Energy Economics 30 (2008) Contents lists available at ScienceDirect. Energy Economics. journal homepage:

Energy Economics 30 (2008) 3077 3094 Contents lists available at ScienceDirect Energy Economics journal homepage: www.elsevier.com/locate/eneco Energy consumption and economic growth: Evidence from China at both aggregated and disaggregated levels Jia-Hai Yuan a,, Jian-Gang Kang b, Chang-Hong Zhao a, Zhao-Guang Hu c a School of Business Administration, North China Electric Power University, China b School of Foreign Languages, North China Electric Power University, China c State Power Economic Research Institute, China article info abstract Article history: Received 11 November 2007 Received in revised form 15 March 2008 Accepted 15 March 2008 Available online 27 March 2008 JEL classification: Q43 C32 Keywords: Cointegration Vector error-correction Energy consumption Economic growth China Using a neo-classical aggregate production model where capital, labor and energy are treated as separate inputs, this paper tests for the existence and direction of causality between output growth and energy use in China at both aggregated total energy and disaggregated levels as coal, oil and electricity consumption. Using the Johansen cointegration technique, the empirical findings indicate that there exists long-run cointegration among output, labor, capital and energy use in China at both aggregated and all three disaggregated levels. Then using a VEC specification, the short-run dynamics of the interested variables are tested, indicating that there exists Granger causality running from electricity and oil consumption to GDP, but does not exist Granger causality running from coal and total energy consumption to GDP. On the other hand, short-run Granger causality exists from GDP to total energy, coal and oil consumption, but does not exist from GDP to electricity consumption. We thus propose policy suggestions to solve the energy and sustainable development dilemma in China as: enhancing energy supply security and guaranteeing energy supply, especially in the short run to provide adequate electric power supply and set up national strategic oil reserve; enhancing energy efficiency to save energy; diversifying energy sources, energetically exploiting renewable energy and drawing out corresponding policies and measures; and finally in the long run, transforming development pattern and cut reliance on resource- and energy-dependent industries. 2008 Elsevier B.V. All rights reserved. Corresponding author. Tel./fax: +86 10 80798654. E-mail address: yuanjh126@126.com (J.-H. Yuan). 0140-9883/$ see front matter 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.eneco.2008.03.007

3078 J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 1. Introduction The role of energy in economic growth has long been a controversial topic in economics literature. The traditional neo-classical growth model, treating energy inputs as intermediate whereas land, labor and capital as basic factors, takes the role of energy in production as neutrality. On the other hand, the biophysical and ecological view is that energy plays an important role in income determination, and thus the economies heavily dependent on energy use will be significantly affected by changes in energy consumption (Cleveland et al., 1984, among others). Beaudreau (1995) criticizes the traditional growth model for treating energy as a secondary factor and points out that for an engineer production is not possible without energy use. From an economist's perspective this calls for considering energy as an important input for production. Theoretical disagreement on the role of energy is matched by mixed empirical evidence. Ever since the seminal work of Kraft and Kraft (1978), a rapidly increasing body of literature has assessed the empirical evidences for both developed and developing countries employing cointegration and Granger causality model (see Lee, 2005, 2006; Chontanawat et al., 2007 for a review of the state of the art). More recent studies mostly in a bivariate model (see Zamani, 2007; Lise and Van Montfort, 2007; Chen et al., 2007; Lee and Chang, 2007; Mehrara, 2007; Narayan, Smyth and Prasad, 2007; Narayan and Singh, 2007; Zachariadis, 2007; among others) still don't come to a consensus on the role of energy in economic development. Though bivariate model has merit that they can be employed with scarce data, recently its limitation to describe energy economy interactions has been criticized. Stern (1993, 1997), Asafu-Adjaye (2000), Glasure (2002) and Stern and Cleveland (2003) point out the importance of omitted variables and argue that multivariate model can offer multiple causality channels which, under a bivariate approach, may remain hidden or can lead to spurious correlations and erroneous conclusions. Employing a four-variable VAR model (capital, labor, energy consumption, and GDP), Stern (1993) found Granger causality that runs from energy to GDP. The results were, in general, consistent with the argument in biophysical models. Adding several other variables into the VAR model, Glasure and Lee (1997) observed a bi-directional causality between energy and GDP growth. Asafu-Adjaye (2000) then included the consumer price index as a proxy for energy prices for Asian economies, but uniform results were not found. Stern (2000) further investigated the USA case using multivariate cointegration tests, and the long-run cointegrated relationships were observed between energy use and GDP. Hondroyiannis et al. (2002) included price development as a proxy for economic efficiency in Greece, and cointegration was found within a multivariate system. Employing a four-variable model (valueadded, capital, labor and energy), Soytas and Sari (2007) also found unilateral Granger cause running from electricity consumption to output in Turkish manufacturing industry. Although multivariate analysis is commonly used in the recent literature, there does not seem to be a theoretical background in most of the studies. Hence, a variety of macro variables have been considered in empirical studies. Probably a more formal treatment is provided in Ghali and El-Sakka (2004) as well as in Soytas and Sari (2006, 2007). Ghali and El-Sakka (2004) assume a neo-classical one-sector production function with three inputs for Canada and find bilateral causality between energy use and output. Soytas and Sari (2007) also assume a neo-classical production for Turkish manufacturing industry. Their results are against the neo-classical assumption of neutrality of energy to growth. As the second largest energy consumer and CO 2 emitter in the world until 2004, relation between energy and economic growth in China has been a worldwide concern in both policy sector and scholarship society. Empirical works at both aggregated energy consumption (Soytas and Sari, 2007, 2006) and disaggregated levels have been presented (Lin, 2003; Shiu and Lam, 2004; Yuan et al., 2007 for electricity, Zhou and Chau, 2006 for oil) but with mixed results. The lack of consensus may be due to the fact that, different studies cover different time periods, use different data (and data manipulation) and more importantly, different test models. As noted by Ghali and El-Sakka (2004), whether proposition of neutrality of energy in income determination is true is best to be tested in a neo-classical aggregated production framework, taking capital, labor, and energy as separate inputs. Soytas and Sari (2007) also point out that different countries have different energy consumption patterns and various sources of energy. Hence, different sources of energy may have varying impacts on the output of an economy. In our opinion, different countries are in utterly different developing stages and developing process may also have significantly different impact on energy and economic growth relation, thus it may be unwise to expect for consensus on the role of energy in economic growth. As for a country to be studied, both aggregated and disaggregated data with long enough time periods should be examined.

