Applied Energy 104 (2013) Contents lists available at SciVerse ScienceDirect. Applied Energy

Transcription

1 Applied Energy 14 (213) Contents lists available at SciVerse ScienceDirect Applied Energy journal homepage: The dynamic links between CO 2 emissions, economic growth and coal consumption in China and India V.G.R. Chandran Govindaraju a,, Chor Foon Tang b a Department of Development Studies, Faculty of Economics and Administration, University of Malaya, 563 Kuala Lumpur, Malaysia b Department of Economics, Faculty of Economics and Administration, University of Malaya, 563 Kuala Lumpur, Malaysia highlights " CO 2 emissions, economic growth and coal consumption relationship in China and India is examined. " The results indicated the presence of cointegration in China, but not in India. " In China, uni-directional causality runs from economic growth to CO 2 emissions. " In the case of India, only a short-run causality is detected. article info abstract Article history: Received 2 March 212 Received in revised form 25 July 212 Accepted 2 October 212 Keywords: Coal consumption CO 2 emissions Economic growth China India In this study, we employ recent and robust estimation techniques of cointegration to provide more conclusive evidence on the nexus of CO 2 emissions, economic growth and coal consumption in China and India. Furthermore, the causal relationships among the variables are further examined using the Granger causality test. Our empirical results suggest that the variables are cointegrated in the case of China but not India. In other words, there is a long-run relationship between CO 2 emissions, economic growth and coal consumption in China. Granger causality test for China reveal a strong evidence of unidirectional causality running from economic growth to CO 2 emissions. Moreover, there is a bi-directional causality between economic growth and coal consumption as well as CO 2 emissions and coal consumption in the short and long run. In the case of India, only a short-run causality is detected. Causality between economic growth and CO 2 emissions as well as CO 2 emissions and coal consumption are bi-directional. Nonetheless, there is only a uni-directional Granger causality running from economic growth to coal consumption in India. The implications of the results are further discussed. Ó 212 Elsevier Ltd. All rights reserved. 1. Introduction Corresponding author. address: vgrchan@gmail.com (V.G.R. Chandran Govindaraju). In achieving rapid developmental goals, countries at large face conflicting policy choices from rapid economic growth, significant consumption of resources and environmental deterioration [1,2]. It is more so in emerging economies such as China and India where both the countries have recorded higher economic growths and significant increases in the consumption of coal [1,3,4]. China and India, together, showed significant increases in their percentage of energy use of total world energy consumption, from 1% in 199 to 21% in 28 and are expected to increase their energy use to 31% in 235 [5]. Robust growth in these two countries, even during recession, is expected to increase the coal consumption and consequently influence the CO 2 emissions. If these countries decide to pursue sustainable developmental goals, it might require a reduction in energy consumption, specifically coal, and an increase in the proportion of renewable energy in primary energy supply. In other words, fear of climatic change may limit the use of coal in the future [6]. Indeed, coal consumption contributes more carbon per tonne of oil equivalent than other resources like natural gas and oil. Nevertheless, currently, coal still plays an important role in economic growth and is the second largest source of world CO 2 emissions [3,7]. Although reduction in energy consumption seems to be a viable option in reducing CO 2 emissions, its impact on economic development can be negative. For instance, China s coal consumption in terms of percentage of total energy consumption is nearly 69% and any attempt to reduce it may have potential reciprocal influence on economic growth. Therefore, there is an urgent need to understand the dynamic links between coal consumption, CO 2 emissions and economic development in these countries. This study is timely given the fact that both countries have recorded high economic development and the consumption of coal is becoming an essential energy mix. On the contrary, pressure to /$ - see front matter Ó 212 Elsevier Ltd. All rights reserved.

2 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) reduce the CO 2 is mounting, forcing policymakers to find alternatives to reduce per capita CO 2 emissions. In addition, as part of the Kyoto Protocol, participating countries are required to reduce CO 2 emissions collectively, about 5% on average over [3]. Nevertheless, in reality, CO 2 emissions between 1992 and 27 have increased by 38% [2]. The real GDP, coal consumption and CO 2 emissions of China and India over the periods of showed an increasing trend (see Fig. 1). China and India recorded a remarkable growth and in the period of 21 29, these economies have been growing at an average rate of 1.5% and 7.4% per annum respectively (see Table 1). China and India being the most populous developing countries have significant influences on global coal consumption and emissions and are expected to have an increasing influence in the future [2,4,6,8,9]. China s and India s per annum average growth for the same periods for coal consumption are 1% and 6.4% respectively and CO 2 emissions are 8.5% and 5.7% respectively. In both countries, with large domestic coal reserves, the coal use for electricity power and industrial processes has increased. Moreover, with increasing coal-fired generation capacity in China and India, coal consumption is expected to increase. Industrial coal consumption from 28 to 235 is expected to grow by 67% in China and 94% in India [1]. Between 23 and 28, China s coal consumption increased by 71% [1]. In 28, China was the top total CO 2 emitter in the world surpassing the United States while India ranked at third place. China s per capita CO 2 emissions increased by two and a half times, while India s per capita CO 2 emissions increased two times [5]. Future forecast shows that annual CO 2 emissions will increase 2.5 times from to 23 reaching 384 million tons (Mt) of CO 2 in 23, recording an annual growth of 3.7% in India [4]. In 1996 China accounted for 13.8% of CO 2 world emissions and the share had increased to 21% in 27 [9]. The biggest challenge for leaders in China and India is on how to maintain economic growth, while keeping CO 2 emissions as well as coal consumption at an acceptable level so that it will not harm economic growth especially if US (Billion) Real GDP (CHINA) 3, 2,5 2, 1,5 1, US (Billion) Real GDP (INDIA) ,6 COAL CONSUMPTION (CHINA) 28 COAL CONSUMPTION (INDIA) 1,4 24 Mtoe 1,2 1, Mtoe Million Tons Carbon Dioxide CO2 EMISSIONS (CHINA) 8, 7, 6, 5, 4, 3, 2, 1, Million Tons Carbon Dioxide CO2 EMISSIONS (INDIA) 1,6 1,4 1,2 1, Fig. 1. The plots of real GDP, coal consumption, CO 2 emissions, China and India,

