sector has become the primary object of energy saving and emission reduction strategies. The energy-saving and emission-reduction performance of the industrial sector will undoubtedly play a crucial role in achieving the energy-saving and emission-reduction targets of the entire national economy.
As shown in Figure 1, since 2001, the energy intensity of China's industrial sector (at the 2000 constant price) has presented a continuous downward trend. Thus, in terms of single factor energy efficiency (the reciprocal of energy intensity), China's EICP has an evident mitigation effect on energy intensity. Such a preliminary result contrasts with those of a number of previous studies that argue that China's public policy is ineffective in EICP's implementation (e.g., Zhang et al., 2017). However, it is noteworthy that, because the single factor energy efficiency fails to consider other production factors (and thus reflect those factors' substitution relationships), single factor energy efficiency is not energy efficiency in a strict sense (Patterson, 1996; Li and Hu, 2012). Therefore, it is still necessary to perform a more detailed and rigorous investigation of EICP's actual effectiveness in improving energy efficiency based on the rational measurements of energy efficiency. Unfortunately, existing studies have paid little attention to this issue.
To fill this gap in existing research, this paper performs the following exploratory research. First, we use a fixed-effect stochastic frontier analysis (SFA) model based on a Translog production function to calculate the total factor energy efficiency growth (TFEEG) (2) rates of China's 36 industrial sub-sectors over 2001-2014. This method is used to avoid the defects of adopting the single factor energy efficiency. Second, for the very first time, we use the difference-in-differences (DID) method to investigate the EICP's (marginal) effect on the TFEEG. Furthermore, we estimate the superposition effect caused by the introduction of China's carbon intensity constraint policy (CICP) on the TFEEG, through the difference-in-difference-in-differences (DDD) method. Finally, through counterfactual, re-grouping and quasi-DID analyses, we conduct a series of robustness tests of the empirical results. Some previous studies have demonstrated that the implementation effectiveness of China's policies is not satisfactory (e.g., Zhang et al., 2017); we provide additional evidence to support those findings in this paper. We find that, after the implementation of the EICP, the total factor energy efficiency (TFEE) significantly declines. Thus, China's mandatory energy-saving policy can actually have a detrimental effect on productivity growth.
The remainder of this paper is organized as follows: Section 2 provides a related literature review; Section 3 introduces the methodology and data, including the estimation model of the TFEEG and the DID model of investigating the effect of the EICP; In Section 4, we report and discuss the estimation results of the TFEEG and the DID model, as well as the superposition effect; In Section 5, we perform a series of the robustness tests of the estimation results; and Section 6 draws main conclusions and provides some policy implications.
2. LITERATURE REVIEW
2.1 How to Measure Energy Efficiency
Depending on the measurement methods used, energy efficiency can be divided into two categories: single factor energy efficiency and total factor energy efficiency. Single factor energy efficiency is usually measured as the ratio of output (e.g., GDP) to energy consumption (Ang, 2006; Ang and Goh, 2018). Since only one input is considered, single factor energy efficiency can be regarded as energy productivity and thus not actual energy efficiency (Patterson, 1996; Li and Hu, 2012). Compared with single factor energy efficiency, TFEE can more rationally reflect the connotation of efficiency, because TFEE expands from just a "single input" to having "multiple inputs"; TFEE also considers different inputs and their substitution relationships (Patterson, 1996; Hu and Wang, 2006). Therefore, TFEE is more widely used (Hu and Wang, 2006; Zhou and Ang, 2008; Zhang et al., 2011; Wang et al., 2012; Fang et al., 2013; Zou et al., 2013; Chang, 2013; Zhao et al., 2014; Honma and Hu, 2014; Makridou et al., 2015). More importantly, because single factor energy efficiency is the reciprocal of energy intensity, we can easily infer that single factor energy efficiency will be improved after the implementation of EICP. However, it is difficult to accurately forecast the impact of EICP on TFEE, due to the substitution effects between the various factors, the structural effects of inputs, and the technical and allocation adjustments that occur during the production process. Hence, there is no doubt that the investigation of the effect of EICP on TFEE is of more value and greater practical significance to policy-makers.
Two types of methods are widely-used for the measurement of TFEE: data envelopment analysis (DEA) and stochastic frontier analysis (SFA). Both methods have their own advantages. The DEA method does not need to set a specific function form and thus can avoid bias from subjective model setting. This allows the DEA method to accurately reflect the production process with a variety of simultaneous outputs and inputs. However, because the DEA method does not consider the effect of random errors, more accurate data is required. Also, the DEA is easily affected by the statistical errors of sample data (Shao et al., 2016). This may lead to a deviation in measured efficiency, to some extent. Hu and Wang (2006) used the DEA method to measure provincial-level TFEE in China. Since then, many types of DEA models have frequently been used to measure China's TFEE (growth). Zhou et al. (2008) provided a comprehensive literature review of this field.
Meanwhile, a number of studies have adopted the SFA method to measure TFEE (growth) (e.g., Murty and Kumar, 2002; Murty et al., 2006; Vaninsky, 2010; Yang et al., 2011; Azadeh et al., 2011; Herrala and Goel, 2012; Zhou et al., 2012; Chen and Paulino, 2013; Wang et al., 2013; Llorca et al., 2017). Compared with DEA, the SFA method has the following advantages. First, the SFA model introduces random disturbance terms. Then, SFA also assumes that deviation from the frontier comes from two parts: the non-negative random disturbance term reflecting ineffective technology, and the system random disturbance term. Second, SFA has a statistical property for testing estimation parameters and model settings. Third, compared with the fixed production frontier used in the DEA method, SFA allows for a stochastic frontier that is closer to reality. Finally, since the measured efficiency based on the DEA model is the "relative" value, it is easy to determine that all efficiency values of effective production units are equal to 1. Thus, performing any further comparative analysis of these units is difficult. In contrast, the SFA method can obtain the "absolute" efficiency value (Coelli et al., 2005). In this paper, we use the statistical data of China's 36 industrial sub-sectors to carry out an empirical analysis. Inevitably, some strong noise and industrial differences exist in these statistical data. Thus, the SFA method is more appropriate for use in the current study.
2.2 Determinants of Energy Efficiency
Existing studies have discussed the determinants of energy efficiency from various perspectives. In general, these determinants can be divided into four categories of effects: scale, structural, technical and institutional effects. The scale effect refers to economic output (Tang and Yang, 2009; Shao et al., 2011; Fisher-Vanden et al., 2016). The structural effect mainly refers to industrial structure (Wei and Shen, 2007) and energy consumption structure (Shao et al., 2011). The technical effect is used to reflect technological progress, which is traditionally measured by energy intensity and R&D investment (Lin and Zhao, 2009; Bointner, 2014; Zha and Kavuri, 2016). The institutional effect involves relevant institutional conditions' effects on energy savings and emission reductions, such as China's opening-up policy (Tang and Yang, 2009; Li et al., 2014) and energy subsidies (Chronopoulos et al., 2016).
In particular, the government's energy-saving policies play a significant role in energy efficiency. For instance, Greening et al. (2000) and Helm (2002) pointed out that technology-related policies could be used to reduce energy use. However, progress made in the field of energy-saving technologies might lead to an increase in energy consumption, i.e., the energy rebound effect. Hence, energy-saving technological progress does not always cause a decline in energy intensity. Liao et al. (2007) found that, from 2003 to 2005, the high investment and over-expansion of China's high energy-consuming industries became the primary contributor of increasing energy intensity. However, thanks to efforts made by the Chinese government, the country's energy intensity level began to decline in 2006.
It is noteworthy that, even though previous studies have paid considerable attention to the determinants of energy efficiency, no special investigation into the effects of China's EICP on energy efficiency has ever been conducted. Compared with existing studies that focus on macroeconomic policies and institutional factors (such as openness and marketization), the EICP mentioned above is a policy specifically designed to improve China's energy efficiency. The effect of the EICP will not only determine whether or not China's energy-saving target, as reflected by single factor energy efficiency, can be successfully achieved; the effect will also determine whether or not the achievement of the energy-saving target has been effective from the perspective of total factor energy efficiency.
2.3 Evaluation on the Effects of Environmental Policies
Unlike existing studies on the measurement and determinants of energy efficiency, specific investigations into the effects of China's energy-saving policies (especially EICP) are scarce. However, a few similar studies, which evaluate the effects of China's environmental policies, can be found. For instance, Bao et al. (2013) used a DID model to analyze the effects of local environmental legislation supervision in China, and found that simple environmental legislation did not significantly inhibit local pollutant emissions in most regions. However, in some regions where there is strict law enforcement or serious pollution problems, environmental legislation has had the desired effect. Li and Chen (2013) also adopted the DID strategy to evaluate the impacts of the revision of the China Air Pollution Control Act of 2000 on China's industrial total factor productivity growth. The total factor productivity of pollution-intensive industrial sub-sectors was found to increase significantly after 2000. As an effective method for evaluating the effects of various policies, the DID method has been widely used in related studies. However, to date, existing studies have not paid sufficient attention to the natural experiment of implementing the EICP in 2006 targeted at reducing China's energy intensity.
Under the backgrounds described above, in this paper, we use the fixed-effect SFA method to estimate the TFEEG of China's 36 industrial sub-sectors. Furthermore, we employ the DID method, to investigate for the first time the average and marginal effects of China's EICP on industrial TFEEG, so that important policy insights can be provided for policy-makers seeking to more effectively implement China's energy conservation policies.
3. METHODOLOGY AND DATA
3.1 Empirical Strategy
The implementation of the EICP, as raised in China's 11th FYP, is bound to have different effects on various industrial sub-sectors with different levels of energy intensity. Thus, such a policy can be regarded as a natural experiment, which in turn allows us to use the DID method to evaluate the actual effect of the policy. The DID strategy has an obvious advantage in investigating the effectiveness of a certain policy by comparing the differences in the effects of a certain policy between the treatment group and the control group (Bertrand et al., 2004). Therefore, it is important to reasonably identify both the treatment group and the control group.
Undoubtedly, the EICP has an important impact on energy intensity. This is especially true in the case of energy-intensive sub-sectors. Hence, sub-sectors with higher energy intensity levels are expected to be more greatly affected by the EICP. Following Li and Chen (2013), we treat the sub-sectors that are heavily affected by the policy as the treatment group, while the remaining sub-sectors are treated as the control group. To avoid the estimation bias caused by the difference between these two groups in terms of sample size, 17 sub-sectors are classified as the treatment group, and 19 sub-sectors constitute the control group. (3) To be very specific, the sub-sectors with an energy intensity of more than 0.35 tee per [10.sup.4] RMB are considered to be part of the treatment group, while the remaining sub-sectors constitute the control group. Thus, we establish the following baseline DID model:
[y.sub.it] = [[beta].sub.0] + [[beta].sub.1][du.sub.it] + [[beta].sub.2][dt.sub.it] + [[beta].sub.3][du.sub.it] x [dt.sub.it] + [beta][CV.sub.it] + [f.sub.i] + [f.sub.t] + [[epsilon].sub.it] (1)
where i and t denote sub-sectors and years, respectively; y represents the estimated TFEEG rate (TFEEG); du is a dummy variable for sub-sectors (1 for the treatment group and 0 for the control group); dt is also a dummy variable for time, and dt= 1 after the EICP was implemented in 2006, otherwise dt=0; [[beta].sub.3] is the DID estimator, representing the net effect of the EICP on the improvement in the TFEE, i.e., the relative changes in TFEEG between the treatment and control groups; CV is the set of controlling variables, i.e., other factors that may affect TFEEG, including innovation ability (rd), capital deepening (kl), asset scale (size), technology spillover from foreign investment (fdi), and the proportion of state-owned enterprises (state); [[beta].sub.0]~[[beta].sub.3] and [beta] are the parameters to be estimated; [f.sub.i] and [f.sub.t] denote the sector fixed effect and the year fixed effect, respectively; [epsilon] is the random disturbance term. The selection reasons and measurement methods of these variables are described as follows.