J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 3079 The aim of this paper is to empirically investigate the causal interactions between energy use and output growth in China. Aggregated level of total energy consumption, as well as disaggregated level of coal, oil and electricity consumption is tested. We use a framework based on the neo-classical one-sector aggregate production technology where capital, labor, and energy are treated as separate inputs. Within this framework, we use the time series properties of the data and employ a vector error-correction (VEC) model to test for multivariate cointegration and Granger causality. Generalized impulse response technique is used in the paper for robustness analysis. The remainder of the paper is organized as follows: Section 2 describes the model and econometric methodology used in the study; Section 3 discusses the data used in the paper and unit root tests; Section 4 presents the cointegration results and Section 5 the vector error-correction model and Granger causality test results. Section 6 presents the generalized impulse response analysis to the results, followed by policy analysis in Section 7 and finally the conclusions. 2. Model and methodology 2.1. Neo-classical production model To investigate the relationship between energy use and output growth, we used the framework proposed in (Ghali and El-Sakka 2004, and also used in Soytas and Sari, 2007, among others) based on the conventional neo-classical one-sector aggregate production technology where capital, labor, and energy are treated as separate inputs. That is: Y t ¼ fðk t ; L t ; E t Þ ð1þ where Y is aggregate output or real GDP; K is the capital stock; L is the level of employment; E is total energy consumption in aggregated level or coal consumption, oil consumption and electricity consumption at disaggregated level, and the subscript t denotes the time period. Taking the differential of Eq. (1) and dividing through by Y t we have: d Y t ¼ a d K t þ b d L t þ c d E t ð2þ where a dot on the top of a variable means that the variable is now in a growth rate form. The constant parameters a, b and c are the elasticities of output with respect to capital, labor and energy, respectively. The relationship between output and capital, labor, and energy inputs described by the production function in Eq. (1) suggests that their long-run movements may be related. Furthermore, if we allow for short-run dynamics in factor-input behavior, the analysis above would also suggest that past changes in capital, labor, and energy could contain useful information for predicting the future changes of output, Ceteris paribus. These implications can be easily examined using tests for multivariate cointegration and Granger causality. 2.2. Test for cointegration and Granger causality Following Granger (1988), and Engle and Granger (1987), we estimated a VEC model for the Granger causality test for our problem at hand. The VEC representation is as follows: DY t ¼ A 1 þ Xr k¼1 a 1;k v k;t p þ Xp g 1;s DY t s þ Xp g 2;s DK t s þ Xp g 3;s DL t s þ Xp g 4;s DE t s þ g 1;t ð3þ DK t ¼ A 2 þ Xr k¼1 a 2;k v k;t p þ Xp h 1;s DY t s þ Xp h 2;s DK t s þ Xp h 3;s DL t s þ Xp h 4;s DE t s þ g 2;t ð4þ DL t ¼ A 3 þ Xr k¼1 a 3;k v k;t p þ Xp u 1;s DY t s þ Xp u 2;s DK t s þ Xp u 3;s DL t s þ Xp u 4;s DE t s þ g 3;t ð5þ DE t ¼ A 4 þ Xr k¼1 a 4;k v k;t p þ Xp q 1;s DY t s þ Xp q 2;s DK t s þ Xp q 3;s DL t s þ Xp q 4;s DE t s þ g 4;t ð6þ

3080 J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 where p is lag length and is decided according to information criterion and final prediction error. The parameters ν k,t p are the cointegrating vectors, derived from the long-run cointegrating relationships (i.e. Y t =β 1 K t +β 2 L t +β 3 ΔE t +β 4 T+ξ where T is trend and ξ is stationary residuals) during cointegration tests and are normalized according to K, L,andE respectively and their coefficients α i,k are the adjustment coefficients. The parameters μ i, i=1, 2, 3 are intercepts and the symbol Δ denotes the difference of the variable following it. The Johansen approach (Johansen 1991,1995; Johansen and Juselius, 1990) estimates the cointegrating relationships between the non-stationary variables using a maximum likelihood procedure which tests for the cointegrating rank r and estimates the parameters α and z. Using the model in Eqs. (3 6), Granger causality tests between the variables can be investigated through the following three channels: (i). The statistical significance of the lagged error-correction terms (ECTs) by applying separate t-tests on the adjustment coefficients. (ii). A joint F-test or a Wald χ 2 -test applied to the coefficients of each explanatory variable in one equation. For example, to test whether energy use Granger-causes output in Eq. (3), we test the following null hypothesis: H0 γ 4,1 =γ 4,2 = =γ 4,p =0. (iii). A joint F-test or a Wald χ 2 -test applied jointly to the terms in (i) and the terms in (ii). 3. The data and unit root tests 3.1. Data and discussions We use total employment and real GDP (1990 YUAN) to stand for Labor force and income and denote them as EM (L), and GDP (Y), respectively. The capital stock (K) data is difficult, if not impossible, to obtain. There is no readily available dataset for it, especially for China. However, National Statistics Bureau provides statistics data for annual balance of net value of fixed assets of all industrial enterprises since 1963. Recognizing that industry has always been the dominant sector in China and assets of industrial enterprises are a good indicator for capital stock, following Jin and Yu (1996) and Shan and Sun (1998) among others, we use annual average balance of net value of fixed assets of all industrial enterprises (K, denoted as AS) as a reliable proxy for growth of capital stock. Of course, we should be cautious with the explanation to empirical results, especially the cointegrating vectors because this proxy ignores assets of other sectors in the economy. For energy consumption, total energy consumption (10 4 tons standard coal equivalent (sce), denoted as TE) is to stand for aggregated energy consumption. For disaggregated level, two of the most important primary energy sources in China are coal (in 10 4 tons, denoted as CL) and oil (in 10 4 tons, denoted as OL) and the most important secondary energy product is electricity (in 10 8 kwh, denoted as EL). Fig. 1 shows the growth trend of the related variables, suggesting that long-run relationship is likely to be present in the study since all the series tend to move very closely together over time. Graphical analysis also reveals that the interested series have linear relationship, with the exception that coal consumption, total energy consumption and employment series display structural break, a hint for the necessity to perform additional unit roots test that accounts for the existence of a structural break. The variables employed in this paper are similar to those commonly used in the literature (for example, Beaudreau (1995), Soytas and Sari (2006, 2007), Ghali and El-Sakka (2004), and Narayan and Smyth (2005)). Note that all variables are in natural logarithms so that their first differences approximate their growth rates. We use L in front of each variable to indicate the natural logarithm and D in front of each variable to indicate the first difference of their natural logarithm. The data of the study are collected and sourced from (National Statistics Bureau of China, 1990 2006a,b,2000 2005). Though there are data for real GDP, total employment and energy consumption from 1953, data of net value of fixed assets is provided only from 1963, so the data sample of our study is 1963 to 2005. The break of total employment in 1990 is largely because of change in employment statistics caliber. Though the data quality of employment statistics and other variables in China has been subject to question, discussion of it is beyond the scope of our paper. Of course, the potential bias in estimation results arising from it should be regarded with caution. Concerning energy consumption, though we are not intended to give overview on energy consumption in China (interested readers are referred to Sinton and Fridley (2000); Paul Crompton and Yanrui Wu (2005) for detailed discussions), some comments on the decline of China's most important fuel coal, and the corresponding decline of total energy consumption during 1997 2000 periods will be discussed here. The decline in total energy consumption has occurred despite robust

