Renewable and Sustainable Energy Reviews: Ping-Yu Chen, Sheng-Tung Chen, Chia-Sheng Hsu, Chi-Chung Chen [PDF]

Citation preview

Renewable and Sustainable Energy Reviews 65 (2016) 420–431



Contents lists available at ScienceDirect



Renewable and Sustainable Energy Reviews journal homepage: www.elsevier.com/locate/rser



Modeling the global relationships among economic growth, energy consumption and CO2 emissions Ping-Yu Chen a, Sheng-Tung Chen b, Chia-Sheng Hsu a, Chi-Chung Chen a,n a b



Department of Applied Economics, National Chung-Hsing University, Taichung, Taiwan Department of Public Finance, Feng-Chia University, Taichung, Taiwan



art ic l e i nf o



a b s t r a c t



Article history: Received 1 October 2014 Received in revised form 18 April 2016 Accepted 28 June 2016



The relationships among economic growth, energy consumption, and greenhouse gas emissions needs to be paid attention since the characteristics of global warming is from economy and greenhouse gas emissions. This study employs a panel cointegration and vector error-correction model to discuss the dynamic economy-energy-environment nexus for 188 countries for the periods of 1993–2010. The empirical results indicate that there exist long-run relationships between economic growth, energy consumption and carbon dioxide emissions for all countries. It is worth noting that energy consumption negatively affects GDP in the world as a whole and developing countries, but not in developed countries. And the unidirectional causality from energy consumption to carbon dioxide emissions exists both on developing and developed countries. That means the developed countries should take more responsibility on the energy efficiency and mitigating Greenhouse Gas Emissions (GHGE). On other hand, environmental regulations related to the prevention of environmental degradation as the economy grows need to be adopted by all countries. & 2016 Published by Elsevier Ltd.



Keywords: Energy consumption Economic growth CO2 emission



Contents 1. 2. 3.



Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Cross-section dependence test and panel unit root test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Panel Cointegration Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Panel long run estimates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Dynamic Panel Granger Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. Conclusions and policy implications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .



1. Introduction Greenhouse gas (GHG) emissions have sharply increased following the industrial development in both developed and developing countries. In particular, carbon dioxide (CO2) emissions account for over half of GHG emissions which are likely to be related to climate



n



change [102,54,98]. The IPCC [54] indicated that GHG emissions have increased by 1.6% annually, while CO2 emissions chiefly arise due to the use of fossil fuels and have also increased by 1.9% annually over the past three decades. Moreover, the IPCC [54] predicted that GHG emissions in 2030 will have increased by 25–90% as compared with the year 2000, and energy-related CO2 emissions in 2030 will have



Correspondence to: Department of Applied Economics, National Chung-Hsing University, 250, Kuo Kuang Rd., Taichung 40246, Taiwan. E-mail addresses: [email protected] (P.-Y. Chen), [email protected] (S.-T. Chen), [email protected] (C.-S. Hsu), [email protected], [email protected] (C.-C. Chen). http://dx.doi.org/10.1016/j.rser.2016.06.074 1364-0321/& 2016 Published by Elsevier Ltd.