3 312 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) Table 1 Annual growth rates, China and India, Source: Authors calculation. Years China India Real GDP Coal consumption CO 2 emissions Real GDP Coal consumption CO 2 emissions Average there is a long-run causal relationship between economic growth, coal consumption and CO 2 emissions. Nevertheless, in terms of CO 2 emissions per unit of GDP, both countries have shown significant reductions between 199 and 29 and this might indicate that there is a potential decoupling of CO 2 emissions from economic growth [5]. However, more investigation is needed to understand the link between CO 2 emissions and economic growth. Indeed, with both countries rapid development, increasing environmental awareness may force the government to reduce CO 2 emissions and coal consumption. China s recent efforts to develop renewable energy sectors may also have implications to reduce coal consumption. Therefore, a wide range of interesting questions emerge in these emerging markets. Will the economic growth stall if coal consumption is controlled or if the policymakers move to alternative sources of energy? Can the control of coal consumption significantly reduce the CO 2 emissions? How do economic progress and coal consumption affect CO 2 emissions and what are their influences? These questions are pertinent to policymakers and require a careful investigation as it will allow a better formulation of strategies and policies. Thus, this study attempts to provide some insights with regards to the above questions. Although numerous studies (e.g. [11 14]) are available in analysing the nexus between energy consumption, economic growth and CO 2 emissions, two fundamental gaps exist in the literature. First, there is limited evidence available on analysing the coal consumption and its implication for economic growth and CO 2 emissions using recent and longer time series data (e.g ). It is important to note that longer series are essential to establish a more robust estimation using unit root and cointegration test. Some studies even used very short time series of 26 years (e.g. [13]). By re-examining the issue, we can provide a more conclusive evidence for policymakers. Second, despite coal being the major energy mix, most of the studies analysing the issues used aggregate energy consumption data and not coal consumption. In fact, in the case of India, evidence is scarce and little data is available for any meaningful insight toward policy direction. Additionally, analysing China and India in a single paper provides insight on whether there are any differences in the links despite both countries being heavy consumers of coal. Indeed, the study will be able to establish reasons of why the results might differ. It may also indicate that country specific information is important since results differ among countries. In this study, we use recent datasets (longer time series data from ) to re-investigate the issues. In terms of methodology, we employ the Augmented Dickey Fuller (ADF) and Ng Perron unit root tests to determine the order of integration of each series. Additionally, we have also performed the Lumsdaine and Papell [17] unit root test with two structural breaks on each variable. This will be useful in examining the environmental Kuznets curve (EKC) hypothesis that posited a non-linear relationship between CO 2 emissions and the level of outputs. Likewise, this study uses the most recently developed combined cointegration tests suggested by Bayer and Hanck [16] to determine the presence of long run equilibrium relationship between CO 2 emissions, economic growth and coal consumption. On the basis of the results of the Monte Carlo experiment, Bayer and Hanck [16] suggested that the combined cointegration tests were more powerful and robust than the existing individual cointegration tests (e.g. [17 2]). Indeed, due to increasing number of studies using conventional methods, Karanfil [21] indicated the need to use new techniques of analysis in examining the energy-growth nexus. There are two main advantages of using the combine cointegration tests. First, the combine cointegration tests have superior performances in small samples. The Engle Granger, Johansen Juselius and ECMbased tests for cointegration are not robust and lead to spurious cointegration results particularly when the sample size is less than 1 observations [22,23]. Second, the combine cointegration tests provide more conclusive result. Gregory et al. [24] found that the p-values for different single cointegration tests are usually not strongly correlated. Moreover, Pesavento [25] showed that the testing powers vary among the single cointegration tests. Hence, the single cointegration tests tend to provide inconclusive results. Owing to these advantages, we choose to use the combined cointegration tests. Finally, the Granger causality test is used to verify the direction of causality among the variables of interest. In doing so, our empirical results will be more comprehensive and robust. Although, some studies use panel data techniques, in this study we limit the use of panel approach due to the following reasons. First, panel data analysis is based upon the homogenous relationship assumption across countries. This assumption leaves the analysis to focus on one perspective but ignoring the complexity and dynamism of an economic behaviour. Solow [26] suggested that an economic model should be dynamic in nature in order to observe the evolution of economic behaviour over time. Likewise, Athukorala and Sen [27] postulated that homogenous relationship may not always exist due to the different nature of economic structure, income and also due to demographic reasons. On the basis of Monte Carlo experiments, Robertson and Symons [28] and Pesaran and Smith [29] revealed that when heterogeneous exist in a panel model with a small cross-section (N) dimension as the case of our study (N = 2), the estimation results are likely to be biased. Maddala et al. [3] added that panel data estimation may not be accurate when the problem of cross-section heterogeneous occurs. Second, Deaton [31] claimed that the quality of data varies across countries, therefore estimation with cross-sectional and panel data studies are more likely not to yield robust results. Therefore, country-specific study with time series data analysis is more suitable when cross-section dimension of a panel data is very small as