J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 3081 Fig. 1. Trends of the variables in the study (before taking logarithms 1963=100) (source: National Statistics Bureau of China, 1990 2006a,b,2000 2005). GDP growth of 8.8% in 1997 and 7.8% in 1998, a puzzling phenomenon of sharply divergent trends in energy and economic growth. Many authors have explained this phenomenon and a number of explanations have been suggested (Sinton and Fridley, 2000; Zhang 2003). In our opinion, the decline should be judged carefully. On the one side, the quality of statistical data for coal production and consumption during the periods is somewhat problematic. In the mid-1990s, the central government began to require that large state-owned mines limit output, part of a package of reforms in the coal mining sector, aimed at turning around the financially troubled state-owned mines, which included laying off large numbers of workers from state-owned mines and closing down dangerous small-scale non-state mines. In 1998, the government directed that 25,000 small non-state mines and 40 state-owned mines be closed. Reportedly, 23,000 small mines were closed by May 1999. Statistics show that most of the fall in output has been from small mines while production at large mines has risen in fact. Some unreported production continues at some small mines that have been officially closed, however. Some consumers have been prohibited form purchasing the output from closed mines, and thus have an incentive to keep quiet about using such coal, which would cause it to disappear entirely from statistics. However the other side of the coin is that there are some sound factors behind the decline of coal consumption. First of all, the enhancement in the heat content of the coal negatively impacts the demand for coal. In the utility sector alone, the average heat content of delivered coal was 200 kcal/kg higher in 1997 than in 1996, and in 1998 the average rose by a further 100 kcal/kg (Jie and Li, 1999). Other factors aside, the increase in average heat content over the two years would have reduced coal consumption in power sector by 30 Mt alone (Sinton and Fridley, 2000). Secondly, the long-term implementation of energy saving policy drives down energy intensity and cuts down energy consumption in China. Economic energy intensity is an overall measure of how much energy is used to produce a unit of economic output in a country or a sector. Fig. 2. depicts the GDP energy intensity and electricity intensity of China from 1963 to 2005. It is readily seen that since 1977 when energy intensity came to a peak, it has dropped phenomenally over the past two decades, with an annual decline of 4.6%. Other factors aside, such decline implies that GDP growth less than 5% will not demand more energy. Considering the average annual GDP growth rate of 12% during 1990 1996, the sudden drop to obvious less than two-digit GDP growth in 1997 witnessed the slowdown even fall in energy consumption is

3082 J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 Fig. 2. GDP energy intensity and electricity intensity in China (GDP in 1990 price). comprehensible. Thirdly, change in the output structure also negatively influenced the energy demand (Fig. 3). In 1997 the secondary industry still accounted for 47.5% of total GDP while in 1998 the ratio dropped to 46.2%. It is estimated that GDP energy intensity will drop by 0.9% if the ratio of secondary industry drops by 1%. And finally, China is undergoing energy transition which is featured as the transition from a user of low efficiency solid fuels to higher efficiency gaseous and liquid fuel and electric power (EIA, 2005). The ratio of coal in total energy consumption dropped from 74.7% in 1996 to 69.09% in 1999, meanwhile the ratio of oil and gas rose from 19.8% to 24.7%, while on the other hand, the ratio of electricity in end-use energy consumption rose from 12.2% in 1996 to 15.2% in 1999, hence the substitution of oil, gas and electricity for coal reduced the demand for coal during the periods. To sum up, there are sound factors resulting in the decline in coal and total energy consumption, but the quality of energy and possible economic statistics data covered up the fact. 3.2. Unit root test Since Granger causality tests are sensitive to the stationarity of the series we first study the stationarity properties of the variables. If the series are integrated of the same order one can proceed with the cointegration tests. There are a variety of unit root tests that sometimes yield conflicting results. Therefore, in order to proceed with cointegration and VEC analyses one needs to be confident as to the order of integration of the series used. Fig. 3. Primary energy structure in China during 1963 2005 (source: National Statistics Bureau of China, 1990 2006a,2000 2005).