(1) Total factor energy efficiency growth rate (TFEEG): Considering the corner-solution estimation problem in Battese and Coelli's (1992, 1995) models (Wang, 2002; Wang and Schmidt, 2002) and the heterogeneity between different industrial sub-sectors, we adopt the fixed-effect SFA model (Greene, 2005) and maximum likelihood method (MLM) to estimate TFEEG. The initial stochastic frontier production function proposed by Aigner et al. (1977) and Meeusen and van den Broeck (1977) can be rewritten as the following panel data form:
[Y.sub.it] = f([X.sub.it],[alpha])exp([v.sub.it] - [u.sub.it]) (2)
where i (i = 1, 2,..., N) still represents industrial sub-sectors; t (t = 1,2,..., T) denotes the years; Y is the industrial output; f([X.sub.it],[alpha]) is the production frontier; X is the inputs vector, and [alpha] is the parameter vector to be estimated; v is a random disturbance term, assuming that v ~ N(0,[[sigma].sup.2.sub.v]) and is independent from u, which reflects the influences of statistical error and various stochastic factors on the frontier output; u (u [greater than or equal to] 0 ) is a unilateral disturbance term that obeys the non-negative truncated normal distribution, and u is a time-varying technical inefficiency term, indicating technical inefficiencies over time.
To estimate TFEEG, we set up a Translog production function for Eq. (2) as follows:
[mathematical expression not reproducible] (3)
where Y is the gross industrial output of each industrial sub-sector; K is capital stock; L is labor; E is energy consumption; [[sigma].sub.i] and [[sigma].sub.t] are the sector fixed effect and the year fixed effect, respectively. (4)
The technical efficiency (TE) of a sub-sector can be determined as the ratio of the expectation of actual output and the expectation of the stochastic production frontier as follows:
[TE.sub.it] = [E[f([X.sub.it],[alpha]) exp([v.sub.it] - [u.sub.it])]/E[f([X.sub.it],[alpha]) exp([v.sub.it] - [u.sub.it])|[u.sub.it] = 0]] = exp([-u.sub.it]) (4)
According to Battese and Coelli (1992), [u.sub.it] = [u.sub.i] exp[-[eta](t - T)], assuming that [u.sub.i] is subject to a non-negative truncated normal distribution, that is [u.sub.i] ~ [N.sup.+]([mu],[[sigma].sup.2.sub.u]). The parameter [eta] represents the change rate of technical efficiency, and [eta] > 0, [eta] = 0, and [eta] < 0 represent the improvement, invariability, and deterioration of technical efficiency, respectively. We employ STATA to estimate such a fixed-effect SFA model.
According to Kumbhakar (2000), the TFEE growth rate can be calculated through the following equation:
[TFEG.sub.it] = [TC.sub.it] + [TEC.sub.it] + [SE.sub.it] (5)
In Eq. (5), technological progress (TC) is defined as the change rate of the production frontier over time when the inputs are controlled:
[TC.sub.it] = [[partial derivative] ln f ([X.sub.it],[alpha])/[partial derivative]t] = [[alpha].sub.1] + [[alpha].sub.2]t + [[alpha].sub.6] ln K + [[alpha].sub.7] ln L + [[alpha].sub.8] ln E (6)
The variable TEC represents the change in technical efficiency and is defined as the change rate of technical efficiency over time:
[TEC.sub.it] = [[partial derivative] ln [TE.sub.it]/[partial derivative]t] = [[partial derivative] ln exp([-u.sub.it])/[partial derivative]t] = -[[partial derivative][u.sub.it]/[partial derivative]t] (7)
The variable SE is the scale efficiency, reflecting the contribution of returns to scale of the factors to productivity growth. This variable can be calculated as follows:
[SE.sub.it] = ([RTS.sub.jit] - 1) [[sigma].sub.j] [[lambda].sub.jit][[??].sub.jit] (8)
where [RTS.sub.jit] = [[SIGMA].sub.j] [[epsilon].sub.jit] represents the scale economies effect; the variable [[epsilon].sub.jit] is the output elasticity of factor j, and [[epsilon].sub.j] = [[partial derivative] ln Y/[partial derivative] ln j] variable [[lambda].sub.jit] = [[epsilon].sub.jit] / [RTS.sub.it] represents the output elasticity of factor j relative to the overall returns to scale; the variable [[??].sub.jit] is the change rate of the input of factor j (j=K, L, and E).
(2) Innovation ability (rd): Existing studies have argued that R&D activities can significantly promote both energy-saving technology and energy efficiency (Bointner, 2014; Yang et al., 2017a). We use R&D intensity, i.e., the ratio of internal expenditure on R&D activities to total industrial output, to reflect industrial innovation ability, and the coefficient of this variable is expected to be positive.
(3) Capital deepening (kl): With reference to Shao et al. (2016), capital deepening has a significant positive effect on green technical change. We measure kl by the ratio of capital stock to labor force (taking a natural logarithm).
(4) Asset scale (size): Expanding the scale of industrial production facilitates a corresponding improvement in equipment productivity and energy intensification in the production process. This may be conducive to improving energy efficiency (Tang and Yang, 2009). Hence, we introduce the natural logarithm of industrial total assets at the 2000 constant price ([10.sub.8] RMB) into the regression model.
(5) Technology spillover of foreign investment (fdi): The technology spillover of foreign investment is beneficial to the improvement in enterprise's productivity in the host country (Javorcik, 2004). Based on data availability and following Yang et al. (2017b), we use the share of the foreign enterprises' sales value in the total sales value to proxy this variable.
(6) Proportion of state-owned enterprises (state): Chen and Golley (2014) argued that state-owned enterprises generally have typical political characteristics and low production efficiency. Hence, we adopt the share of the state-owned enterprises' output value in the total industrial output value to measure this variable.
Thus, the final baseline regression model based on Eq. (1) is as follows:
[y.sub.it] = [[beta].sub.0] + [[beta].sub.1][du.sub.it] + [[beta].sub.2][dt.sub.it] + [[beta].sub.3][du.sub.it] x [dt.sub.it] + [[beta].sub.4][rd.sub.it] + [[beta].sub.5][kl.sub.it] + [[beta].sub.6][size.sub.it] + [[beta].sub.7][fdi.sub.it] + [[beta].sub.8][state.sub.it] + [f.sub.i] + [f.sub.t] + [[epsilon].sub.it] (9)
3.2 Data Description
Based on the availability and integrality of the relevant data, we use the panel data of 36 industrial sub-sectors in China, covering the years from 2001 to 2014, as our research sample. The input and output data of Eq. (3) are from the China Industrial Economics Statistical Yearbook, China Labor Statistical Yearbook, and China Energy Statistical Yearbook. To ensure the comparability of data, we deflate all the raw data at the 2000 constant price.
(1) Gross industrial output (Y): Following existing studies (e.g., Yang et al., 2017b), we use gross industrial output value, which contains the intermediate input. (5)
(2) Industrial capital stock (K): We adopt the perpetual inventory method to estimate the capital stock of each sub-sector as [K.sub.t] =(1 - [[delta].sub.t])[K.sub.t-1] + [I.sub.t], where [K.sub.t] represents the capital stock at time t, and the initial stock is estimated by referring to Chen (2011); [I.sub.t] denotes the amount of annual physical capital investment, and the amount of gross investment in fixed assets is chosen as its proxy; [[delta].sub.t] is the depreciation rate, which is calculated from the ratio of current depreciation to the original value of fixed assets, as reported in statistical yearbooks (Chen, 2011).
(3) Industrial labor input (L): We take the annual average employment of each sub-sector as the proxy of the labor input.
(4) Industrial energy consumption (E): We use the total industrial energy consumption, with the unit of tons of coal equivalent (tce), to measure this variable.
Because of the alteration of the statistical scope, three sub-sectors are excluded from our sample, i.e., other mining industry, handicrafts and other manufacturing, and waste resources and materials recycling. Thus, our sample consists of 36 industrial sub-sectors in China. For convenience, these various sub-sectors are labeled by Nos. S1-S36, in turn (see Table A1). The descriptive statistics of these variables are shown in Table 1. (6)
4. RESULTS AND DISCUSSION
4.1 Estimation Results of Total Factor Energy Efficiency Growth Rate
The estimated parameters of Eq. (3) are shown in Table 2. Most coefficients are significant at the 1% level. The logarithmic likelihood function value is valid, indicating that the model is well set.
In Table 2, [micro] is the average of the estimated inefficiency term u in Eq. (3); a positive [micro] value indicates an overall decline in the technical efficiency of industrial production. The overall variance ([[sigma].sup.2] = [[sigma].sub.v.sup.2] + [[sigma].sub.u.sup.2]) is 8.7513, which indicates that the overall fluctuation in productivity is affected by random and inefficient factors. The variance of the random disturbance term ([[sigma].sub.v.sup.2]) is far less than that of the technical inefficiency term ([[sigma].sub.u.sup.2]). This indicates that the gap between the actual output and the production frontier is mainly caused by technical inefficiency. Because full consideration has been given to the loss of technical efficiency and the production heterogeneity of different sub-sectors, compared with the traditional production function, the fixed-effect stochastic frontier production function can more rationally describe the level of technical efficiency and the changes in technical efficiency in the industrial production process.
Based on Eqs. (5)-(8), we further calculate the TFEE's growth rates (TFEEG rates) of the 36 studied industrial sub-sectors during the 2001-2014 study period; the detailed results are listed in Table A2. As shown in Figure 2, the TFEEG rates of the three types of industrial sub-sectors and the entire industrial sector all show an overall declining trend during the study period. The TFEEG rate of "Production and Supply of Electricity, Gas and Water" sub-sector is higher than the growth rates of the other two types of sub-sectors, except in 2014. The trend of the manufacturing's TFEEG rate is the closest to that of the entire industrial sector, thereby indicating that the TFEEG of the entire industrial sector is largely determined and significantly influenced by manufacturing.
We separately depict the trend of the TFEEG rate of the entire industrial sector in Figure 3. We find that the TFEEG rate remained positive but continued to decline from 2001 to 2014. This finding indicates that the production technology of the entire industrial sector has been improving, but with a declining growth rate. It can clearly be seen that the TFEEG shows obvious fluctuations after 2006 (the year of the implementation of the EICP). This trend was in line with China's macroeconomic development and environmental policy at that time. Since the implementation of the EICP in 2006, industrial enterprises have faced mandatory constraints on energy use and have thus had to adjust their factor input structures. Undoubtedly, obtaining and implementing the optimal combination of production factors in such a short timeframe would be difficult for China's industrial enterprises. Hence, the changes caused some fluctuations in the TFEEG rate in the industrial sector. This provides preliminary evidence of the significant effects of the EICP. The real impact of the EICP on the TFEEG will be discussed in detail, later in this paper.