4 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) Table 2 Summary of selected literatures and major findings. Authors Variables Time period China Li et al. [35] Energy, GDP, CO Wang et al. [14] CO 2, energy, GDP Li and Leung [36] Coal, GDP Bloch et al. [37] DD: coal, GDP, price, CO 2 SS: GDP, 1965 labour, capital, coal 28 Yuan et al. [38] Coal, GDP, labour, capital 1963 Li et al. [3] Coal, GDP, CO Wolde-Rufael [7] Coal, GDP, labour, capital 1965 Ang [39] CO 2, energy, GDP Jalil and Mahmud GDP, CO [12] Zhang and Cheng GDP, energy 196 [11] 27 Chang [13] Energy, GDP, CO India Alam et al. [4] Energy, GDP, CO 2, labour, capital Wolde-Rufael [7] Coal, GDP, labour, capital 1965 Li et al. [3] Coal, GDP, CO Kulshreshtha and Coal, GDP 197 Parikh [41] 1995 Paul and Bhattacharya [42] Energy, GDP Ghosh [43] CO 2, GDP Methodology Coint. Causation Pedroni cointegration; DOLS Yes Pedroni cointegration; Granger causality Yes SR: E M GDP; CO 2 M ELR: E M CO 2 ; GDP? CO 2 Pedroni cointegration; Granger Yes C M GDP (costal & eastern region) causality GDP? C (western region) Johansen Juselius; Granger causality Yes DD: C? GDP; C M CO 2 SS: C? GDP Johansen s test; Granger causality Yes GDP? C Johansen Juselius; Granger Causality Yes GDP? C MWALD causality GDP? C ARDL; FM-OLS; FM-UECM; DOLS ARDL; Granger causality Yes GDP? CO 2 MWALD causality GDP? E; GDP CO 2 Johansen Juselius; Granger causality Yes E? CO 2 ; GDP? C; GDP? CO 2 MWALD causality E M CO 2 ;E GDP MWALD causality C? GDP Johansen Juselius; Granger Causality Yes GDP C Johansen Juselius; Granger causality Yes Engle Granger; Johansen Juselius; Granger causality test Yes E M GDP ARDL; Johansen Juselius Yes CO 2 M GDP; E? CO 2 Note for the abbreviation: DD = Demand-side model; SS = Supply-side model; DOLS = Dynamic OLS; ARDL = Autoregression distributed lag; FM-OLS = Fully modified OLS; FM-UECM = Fully modified unrestricted error-correction model; MWALD = Modified Wald test; SR = Short run; LR = Long run; E = Energy consumption; C = Coal consumption; CO 2 = Carbon dioxide emission; GDP = Income. The arrows (? and M) represent the direction of causality while indicates no causality. the case of our study (N = 2). 1 Subsequently, separate policies can be suggested for China and India. The rest of the paper is organised as follows. Section 2 briefly reviews the previous empirical studies. Section 3 deliberates the data, empirical model and econometric methods. Section 4 discusses the empirical results whereas the conclusion and policy implications are presented in Section Brief literature review The existing literatures, to our knowledge, revealed that very few studies are available on the issues of coal consumption and its link to economic development and CO 2 emissions. In the case of China, except for a few, most of the studies examined the relationship between total energy consumption, economic growth and CO 2 emissions whereas for India, literature was scant. Generally, these literatures tended to examine: first, the relationship between energy consumption and economic growth; second, energy consumption and CO 2 emissions and third, economic growth and 1 We are aware of the fact that Pesaran et al. [32] and Pedroni [33] suggested panel data approaches that released the assumption of homogenous. Nevertheless, the application of panel data analysis in a very small cross-section dimension (N = 2)as the case of this study may not significantly enhance the robustness of the testing results. CO 2 emissions, the so-called environmental Kuznets curve (EKC) hypothesis [34]. Table 2 summaries and compares the recent literatures that examine the link between energy consumption, economic growth and CO 2 emissions in China and India. The findings of the previous studies (as depicted in Table 2) suggested that the results are inconclusive depending on the methods and time period employed. In the case of China, Li et al. [35] used panel unit root, heterogeneous panel cointegration and dynamic OLS methods to examine the relationship between energy consumption and economic growth for 3 provinces in China. Their results showed that there was a positive long-run relationship between energy consumption, CO 2 emissions and economic growth. They found that a 1% increase in GDP per capita increased the energy consumption and CO 2 emissions by.5% and.43% respectively. Similarly, Wang et al. [14] examined the relationship between CO 2 emissions, energy consumption and economic development using a panel data of 28 provinces in China. The result confirmed the cointegration relationship among the three variables. Moreover, the study found evidence of bi-directional causality between CO 2 emissions and energy consumption as well as between energy consumption and economic growth. In addition, energy consumption and economic growth caused CO 2 emissions in the long run while CO 2 emissions and economic growth were the long-run causes of energy consumption. Li and Leung [36] discovered that coal consumption and GDP were cointegrated and