J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 3083 In order to have robust results, we conducted five different unit root tests, namely augmented Dickey Fuller (ADF), Elliot Rothenberg Stock Dickey Fuller GLS detrended (DF-GLS), Phillips Perron (PP), Kwiatkowski Phillips Schmidt Shin (KPSS), and Ng Perron MZα(NP). We do not discuss the details of the unit root tests here to conserve space (see Maddala and Kim (1998) for excellent treatment of ADF, DF- GLS, PP and KPSS; and Ng and Perron (2001) for NP). ADF and PP tests are often criticized due to their low power properties (Hubrich, Lutkepohl and Saikkonen, 2001), but we included them in our analysis because most of the studies in the literature still use them. It is also well known that the unit root tests are also sensitive to different lag structures. Hence, we employed two lag selection information criteria often employed in the literature, namely the Akaike Information Criterion (AIC) and Schwarz Information Criterion (SIC). All unit root tests (except the KPSS) employed in our study have a null hypothesis that the series in question has a unit root against the alternative of stationarity. The null of KPSS, on the other hand, states that the variable is stationary. In the literature, KPSS is sometimes used to verify the results of commonly used ADF and PP tests although it also suffers from the same low power problems (Maddala and Kim, 1998). The result of unit root tests is presented in Table 1. All null hypotheses except KPSS are unit root; whereas, in KPSS null is stationarity. The result reveals that all the five tests almost unanimously indicate that all the variables are non-stationary in their level data (with or without trend). However, the stationarity property is found in the first difference of the variables (with or without trend) in 5% or stricter Table 1 Unit root test results of all variable in the study ADF DF-GLS PP KPSS NP (MZ a ) Panel A: Level Intercept LEM 2.04(0; AIC,SIC) 0.32(2;AIC) 2.00 0.81 a 0.99 (1 ;SIC) LTE 1.00(1;AIC,SIC) 0.47(1;AIC,SIC) 1.32 0.81 a 0.56(1 ;SIC) LEC 0.91(0;AIC,SIC) 2.03(9;AIC) 1.14 0.83 a 1.60(2;AIC) LCL 0.75(1;AIC,SIC) 0.55(1;AIC) 0.80 0.81 a 0.83(1;SIC) LOL 2.36(1;AIC,SIC) 0.08(1;AIC,SIC) 3.85 a 0.77 a 0.22(1;AIC) LAS 0.95(3;AIC,SIC) 0.14(1;AIC,SIC) 1.33 0.81 a 2.25(1;SIC) LGDP 1.00(1;AIC,SIC) 0.21(8;AIC),1.2(1;SIC) 1.31 0.82 a 1.45(1;SIC) Int. and trend LEM 0.91(0 ;SIC,AIC) 1.06(0 ;AIC) 1.04 0.119 2.92 (0 ;SIC) LTE 2.86(1;AIC,SIC) 2.55(1;AIC,SIC) 1.94 0.17 b 12.29(1 ;AIC) LEC 2.50(9;AIC) 2.19(0;AIC,SIC) 2.78 0.16 b 6.43(0;SIC) LCL 2.93(1;AIC,SIC) 2.85(1;AIC,SIC) 2.10 0.18 b 15.80(1;SIC) LOL 1.76(2;SIC) 0.89(2;SIC) 2.91 0.15 b 2.2(2;SIC) LAS 2.06(1;AIC,SIC) 1.66(1;AIC,SIC) 1.81 0.18 b 5.84(1;SIC) LGDP 2.61(1;AIC,SIC) 2.53(1;AIC,SIC) 1.17 0.19 b 9.87(1;SIC) Panel B: First difference Intercept LEM 5.89(0 ;AIC,SIC) a 5.27(0 ;SIC) a 5.90 a 0.32 a 18.17(0,SIC) a LTE 3.60(0;AIC,SIC) a 3.64(0;AIC,SIC) a 3.04 b 0.16 a 16.19(0 ;AIC) a LEC 5.07(0;SIC) a 5.16(1,AIC) a 5.80 a 0.20 29(1;SIC) a LCL 3.90(0;AIC,SIC) a 4.01(0;AIC,SIC) 3.47 a 0.09 17.79(0;SIC) a LOL 3.40(0;AIC,SIC) b 3.63(1;AIC,SIC) a 3.39 a 0.48 b 12.30(1;SIC) b LAS 2.93(2;AIC,SIC) b 2.61(0;SIC) b 2.83 c 0.32 10.08(0;SIC) b LGDP 4.34(0;AIC,SIC) a 3.17(0;AIC,SIC) a 4.15 a 0.19 12.8(0;SIC) b Int. and trend LEM 6.22(0;AIC,SIC) a 6.26(0;AIC,SIC) a 6.26 a 0.10 a 20.57(0 ;AIC) b LTE 3.59(0;AIC,SIC) a 3.68(0;SIC) a 2.94 0.09 15.73(0;AIC) c LEC 5.01(0;SIC) a 5.10(0,SIC) a 6.19 a 0.17 b 20.27(0,SIC) b LCL 3.91(0;AIC,SIC) b 4.00(0;AIC,SIC) a 3.36 c 0.07 17.17(0;SIC) c LOL 3.74(0;AIC,SIC) b 3.82(0;AIC,SIC) a 3.70 b 0.17 b 19.63(0;AIC) b LAS 3.20(2;AIC,SIC) c 3.29(2;AIC,SIC) b 2.99 0.09 39.8(2;SIC) a LGDP 4.51(0;AIC,SIC) a 3.87(0;AIC,SIC) a 6.15 a 0.19 b 15.10(1;AIC) c Note to Table 1: In the table superscript a, b, c denote significance at 1%, 5% and 10% critical level, denotes insignificance by Schwarz Information Criterion (SIC) but significant by modified Akaike Information Criterion (AIC) at 1% critical level while denotes insignificance by SIC, but significance by modified AIC at 5% critical level. Lag lengths are in parentheses, along with lag selection criteria as AIC and SIC. ADF, PP, and DF-GLS (intercept only) critical values are sourced from MacKinnon (1991). KPSS critical values are from Kwiatkowski et al. (1992) and MZ α critical values are from Ng and Perron (2001). DF-GLS test with intercept and trend critical values are for 50 observations and may not be appropriate for our sample size of 44.