4.2 Estimation Results of DID Model
In the DID model, the treatment group and the control group should meet the parallel trend assumption, i.e., the dependent variables of both the treatment and control groups should display the same trends before the implementation of the EICP. As shown in Figure 4, before the introduction of the EICP, the TFEEG rate in the treatment group was higher than that the rate in the control group. Also, the TFEEG rates in these two groups showed similar trends. After the introduction of the EICP in 2006, the treatment group's TFEEG rate experienced a more significant level of decrease, while the control group's rate showed a slower declining trend and then exceeded the rate of the treatment group for the first time in 2008. This indicates that the EICP had a stronger influence on the treatment group than on the control group. However, a scientific and rigorous empirical analysis is still required, in order to test whether such differences in the effects of the EICP on the TFEEG (between the treatment group and the control group) are significant.
The estimation results of Eq. (9) are shown in Table 3. We first use the OLS to estimate the model. The control variables (CV) are introduced in Model 2. We find that the coefficients of du x dt are significantly negative in both models. Considering that the panel data have both cross-sectional and time-series data features, we adopt the Woodridge test and the White test to verify the autocorrelation and heteroscedasticity of the residuals. We find that significant first-order autocorrelation and heteroscedasticity exists, regardless of whether or not the control variables are considered. This finding indicates that the OLS estimation results are biased. Hence, we also adopt the panel corrected standard errors (PCSE) method to correct such a bias. The estimation results of the PCSE method are reported in Model 3 and in Model 4 in Table 3. According to the principles of the PCSE model, the model's estimation results are obtained by correcting the standard errors and significance levels of coefficients based on the results of the OLS model. Therefore, the estimation parameter values of the PCSE model are the same to those of the OLS model, but more robust.
Regarding Model 3, without any control variables, the coefficient of du x dt is significantly positive at the 1% level. In Model 4, with the control variables, the corresponding coefficient is also significant at the 1% level. These results show that the EICP has a negative effect on China's industrial TFEEG. After the implementation of the policy, the TFEEG rate in the treatment group declined by 4.31%, compared to the rate of the control group. Such a result is consistent with Yang et al. (2017b), whose study concludes that a coercive environmental policy could have detrimental effects on China's industrial productivity growth. After the implementation of the EICP, industrial enterprises were required to decrease energy use and increase the use of other factors for the substitution relationships between factor inputs (Kumar et al., 2015), especially capital. China's statistical data show that the average annual growth rate of capital input after 2006 has been 10.53% higher than that the average annual growth rate before 2006. Therefore, the low efficiency of capital input in the production process leads to a decline in the TFEEG (Chen et al., 2011).
Price et al. (2011) pointed out that China made some efforts to achieve the energy intensity constraint targets set in the 11th FYP. In fact, the country has made obvious progress in improving energy efficiency. This study, however, provides empirical evidence contrasts with such an argument. After the EICP was put forward, China's central and local governments adopted a series of policies and measures designed to improve energy efficiency. However, the industrial TFEEG did not significantly improve and, in fact the effectiveness of these new policies and measures was unsatisfactory.
In addition, the coefficients of most control variables in Model 4 are statistically significant, as expected. Specifically, the coefficient of rd is significantly positive, which shows that sub-sectors with a higher proportion of R&D investments have a better performance in terms of the TFEEG. The same conclusion can be found in Bointner (2014) and Song et al. (2018). Also, the coefficient of kl is significantly positive, indicating that capital deepening can improve the TFEEG. This is because technological progress is usually embodied in the investment in physical capital, such as new equipment. This finding is consistent with that of Greenwood et al. (1997). Similarly, the coefficient of size is significantly positive at the 1% level, clearly indicating that the expansion of the industrial assets scale is conducive to intensive energy use and improvements in the TFEE. This conclusion is also consistent with that of Tang and Yang (2009). Since the coefficient of fdi is positive but not significant, the effect of foreign investments on the TFEEG is not obvious. According to the "pollution haven" hypothesis, foreign enterprises tend to increase energy use and corresponding pollutant emissions to impair the improvement in green productivity (Shahbaz et al., 2015; Yuan and Zhang, 2017). The coefficient of state is negative but not significant. This is consistent with the results of some existing studies (e.g., Boycko et al., 1996) and may be attributed to the low production efficiency of state-owned enterprises (Ohene-Asare, 2017).
4.3 Estimation Results of Marginal Effect
The estimation results in Table 3 only reflect the degree of annual average influence of the EICP on the TFEEG during the years from 2007 to 2014, but cannot reveal the marginal effect of the EICP. However, the estimation of the marginal effect is of great importance in terms of evaluating the actual effects of a certain policy, because the marginal effect can help us capture the policy's yearly effect. As such, referring to some existing studies (e.g., Nunn and Qian, 2011; Li and Chen, 2013), we re-write Eq. (9) as follows:
[y.sub.it] = [[theta].sub.0] + [[theta].sub.1][du.sub.it] + [[theta].sub.2][dt.sub.it] + [[theta].sub.3][du.sub.it] x [dt.sub.it] + [[theta].sub.4][rd.sub.it] + [[theta].sub.5][kl.sub.it] + [[theta].sub.6][state.sub.it] + [[theta].sub.7][fdi.sub.it] + [[theta].sub.8][size.sub.it] + [2014.summation over (j = 2007)][[tau].sub.j][du.sub.it] x [dt.sub.it] x [year.sup.j] + [f'.sub.i]+ [f'.sub.t] + [[epsilon]'.sub.it] (10)
where [[theta].sub.0]~[[theta].sub.8] and [[tau].sub.j] are the parameters to be estimated; [year.sup.j] is a year dummy variable that is equal to 1 in year j and is otherwise 0; [f'.sub.i] and [f'.sub.t] denote the sector fixed effect and year fixed effect, respectively; [epsilon]' is the random disturbance term.
Through the estimation of Eq. (10), we can calculate the marginal effect of EICP on industrial TFEEG. For year j, the marginal effect is [[theta].sub.3] + [[tau].sub.j]. We also use the PCSE method to estimate Eq. (10). The results are shown in Figure 5 and Table 4. As shown in Figure 5, the marginal effect experiences an overall inverted N-shaped trend. The negative value of the marginal effect in all years indicates the existence of a negative influence of the EICP on the TFEEG. After the implementation of the EICP, industrial firms were inevitably faced with some shocks and "labor pains" in the short term. Under conditions of the intensifying of energy constraints, industrial firms were forced to adjust their production factor structures, in order to maintain or raise their output levels. Thus, the original equilibrium of factor allocation was broken. Before reaching a new equilibrium, it was inevitable that the marginal growth of the TFEE would slow down at the beginning of the introduction of the new policy. Generally speaking, an up-bottom policy will typically need to experience a refined or revised process before achieving the expected effect. The overall process will eventually lead to a unified state between the central policy and any diversified local auxiliary measures. Moreover, the implementation of a specific policy will generally involve multiple domains, including economy, politics, and culture. As such, the success of a policy will ultimately depend on multi-sector cooperation and other supporting polices (He and Kong, 2011). It is fair to say that the EICP has to face the challenges of the existing inertia of production patterns and traditional factor structures.
In addition, we find that the marginal effects experienced in 2007 and 2008 were significantly negative (see Table 4), but the effect in 2009 was not significant. This suggests that the negative impact of the EICP came into effect soon after the policy's implementation. However, the evident effect only lasted for the first two years. Moreover, the magnitude of the 2008 coefficient is less than that in 2007, thereby indicating that the marginal effect has a decreasing trend. In particular, the influence of the policy was not significant in 2009. However, a significantly negative effect (cause by the policy) appeared again from 2010 onwards. Such results are consistent with the actual policy situations of the time. As mentioned above, China officially raised the carbon intensity constraint policy (CICP) at the end of 2009, but the signal effect and practical implementation of the policy came into force in 2010. Existing studies have confirmed that the CICP can and does reduce the allocative efficiency of resources in China (Yang et al., 2017b). That is to say, the implementation of the CICP may also have a negative effect on the improvement in energy efficiency. Obviously, the energy intensity targeted by the EICP is related closely to the carbon intensity targeted by the CICP. As such, it is not surprising that the negative impact of the EICP on the TFEEG could have been triggered by the implementation of the CICP in 2010. In other words, the implementation of the CICP could have enhanced the negative effect of the EICP on the TFEEG. Hence, we have reasons to believe that the significantly negative effect of the EICP on the TFEEG as seen after 2010 is due to the superposition effect of the CICP. Next, we will perform a detailed investigation of the superposition effect.
4.4 Superposition Effect of Carbon Intensity Constraint Policy
In 2009, China officially introduced the carbon intensity constraint policy (CICP), which stated that carbon intensity levels (carbon dioxide emissions per GDP) would be reduced by 40% to 45% in 2020, compared to the country's 2005 level. A recent study by Yang et al. (2017b) presented empirical evidence that the CICP had a negative effect on China's industrial green total factor productivity. Considering that our sample period covers the implementation stage of the CICP, it seems possible that both the CICP and the EICP may have simultaneously affected industrial TFEEG after the introduction of the CICP since 2010. (7) In other words, the CICP may have a superposition effect on the TFEEG, based on the effect of the EICP. Therefore, we need to adopt the difference-in-difference-in-differences (DDD) method to investigate the superposition effect of the CICP. Obviously, since the sub-sectors with higher carbon intensities are more affected by the CICP, we consider these sub-sectors as the treatment group; the remaining sub-sectors are considered to be the control group. To be specific, those sub-sectors with carbon intensities of more than 0.30 tons per [10.sup.4] RMB are considered to be the treatment group, (8) the remaining sub-sectors constitute the control group. To avoid the estimation bias caused by the difference in sample size between these two groups, 18 sub-sectors are classified as the treatment group, and the other 18 sub-sectors are classified as the control group. The detailed classification of industrial sub-sectors is also listed in Table Al. We set a dummy variable dm=1 to indicate the treatment group, while dm=0 represents the control group. Thus, following Moser and Voena (2012), the DDD model is set as follows:
[y.sub.it] = [[lambda].sub.0] + [[lambda].sub.1][du.sub.it] + [[lambda].sub.2][dt.sub.it] + [[lambda].sub.3][du.sub.it] x [dt.sub.it] + [[lambda].sub.4]dm + [[lambda].sub.5]dm x du + [[lambda].sub.6]dm x dt + [[lambda].sub.7]dm x du x dz + [[lambda].sub.8][rd.sub.it] + [[lambda].sub.9][kl.sub.it] + [[lambda].sub.10][size.sub.it] + [[lambda].sub.11][fdi.sub.it] + [[lambda].sub.12][state.sub.it] + [[omega].sub.i] + [[omega]sub.t] + [[phi]sub.it] (11)
where [[lambda].sub.0]~[[lambda].sub.12] are the parameters to be estimated, and [[lambda].sub.7] indicates the superposition effect of the CICP; dz is a time dummy variable, and dz=0 indicates the years before 2009, while dz = 1 represents the years since 2010, when the CICP came into effect; [[omega].sub.i] and [[omega].sub.t], denote the sector fixed effect and the year fixed effect, respectively; [phi] is the random disturbance term. We also adopt the PCSE method to estimate Eq. (11) and show the estimation results in Table 5.