5 314 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) confirmed bi-directional causality between coal consumption and GDP for the coastal and central regions where most coal intensive industries are situated. However, they found that in the Western region there was uni-directional Granger causality running from GDP to coal consumption. Bloch et al. [37] used the demand and supply-side frameworks to analyse the relationship between coal consumption and output in China. The results of the supply-side analysis showed that there was uni-directional causality running from coal to output in the short and long run. However, the results of the demand-side analysis showed that there was a uni-directional causality running from output to coal consumption in the short and long run. In addition, they found that there was a bidirectional causality between coal consumption and CO 2 emissions in the short and long run. Likewise, Yuan et al. [38] examined the cointegration and causal relationship between energy consumption and economic growth at both aggregated and disaggregated levels such as coal, oil and electricity in a multivariate framework using labour and capital as additional inputs. Their results indicated the existence of cointegration between the variables. In terms of causal relationship, the authors found uni-directional causality running from GDP to coal consumption in the short run, but not vice versa. Other studies for China such as Li et al. [3] and Wolde-Rufael [7] also found uni-directional causality running from GDP to coal consumption. In contrast, Yuan et al. [38] discovered uni-directional causality running from GDP to coal consumption in the short run and bi-directional causality in the long run. Other relevant studies in the case of China included Ang [39], Jalil and Mahmud [12], Zhang and Cheng [11] and Chang [13]. A result from Ang s [39] study indicated that CO 2 emissions could be influenced by energy consumption and output. Jalil and Mahmud [12] found that GDP Granger caused CO 2 emissions while Zhang and Cheng [11] found that GDP Granger caused energy consumption and energy consumption to Granger caused CO 2 emissions respectively. Chang [13] revealed that coal consumption had an impact on CO 2 emissions in the long run and thus reduction in coal consumption would have a positive impact on the environment. On the same note, GDP growth increased CO 2 emissions in China. In the case of India, empirical evidence seemed to be very limited (see Table 2). Based on our knowledge, only a few studies examined the link between energy consumption, economic growth and CO 2 emissions in India. Alam et al. [4] investigated the causal relationship between energy consumption, CO 2 emissions and income in India. Evidence supported bi-directional causality between energy consumption and CO 2 emissions in the long run. However, there was no causality between energy consumption and income as well as between CO 2 emissions and income. Thus, the authors conclude that energy conservation policies could be implemented without affecting economic growth. In addition, reducing CO 2 would be less easy in India due to the absence of causality in any direction. However, Wolde-Rufael [7] found a uni-directional causality running from coal consumption to economic growth in the case of India. Li et al. [3], in contrast, found no evidence of causal relationship between coal consumption and GDP for India. Using sectoral data, Kulshreshtha and Parikh [41] investigated the long run relationship between coal demand and economic growth. The results indicated that economic growth had been the major drive for coal demand in the long run in all sectors, but the income elasticity varied across sectors. Further, Paul and Bhattacharya [42] found a bi-directional causality between energy consumption and economic growth in India. Ghosh [43] failed to find any long-run cointegration and causal relationship between CO 2 emissions and economic growth in India. Instead, only short-run bi-directional causality was detected between the two variables. Likewise, a number of studies (limited studies available) also examined the EKC hypothesis (inverse U-shaped relationship between emissions and economic growth) in the case of China and India [12,44,45]. Song et al. [44], for instance, found support of the EKC hypothesis in China using the Chinese provincial data from 1985 to. Similarly, Jalil and Mahmud [12] also supported the EKC hypothesis indicating a quadratic relationship between income and CO 2 emissions. Managi and Jena [45], in the case of India, also found evidence for EKC hypothesis. As a whole, although there seemed to be an agreement on the existence of EKC hypothesis at least, studies on China and India were limited in the sense that evidence on the cointegration relationship and direction of causality between energy consumption, CO 2 emissions and economic growth appeared to be mixed and inconclusive. Therefore, examining the dynamic relationship requires more robust investigation using extended time series data as well as new estimation techniques as proposed in this study. This study fills the gap and reassesses the less explored link specifically between coal consumption, economic growth and CO 2 emissions in China and India. 3. Data, model and econometric methods 3.1. Data and empirical model This study uses annual data of per capita CO 2 emissions in metric tonnes, per capita real gross domestic product (GDP) in US dollar, and per capita coal consumption in million tonnes of oil equivalent (Mtoe). This study covers the annual sample period from 1965 to 29 for China and India. This dataset is collected from the World Development Indicators (WDI) and the Review of World Energy 211. All variables are converted into natural logarithms, thus the variables can be interpreted as growth after taking the first difference. Following the existing literatures on environmental economic, output and energy consumption are two main determinants of CO 2 emissions. More specifically, the EKC hypothesis claims that the relationship between CO 2 emissions and output is non-linear. According to the EKC hypothesis, CO 2 emissions and output should be in the form of inverted U-shape relationship. Hence, the econometric model of CO 2 emissions for China and India can be expressed as below: ln CO 2t ¼ a þ a 1 ln GDP t þ a 2 ln GDP 2 t þ a 3 ln COAL t þ e t ð1þ Here, ln denotes the natural logarithm, ln CO2 t is the per capita CO 2 emissions, ln GDP t is the per capita real GDP, ln GDP 2 t is the squared of per capita real GDP and ln COAL t is the per capita coal consumption (a proxy for energy consumption). The disturbance term e t is assumed to be normally distributed and white noise. a 1, a 2 and a 3 are the long run parameters of ln CO2 t with respect to ln GDP t, lngdp 2 t and ln COAL t. Based on the EKC hypothesis, a 1 > and a 2 < form the inverted U-shape relationship, while a 3 > Cointegration analysis Engle and Granger (EG, [17]) proposed the first version of cointegration test based on the estimated residuals of a long run regression model. This method was then termed as the residualsbased test for cointegration. After a decade, various cointegration tests were developed such as the system-based test of Johansen (JOH, [18]), the ECM-based F-test of Boswijk (BO, [19]) and the ECM-based t-test of Banerjee et al. (BDM, [2]). Unfortunately, different cointegration tests might suggest different conclusions because no one cointegration test was perfect and completely robust in all applications [46]. To enhance the power of the cointegration tests, this study uses the newly-developed cointegration test suggested by Bayer and Hanck [16] to check the presence of cointegrating relationship between CO 2 emissions and its