3084 J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 Table 2 Results of the ZA Unit root tests with a structural break Series Level(t(λˆ inf )) Break First difference (t(λˆ inf )) Break LEM 1.86(A) 1989 5.39(A) 1990 LCL 2.15(C) 1997 4.95(B) 2000 LTE 2.75(C) 1998 4.73(B) 1998 Note to Table 2: The critical values for the 5% levels are 4.80, 4.42, and 5.08 for Model A, Model B and Model C, respectively, from Zivot and Andrews (1992). Model A allows for a change in the level of the series; Model B allows for a change in the slope of the trend of a series, while Model C combines both the changes. The letters in parentheses indicate the appropriate model based on the results from the ADF test. Break denotes the time of the structure change. indicates significance at the 5% level. 1% critical level. Although some conflicting results on the stationarity property in the first difference are found for series EM and OL (with and without trend), TE (without trend) and GDP (with trend) between KPSS and other tests, in general it seems to indicate that all variables are integrated of order 1, i.e., I(1). As argued by Perron (1989), many structural break macroeconomic time series are in fact stationary fluctuations around a deterministic trend function if allowance is made for a possible change in intercept and slope. So in practice particular attention should be paid to the series with structural change. However, standard unit root tests are not appropriate for testing for the stationarity of a series which encounters a structural change. Thus, we further take the structure break into account when employing unit root test. Zivot and Andrews (1992) (hereafter ZA) developed a unit root test with an endogenous structural break, which is widely regarded as suitable to test for the order of integration of a series. Table 2 presents the results of ZA unit root tests for the TE, CL and EM series. ZA tests indicate that all the three series are integrated of I(1) at 5% critical level. 4. Cointegration tests Given the results of unit roots, we now use, Johansen (1991), Johansen (1995) and Johansen and Juselius (1990) techniques to test for cointegration between the variables within a VEC model as specified in Eqs. (3 6). Before applying the Johansen's procedure to estimate cointegration vector and adjustment factors, it is necessary to determine the lag length in the VAR, which should be high enough to ensure that the errors are approximately white noise, but small enough to allow estimation. Since the Johansen procedure is sensitive to the choice of the lag length, we based our decision on the AIC criterion and final prediction error to select the lag number. The lag length is further validated by test for normality and absence of serial correlation in the residuals in VAR to make sure that none of them violates the standard assumptions of the model. The results of testing for the number of cointegrating vectors for (LGDP, LAS, LEM, LTE) are reported in Table 3, which presents the maximum eigenvalue (λ max ), the trace statistics and the 5% critical value, as Table 3 Cointegration test results for (LGDP, LAS, LEM, LTE) Trace λ max H 0 H 1 Statistics 5% H 0 H 1 Statistics 5% r=0 r 1 72.65 47.85 r=0 r=1 31.31 27.58 r 1 r 2 41.34 29.79 r 1 r=2 30.02 21.13 r 2 r 3 11.31 15.49 r 2 r=3 11.19 14.26 r 3 r 4 0.119 3.841 r 3 r=4 0.119 3.841 Normalized cointegrating equation on LGDP LGDP LAS LEM LTE T 1.00 0.51 0.35 0.17 0.008 (0.03) (0.23) (0.07) (0.002) (15.3 a ) (1.54) (2.32 a ) (4.00 a ) Notes to Tables 3 6: Numbers in parenthesis are standard errors and t-statistics, respectively. Superscripts a and b denote 1% and 5% significance.

J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 3085 Table 4 Cointegration test results for (LGDP, LAS, LEM, LEL) Trace λ max H 0 H 1 Statistics 5% H 0 H 1 Statistics 5% r=0 r 1 77.05 63.87 r=0 r=1 33.31 32.11 r 1 r 2 43.73 42.91 r 1 r=2 23.96 25.82 r 2 r 3 19.77 25.87 r 2 r=3 14.06 19.38 r 3 r 4 5.71 12.51 r 3 r=4 5.716 12.51 Normalized cointegrating equation on LGDP LGDP LAS LEM LEL T 1.00 0.45 0.19 0.25 0.007 (0.03) (0.20) (0.06) (0.0022) (13.2 a ) (0.99) (3.96 a ) (3.18 a ) well as the tested cointegrating equation normalized on LGDP. The lag interval is determined as 3 and linear trend is tested to exist in the cointegration space according to Johansen (1995) VEC sequential specification test. From Table 3 we can see that both tests suggest the existence of two cointegrating vectors driving the series with two common stochastic trends in the data. That is, real GDP, net value of fixed asset, total employment and total energy consumption share common trend in the long run. As far as results of cointegrating vector normalized on GDP growth is concerned, the coefficients of fixed assets growth and total energy consumption growth are found to be affecting the level of development significantly and positively by 0.51% and 0.17%, respectively. This indicates that in China process of economic development is heavily dependent on investment as well as level of energy use. The coefficient of employment growth to GDP growth is positive but insignificant in t-statistics, implying that it should be excluded from the GDP equation, which deviates from literature and also presents in the other three models using disaggregated level of energy measures (except for oil). However, employment is significant in the second cointegrating equation normalized on LAS, indicating that it cannot be removed entirely from the study. The exclusion of employment in GDP equation is somewhat at odds, though it also happens in Ghali and El-Sakka (2004) study of Canada case. In view of the enormous population, inefficiency in employment and the potential data quality problem of employment statistics in China, it sounds more acceptable than initially. The results of testing for the number of cointegrating vectors for disaggregated level of energy measures with other three variables in production model setting as well the cointegrating equation normalized on LGDP are reported in Tables 4 6.ForTable 4, the lag interval is determined as 2 and linear trend is tested to exist in the cointegration space. Trace test statistics indicates existence of two cointegration vectors at 5% significant level while λ max test indicates one cointegration vector at critical level of 5% and two at 10% level. Considering the sampling size, specification of two cointegration vectors is set in the following VEC model. For robustness, one cointegration vector specification is also tested and the qualitative result is almost the same. For spare of space, the result of one cointegration vector specification is not reported in Table 5 Cointegration test results for (LGDP, LAS, LEM, LCL) Trace λ max H 0 H 1 Statistics 5% H 0 H 1 Statistics 5% r=0 r 1 118.19 63.87 r=0 r=1 60.35 32.11 r 1 r 2 57.84 42.91 r 1 r=2 35.23 25.82 r 2 r 3 22.61 25.87 r 2 r=3 12.81 19.38 r 3 r 4 9.792 12.51 r 3 r=4 9.792 12.51 Normalized cointegrating equation on LGDP LGDP LAS LEM LCL T 1.00 0.52 0.19 0.25 0.009 (0.03) (0.24) (0.08) (0.004) (16.0 a ) (0.80) (2.95 a ) (2.25 a )