We find that both the EICP and the CICP have a significantly negative superposition effect on the TFEEG, as a result of the significant negative coefficient of dm x du x dz (-0.0150). Also, even though the EICP and the CICP have achieved the desired targets of reducing energy intensity and carbon intensity (Price et al., 2011; Yang et al., 2017b), respectively, these two policies have a negative synergy effect on the TFEEG.
Such a result provides a reasonable explanation for the different degrees of significances of EICP's negative impacts on the TFEEG in 2009 and 2010, as shown in Table 4. After the effect of the EICP was no longer significant in 2009, the introduction of the CICP forced industrial enterprises to further adjust their factor input structures. The result was a reduction in the efficiency of resource allocation and in the total factor productivity. Our findings further confirm that energy-saving and emission-reduction policies can decrease energy consumption at the expense of reducing the allocative efficiency of resources. Meanwhile, market-based energy-saving and emission-reduction policies are based on explicit price signals and are generally considered to be more cost-effective (Yang et al., 2017b). After taking the CICP into consideration, the inhibitory effect of the EICP on the TFEEG remains significant. This shows that the results of the previous analysis conducted in this paper are robust, to a certain extent.
5. ROBUSTNESS TEST
Figure 4 illustrates that the decline in the TFEEG in the treatment group is more obvious after the introduction of the EICP. However, our DID strategy still needs to be further verified, in order to meet the parallel trend assumptions. In the absence of the policy shock, a precondition of the DID method is that the treatment group and the control group must have a common trend. With regard to this study, if the EICP had not been introduced in 2006, then we can assume that the TFEEG rates of the treatment group and the control group would have had the same trend. However, because the EICP was introduced, it is not possible to be completely certain of what the trends of TFEEG rates in the treatment group and the control group would have been had the EICP not been implemented. To solve such a problem, existing studies usually employ the following methods to conduct robustness tests.
5.1 Counterfactual Analysis
We assume that the EICP was implemented ahead of schedule. That is to say, supposing that the counterfactual event occurred in 2005 or 2004, we investigate the effect of the virtual EICP on the TFEEG. If the virtual EICP that was implemented in 2004 or 2005 still has a significant effect on the TFEEG, then we can conclude that the previously obtained results are not robust. Otherwise, the results already obtained are robust. Thus, following some previous studies (e.g., Moser and Voena, 2012; Shao et al., 2017), we establish the following DID model to perform the counterfactual analysis. (9)
[y.sub.it] = [[beta]'.sub.0] + [summation over (t = 2004,2005)][[beta]'.sub.1][du.sub.it] x [Preyear.sub.t] + [[beta]'.sub.2][rd.sub.it] + [[beta]'.sub.3][kl.sub.it] + [[beta]'.sub.4][size.sub.it] + [[beta]'.sub.5][fdi.sub.it] + [[beta]'.sub.6][state.sub.it] + [f'.sub.i]+ [f.sub.t] + [[epsilon]'.sub.it] (12)
where preyear is a time dummy variable, and preyear= 1 when t=2004 or 2005; otherwise, pre-year = 0, when supposing that the EICP was introduced in 2004 or 2005. We still use the PCSE method to estimate Eq. (12). We present the results of the counterfactual analysis for 2004 and 2005 in Test 1 and Test 2, respectively, of Table 6.
We find that the interaction terms between preyear and du in Test 1 and Test 2 of Table 6 are not significant. This finding indicates that the parallel trend assumption between the treatment group and the control group is met (Moser and Voena, 2012; Shao et al., 2017). Hence, the previous results are both robust and reliable.
5.2 Re-grouping Analysis
In the previous analysis, to avoid the estimation bias caused by the difference in the sample size between the treatment group and the control group, we subjectively set the close numbers of the two groups. However, we still need to rule out the possibility that the EICP's effect could change in line with corresponding changes in the sample size of the treatment group. Therefore, we re-divide the treatment group and the control group according, this time using a higher grouping standard. If the grouping standard of energy intensity is raised to 0.5 tee per [10.sup.4] RMB and 0.7 tee per [10.sup.4] RMB (the numbers of the treatment group are 15 and 13, respectively), (10) and if the EICP still has a significant effect on the TFEEG, then it can be concluded that the results obtained earlily in this study are robust. We present the results of the re-grouping analysis in Test 3 and Test 4, respectively, of Table 6.
We continue to use the PCSE method to estimate the DID model in the re-grouping analysis. In Test 3 and Test 4 of Table 6, the results suggest that the EICP's negative effect on the TFEEG does not change with the abovementioned corresponding changes in the sample size of the treatment group. This further confirms that our previous results are robust. In addition, it is noteworthy that the signs and significance levels of the coefficients of the various control variables in Table 6 are similar to those in Tables 3-5. This finding indicates that the control variables we selected for this study are reasonable.
5.3 Quasi-DID Analysis
We have proved that the EICP has a significant negative effect on the TFEEG by using the DID strategy. In the DID analysis, the treatment group and the control group are classified based on the energy intensities of different industrial sub-sectors. The implicit premise is that the EICP has the same effect on all sub-sectors, while the policy's impact on the sub-sectors with higher energy intensities is more significant. However, we cannot rule out the extreme situation in which the policy only affects sub-sectors with higher energy intensities. The potential of this reality being true may affect the reliability of this study's previous results, to some extent. Therefore, we further use the quasi-DID strategy (Nunn and Qian, 2011; Yang et al., 2017b) and the PCSE method to investigate the effect of the EICP. Thus, Eq. (9) is rewritten as follows:
[y.sub.it] = [[chi].sub.0] + [[chi].sub.1][EI.sub.it] + [[chi].sub.2][dt.sub.it] + [[chi].sub.3][EI.sub.it] x dt + [[chi].sub.4][rd.sub.it] + [[chi].sub.5][kl.sub.it] + [[chi].sub.7][size.sub.it] + [[chi].sub.8][fdi.sub.it] + [[chi].sub.9]state + [[phi].sub.i] +[[phi].sub.it+ [[xi].sub.it] (13)
where El represents energy intensity; and the coefficient of [EI.sub.it]xdt reflects the effect of the change in energy intensity on the TFEEG after the introduction of the EICP. Also, Eq. (13) can reflect the impact of the improvement in the single factor energy efficiency on the TFEEG, after the introduction of the EICP.
The difference between Eq. (13) and Eq. (9) is that, in Eq. (13), we no longer group sub-sectors. Thus, there is no need to set the dummy variable for sub-sectors, and we only need to set a time dummy variable. In this way, we can conduct a more comprehensive analysis of the effect of the EICP on the overall industrial TFEEG. As shown in Table 7, we find that the coefficients of [EI.sub.it] x dt are significantly negative, thereby indicating that the implementation of thr EICP in 2006 caused a negative effect on the TFEEG. This further confirms that our previous results are robust.
6. CONCLUSIONS AND POLICY IMPLICATIONS
To realize green and sustainable development, as the largest developing country in the world, China made the decision to implement a mandatory energy intensity reduction target in 2006. Based on the background of such a policy, in this paper, we first use the fixed-effect SFA model to estimate the total factor energy efficiency growth rate of China's 36 industrial sub-sectors, during the period from 2001 to 2014. Then, we investigate the effect of the energy intensity constraint policy (EICP) on the total factor energy efficiency growth (TFEEG), based on the DID strategy. Furthermore, this paper also estimates the EICP's marginal effect and the superposition effect caused by the introduction of a carbon intensity constraint policy (CICP) through the DDD strategy. Finally, through counterfactual, re-grouping and quasi-DID analyses, we conduct a series of robustness tests to obtain the empirical results. The main conclusions can be drawn as follows.
(1) The TFEEG of China's industrial sector experienced an obvious declining trend during the 2001 to 2014 study period. This declining trend was in accordance with China's macroeconomic development and environmental policies. An obvious difference in the TFEEG exists between different types of industrial sub-sectors. The trend of the overall industrial sector's TFEEG is closest to that of the manufacturing sub-sector. This finding indicates that the TFEEG of the overall industrial sector is mainly determined by manufacturing.
(2) The implementation of the EICP in 2006 had a significant negative effect on the levels of improvement in the total factor energy efficiency of the treatment group, i.e., the sub-sectors with higher energy intensities. Compared to the rate of the control group, the rate of the TFEEG in the treatment group declined by an overall average of 4.31%. This indicates that the EICP does not necessarily improve industrial TFEE, and thus results in unsatisfactory policy effectiveness.
(3) The negative marginal effect of the EICP on industrial TFEEG experiences an overall inverted N-shaped trend. The effect of the EICP is significant in the first two years but not significant in the third year. After the introduction of the CICP, these two policies have a significantly negative superposition effect on the TFEEG. Therefore, energy-saving and emission-reduction policies can reduce energy consumption at the expense of reducing the allocative efficiency of resources.
The above findings have significant policy implications. First, through a series of empirical analyses, we find that the effect of China's energy-saving policy is not as satisfactory as previous studies have illustrated. Although the EICP can reduce energy intensity and even energy consumption per se, industrial enterprises inevitably need some time to make the adjustments necessary to achieve their expected targets. Hence, to shorten the time required and to actually achieve the expected effectiveness of energy-saving policies, the government should play a key role in the coordination of various involved subjects. The government must act to avoid policy segmentation and to decrease the costs of friction between various institutions and at different levels of government.
Second, the results of this study suggest that the non-market command-and-control policies can lead to the loss of the allocative efficiency of resources, to some extent. Hence, it is neither efficient nor sustainable for energy-saving endeavors to heavily depend on a command-and-control policy (such as the EICP), while simultaneously ignoring the important role of the market mechanism. Compared to the command-and-control approaches, market-oriented approaches are generally more efficient, because they can provide firms with greater flexibility and more incentives when engaging in energy-saving processes (Albrizio et al., 2017). Therefore, it is necessary to implement some market-oriented policies and measures (such as price and tax policies), to stimulate and encourage industrial enterprises to conserve energy by improving their resource allocation efficiencies and by employing more energy-saving technologies. In particular, the Chinese government should deeply reform the current energy pricing mechanism, i.e., establish an energy price formation mechanism that will accurately reflect the supply and demand relationship, scarcity of energies, and external costs of energy use.
Last but not least, before implementing any energy-saving and emission-reduction policy, the government should comprehensively prejudge the potential impacts of such policies on the economic system. The government should also mitigate the negative effects of such policies - as much as possible - through auxiliary measures. Meanwhile, the government should avoid implementing fragmented policies and should also make a greater effort to promote the synergy of different policies. As Price et al. (2011) argued, the Chinese government's energy intensity constraint policy has been successful in maintaining and reinforcing existing energy-saving effects and plans, but the policy still needs some necessary modifications, as well as new auxiliary policies, to achieve higher targets and more reasonable plans.
We acknowledge four referees and the financial support from the National Natural Science Foundation of China (Nos. 71373153,71773075, 71503168 and 71533004), the National Social Science Foundation of China (Nos. 18ZDA051 and 18ZDA102), the National Top-Notch Young Talent Support Program of China, the National Key Research and Development Program of China (No. 2016YFA0602500), and Shanghai Philosophy and Social Science Fund Project (No. 2015BJB005).