6 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) determinants in China and India. A unique aspect of this new cointegration test is it allows us to combine various individual cointegration test results to provide a more conclusive finding. With respect to this, Bayer and Hanck [16] proposed to combine the computed significance level (p-values) of the individual cointegration test with the following Fisher s formulas: EG JOH ¼ 2½lnðp EG Þþlnðp JOH ÞŠ EG JOH BO BDM ¼ 2½lnðp EG Þþlnðp JOH Þþlnðp BO Þ þ lnðp BDM ÞŠ where p EG, p JOH, p BO and p BDM are the p-values of various individual cointegration tests respectively. 2 If the calculated Fisher statistics exceed the critical values provided by Bayer and Hanck [16], the null hypothesis of no cointegration can be rejected Granger causality analysis Testing for the presence of cointegration is necessary to obtain reliable results, but it is insufficient for modelling the causation. This is because the direction of causality remains uncertain when it comes to economic growth, coal consumption and CO 2 emissions. With respect to this, we apply the Granger causality test to determine the direction of causality between CO 2 emissions, real GDP and coal consumption in China and India. If the variables are found to be cointegrated, the vector error-correction model (VECM) must be used to test for Granger causality. On the other hand, if the variables are not cointegrated, the first difference vector autoregressive (VAR) model will be used. Assuming that the variables are cointegrated, the Granger causality test can be conducted by estimating the following VECM: ln CO2 t b 1 ln GDP t ð1 LÞ6 4 ln GDP ¼ b t b 3 5 ln COAL t b B 11;k B 12;k B 13;k B 14;k þ Xp B 21;k B 22;k B 23;k B 24;k ð1 LÞ6 7 4 B k¼1 31;k B 32;k B 33;k B 34;k 5 B 41;k B 42;k B 43;k B 44;k ln CO2 t k d 1 l 1t ln GDP t k 6 4 ln GDP þ d t k d 3 5 e l ½ t 1Šþ 2t l 3t 5 ð4þ ln COAL t k d 4 Here, (1 L) is the first difference operator and l it is the disturbance term which is assumed to be normally distributed and white noise. e t 1 is the one period lagged error-correction term derived from the cointegrating vector. It is also important to point out here that if the variables are not cointegrated, the e t 1 will be removed from the equations. In other words, it will become the first difference VAR model. If the cointegration is detected, there will be short and long run Granger causality relationships; otherwise it can only have short run Granger causality relationships. Given that there are two sources of Granger causality, we can test the short run Granger causality by restricting the first difference explanatory variables in the system. For example, to investigate whether real GDP growth Granger causes growth of CO 2 emissions, we test the null hypothesis of B 12,k = B 13,k = with the Wald tests. Similarly, if we 2 To conserve space, the testing procedure for individual cointegration tests will not be presented here. Nevertheless, interested readers can refer to Maddala and Kim [47] for detailed testing procedures. l 4t ð2þ ð3þ attempt to examine whether growth of CO 2 emissions Granger causes real GDP growth, we can test the null hypothesis of B 21,k = B 31,k = with the Wald tests as well. For the long run Granger causality, we can test the null hypothesis of B 12,k = B 13,k = d 1 = using the Wald test. Rejection of this hypothesis implies that there is an overall long run Granger causality running from Dln GDP t and D ln GDP 2 t to Dln CO2 t. Likewise, if the hypothesis B 21,k = B 31,k = d 2 = d 3 = is rejected, it means that there is an overall long run Granger causality running from Dln CO2 t to D ln GDP t and D ln GDP 2 t. Finally, the same testing step applies to the growth of coal consumption and its causal relationship with real GDP growth and growth of CO 2 emissions in the system. 4. Empirical results 4.1. Unit root and cointegration results Table 3 reports the results of unit root tests for China in Panel A and India in Panel B. At the 5% significance level, the ADF test statistics for all variables in both countries cannot reject the null hypothesis of a unit root at level. Hence, the variables are non-stationary at level. Nevertheless, when we transform the variables into first difference, the ADF statistics automatically reject the null hypothesis of a unit root at the 1% significance level. Therefore, the ADF results suggest that the variables under investigation for the two countries follow the I(1) process. Given the sample size of this study and the low power of ADF test in handling small samples, it is necessary to apply the Ng and Perron [48] unit root test to confirm the order of integration of each series. A family of unit root tests are reported in Table 2 MZ a, MZ t, MSB and MPT. At the level, the aforementioned Ng Perron statistics cannot reject the null hypothesis of a unit root at the 5% significance level. On the contrary, the statistics of all variables consistently reject the hypothesis of a unit root at the 5% significance level. With these unit root evidences, we conclude that ln CO2 t,lngdp t,lngdp 2 t and ln COAL t for China and India are integration of order one, I(1). Similarly, for robustness check, we have also performed the Lumsdaine and Papell [15] unit root test with two structural breaks on each Table 3 The results of unit root tests. Variables ADF test Ng Perron test MZ a MZ t MSB MPT Panel A: Unit root tests for China ln CO2 t (1) D ln CO2 t 4.49 (2) *** *** 4.91 ***.12 *** *** ln GDP t (2) D ln GDP t (1) *** *** 4.64 ***.123 *** *** ln GDP 2 t.518 (2) D ln GDP 2 t 6.77 (1) *** *** 4.94 ***.122 *** *** ln COAL t (2) D ln COAL t () *** ** **.157 ** ** Panel B: Unit root tests for India ln CO2 t () D ln CO2 t () *** *** 3.26 **.155 ** ** ln GDP t.197 () D ln GDP t 5.23 (3) *** *** ***.61 ***.77 *** ln GDP 2 t.291 () D ln GDP 2 t 5.58 (3) *** *** ***.59 ***.659 *** ln COAL t () D ln COAL t () *** *** **.154 ** ** Note: The numbers in the parenthesis () indicate the optimal lag order selected by the Akaike s Information Criterion (AIC). *** Denote the significance level at the 1%. ** Denote the significance level at the 5%.