3086 J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 Table 6 Cointegration test results for (LGDP, LAS, LEM, LOL) Trace λ max H 0 H 1 Statistics 5% H 0 H 1 Statistics 5% r=0 r 1 66.05 47.85 r=0 r=1 29.31 27.58 r 1 r 2 36.74 29.79 r 1 r=2 24.07 21.13 r 2 r 3 12.66 15.49 r 2 r=3 10.71 14.26 r 3 r 4 1.953 3.841 r 3 r=4 1.953 3.841 Normalized cointegrating equation on LGDP LGDP LAS LEM LOL T 1.00 0.55 0.37 0.07 (0.017) (0.016) (0.03) (32.3 a ) (22.5 a ) (2.33 a ) Table 7 Granger causality test results for model (DGDP, DAS, DEM, DTE) Equation Short-run Long-run DGDP DAS DEM DTE ECT 1,t 1 ECT 2,t 1 ECT DGDP ECT DAS ECT DEM ECT DTE F-statistics T-statistics Joint F-statistics DGDP 1.27 2.06 0.24 0.63 0.87 4.32 a 2.62 3.68 a DAS 0.01 6.83 a 2.23 0.93 3.1 a 8.20 a 9.58 a 7.78 a DEM 0.83 0.86 0.10 1.25 0.93 1.29 0.56 0.87 DTE 17.1 a 0.91 5.83 a 2.49 a 3.49 a 10.98 a 5.99a 4.46 a Notes to Table 7: superscripts a and b denote significance at 1% and 5% critical level and so on in the following tables. the paper. For Table 5, lag interval is determined as 4 and linear trend is tested to exist in the cointegration space. Both tests suggest the existence of two cointegrating vectors at 5% level. For Table 6, lag interval is determined as 3 and no trend is tested to exist in the cointegration space. Both tests suggest the existence of two cointegrating vectors at 5% level. The results indicate, without exception, that real GDP, net value of fixed assets, total employment and three disaggregated energy consumptions all are cointegrated in the long run, corroborating the proposition that China is heavily energy-dependent for economic development. It is worthwhile pointing out the relative lower output effect of oil comparable to other energy product. The explanation is that though oil becomes more and more important to Chinese economy, due to the fact that China is coal dependent country and share of oil consumption in total energy consumption is small, the overall effect of oil to production is relatively small 1. Electricity has the largest output effect due to the fact that electricity is closely intertwined with all sectors of the society and is efficient way of energy utilization. 5. Vector error-correction and Granger causality analysis Confirming the existence of cointegration, we proceed to test for Granger causality according to the test method discussed above. The short-run F-statistics, long-run t-statistics and joint F-statistics are reported. Before confirming the results, the models in Eqs. (3 6) are subjected to a battery of diagnostic tests for normality (Jarque Bera), serial correlation (LM), and parameter instability (CUSUM and CUSUM square). The error-correction terms in all the models are also checked for unit roots. The Granger causality test results are reported in Tables 7 10. From Table 7, it is worth noting that first error-correction term (ECT) (normalizing on LGDP) is only significant in total energy consumption equation, while the second ECT (normalizing on LAS) is significant 1 However, the limiting effect of oil shortage to GDP is remarkable in China. Since 2003 the high price of oil arise oil shortage in Chinese domestic market. It is reported that the negative influence of oil price to China economy can be as high as 0.5 1% GDP growth since 2003.

J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 3087 Table 8 Granger causality test results for model (DGDP, DAS, DEM, DEL) Equation Short-run Long-run DGDP DAS DEM DEL ECT 1,t 1 ECT 2,t 1 ECT DGDP ECT DAS ECT DEM ECT DEL F-statistics T-statistics Joint F-statistics DGDP 4.25 a 1.82 4.63 b 3.96 a 0.61 8.23 a 7.56 a 10.39 a DAS 0.33 3.72 b 0.22 1.06 4.30 a 8.95 a 8.07 a 7.88 a DEM 2.09 0.79 0.32 0.61 0.59 1.60 0.80 1.17 DEL 0.74 2.56 2.15 2.63 a 1.41 2.28 b 3.50 b 2.50 b Table 9 Granger causality test results for model (DGDP, DAS, DEM, DCL) Equation Short-run Long-run DGDP DAS DEM DCL ECT 1,t 1 ECT 2,t 1 ECT DGDP ECT DAS ECT DEM ECT DCL F-statistics T-statistics Joint F-statistics DGDP 0.23 0.08 0.72 3.29 a 1.09 4.81 a 4.24b 5.47 a DAS 2.53 1.33 5.75 a 2.44 a 3.28 a 6.95 a 10.15 a 6.27 a DEM 0.01 0.29 0.10 1.40 0.70 1.54 0.99 1.23 DCL 13.1 a 7.90 a 5.29 a 0.97 5.17 a 17.4 a 16.07 a 10.07 a Table 10 Granger causality test results for model (DGDP, DAS, DEM, DOL) Equation Short-run Long-run DGDP DAS DEM DOL ECT 1,t 1 ECT 2,t 1 ECT DGDP ECT DAS ECT DEM ECT DOL F-statistics T-statistics Joint F-statistics DGDP 0.62 0.41 3.09 b 1.06 0.31 2.73 b 1.98 6.42 a DAS 2.22 4.42 a 2.36 1.48 3.87 a 9.01 a 10.53 a 8.37 a DEM 0.01 0.02 2.13 2.13 a 1.53 2.76 b 1.76 2.45 b DOL 4.20 a 0.73 0.97 2.44 a 2.16 a 5.06 a 2.46 b 2.44 b in both fixed assets and total energy consumption equations. This suggests that all variables dynamically interact to return to the long-run equilibrium whenever there is a deviation from the cointegrating relationship. When we consider the Wald test statistics for the joint significance of the sum of the lags of the explanatory variable and the error-correction term, we see that the results indicate Granger endogeneity of all variables, except for total employment, an interesting result when considering that population in China is enormous and the employment efficiency in China has long been doubted. This suggests that there is bilateral Granger causality between total energy consumption and real GDP. In the short-run dynamics, real GDP, fixed asset and total employment are all significant in total energy consumption equation, but none are significant in real GDP equation. This suggests that there is only unidirectional Granger causality running from GDP to total energy consumption in the short run. Note also that in the short-run uni-directional Granger causality running from total employment and total energy consumption to fixed assets, but not vice versa. From Table 8, it is worth noting that first ECT (normalizing on LGDP) is significant in both GDP and electricity consumption equations, while secondary ECT (normalizing on LAS) is significant in both fixed assets and electricity consumption equations. This suggests that all variables dynamically interact to return to the long-run equilibrium whenever there is a deviation from the cointegrating relationship. Considering the joint Wald test, all variables except total employment are confirmed of Granger endogeneity. This suggests that there exists bi-directional Granger causality between electricity consumption and real GDP in the long run. In the short-run dynamics, fixed asset and electricity consumption are significant in GDP equation, but none are significant in electricity consumption equation, indicating the uni-directional Granger causality running from electricity consumption to real GDP in the short run.