Aigner, D., C.A.K. Lovell, and P. Schmidt (1977). "Formulation and Estimation of Stochastic Frontier Production Function Models." Journal of Econometrics 6(2): 21-38. https://doi.org/10.1016/0304-4076(77)90052-5.
Albrizio, S., T. Kozluk, and V. Zipperer (2017). "Environmental Policies and Productivity Growth: Evidence across Industries and Firms." Journal of Environmental Economics and Management 81: 209-226. https://doi.org/10.1016/j.jeem.2016.06.002.
Ang, B.W. (2006). "Monitoring Changes in Economy-wide Energy Efficiency: From Energy-GDP Ratio to Composite Efficiency Index." Energy Policy 34: 574-582. https://doi.org/10.1016/j.enpol.2005.11.011.
Ang, B.W. and T. Goh (2018). "Bridging the Gap between Energy-to-GDP Ratio and Composite Energy Intensity Index." Enetgy Policy 119: 105-112. https://doi.org/10.1016/j.enpol.20I8.04.038.
Ang, B.W., A.R. Mu, and P. Zhou (2010). "Accounting Frameworks for Tracking Energy Efficiency Trends." Energy Economics 32: 1209-1219. https://doi.org/10.1016/j.eneco.2010.03.011.
Azadeh, A., S.M. Asadzadeh, M. Saberi, V. Nadimi, A. Tajvidi, and M. Sheikalishahi (2011). "A Neuro-fuzzy-stochastic Frontier Analysis Approach for Long-term Natural Gas Consumption Forecasting and Behavior Analysis: The Cases of Bahrain, Saudi, Arabia, Syria, and UAE." Applied Energy 88: 3850-3859. https://doi.org/l 0.1016/j.apenergy.2011.04.027.
Bao, Q., M. Shao, and D.Yang (2013). "Environmental Regulation, Provincial Legislation and Pollution Emissions in China." Economic Research Journal (12): 42-54.
Battes, G.E. and T.J. Coelli (1992). "Frontier Production Functions, Technical Efficiency and Panel Data: With Application to Paddy Fanners in India." Journal of Productivity Analysis 3: 149-165.
Battese, G.E. and T.J. Coelli (1995). "A Model for Technical in Efficiency Effects in a Stochastic Production Frontier for Panel Date." Empirical Economics 20(2): 325-332. https://doi.org/10.1007/BF01205442.
Bertrand, M., E. Duflo, and S. Mullainathan (2004). "How Should We Trust Differences-in-differences Estimates." Quarterly Journal of Economics 119: 249-275. https://doi.org/10.H62/003355304772839588.
Bointner, R. (2014). "Innovation in the Energy Sector: Lessons Learnt from R&D Expenditures and Patents in Selected IEA Countries." Energy Policy 73: 733-747. https://doi.org/10.1016/j.enpol.2014.06.001.
Boycko, M., A. Shleifer, and R.W. Vishny (1996). "A Theory of Privatisation." Economic Journal 106: 309-319. https://doi.org/10.2307/2235248.
Greenwood, J., Z. Hercowitz, and P. Krusell (1997). "Long-run Implications of Investment-specific Technological Change." American Economic Review 87(3): 342-362.
Chang, M.C. (2013). "A Comment on the Calculation of the Total-factor Energy Efficiency (TFEE) Index." Energy Policy 53: 500-504. https://doi.org/l0.1016/j.enpol.2012.10.025.
Chen, S. (2011). "Reconstruction of Sub-industrial Statistical Data in China: 1980-2008." China Economic Quarterly 10(3): 735-776.
Chen, S.Y. and J. Golley (2014). "'Green' Productivity Growth in China's Industry Economy." Energy Economics 44: 89-98. https://doi.org/10.1016/j.eneco.2014.04.002.
Chen, S. and A.U. Paulino (2013). "Energy Consumption Restricted Productivity Re-estimates and Industrial Sustainability Analysis in Post-reform China." Energy Policy 57: 52-60. https://doi.org/10.1016/j.enpol.2012.08.060.
Chen, S.M., Z. Sun, S. Tang, and D.H. Wu (2011). "Government Intervention and Investment Efficiency: Evidence from China." Journal of Corporate Finance 17(2): 259-271. https://doi.org/10.1016/j.jcorpfin.2010.08.004.
Chronopoulos, M., V. Hagspiel, and S.E. Fleten (2016). "Stepwise Green Investment under Policy Uncertainty." The Energy Journal 37(4): 87-108. https://doi.org/10.5547/01956574.37.4.mchr.
Coelli, T.J., D.S.P. Rao, C.J. O'Donnell, and G. E. Battese (2005). An Introduction to Efficiency and Productivity Analysis (2nd Edition). New York: Springer.
Fang, C.Y., J.L. Hu, and T.K. Lou (2013). "Environment-adjusted Total-factor Energy Efficiency of Taiwan's Service Sectors." Energy Policy 63: 1160-1168. https://doi.org/10.1016/j.enpol.2013.07.124.
Fei, R. and B. Lin (2016). "Energy Efficiency and Production Technology Heterogeneity in China's Agricultural Sector: A Meta-frontier Approach." Technological Forecasting & Social Change 109: 25-34. https://doi.org/10.1016/j.teeh-fore.2016.05.012.
Fisher-Vanden, K., Y. Hu, G. H. Jefferson, M. T. Rock, and M. Toman (2016). "Factors Influencing Energy Intensity in Four Chinese Industries." The Energy Journal Ti'(1): 153-178. https://doi.org/10.5547/01956574.37.SIl.kfis.
Greene, W. (2005). "Reconsidering Heterogeneity in Panel Data Estimators of the Stochastic Frontier Model." Journal of Econometrics 126: 269-303. https://doi.org/10.1016/j.jeconom.2004.05.003.
Greening, L.A., D.L. Greene, and C. Difiglio (2000). "Energy Efficiency and Consumption-the Rebound Effect-a Survey." Energy Policy 28: 389-401. https://doi.org/10.1016/S0301-4215(00)00021-5.
He, D. and F. Kong (2011). "China's Experience in Public Policy Implementation." Social Sciences in China (5): 61-79.
Helm, D., (2002). "Energy Policy: Security of Supply, Sustainability and Competition." Energy Policy 30: 173-184. https://doi.org/10.1016/S0301-4215(01)00141-0.
Herrala, R. and R.K. Goel (2012). "Global C[O.sub.2] Efficiency: Country-wise Estimates Using a Stochastic Cost Frontier." Energy Policy 45: 762-770. https://doi.Org/I0.1016/j.enpol.2012.03.007.
Honma, S. and J.L. Hu (2014). "Industry-level Total-factor Energy Efficiency in Developed Countries: A Japan-centered Analysis." Applied Energy 119: 67-78. https://doi.org/10.1016/j.apenergy.2013.12.049.
Hu, H.L. and S.C. Wang (2006). "Total-factor Energy Efficiency of Regions in China." Energy Policy 34: 3206-3217. https://doi.org/10.1016/j.enpol.2005.06.015.
IEA (2009). C[O.sub.2] Emissions from Fuel Combustion-highlights (2009 Edition) [EB/OL]. (2009-12-18) [2012-04-08]. http://www.iea.org/co2highlights/co2highlights.pdf.
Javorcik, B. (2004). "Does Foreign Direct Investment Increase the Productivity of Domestic Firms'? In Search of Spillovers through Backward Linkages." American Economic Review 94(3): 605-627. https://doi.org/10.1257/0002828041464605.
Kumar, S., H. Fujii, and S. Managi (2015). "Substitute or Complement? Assessing Renewable and Non-renewable Energy in OECD Countries." Applied Economics 47(14): 1438-1459. https://doi.org/10.1080/00036846.20I4.997922.
Kumbhakar, S.C. (2000). "Estimation and Decomposition of Productivity Change When Production Is Not Efficient: A Panel Data Approach." Econometric Review 19(4): 425-460. https://doi.org/10.1080/07474930008800481.
Li, S. and G. Chen (2013). "Environmental Regulation and the Growth of Productivity in China: Evidence from the Revision of Air Pollution Prevention and Control Law in 2000." Economic Research Journal (1): 17-31.
Li, L., J.J. Wang, Z.F. Tan, X.Q. Ge, J. Zhang, and X.Z. Yun (2014). "Policies for Eliminating Low-efficiency Production Capacities and Improving Energy Efficiency of Energy-intensive Industries in China." Renewable and Sustainable Energy Reviews 39: 312-326. https://doi.org/10.1016/j.rser.2014.07.099.
Li, L. and J. Hu (2012). "Ecological Total-factor Energy Efficiency of Regions in China." Energy Policy 46: 216-224. https://doi.org/10.1016/j.enpol.2012.03.053.
Liao, H., Y. Fan, and YM. Wei (2007). "What Induced China's Energy Intensity to Fluctuate: 1997-2006?" Energy Policy 35: 4640-4649. https://doi.org/10.1016/j.enpol.2007.03.028.
Lin, B. and H. Zhao (2016). "Technology Gap and Regional Energy Efficiency in China's Textile Industry: A Non-parametric Meta-frontier Approach." Journal of Cleaner Production 137: 21-28 https://doi.org/l 0.1016/j.jclepro,2016.07.055.
Llorca, M., J. Ba-os-Pino, J. Somoza, and P. Arbues (2017). "A Stochastic Frontier Analysis Approach for Estimating Energy Demand and Efficiency in the Transport Sector of Latin America and the Caribbean." The Energy Journal 38(5): 153-174. https://doi.org/10.5547/01956574.38.5.mllo.
Ma, C. (2014). "A Multi-fuel, Multi-sector and Multi-region Approach to Index Decomposition: An Application to China's Energy Consumption 1995-2010." Energy Economics 42: 9-16. https://doi.org/10.1016/j.eneco.2013.11.009.
Makridou, G., K. Andriosopoulos, M. Doumpos, and C. Zopounidis (2015). "A Two-stage Approach for Energy Efficiency Analysis in European Union Countries." The Energy Journal 36(2): 47-69. https://doi.org/10.5547/019565184.108.40.206.
Meeusen, W. and J. Broeck (1977). "Efficiency Estimation from Cobb-Douglas Production Functions with Composed Error." International Economic Review 18(2): 435-444. https://doi.org/10.2307/2525757.
Moser, P. and A. Voena (2012). "Compulsory Licensing: Evidence from the Trading with the Enemy Act." American Economic Review 102(1): 396-427. https://doi.org/10.1257/aer.102.1.396.
Murty, M.N. and S. Kumar (2002). "Measuring the Cost of Environmental Sustainable Industrial Development in India: A Distance Function Approach." Environmental and Development Economics 7(3): 467^186. https://doi.org/10.1017/S1355770X02000281.
Murty, M.N., S. Kumar, and K. Dhavala (2006). "Measuring Environmental Efficiency of Industry: A Case Study of Thermal Power Generation in India." Environmental and Resource Economics 38( 1): 31-50. https://doi.org/10.1007/s10640-006-9055-6.
Nunn, N. and N. Qian (2011). "The Impact of Potatoes on Old World Population and Urbanization." Quarterly Journal of Economics 126(2): 593-650. https://doi.org/10.1093/qje/ajr009.