7 316 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) Table 4 Lumsdaine and Papell [15] unit root test with two breaks. ln CO2 t ln GDP t ln GDP 2 t ln COAL t Model AA Model CC Model AA Model CC Model AA Model CC Model AA Model CC China TB TB tð^k inf Þ DU1 t 3.32 *** 3.59 *** 2.85 *** *** 5.11 *** DU2 t 4.34 *** 5.1 *** ** 2.6 ** 2.29 ** 3.43 *** 2.53 ** DT1 t 5.3 *** 1.71 * 2.59 ** 6.46 *** DT2 t ** 2.6 **.93 Lag length India TB TB tð^k inf Þ DU1 t 2.61 ** 3.14 *** 3.1 *** 3.81 *** 2.98 *** ** 3.23 *** DU2 t 3.96 *** 3.88 *** 2.96 *** ** 3.65 *** 2.47 **.93 DT1 t 2.98 *** 4.52 *** 4.17 *** 3.2 *** DT2 t *** 4.28 *** 1.86 * Lag length Critical values: Model AA: 6.94 (1%), 6.24 (5%) and Model CC: 7.34 (1%), 6.82 (5%) Note: The lag length is determined by Akaike s Information Criterion (AIC). Model AA allow for two breaks in intercept, while Model CC allow for two breaks in both the intercept and slope of the trend function. DU1 and DU2 are the dummy variables for breaks in intercept, while DT1 and DT2 are the dummy variables for trend breaks. *** Denote significance level at 1% level. ** Denote significance level at 5% level. * Denote significance level at 1% level. Table 5 The results of Bayer and Hanck [16] combine cointegration tests. Countries Fisher statistics EG JOH EG JOH BO BDM Conclusion China [1] ** *** Cointegrated [2] ** ** Cointegrated [3] ** *** Cointegrated India [1] Not cointegrated [2] Not cointegrated [3] Not cointegrated Significance level Critical values 1% % % Note: The numbers in the parentheses [ ] indicate the lag order for testing the cointegration. EG JOH is the combination of two cointegration tests proposed by Engle and Granger (EG, [17]) and Johansen (JOH, [18]). However, EG JOH BO BDM is the combination of four cointegration tests proposed by Engle and Granger (EG, [17]), Johansen (JOH, [18]), Boswijk (BO, [19]) and Banerjee et al. (BDM, [2]). *** Denote the significance level at the 1%. ** Denote the significance level at the 5%. variable for China and India (see Table 4). We find that there is structural break(s) in the variables and that strengthen the use of non-linear model for testing the presence of EKC. Moreover, the Lumsdaine Papell unit root test with two structural breaks also confirm that the variables are integrated of order one. As our unit root results reveal that all variables follow the I(1) process, we can proceed to implement the combined cointegration tests proposed by Bayer and Hanck [16]. Table 5 exhibits the results of combined cointegration tests, namely EG JOH and EG JOH BO BDM. Given the cointegration test results are sensitive to the choice of lag order; we perform the combined cointegration tests in three different lag orders. For the case of China, the Fisher statistics for EG JOH and EG JOH BO BDM tests are greater than the 5% critical values regardless of which lag order is employed. Therefore, both EG JOH and EG JOH BO BDM tests consistently reject the null hypothesis of no cointegrating relationship between the variables. However, the combined cointegration results for the case of India totally contradict with the findings in China s case. Both combined cointegration tests fail to reject the null hypothesis of no cointegration regardless of which lag order is employed. Therefore, we surmise that there is a long run equilibrium relationship between CO 2 emissions and its determinants in China, but not for the case of India. 3 Our study adds validity and reconfirms the previous studies for China (i.e. [14,35,36]) and India (i.e. [43]) on the presence of cointegration for China but not for India. One possible explanation in the case of India is that access to energy is still limited to the majority of the population despite an increase in income. In addition, Alam et al. [4] argued that due to the structural nature of India that largely depended on service sectors, less energy input was required. Similarly, the agriculture sector relied more on other production factors such as water supply than energy sources. Indeed, Wolde-Rufael [5] found nuclear energy to be the main drive of economic growth in India Granger causality results At this stage, we investigate the causal relationship between CO 2 emissions, real GDP and coal consumption in China and India. Granger [51] stipulated that if the variables were cointegrated, the error-correction model should be used to examine the direction of causality because it contained both short and long run causality information. Based on our cointegration findings in Table 5, we conducted the Granger causality test for China and India using the VECM and the first difference VAR models, respectively. We used the first difference VAR model to test the Granger causality for India because the variables were not cointegrated. Table 6 reports the Granger causality results of China and India. 3 We also assess the variables with the residual-based test for cointegration with regime shift proposed by Gregory and Hansen [49] for both countries. Likewise, the cointegration inferences are no different. To conserve space, the full results of this cointegration test will not be reported here, but it is available upon request.

8 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) Table 6 The results of Granger causality tests. Null hypothesis Short run Granger causality test Overall long run Granger causality test Panel A: Granger causality tests for China in the VECM framework D ln GDP t; D ln GDP 2 t 9D ln CO 2t *** *** D ln CO 2t 9D ln GDP t; D ln GDP 2 t D ln COAL t9d ln CO 2t *** *** D ln CO 2t 9D ln COAL t *** *** D ln COAL t9d ln GDP t; D ln GDP 2 t *** *** D ln GDP t; D ln GDP 2 t 9D ln COALt ** *** Panel B: Granger causality tests for India in the first difference VAR framework D ln GDP t; D ln GDP 2 t 9D ln CO 2t ** D ln CO 2t 9D ln GDP t; D ln GDP 2 t *** D ln COAL t9d ln CO 2t *** D ln CO 2t 9D ln COAL t *** D ln COAL t9d ln GDP t; D ln GDP 2 t.656 D ln GDP t; D ln GDP 2 t 9D ln COALt ** Note: The optimal lag order is determined by the Akaike s Information Criterion (AIC). *** Denote the significance level at the 1%. ** Denote the significance level at the 5%. In the case of China, as the results depicted in Panel A of Table 6, we found strong evidence of uni-directional Granger causality running from real GDP growth to growth of CO 2 emissions