3088 J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 Fig. 4. Impulse responses between GDP and energy consumption in the four models.

J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 3089 From Table 9, it is worth noting that first ECT (normalizing on LGDP) is significant in both GDP and coal consumption equations, while secondary ECT (normalizing on LAS) is significant in both fixed asset and coal consumption equations. Considering the joint Wald test, again, all variables except total employment are confirmed of Granger endogeneity. This suggests that there exists bi-directional Granger causality between coal consumption and real GDP in the long run. In the short-run dynamics, real GDP, fixed asset and total employment are all significant in coal consumption equation, but none are significant in real GDP equation. This suggests that there is only uni-directional Granger causality running from GDP to coal consumption in the short run. Note also that, in the short run, there exists uni-directional Granger causality running from coal consumption to fixed assets. From Table 10, it is worth noting that first ECT (normalizing on LGDP) is significant in both employment and oil consumption equations, while secondary ECT (normalizing on LAS) is significant in both fixed assets and oil consumption equations. Considering the joint Wald test, all variable are confirmed of Granger endogeneity, somewhat different from the above. This suggests that there exists long-run bi-directional Granger causality between oil consumption and real GDP. In the short-run dynamics, real GDP and oil consumption are symmetrically significant in respective equations, suggesting the existence of bidirectional Granger causality in the short run. 6. Generalized impulse response analysis and discussions of the empirical results The direction of Granger causality can be determined via the VEC framework; however, the importance of the causal impact is also of interest. The impulse response functions (IRFs) are based on a moving average representation of the VAR model, and the dynamic responses of one variable to another are evaluated over horizons. In order to assess how a shock to one variable affects another variable and how long the effect lasts we utilize generalized impulse response (Koop, Pesaran and Potter, 1996; Pesaran and Shin, 1998). Impulse response results are plotted out in Fig. 4. Point estimates of the IRFs are plotted with a solid line while the dotted lines represent a two standard deviation band around the point estimates. If the bands cross zero, point estimates are considered to be significant our rule of thumb. For spare of space, we only report the impulse responses among GDP and energy consumption at both levels for all the models. From Fig. 4a), we can see that the output effect of total energy consumption shock is found to be insignificant over the horizons. While the real GDP shock to total energy consumption is significant and positive, and slowly Table 11 Comparison of Granger causality test results with related earlier works Relationship tested Total energy and real GDP Electricity and real GDP Coal and real GDP Oil and real GDP Previous works Authors Empirical method and model Soytas and Sari (2006) Lin (2003) Shiu and Lam (2004) Yuan et al. (2007) Zhang and Li (2007) Zhou and Chau (2006) Toda Yamamoto procedure in multivariate setting of (Y,K,L,E) Error-correction model in a multivariate demand function setting including GDP, energy price, economic structure, population and efficiency enhancement Error-correction model in abivariatemodel Error-correction model in abivariatemodel Error-correction model in abivariatemodel Error-correction model in abivariatemodel Data period 1971 2002 (World Development Indicators data) Results Non-cointegrated 1978 2001 GDP E in the long run 1971 2000 GDP E 1978 2004 GDP E inboth short and long run 1980 2004 GDP E inboth short and long run 1953 2002 GDP E in the long run GDP E intheshortrun Notes to Table 11: GDP E denotes Granger causality runs from GDP to energy consumption. GDP E denotes Granger causality runs from energy consumption to GDP. GDP E denotes bi-directional Granger causality between GDP and energy consumption. Our results GDP E inthelongrun GDP E intheshortrun GDP E inthelongrun GDP E intheshortrun GDP E inthelongrun GDP E intheshortrun GDP E inboththe short and long run

3090 J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 enervates over the horizons. Concerning GDP and electricity consumption (Fig. 4b)), shocks in both directions are significant and positive and enervate quickly (in 2 3 years) over the horizons. Concerning GDP and coal consumption (Fig. 4c)) similar responses are found. Concerning GDP and oil consumption (Fig. 4d)), shocks in both directions are significant and positive, but enervate quickly over the horizons. The impulse response results corroborate the conclusions of Granger causality analysis 2. Granger causality test, along with generalized impulse response analysis, seems to confirm that energy consumption, whether at aggregated or disaggregated level, has significant impact on economic growth, and vice versa. The Granger causality found in the paper between economic growth and total energy, coal, electricity and oil consumption respectively, is summarized in Table 11 and compared with previous works. Our empirical results seem to support the empirical result of Shiu and Lam (2004), Yuan et al. (2007), but contradict Lin (2003) on electricity consumption and economic development relationship, support Zhou and Chau (2006) on oil and economic development relationship. At aggregated level, our results differ with Soytas and Sari (2006) result of no cointegration between energy consumption and economic growth in China, and the hereafter optimistic conclusion that China can achieve sustainable growth without putting too much pressure on the environment. Though difference in empirical results is perhaps because of different data and model used (Shrestha, 2000), our empirical results seem to have more robust base according to the experiences with the status quo of energy and economy reality in China. 3 7. Policy implications The energy-income nexus poses important challenges to Chinese policy makers, considering the high energy consumption growth rate, high CO 2 emissions level and its growth rate. Economic growth rate is expected to keep as high as 7 8% in the next 20 years. In light of the close energy-income nexus, how can China realize sustainable development and cut down GHG emissions? Since the emissions mainly result from consumption of fossil fuels, reducing energy consumption seems to be the direct way of handling the emissions problem. However, due to the negative impact on economic growth, direct measure to reduce energy consumption is not viable in China. On the other hand, in China pure development itself may not be a solution to environmental and ecological problem. Hence, active policies and measures must be implemented. First of all, enhancing energy supply security and guaranteeing energy supply is of uttermost importance to China. Particularly in the short run, proper supply of electric power and oil is vital to the function of economic activity. Concerning electric power, due to the 3 4 year lead time of power infrastructure construction, proper and active strategic planning for electric power supply should be drawn. The policy of electricity supply leads economic growth has been implemented by China Government in the long period of power shortage and should be insisted for a long period. As far as oil supply is concerned, China is a country short of oil resource and currently nearly half of domestic oil consumption comes from import, while it is estimated that in 2020 the number will be as high as 60% if the current trend continues 4. Oil supply has always been an international dispute and is sensitively disrupted by international political affairs, geostrategic riffle and even subtle sentimental change. Up to now the national level strategic oil reserve in China is yet to set up. Without sufficient pre-action China domestic economy is vulnerable to internal oil price shock. For example, since 2003 with the abrupt rise of international oil price, shock of oil price hike to domestic economy has called serious concern of all levels of the society. In the end of 2007, with the unprecedented international oil price as high as nearly 100 USD/barrel, the unwillingness of main oil refinery producer to import crude oil aroused domestic commodity oil shortage again. The estimated price increase and intense demand for oil foretells that oil shortage will be normal status in China for the future and China has to draw out effective long-range measures. National strategic oil conservation system can serve as a buffer to price hike and leave the domestic industry more space for adjustment. In the long 2 Readers are referred to the appendix of the website version of the paper for the responses among all variables. 3 China central government has stipulated the energy efficiency goal of reducing energy consumption by 20% in the eleventh fiveyear plan period. The goal for the first year of eleventh five-year, 2006, is to reduce energy intensity by 4% and main pollutant emission by 2%. However, the fact is that energy intensity only decreased by 1.23% while SO 2 emission increased by 1.8% as to 2005, which means that the energy efficiency goal is not achieved for the first year. The factors behind may be manifold, but one of the most important is development pattern and substantial change in economic output structure. 4 In 1993 China became the net oil importer and since then dependence of domestic oil consumption on import increased year by year. In 2002 the number is 33% while in 2005 the number rises to 43% abruptly and 2006 to 47%.