Ohene-Asare, K., C. Turkson, and A. Afful-Dadzie (2017). "Multinational Operation, Ownership and Efficiency Differences in the International Oil Industry." Energy Economics 68: 303-312. https://doi.org/10.1016/j.eneco.2017.10.006.
Patterson, M.G. (1996). "What Is Energy Efficiency? Concepts, Indicators and Methodological Issues." Energy Policy 24: 377-390. https://doi.org/l0.1016/0301-4215(96)00017-1.
Price, L., M.D. Levine, N. Zhou, D. Fridley, N. Aden, HY. Lu, M. McNeil, N.N. Zheng, Y.N. Qin, and P. Yowargana (2011). "Assessment of China's Energy-saving and Emission-reduction Accomplishments and Opportunities during the 11th Five Year Plan." Energy Policy 39: 2165-2178. https://doi.org/10.1016/j.enpol.2011.02.006.
Shahbaz, M., S. Nasreen, F. Abbas, and O. Anis (2015). "Does Foreign Direct Investment Impede Environmental Quality in High, Middle and Low-income Countries?" Energy Economics 51: 275-287. https://doi.org/10.1016/j.eneco.2015.06.014.
Shao, S., R. Luan, Z. Yang, and C. Li (2016). "Does Directed Technological Change Get Greener: Empirical Evidence from Shanghai's Industrial Green Development Transformation." Ecological Indicators 69: 758-770. https://doi.org/10.1016/j.ecolind.2016.04.050.
Shao, S., L. Yang, M.B. Yu, and M.L. Yu (2011). "Estimation, Characteristics, and Determinants of Energy-related Industrial CO2 Emissions in Shanghai (China), 1994-2009." Energy Policy 39: 6476-6494. https://doi.org/10.1016/j.en-pol.2011.07.049.
Shao, S., Z. Tian, and L. Yang (2017). "High Speed Rail and Urban Service Industry Agglomeration: Evidence from China's Yangtze River Delta Region." Journal of Transport Geography 64: 174-183. https://doi.org/I0.l016/jjtrangeo.2017.08.019.
Song, M., S. Wang, and J. Sun (2018). "Environmental Regulations, Staff Quality, Green Technology, R&D Efficiency, and Profit in Manufacturing." Technological Forecasting & Social Change 133: 1-14. https://doi.org/10.1016/j.techfore.2018.04.020.
Tang, L. and Z. Yang (2009). "Energy Efficiency and Industrial Economic Transition." Journal of Quantitative & Technical Economics (10): 34-48.
The Intergovernmental Panel on Climate Change (IPCC) (2006). 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Hayama, Kanagawa, Japan: Institute for Global Environmental Strategies (IGES).
Vaninsky, A. (2010). "Prospective National and Regional Environmental Performance: Boundary Estimations Using a Combined Data Envelopment-stochastic Frontier Analysis Approach." Energy 35: 3657-3665. https://doi.org/10.1016/j.energy.2010.05.010.
Wang, H.J. (2002). "Heteroscedasticity and Non-monotonic Efficiency Effects of a Stochastic Frontier Model." Journal of Productivity Analysis 18: 241-253. https://doi.org/10.1023/A:1020638827640.
Wang, H., B.W. Ang, Q.W. Wang, and P. Zhou (2017). "Measuring Energy Performance with Sectoral Heterogeneity: A Non-parametric Frontier Approach." Energy Economics 62: 70-78. https://doi.org/l 0.1016/j.eneco.2016.12.005.
Wang, H.J. and C.W. Ho (2010). "Estimating Fixed-effect panel Stochastic Frontier Models by Model Transformation." Journal of Econometrics 157(2): 289-296. https://doi.org/10.1016/j.jeconom.2009.12.006.
Wang, H.J. and P. Schmidt (2002). "One-step and Two-step Estimation of the Effects of Exogenous Variables on Technical Efficiency Levels." Journal of Productivity Analysis 18:129-144. https://d0i.0rg/l 0.1023/A: 1016565719882.
Wang, Z.H., H.L. Zeng, Y.M. Wei, and Y.X. Zhang (2012). "Regional Total Factor Energy Efficiency: An Empirical Analysis of Industrial Sector in China." Applied Energy 97: 115-123. https://doi.org/10.1016/j.apenergy.2011.12.071.
Wang, Q. W., P. Zhou, N. Shen, and S.S. Wang (2013). "Measuring Carbon Dioxide Emission Performance in Chinese Provinces: A Parametric Approach." Renewable and Sustainable Energy Reviews 21: 324-330. https://doi.org/10.1016/j.rser.2012.12.061.
Wei, C. and M. Shen (2007). "Energy Efficiency and Its Influencing Factors: An Empirical Analysis Based on DEA." Management World (8): 66-76.
Yang, M., F. Yang, and X. Chen (2011). "Effects of Substituting Energy with Capital on China's Aggregated Energy and Environmental Efficiency." Energy Policy 39: 6065-6072. https://doi.org/10.1016/j.enpol.2011.07.006.
Yang, Z., S. Shao, L. Yang, and J. Liu (2017a). "Differentiated Effects of Diversified Technological Sources on Energy-saving Technological Progress: Empirical Evidence from China's Industrial Sectors." Renewable and Sustainable Energy Reviews 72: 1379-1388. https://doi.org/10.1016/j.rser.2016.11.072.
Yang, Z., S. Shao, L. Yang, and Z. Miao (2018). "Improvement Pathway of Energy Consumption Structure in China's Industrial Sector: from the Perspective of Directed Technical Change." Energy Economics 72: 166-176. https://doi.org/10.1016/). eneco.2018.04.003.
Yang, Z., M. Fan, S. Shao, and L. Yang (2017b). "Does Carbon Intensity Constraint Policy Improve Industrial Green Production Performance in China? A Quasi-DID Analysis." Energy Economics 68: 271-282. https://doi.org/10.10l6/j.eneco.2017.10.009.
Yuan, B.L. and K. Zhang (2017). "Can Environmental Regulation Promote Industrial Innovation and Productivity? Based on the Strong and Weak Porter Hypothesis." Chinese Journal of Population, Resources and Environment 15(4): 322-336. https://doi.org/10.1080/10042857.2017.1416042.
Zha, D.L. and A.S. Kavuri (2016). "Effects of Technical and Allocative Inefficiencies on Energy and Nonenergy Elasticities: An Analysis of Energy-intensive Industries in China." Chinese Journal of Population, Resources and Environment 14(4): 292-297. https://doi.org/10.1080/10042857.2016.1258805.
Zhang, X.P., X.M. Cheng, J.H. Yuan, and X.J. Gao (2011). "Total-factor Energy Efficiency in Developing Countries." Energy Policy 39: 644-650. https://doi.org/10.1016/j.enpol.2010.10.037.
Zhang, X., Y. Liang, E. Yu, R. Rao, and J. Xie (2017). "Review of Electric Vehicle Policies in China: Content Summary and Effect Analysis." Renewable and Sustainable Energy Reviews 70: 698-714. https://doi.org/10.1016/j.rser.2016.11.250.
Zhang, Z. and J. Ye (2015). "Decomposition of Environmental Total Factor Productivity Growth Using Hyperbolic Distance Functions: A Panel Data Analysis for China." Energy Economics Al: 87-97. https://doi.org/10.1016/j.eneco.2014.10.023.
Zhao, X.L., R. Yang, and M. Qian (2014). "China's Total Factor Energy Efficiency of Provincial Industrial Sectors." Energy 65(2): 52-61.
Zhou, P. and B.W. Ang (2008). "Linear Programming Models for Measuring Economy-wide Energy Efficiency Performance." Energy Policy 36: 2911-2916. https://doi.org/10.1016/j.enpol.2008.03.041.
Zhou, P., B. W. Ang, and D. Q. Zhou (2012). "Measuring Economy-wide Energy Efficiency Performance: A Parametric Frontier Approach." Applied Energy 90: 196-200. https://doi.org/10.1016/j.apenergy.2011.02.025.
Zhou, P., B.W. Ang, and K.L. Poh (2008). "A Survey of Data Envelope Analysis in Energy and Environmental Studies." European Journal of Operational Research 189: 1-18. https://doi.org/10.1016/j.ejor.2007.04.042.
Zou, G.F., L. M. Chen, W. Liu, X.X. Hong, G.J. Zhang, and Z.Y. Zhang (2013). "Measurement and Evaluation of Chinese Regional Energy Efficiency Based on Provincial Panel Data." Mathematical and Computer Modelling 58(5): 1000-1009. https://doi.org/10.1016/j.mcm.2012.09.020.