in the short and long run. However, when we tested the Granger causality on real GDP growth and growth of coal consumption, we found evidence of bi-directional Granger causality in the short and long run. Likewise, the results also revealed that growth of CO 2 emissions and growth of coal consumption in China were bi-directional in the short and long run. Our results support Li and Leung [36] that suggested bi-directional causality between GDP and coal consumption and Bloch et al. [37] that suggested bi-directional causality between CO 2 and coal consumption, respectively. Similarly, Wang et al. [14] and Chang s [13] assertion that GDP Granger causes CO 2 emission is also supported. Nevertheless, when comparing with those studies that use aggregate energy consumption data (e.g. Wang et al. [11]), our study also seems to support some of the claims especially with regards to short run causality between energy and GDP, CO 2 and energy and the long run causality between GDP and CO 2. This may be due to the fact that coal represents a larger energy mix of the entire energy consumption and their influences are well captured even if aggregate energy consumption data is used. However, our results also contradict with that of Li et al. [3] and Wolde-Rufael [7] especially with regards to the relationship between GDP and coal consumption. These studies suggested a uni-directional causality running from GDP to coal consumption. Yet, when comparing our results with the previous studies, what is apparent is that there is some form of consensus on the direction of causality. The consensus warrants policymakers to further validate the direction of causality between CO 2, coal consumption and GDP in China for future energy planning. Contradictory to the findings in China, we found that real GDP growth and growth of CO 2 emissions for India (Panel B, Table 6) were bi-directional Granger causality in the short run. In addition, growth of CO 2 emissions and growth of coal consumption also Granger cause each other in the short run (i.e. bi-directional causality). Nevertheless, we could only find evidence of uni-directional Granger causality running from real GDP growth to growth of coal consumption in the short run. The results of our study are consistent with Ghosh [43] who found no cointegration between the variables and bi-directional causality between GDP and CO 2. It also supports Alam s et al. [4] findings that GDP Granger caused coal consumption. However, it contradicts with the findings of Wolde-Rufael [7] that found uni-directional causality running from coal consumption to economic growth. This might be due to the application of cross country and panel data analysis. The preferred country specific analysis in this study captures and accounts for the complexity of economic environment and histories of energy development in China and India respectively, of which panel analysis is unable to capture. A more country-specific analysis such as ours is needed to provide consistent results [52]. As a whole, a consensus emerges in that there is no long run relationship between the variables in the case of India and as such policymakers can be confident that there will not be any long run implication of the energy conservation policies for India. In contrast, based on the findings of the Granger causality tests, we may summarise that the degradation of environment will affect the process of economic growth and development in India, but this does not apply to China. Therefore, more environmental protection policies can be applied to China compared to India. Furthermore, the results of this study also show that China is a coal-dependent country whereas India is a coal-independent country. It is therefore a certainty that coal conservation policies will deteriorate the development process in China. Nonetheless, the coal conservation policies will effectively reduce the CO 2 emissions because coal consumption Granger causes CO 2 emissions in these two countries. 5. Conclusion and policy implications This study attempts to investigate the relationship between CO 2 emissions, economic growth and coal consumption in China and India using newly-developed cointegration analysis for an extended time period, i.e. from There exists a long-run relationship between the variables in the case of China but not for India. Bi-directional causality, in the short and long run, is detected between economic growth and coal consumption as well as between coal consumption and CO 2 emissions in China. In addition, uni-directional causality is detected from economic growth to CO 2 emissions. For India, a short-run bi-directional causality exists between economic growth and CO 2 emissions and CO 2 and coal consumption. However, economic growth Granger causes coal consumption in the short run in India. Since it is confirmed that there are consensus in results (our study and others), the policy message is clear for both China and India. China should be cautious in implementing any conservation policy while India should implement the policy without a destabilisation of a long run economic growth. As a whole, China may have to put in more effort to devise alternative choices of policy options than India. Since coal consumption impacts the CO 2 emissions and economic growth in the long run, any coal conservation policy might reduce CO 2 emissions but has negative consequences on economic growth in China. Since China s electricity generation is mostly from coal (78% of the total capacity in 27: [53]), any reduction in coal consumption will adversely affect its electricity supply. One policy option is to improve coal utilisation efficiency. However, this viable policy option will be challenging in that the Chinese government has to devise policies to improve the efficiency options to increase the GDP-coal intensity. In this way, China will be able to reduce the CO 2 emissions without any adverse effect on economic growth. Another viable policy option is to increase the consumption of renewable energy. Fang [54] claimed that China has the potential of increasing renewable energy; however, it requires market openness, institutional innovation and policy integration. With respect to this, investment and institutional arrangements should be intensified to speed up the development of renewable energy sectors.