420 421 422 422 422 422 424 425 426 429 430



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



increased by 40–110%. How to face the threat of climate change has become a priority task worldwide. To protect human life and health from dramatic climate change, the Kyoto protocol in 1997 set a goal to reduce GHG emissions in developed countries to 5.2% below the 1990 level during 2008– 2012, and has executed such an agreement since 2005. However, the Kyoto Protocol was amended to be extended to 2020 at the 2012 United Nations’ Climate Change Conference. Japan, Canada and Russia have withdrawn from the agreement, and the United States has not yet ratified this Protocol. Besides, the GHG emissions of the countries that have signed the Kyoto Protocol Extension only account for 15% of annual global GHG emissions. On the other hand, developing countries also have the responsibility to reduce GHG emissions. Recently, major developing countries such as Brazil, China, India, India, Malaysia, Mexico, the Philippines, South Africa, Thailand and Turkey have rapidly grown in terms of their energy use and this has resulted in an increase in GHG emissions. The amount of CO2 emissions from energy use in the above developing countries accounted for about 13.69% of the world's emissions in 1980, but this share had grown to 40.76% in 2009 [55]. Under the pressure of mitigating climate change, the developing countries are faced with a dilemma of increasing economic growth or reducing energy consumption and GHG emissions [101]. Adamantiades and Kessides [2], DeCanio [31] and Reddy and Assenza [90] have indicated that if we do not take action to solve the problem of climate change due to global warming, there will not only be economic loss but also an environmental disaster. Therefore, how to formulate a corresponding policy to maintain economic growth with stable economic development while addressing the issues related to CO2 emissions is extremely important. Therefore, firstly, the energy-economy-environment nexus issue should be focused on the world as a whole instead of an individual country or region due to the characteristics of global warming and greenhouse gas emissions. This is why we will utilize a data set encompassing 188 countries to investigate this issue. Moreover, in view of the fact that each country has alternative technologies related to energy consumption, this study also takes account of this nexus issue for different levels of economic development. Hence, both the developed and developing countries are considered here to develop suitable corresponding policies and the empirical results show that such relationships in the energyeconomy-environment nexus will be different among countries at different stages of economic development. Recent research from Akhmat et al. [5] and Liddle and Lung [66] also discuss the energyeconomic nexus in different economic development countries. Finally, in considering a possible problem that arises with panel data, namely, cross-sectional dependence, we apply the recent empirical estimation approach in view of the cross-sectional dependence to acquire more reliable estimates. The remaining parts of this study are arranged as follows. Section 2 present the literature review to introduce prior research related to the environment-energy-economic nexus. Section 3 describes the methodology that employs two econometric models to estimate the relationships among the economy, energy and the environmental nexus and the data sets are also analyzed in this section. Section 4 presents the empirical results for all of the estimation procedures and the Granger causality tests. Explanations for these three variables are provided in Section 5. The final section offers conclusions and corresponding policy recommendations.



2. Literature review The traditional environmental and economic nexus may be found in the Environmental Kuznets Curve (EKC), which provides the theoretical background and empirical analysis for this



421



relationship. The literature principally concentrates on testing the validity of the EKC hypothesis, which sets emissions of environmental pollutants to be a function of income, and the relationship between income and environmental quality is depicted by an inverted U-shaped curve. Since Grossman and Krueger [43] began estimating the EKC hypothesis, a significant number of studies have assessed this context [3,32,37,38,47,48,50,69,70,74,91,93,97]. Nevertheless, the EKC hypothesis lacks feedback on the effects of environmental pollutants on economic growth. Coondoo and Dinda [30], Dinda and Coondoo [32], Akbostanci et al. [4], Lee and Lee [64], Fodhaa and Zaghdoud [36], Esteve and Tamarit [35], and Sephtona and Mann [94] resolve this issue by examining the dynamic causality between economic growth and environmental pollutants. The EKC hypothesis not only deals with the issue of the unidirectional causality from economic growth to environmental pollutants, but also focuses on the problem of omitted variables bias as well as the reverse effects [12,99]. On the other hand, the economic growth and energy consumption nexus has been investigated by a great number of studies. For example, Kraft and Kraft [62] is the original empirical study on the relationship between output and energy consumption, and has been followed by numerous studies (see Chen et al. [29] for a recent review) that have started to apply Granger causality tests to assess the economic growth/energy consumption nexus. Chen et al. [29] survey 174 studies using a meta analysis with a multinomial logit model to examine the relationship between economic growth and energy consumption. They have demonstrated how the time spans, the subject selections including GDP and energy consumption, the econometric models, and the tools regarding the greenhouse gases emission reduction characteristics significantly affect these controversial outcomes. This implies that economic growth, energy consumption, and greenhouse gas emissions need to be examined simultaneously. By combining the economic growth and energy consumption nexus with the economic and environmental nexus, Ang [12] and Soytas et al. [99] were the first to explore economic growth, energy consumption, and environmental pollutant emissions simultaneously within a Granger causality multivariate framework, and formed a new area of research. Recent studies, for instance, Ang [13], Halicioglu [44], Jalil and Mahmud [57], Soytas and Sari [98], Zhang and Cheng [108], Menyah and Wolde-Rufael [71,72], Acaravci and Ozturk [1], Chang [27], Hatzigeorgiou et al. [46], Bloch et al. [21], Jayanthakumaran et al. [58], Govindaraju and Tang [41], and Saboori and Sulaiman [92] have all empirically tested the relationships between economic growth, energy consumption, and environmental pollutant emissions for a single country. Normally, the macroeconomic variables data cover only a small or medium sample span. However, unit root tests, cointegration tests and causality tests require a large span of data, for a small or medium sample span significantly reduces the power of these tests [8,39]. Therefore, panel data provide more informative data, more degrees of freedom, and more efficient estimates. To the best of our knowledge, there has so far been no investigation that has estimated the economic growth, energy consumption, and environmental pollution nexus using global data sets. As for the directions of the causal relationships between energy consumption and economic growth, Ozturk and Acaravic [76] and Payne [83] have briefly categorized these into four types, no causality, uni-directional causality running from economic growth to energy consumption, uni-directional causality running from energy consumption to economic growth, and bi-directional causality between energy consumption and economic growth. Although the controversial and inconsistent conclusions from the literature have not been resolved, the relationship between income and energy consumption may be described as one of these four types of relationships. We also observe that such alternative conclusions or