Table A1: Classification and Code of Industrial Sub-sectors Code Sub-sector EICP CICP S1 Mining and Washing of Coal 1 1 S2 Extraction of Petroleum and Natural Gas 1 1 S3 Mining and Processing of Ferrous Metal Ores 1 1 S4 Mining and Processing of Non-Ferrous Metal Ores 1 1 S5 Mining and Processing of Non-metal Ores 1 1 S6 Processing of Food from Agricultural Products 0 0 S7 Manufacture of Foods 0 1 S8 Manufacture of Liquor, Beverages and Refined Tea 0 1 S9 Manufacture of Tobacco 0 0 S10 Manufacture of Textile 1 0 Sll Manufacture of Textile, Wearing Apparel and 0 0 Accessories S12 Manufacture of Leather, Fur, Feather and Related 0 0 Products and Footwear S13 Processing of Timber, Manufacture of Wood, Bamboo, 0 1 Rattan, Palm and Straw Products S14 Manufacture of Furniture 0 0 S15 Manufacture of Paper and Paper Products 1 1 S16 Printing and Reproduction of Recording Media 0 0 S17 Manufacture of Articles for Culture, Education, 0 0 Arts and Crafts, Sport and Entertainment Activities S18 Processing of Petroleum, Coking and Processing of Nuclear Fuel S19 Manufacture of Raw Chemical Materials and Chemical 1 1 Products S20 Manufacture of Medicines 0 0 S21 Manufacture of Chemical Fibers 1 1 S22 Manufacture of Rubber 1 1 S23 Manufacture of Plastic 0 0 S24 Manufacture of Non-metallic Mineral Products 1 1 S25 Smelting and Pressing of Ferrous Metals 1 1 S26 Smelting and Pressing of Non-ferrous Metals 1 1 S27 Manufacture of Metal Products 0 0 S28 Manufacture of General Purpose Machinery 0 0 S29 Manufacture of Special Purpose Machinery 0 0 S30 Manufacture of Transport Equipment 0 0 S31 Manufacture of Electrical Machinery and Apparatus 0 0 S32 Manufacture of Computers, Communication and 0 0 Other Electronic Equipment S33 Manufacture of Measuring Instruments and Machinery 0 0 S34 Production and Supply of Electric Power and 1 1 Heat Power S35 Production and Supply of Gas 1 1 S36 Production and Supply of Water 1 0 Notes: "1" and "0" denote the treatment group and the control group, respectively. Table A2: TFEEG Rate of Industrial Sub-sectors (unit: %) Sub-sector 2001 2002 2003 2004 2005 2006 2007 2008 S1 27.9 22.1 24.3 11.8 14.5 13.4 14.5 10.6 S2 24.1 20.0 11.8 4.1 9.9 6.5 8.2 3.9 S3 21.9 22.1 20.8 17.4 15.5 14.1 14.4 12.1 S4 20.9 18.5 20.3 15.5 12.4 10.2 7.3 9.5 S5 20.9 19.2 18.5 16.3 14.7 14.5 13.0 11.7 S6 20.4 17.4 13.8 18.2 13.2 11.9 11.7 8.3 S7 18.6 15.3 14.6 16.3 13.6 13.1 11.5 10.5 S8 25.4 23.6 20.8 26.4 17.6 14.1 11.9 10.6 S9 24.1 22.9 22.6 12.1 23.3 16.3 16.9 15.5 S10 22.7 21.5 18.7 16.9 13.6 13.6 12.4 10.3 S11 7.8 7.8 8.2 4.6 7.1 7.6 9.5 10.1 S12 7.2 5.6 4.1 0.9 3.1 0.0 7.1 7.7 S13 16.9 16.4 10.7 11.7 11.5 11.7 11.1 10.2 S14 7.1 8.9 2.1 5.9 2.2 0.9 8.7 7.3 S15 22.4 20.9 18.0 18.9 15.2 13.1 11.3 10.9 S16 19.7 17.9 16.8 13.6 13.5 14.3 13.1 10.8 S17 4.6 1.9 10.9 -1.5 5.5 3.7 8.3 7.9 S18 15.3 14.9 15.0 17.6 6.9 7.8 7.2 1.7 S19 26.5 27.4 25.9 25.7 18.8 16.6 14.6 9.2 S20 20.3 20.9 20.4 18.9 18.4 16.1 14.1 12.6 S21 19.9 20.1 13.0 14.5 14.5 12.5 11.6 8.5 S22 18.4 14.5 14.1 11.3 10.4 9.9 9.0 8.1 S23 14.2 12.1 12.6 10.0 9.5 9.5 8.7 8.2 S24 24.1 21.1 24.1 23.3 17.3 14.2 14.2 12.6 S25 19.8 25.1 26.0 20.7 24.5 16.3 15.5 8.8 S26 17.5 20.1 20.1 16.1 14.5 15.6 13.0 7.8 S27 14.2 12.5 13.5 9.5 9.8 10.8 10.6 8.9 S28 26.8 24.9 18.7 15.6 16.4 17.0 15.3 13.2 S29 29.8 25.0 16.3 19.4 19.5 17.9 16.0 12.3 S30 30.1 26.7 21.1 20.3 17.3 19.0 16.7 13.0 S31 21.7 17.4 13.5 8.6 9.4 11.2 10.6 8.8 S32 25.5 20.3 15.0 2.5 6.2 11.9 8.0 7.0 S33 17.3 14.3 6.8 12.6 9.4 8.1 9.1 9.5 S34 49.4 48.1 40.4 19.6 23.7 25.7 22.8 19.5 S35 26.9 21.0 23.3 20.3 18.1 18.9 20.0 13.1 S36 20.3 15.1 16.0 19.1 13.7 12.6 11.7 8.2 Sub-sector 2009 2010 2011 2012 2013 2014 S1 15.7 8.7 11.1 8.4 7.7 4.6 S2 2.6 2.6 -1.5 -1.6 -1.0 -5.7 S3 12.3 16.4 12.6 11.2 11.6 10.1 S4 8.7 7.8 8.7 4.9 4.0 2.8 S5 11.8 10.7 9.8 9.0 S.4 7.9 S6 8.2 6.0 8.3 5.9 5.7 5.3 S7 10.0 7.8 9.2 6.6 6.0 5.2 S8 9.4 5.4 8.8 7.0 6.1 4.6 S9 13.2 12.8 19.4 6.1 8.7 3.4 S10 10.6 8.2 9.2 8.0 5.6 4.0 S11 12.4 10.0 10.8 7.4 8.3 8.3 S12 9.6 6.8 7.6 -6.3 5.9 5.5 S13 13.0 10.9 11.2 8.9 8.6 7.9 S14 11.9 6.2 9.3 7.9 5.9 -0.3 S15 10.1 7.4 9.2 6.3 6.2 5.2 S16 12.5 10.8 11.9 8.6 5.8 6.8 S17 9.6 7.9 3.7 4.9 7.1 7.2 S18 3.8 -2.3 2.9 -1.8 -2.3 3.1 S19 9.7 10.2 11.8 7.7 6.4 6.9 S20 10.2 10.4 10.6 7.8 7.3 6.2 S21 9.2 7.4 8.0 5.9 5.1 3.5 S22 8.2 7.2 7.2 7.7 5.9 4.7 S23 9.2 7.4 8.4 6.3 4.8 4.6 S24 12.1 12.3 15.2 9.0 7.6 7.7 S25 11.2 6.4 9.7 5.3 5.0 3.0 S26 7.3 8.5 8.6 6.6 5.7 5.1 S27 11.6 10.5 10.2 7.7 8.3 6.9 S28 15.2 13.0 16.6 10.0 10.0 9.3 S29 15.8 14.5 14.4 13.5 9.3 11.3 S30 15.3 14.0 14.0 11.2 11.1 4.0 S31 12.3 9.4 12.1 10.1 9.6 8.3 S32 11.6 9.1 6.4 2.5 9.6 5.1 S33 11.0 6.5 10.3 10.1 7.1 7.2 S34 17.4 18.6 21.7 6.7 14.3 6.3 S35 14.3 16.1 14.4 16.1 16.9 9.6 S36 6.9 7.4 6.1 2.5 -0.03 1.2
Shuai Shao, (a) Zhenbing Yamg, (b) Lili Yang, (c) and Shuang Ma (d)
(a) School of Urban and Regional Science, Institute of Finance and Economics Research, Shanghai University of Finance and Economics, Shanghai 200433, China. E-mail: email@example.com.
(b) Corresponding author. School of Economics, Nanjing University of Finance and Economics, Nanjing 210023, China. E-mail: firstname.lastname@example.org.
(c) School of International Economics and Trade, Shanghai Lixin University of Accounting and Finance, Shanghai 201209, China. E-mail: email@example.com.
(d) School of Economics and Statistics, Guangzhou University, Guangzhou 510006, China. E-mail: firstname.lastname@example.org.
(1.) See http://www.nea.gov.cn/2014-02/12/c_l33l09179.htm.
(2.) Although no country/region in the world adopts the total factor energy efficiency (TFEE) indicators to track national/regional energy efficiency in practical policy-making process (Ang, 2006; Ang et al., 2010; Ang and Goh, 2018), this situation has recently changed in China. In 2017, the 19th National Congress of Communist Party of China proposed new development goals of "promoting the quality, efficiency and impetus transformations of economic development to improve the total factor productivity". This was the first time the Chinese government introduced the term "total factor productivity" (TFP) into the country's official documents, indicating the Chinese government's ambition to improve the country's TFP. Considering energy input in addition to capital and labor inputs, the TFEE can better reflect the TFP under energy constraints. The introduction of the term "TFP" in China's official nuclear document means that the TFP and even the TFEE will receive more attention in the policy-making process of the Chinese government in the future. Under such a background, this study is of practical significance as the first study to investigate the effect of the EICP on China's industrial TFEEG.
(3.) The energy intensities of the 17th, 18th and 19th sub-sectors in rank are 0.370 tce per [10.sup.4] RMB, 0.323 tee per [10.sup.4] RMB and 0.316 tee per [10.sup.4] RMB, respectively. Since the energy intensity of the 18th sub-sector in rank is very close to that of the 19th sub-sector, we also classify the 18th sub-sector into the control group. In Section 5, we will re-group sub-sectors to conduct a robustness test.
(4.) Ma (2014) argued that there was regional heterogeneity in energy use in China. Due to the differences in production process and technology as well as energy usage mode between different industrial sub-sectors, the heterogeneity between different industrial sub-sectors is obvious, which has been discussed in some existing studies (e.g., Wang et al., 2017; Yang et al., 2018). Hence, following Yang et al. (2018), we consider the sector fixed effect in the SFA model to capture this heterogeneity and thus reinforce the accuracy and credibility of our research results. Such a fixed-effect SFA model can capture the individual effect reflecting the heterogeneity to avoid biased estimation results (Greene, 2005; Wang and Ho, 2010).
(5.) The data of industrial added value were no longer issued in the China Industrial Economics Statistical Yearbook after 2008. Thus, based on the availability and consistency of data, we have to select the gross industrial output as the proxy variable of industrial output, which contains the intermediate input. The industrial energy consumption we use can be regarded as a kind of intermediate inputs. Although the data of non-energy intermediate inputs may be estimated based on the input-output table, China's input-output table is released once every five years and thus cannot meet the need for the annual panel data used in this study. Therefore, similar to most existing studies (e.g., Chen and Golley, 2014; Zhang and Ye, 2015; Fei and Lin, 2017; Yang et al., 2017b), the non-energy intermediate inputs are not considered in this study.
(6.) Note that the number of some variables (Y, K, L and E) is 540, because their time interval is from 2000 to 2014, but the number of other variables is 504, because their time interval is from 2001 to 2014. The reason is that, based on the input-output variables (Y, K, L and E) covering the years from 2000 to 2014, the TFEE growth rate we get just covers the years from 2001 to 2014.
(7.) There are some differences between these two policies. First, their introduction time points are different. The EICP was introduced in 2006, while the CICP was proposed in 2009. Second, they have different objectives. The EICP aims to reduce energy intensity, while the CICP plans to mitigate carbon intensity. Third, the industrial sub-sectors mainly affected by these two policies are different. As shown in Table Al, sub-sectors with higher levels of energy intensity are not entirely consistent with sub-sectors with higher levels of carbon intensity. Some sub-sectors belong to high carbon intensity group, but their energy intensities are relatively low (e.g., S7, S8 and S13); the other sub-sectors present a higher level of energy intensity, while their carbon intensities are relatively low (e.g., S10 and S36). Obviously, such a disparity can be attributed to different energy consumption structures between these sub-sectors. Therefore, the EICP and the CICP are expected to have different implementation effects.
(8.) Based on the reference method proposed by 1PCC (2006), carbon emissions are the product of final energy consumption, corresponding carbon content, low calorific value, and the molecular weight of the carbon oxidation factor. Following some previous studies (e.g., Shao et al., 2011; Yang et al., 2017b), we adopt the priority principle to select the related parameters officially released in China, and the second choice of the defaults provided by IPCC (2006). To assure the accuracy of the estimation result, we take into account all 16 types of fossil fuels reported consecutively in the statistical yearbooks.
(9.) This is a simplified specification of the DID model (Moser and Voena, 2012; Shao et al., 2017). When we add du and dt into Eq. (12) using a standard DID model, the signs and significance levels of all coefficients remain unchanged.
(10.) With the rise in the grouping level, the number of sub-sectors in the treatment group will be reduced, and the sub-sectors affected by the EICP in the DID model will also be reduced. Thus, the range of influence of the policy investigated in the DID model will be narrowed. This is obviously inconsistent with reality. In contrast, too many sub-sectors in the treatment group may cause an estimation bias for the DID strategy due to the lack of a sufficient number of sub-sectors in the control group. Therefore, we attempt to keep the number of sub-sectors in the treatment group close to the number of sub-sectors in the control group. In a particular case, we conduct a re-grouping analysis considering that the number of sub-sectors in the treatment group is changed to 10, and we find that the empirical results are still robust.