J.-H. Yuan et al. / Energy Economics 30 (2008) 3077 3094 3091 run, proper strategic planning for coal supply is the prerequisite for energy security because of China's energy gift. Another viable policy as well as challenging task is to enhance energy efficiency. Though energy efficiency in China has improved significantly, the absolute level is lower than international level. It is estimated that the GDP energy intensity has been decreased from 5320 in 1990 to 2680 kg sce/10 4 Yuan (in 1990 constant price) in 2002, with annual decrease rate of 5.6%. However, it is still distinctly higher than international levels. According to the current exchange rate, the GDP energy intensity for China in 2000 is 1274 tons sce/million dollars, which is 2.4 times of international average, 2.5 times of USA level, 4.9 times of European Union level and 8.7 times of Japan level (State Development and Reform Commission of China, 2004). Considering the product level, the average energy intensity for main product in eight industry sectors of electric power, oil, non-ferrous metals, construction material, textile and others is 40% higher than international average. For example, power supply energy intensity in coal power plant is 22.5% higher, steel product 21.4% higher, and cement product 45.3% higher than international average. The efficiency of main energy-consuming equipments is low too. For example, the average operation efficiency in industrial coal firing boiler is 65%, less than international average by 15 20%. The efficiency of small electromotor is 87% and 75% for wind turbine and water pump, 5% lower than international levels. Realizing the big gap in energy efficiency, China State Development and Reform Commission stipulated Medium and Long Term Energy Saving Plan in 2004, regulating that GDP energy intensity decrease from 2680 kg sce/10 4 Yuan in 2002 to 2250 kg sce/10 4 Yuan (in 1990 constant price) in 2010, which can save 400 million tons sce. The 2020 goal is to further decrease the GDP energy intensity to 1540 kg sce/10 4 Yuan and the accumulated energy saving will be 1.4 billion tons sce. If energy saving goal for regular fossil fuels is achieved, it will approximate 2.8 billion tons reduction of CO 2 emissions, a great contribution to global fighting against GHG. The energy saving goal declared in Medium and Long Term Energy Saving Plan is further accentuated in Eleventh Five-Year Plan for Economic and Social Development, the top-level economic and social development creed in China (the Central People's Government of the People's Republic of China, 2006) and Eleventh Five-Year Plan for Energy Development (State development and reform commission of China, 2007b). However, the fact that energy efficiency goal (4% decease of GDP energy intensity as to 2005) for 2006 is not achieved has called serious attention of Central Government. In 2007, the government executed strict administrative measures to ensure the goal to be achieved. For example, in power generation sector, 10 GW small-scale low efficiency coal power plants are forced to phase out and total of 50 GW such coal power plants will be closed during 2006 2010 period. State development and reform commission also stipulated specific production capacity optimization and efficiency enhancement goals for those high energy intensity industries, such as cement, steel, colored metals and others. The recently revised Energy Saving Act (The National People's Congress of the People's Republic of China, 2007) also highlights the responsibility of local government in energy saving and takes energy saving as basic national policy. In the future, China government must implement more economic incentive policies to realize energy efficiency enhancement. One effective long-term policy is to diversify energy supply with preference on renewable energy such as wind power, solar cell and others. China is heavily dependent on coal for energy source. During 1953 2005 periods, the ratio of coal in total energy consumption was as high as 78% in annual average. With the slowdown of economic growth because of 1997 Asian Crisis, the demand for coal began to fall and the ratio in 1998 was 69.6%, for the first time below 70%. Since then it continued to go down and the number was 66.32% in 2002. However since 2002 with the vigor of economy recovering, the demand for coal increased abruptly and the number rose back to 68.9% in 2005. Roughly 60% of coal consumption in China is for electricity production; so to reduce the reliance on coal, China has to diversify its energy source. In 2005 China firstly enacted the Renewable Energy Act (the National People's Congress of the People's Republic of China, 2005) to provide legal base for the development of renewable energy in China and formulate the principle of R&D, industrialization, popularization and application and economic incentive for renewable energy exploration and usage. In 2007, the Medium and Long Term Development Plan for Renewable Energy (State development and reform commission of China, 2007a) stipulated more concrete goal that the ratio of renewable energy in total energy consumption should be no less than 10% in 2010 and 15% in 2020, while renewable should account for no less than 30% of total power generation capacity in 2020. The saving of fossil fuels is expected to be tremendous. It is estimated if the goal stipulated be achieved, 300 and 600 million tons sce will be substituted in 2010 and 2020 respectively; coal and oil consumption will be