422



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



these four types of relationships within the literature may be due to the subject/country selections, data time spans, empirical econometric model settings or other explanatory variable selections. However, bi-directional causality between energy consumption and economic growth is found to exist when the global data set is applied. Such findings are consistent with the study by Chen et al. [29] which has analyzed the relationship between energy consumption and economic growth based on the findings of 174 studies. The other major reason why we apply a global data set instead of data for a single country or a set of pooling countries is due to the characteristics of global warming and CO2 concentrations when the issue of economic growth, energy consumption and the environmental pollution nexus is investigated. Since the combustion of fossil fuels has induced an increase in CO2 concentrations from 280 ppm to 390 ppm, and the greenhouse effect has resulted in global warming and caused economic losses worldwide, such logic may be suitable from a global point of view instead of for an individual country or a regional area. For instance, the increase in CO2 emissions due to the increase in energy consumption could occur in any country, but the effects of climate change due to an increase in greenhouse gas emissions is a global issue. This implies that exploring the relationships among energy use, economic growth and the environment should be considered from a global perspective. Therefore, this study investigates the dynamic relationships among economic growth, energy consumption and CO2 emissions both for the world as a whole, for developed countries and for developing countries during the period from 1993 to 2010 using newly-developed econometric models.



3. Methods Such a nexus in terms of the relationships among energy, the economy and the environment could be investigated for the whole world, developed countries, and developing countries, which are referred to in Panel A, Panel B, and Panel C, respectively. Owing to the characteristics of our data with large numbers of individuals (i.e., N) and with small time periods (i.e., T), a recently developed econometric method is applied to verify the validity of the estimators in the empirical panel data model. Three estimation procedures are established as follows. First, we employ the panel unit root test [89] to inspect the stationarity and the order of integration for the variables. Then, we adopt the panel cointegration test [40] to certify whether these variables possess cointegration relationships while the variables have the same order of integration [59,68,84,85]. When a panel cointegration relationship is found to exist between any two of these variables, an estimation using common correlated effects (hereafter CCE) is conducted in assessing the long-run relationships between these variables. Finally, we construct a panel vector error correction model and the generalized method of moments (hereafter GMM) approach is employed to test the dynamic panel Granger causality. 3.1. Cross-section dependence test and panel unit root test The traditional panel unit root tests such as the LLC test [65] and IPS test [53] loosen the homogeneous form of the individual deterministic effects and serial correlation structure of the error terms, but assume that the cross-sections are independent. Pesaran [87] has pointed out that the cross-sectional dependence (hereafter CSD) could arise because of unobserved common factors, externalities, regional and macroeconomic linkages, and unaccounted for residual interdependence. As ignoring CSD would likely result in imprecise estimates and a severe identification issue, the Pesaran [87] crosssectional dependence test (hereafter CD test) could have better power under the circumstance where N 4 T (as fits our research)



with balanced or unbalanced panel data. Here, we consider the Pesaran [89] Cross-Sectionally Augmented IPS (CIPS) test which relaxes the assumption of cross-sectional independence to test the stationarity of these variables and simultaneously allow for the heterogeneity. His Cross-Sectionally Augmented Dickey-Fuller (hereafter CADF) regression based on the OLS method for the ith cross-section in the panel is expressed as follows: k