9 318 V.G.R. Chandran Govindaraju, C.F. Tang / Applied Energy 14 (213) Acknowledgement We would like to thank the two anonymous reviewers because their comments and suggestions have significantly improved the earlier version of this research. In addition, our appreciation also goes to Christoph Hanck for sharing knowledge and programming codes. The remains flaws in this research are solely our responsibility and the usual disclaimer applies. References [1] Apergis N, Payne JE. The causal dynamics between coal consumption and growth: evidence from emerging market economies. Appl Energy 21;87: [2] Singh K. India s emissions in a climate constrained world. Energy Policy 211;39: [3] Li J, Song H, Geng D. Causality relationship between coal consumption and GDP: difference of major OECD and non-oecd countries. Appl Energy 28;85: [4] Garg A, Shukla PR. Coal and energy security for India: role of carbon dioxide (CO 2 ) capture and storage (CCS). Energy 29;34: [5] EIA. CO 2 emissions from fuel combustion: highlights. Washington, DC: US Energy Information Administration, US Department of Energy; 211. [6] Parikh J, Parikh K. India s energy needs and low carbon options. Energy 211;36: [7] Wolde-Rufael Y. Coal consumption and economic growth revisited. Appl Energy 21;87:16 7. [8] Lu W, Ma Y. Image of energy consumption of well off society in China. Energy Convers Manage 24;45: [9] Rout UK, Vob A, Singh A, Fahl U, Blesl M, Gallachóir BPO. Energy and emissions forecast of China over a long-time horizon. Energy 211;36:1 11. [1] EIA. International Energy Outlook. Washington, DC: US Energy Information Administration, US Department of Energy; 211. [11] Zhang X-P, Cheng X-M. Energy consumption, carbon emissions, and economic growth in China. Ecol. Econ. 29;68: [12] Jalil A, Mahmud SF. Environment Kuznets curve for CO2 emissions: a cointegration analysis for China. Energy Policy 29;37: [13] Chang C-C. A multivariate causality test of carbon dioxide emissions, energy consumption and economic growth in China. Appl Energy 21;87: [14] Wang SS, Zhou DQ, Zhou P, Wang QW. CO2 emissions, energy consumption and economic growth in China: a panel data analysis. Energy Policy 211;39: [15] Lumsdaine RL, Papell DH. Multiple trend breaks and the unit-root hypothesis. Rev Econ Stat 1997;79: [16] Bayer C, Hanck C. Combining non-cointegration tests. Maastricht Research School of Economics of Technology and Organization, Maastricht University; 21. [17] Engle RF, Granger CWJ. Co-Integration and error correction: representation, estimation, and testing. Econometrica 1987;55: [18] Johansen S. Statistical analysis of cointegration vectors. J Econ Dynam Control 1988;12: [19] Boswijk HP. Testing for an unstable root in conditional and structural error correction models. J Econom 1994;63:37 6. [2] Banerjee A, Dolado JJ, Mestre R. Error-correction mechanism test for cointegration in a single-equation framework. J Time Ser Anal 1998;19: [21] Karanfil F. How many times again will we examine the energy-income nexus using a limited range of traditional econometric tools? Energy Policy 29;37: [22] Podivinsky JM. Testing misspecified cointegration relationships. Econ Lett 1998;6:1 9. [23] Boswijk HP, Franses PH. Dynamic specification and cointegration. Oxford Bull Econ Stat 1992;54: [24] Gregory AW, Haug AA, Lomuto N. Mixed signals among tests for cointegration. J Appl Econom 24;19: [25] Pesavento E. Analytical evaluation of the power of tests for the absence of cointegration. J Econom 24;122: [26] Solow RM. Applying growth theory across countries. World Bank Econom Rev 21;15: [27] Athukorala P, Sen K. Saving, investment, and growth in India. Oxford: Oxford University Press; 22. [28] Robertson D, Symons J. Some strange properties of panel data estimators. J Appl Econom 1992;7: [29] Pesaran MH, Smith R. Estimating long-run relationship from dynamic heterogeneous panels. J Econom 1995;68: [3] Maddala GS, Trost RP, Li H, Joutz F. Estimation of short-run and long-run elasticities of energy demand from panel data using shrinkage estimators. J Bus Econ Stat 1997;15:9 1. [31] Deaton AS. Saving in developing countries: theory and review. In: Fisher S, de Tray D, editors. Proceedings of the World Bank Annual Conference on Development Economics. Washington, DC: World Bank; [32] Pesaran MH, Shin Y, Smith R. Pooled mean group estimation of dynamic heterogeneous panels. J Am Stat Assoc 1999;94: [33] Pedroni P. Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxford Bull Econ Stat 1999;61: [34] Acaravci A, Ozturk I. On the relationship between energy consumption, CO 2 emissions and economic growth in Europe. Energy 21;35: [35] Li F, Dong S, Li X, Liang Q, Yang W. Energy consumption-economic growth relationship and carbon dioxide emissions in China. Energy Policy 211;39: [36] Li R, Leung GCK. Coal consumption and economic growth in China. Energy Policy 212;4: [37] Bloch H, Rafiq S, Salim R. Coal consumption, CO 2 emission and economic growth in China: empirical evidence and policy responses. Energy Econom 212;34: [38] Yuan J-H, Kang J-G, Zhao C-H, Hu Z-G. Energy consumption and economic growth: evidence from China at both aggregated and disaggregated levels. Energy Econom 28;3: [39] Ang JB. CO 2 emissions, research and technology transfer in China. Ecol Econ 29;68: [4] Alam MJ, Begum IA, Buysse J, Rahman S, Van Huylenbroeck G. Dynamic modeling of causal relationship between energy consumption, CO 2 emissions and economic growth in India. Renew Sustain Energy Rev 211;15: [41] Kulshreshtha M, Parikh JK. Modeling demand for coal in India: vector autoregressive models with cointegrated variables. Energy 2;25: [42] Paul S, Bhattacharya RN. Causality between energy consumption and economic growth in India: a note on conflicting results. Energy Econom 24;26: [43] Ghosh S. Examining carbon emissions economic growth nexus for India: a multivariate cointegration approach. Energy Policy 21;38: [44] Song T, Zheng T, Tong L. An empirical test of the environmental Kuznets curve in China: a panel cointegration approach. China Econ Rev 28;19: [45] Managi S, Jena PR. Environmental productivity and Kuznets curve in India. Ecol Econ 28;65: [46] Elliott G, Jansson M, Pesavento E. Optimal power for testing potential cointegrating vectors with known parameters for nonstationary. J Bus Econ Stat ;23: [47] Maddala G, Kim I. Unit roots, cointegration, and structural change. Cambridge University Press; [48] Ng S, Perron P. Lag length selection and the construction of unit root tests with good size and power. Econometrica 21;69: [49] Gregory AW, Hansen BE. Residual-based tests for cointegration in models with regime shifts. J Econom 1996;7: [5] Wolde-Rufael Y. Bounds test approach to cointegration and causality between nuclear energy consumption and economic growth in India. Energy Policy 21;38:52 8. [51] Granger CWJ. Some recent development in a concept of causality. J Econom 1988;39: [52] Chandran VGR, Sharma S, Madhavan K. Electricity consumption-growth nexus: the case of Malaysia. Energy Policy 21;38: [53] Pitman R, Zhang VY. Electricity restructuring in China: how competitive will generation market be? Singapore Econ Rev 21;55: [54] Fang Y. Economic welfare impacts from renewable energy consumption: the China experience. Renew Sustain Energy Rev 211;15:512 8.