Table 1: Descriptive Statistics of Variables Variable (unit) Sample size Mean Standard deviation Eq.(3) Y ([10.sup.8] RMB) 540 11509.0 16362.5 K ([10.sup.8] RMB) 540 4592.35 6727.42 L ([10.sup.4] persons) 540 211.764 178.916 E ([10.sup.4] tce) 540 5455.65 11154.9 Eq. (9) TFEEG (%) 504 12.0460 6.8216 rd (%) 504 0.7359 0.5574 kl ([10.sup.4] 504 30.1632 36.3032 RMB per capita) size ([10.sup.8] RMB) 504 10543.8 12936.8 fdi (%) 504 21.6776 13.5615 state (%) 504 28.8940 28.4906 Variable (unit) Minimum Maximum Eq.(3) Y ([10.sup.8] RMB) 164.860 131645 K ([10.sup.8] RMB) 87.3957 58580.3 L ([10.sup.4] persons) 14.5400 977.505 E ([10.sup.4] tce) 103.860 80336.1 Eq. (9) TFEEG (%) -6.2672 49.3893 rd (%) 0.0081 2.5581 kl ([10.sup.4] 1.2149 226.989 RMB per capita) size ([10.sup.8] RMB) 339.811 91197.3 fdi (%) 0 74.3682 state (%) 0.3010 99.4886 Table 2: Estimation Results of Eq. (3) variable Coefficient t-value [[alpha].sub.0] 4.0614 (***) (0.9339) 4.35 [[alpha].sub.1] -0.0116 (0.0442) -0.26 [[alpha].sub.2] -0.0122 (***) (0.0024) 5.03 [[alpha].sub.3] 0.6732 (***) (0.2292) 2.94 [[alpha].sub.4] 1.1839 (***) (0.1989) 5.95 [[alpha].sub.5] -0.6789 (***) (0.1687) -4.02 [[alpha].sub.6] 0.0679 (***) (0.0070) 9.66 [[alpha].sub.7] -0.0156 (**)(0.0067) -2.32 [[sigma].sub.v.sup.2] 0.0046 Sector fixed effect Yes Log likelihood function variable Coefficient t-value [[alpha].sub.8] -0.0205 (***) (0.0053) -3.84 [[alpha].sub.9] -0.2205 (***) (0.0605) -3.65 [[alpha].sub.10] 0.4023 (***) (0.0669) 6.02 [[alpha].sub.11] -0.2930 (***) (0.0631) -4.64 [[alpha].sub.12] 0.2693 (***) (0.0503) -5.35 [[alpha].sub.13] 0.2855 (***) (0.0467) 6.12 [[alpha].sub.14] 0.0195 (***)(0.0306) 0.64 [mu] 0.1971 (2.5684) 0.08 [[sigma].sub.u.sup.2] 8.7467 Year fixed effect Yes Log likelihood function 5575.0486 Notes: Standard errors for coefficients are in parentheses; (***), (**), and (*) denote statistical significance at the levels of 1%, 5%, and 10%, respectively. Table 3: Estimation Results of Eq. (9) variable OLS PCSE Model 1 Model 2 Model 3 du x dt -0.0559 (***) -0.0431 (***) -0.0559 (***) (0.0098) (0.0054) (0.0067) du 0.0470 (***) -0.0060 0.0470 (***) (0.0076) (0.0268) (0.0052) dt -0.0506 (***) -0.2889 (***) -0.0506 (***) (0.0068) (0.0187) (0.0124) rd 0.0331 (***) (0.0075) kl 0.0931 (***) (0.0091) size 0.0326 (***) (0.0075) fdi 0.0088 (0.0420) state -0.0610 (*) (0.0329) Constant 0.1397 (***) -0.2433 0.1397 (***) (0.0052) (0.0329) (0.0096) Sector fixed effect No Yes No Year fixed effect No Yes No [R.sup.2] 0.3447 0.8488 0.3484 Woodridge test 16.654 (0.0002) 8.276 (0.0068) N.A. White test 23.14(0.0000) 177.34 (0.0000) N.A. N 504 504 504 variable Model 4 du x dt -0.0431 (***) (0.0072) du -0.0060 (0.0333) dt -0.2889 (***) (0.0207) rd 0.0331 (***) (0.0092) kl 0.0931 (***) (0.0116) size 0.0326 (***) (0.0075) fdi 0.0088 (0.0522) state -0.0610 (0.0501) Constant 0.0536 (0.1360) Sector fixed effect Yes Year fixed effect Yes [R.sup.2] 0.8488 Woodridge test N.A. White test N.A. N 504 Notes: Standard errors for coefficients are in parentheses; (***), (**), and (*) denote statistical significance at the levels of 1%, 5%, and 10%, respectively. Table 4: Estimation Results of Marginal Effects Variable Coefficient Standard error Probability du x dt -0.0648 (***) 0.0078 0.0000 du x dt x [year.sup.2007] 0.0462 (***) 0.0096 0.0000 du x dt x [year.sup.2008] 0.0362 (***) 0.0096 0.0000 du x dt x [year.sup.2009] 0.0132 0.0094 0.1570 du x dt x [year.sup.2010] 0.0217 (**) 0.0093 0.0200 du x dt x [year.sup.2011] 0.0170 (*) 0.0093 0.0680 du x dt x [year.sup.2012] 0.0292 (***) 0.0092 0.0020 du x dt x [year.sup.2013] 0.0161 (*) 0.0092 0.0790 du x dt x [year.sup.2014] 0.0083 0.0091 0.3600 du 0.0055 0.0326 0.8650 dt -0.2812 (***) 0.0209 0.0000 rd 0.0322 (***) 0.0093 0.0010 kl 0.0972 (***) 0.0117 0.0000 size 0.0310 (***) 0.0063 0.0000 fdi 0.0301 0.0512 0.5560 state -0.0707 0.0504 0.1600 Sector fixed effect Yes [R.sup.2] 0.8543 Year fixed effect Yes Variable Marginal effect du x dt du x dt x [year.sup.2007] -0.0186 (***) du x dt x [year.sup.2008] -0.0286 (***) du x dt x [year.sup.2009] -0.0516 du x dt x [year.sup.2010] -0.0431 (***) du x dt x [year.sup.2011] -0.0477 (*) du x dt x [year.sup.2012] -0.0356 (***) du x dt x [year.sup.2013] -0.0487 (*) du x dt x [year.sup.2014] -0.0564 du dt rd kl size fdi state Sector fixed effect Year fixed effect Notes: (***), (**), and (*) denote statistical significance at the levels of 1%, 5%, and 10%, respectively. Table 5: Estimation Results of Eq. (11) Variable Coefficient t-value Variable dm x du x dz -0.0150 (**)(0.0070) -2.14 rd dm 0.0275 (*) (0.0166) 1.66 kl dm x du 0.0810 (***) (0.0298) 2.72 size dm x dt -0.0021(0.0071) -0.29 fdi du -0.1162 (***) (0.0411) -2.83 state dt -0.2942 (***) (0.0214) -13.77 Constant du x dt -0.0316 (***) (0.0086) -3.67 [R.sup.2] Sector fixed effect Yes Year fixed effect variable Coefficient t-value dm x du x dz 0.0323 (***) (0.0091) 3.57 dm 0.0938 (***) (0.0120) 7.79 dm x du 0.0354 (***) (0.0063) 5.58 dm x dt 0.0026 (0.0554) 0.05 du -0.0693 (0.0497) -1.4 dt -0.2586 (***) (0.0594) -4.35 du x dt 0.8504 Sector fixed effect Yes Notes: Standard errors for coefficients are in parentheses; (***), (**), and (*) denote statistical significance at the levels of 1%, 5%, and 10%, respectively. Table 6: Estimation Results of Robustness Tests Variable Test 1 Test 2 Test 3 du x [preyear.sup.2004] 0.0315 (0.0210) du x [preyear.sup.2005] 0.0175 0.0147 (0.0208) (0.0219) du x dt -0.0440 (***) (0.0073) du -0.0184 (0.0328) dt -0.2951 (***) (0.0204) rd 0.0380 (***) 0.0366 (***) 0.0320 (***) (0.0097) (***) (0.0093) (0.0096) kl 0.1001 (***) 0.1024 (***) 0.0900 (***) (0.0115) (0.0113) (0.0115) size 0.0313 (***) 0.0305 0.0384 (***) (0.0068) (0.0068) (0.0064) fdi -0.0525 -0.0646 0.0182 (0.0528) (0.0520) (0.0547) state -0.0363 -0.0319 -0.0455 (0.0489) (0.0510) (0.0519) Constant -0.2246 (***) -0.2187 (***) -0.2831 (***) (0.0642) (0.0654) (0.0625) Sector fixed effect Yes Yes Yes Year fixed effect Yes Yes Yes [R.sup.2] 0.8326 0.8295 0.8483 variable Test 4 du x [preyear.sup.2004] du x [preyear.sup.2005] du x dt -0.0387 (***) (0.0078) du 0.1498 (***) (0.0351) dt -0.3167 (***) (0.0199) rd 0.0350 (***) (0.0095) kl 0.0975 (***) (0.0116) size 0.0425 (***) (0.0067) fdi 0.0058 (0.0561) state -0.0630 (0.0490) Constant -0.4937 (***) (0.0988) Sector fixed effect Yes Year fixed effect Yes [R.sup.2] 0.8427 Notes: Standard errors for coefficients are in parentheses; (***), (**), and (*) denote statistical significance at the levels of 1%, 5%, and 10%, respectively. Table 7: Quasi-DID Estimation Results Variable Coefficient t-value [EI.sub.it]xdt -0.0480 (***) (0.0096) -5.01 EI 0.0371 (***) (0.0045) dt -0.0402 (***) (0.0137) -2.94 rd kl size fdi state Constant 0.1297 (***) (0.0106) 12.2 Sector fixed effect Yes Year fixed effect Yes [R.sup.2] 0.3818 Variable Coefficient t-value [EI.sub.it]xdt -0.0338 (***) (0.0091) -3.7 EI 0.0282 (***) (0.0062) 4.56 dt -0.3094 (***) (0.0210) -14.72 rd 0.0278 (***) (0.0089) 3.14 kl 0.1000 (***) (0.0116) 8.65 size 0.0416 (***) (0.0063) 6.64 fdi -0.0133 (0.0482) -0.28 state -0.1069 (**) (0.0481) -2.23 Constant -0.3558 (***) (0.0672) -5.3 Sector fixed effect Yes Year fixed effect Yes [R.sup.2] 0.8555 Notes: Standard errors for coefficients are in parentheses; (***), (**), and (*) denote statistical significance at the levels of 1%, 5%, and 10%, respectively.
|Printer friendly Cite/link Email Feedback|
|Author:||Shao, Shuai; Yamg, Zhenbing; Yang, Lili; Ma, Shuang|
|Publication:||The Energy Journal|
|Date:||Jul 1, 2019|
|Previous Article:||Co-firing Coal with Biomass under Mandatory Obligation for Renewable Electricity: Implication for the Electricity Mix.|
|Next Article:||Merchant Storage Investment in a Restructured Electricity Industry.|