Δx it = α i0 + α i1t + βi1x i, t − 1 + βi2 xt − 1 +



k



∑ dij Δxt − j + ∑ δ ij Δx i, t − j + εit j =0



j =0



(1)



where xit could represent all variables including CO2, EC, real GDP N N and the square of real GDP. xt = (1/N ) ∑i = 1 xit , Δxt = (1/N ) ∑i = 1 Δxit , and εit is the regression error. The optimal lag-length is selected by the Akaike Information Criterion (hereafter AIC) or the SchwarzBayesian information criterion (hereafter SBC). The CIPS statistic is a cross-sectional average of ti (N , T ), which is shown as follows. N



CIPS = ( 1/N ) ∑ ti (N , T )



(2)



i=1



where ti(N,T) is the t-statistic for the estimate of βi1 in Eq. (1) which is used to compute the individual ADF statistics. Pesaran supposes that the CIPS statistic is based on the average of the individual CADF statistics. 3.2. Panel Cointegration Test Following the findings that CO2, EC, Real GDP, and the square of real GDP in the three panels are all integrated of order 1. In the past few years, a number of studies [104–107,18,19,28,40,42,73,86] have addressed the panel cointegration tests by allowing for error cross-sectional dependence (hereafter CSD). The test proposed by Bai and Kao [18] gives consideration to CSD but no consideration to cross-sectional heterogeneity [49]. The panel cointegration test with the conditional panel error correction model (hereafter ECM) by Gengenbach et al. [40] has postulated a common factor construction for CSD which will provide more robust outcomes. Gengenbach et al. [40] panel cointegration test employs two tests, the t-statistic (tαi ) and the Wald test ( wδi ) for the null hypothesis of no error correction. 3.3. Panel long run estimates ∂Y



X̅



According to the elasticity formula: η = ∂X × Y ̅ , where Y is the dependent variable and X is the independent variable of the estimate of Eq. (1a), (1b) and (1c). For instance, we can get the elasticity from



∂CO2 ∂EC



×



EC , CO2



where EC and CO2 are the averages of



energy consumption per capita and CO2 emissions per capita for 188 countries during the period from 1993 to 2010. Two empirical econometric models will be constructed in this study. The first one investigates the long-run relationships among economic growth, energy consumption and CO2 emissions in the world following the studies by Ang [12] and Apergis and Payne [14]. The second one examines the Granger causality among these three variables by applying the panel error correction model (hereafter ECM) as in Gengenbach et al. [40]. The long-run relationships among CO2, energy consumption and real GDP are established based on Ang [12] and Apergis and Payne [14] and are expressed as follows:



CO2it = f (ECit , GDPit , GDPit2 ) + ε1it



(3a)



EC2it = g (CO2it , GDPit , GDPit2 ) + ε2it



(3b)



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



GDPit = h (CO2it , ECit ) + ε3it



(3c)



where i ¼1, 2, …, N denotes the country, t ¼1, …, T denotes the time period, CO2 represents CO2 emissions per capita (calibrated in metric tons), EC is primary energy consumption per capita (calibrated in kgs of oil equivalent), and GDP is real GDP per capita (calibrated in 2005 US dollars). The second empirical model is referred to as the expanded conditional ECM following the approach by Gengenbach et al. [40] and is expressed as follows: q



q



∑ β11ik ΔCO2it − k + ∑ β12ik ΔECit − k



ΔCO2it = α1i +



k=1 q



k=1 q



∑ β13ik ΔGDPit − k + ∑ β14ik ΔGDPit2− k



+



k=1 q



k=1 q



q



∑ ψ11ik ΔCO2t − k + ∑ ψ12ik ΔECt − k + ∑ ψ13ik ΔGDPt − k



+



k=1 q



∑



+



k=1 2 ψ14ik ΔGDPt − k



k=1



+ γ1i ECTit − 1 + ϕ11i CO2t − 1



k=1 2



+ ϕ12i ECt − 1 + ϕ13i GDPt − 1 + ϕ14i GDPt − 1 + ε1it



q



q



∑ β21ik ΔCO2it − k + ∑ β22ik ΔECit − k



ΔECit = α2i +



k=1 q



+



∑ β23ik ΔGDPit − k + ∑ β24ik ΔGDPit2− k k=1 q



∑ ψ11ik ΔEC2t − k + ∑ ψ12ik ΔCO2t − k k=1 q



+



k=1 q



k=1 q



+



(4a)



k=1 q



∑ ψ13ik ΔGDPt − k + ∑ ψ14ik ΔGDPt2− k k=1



k=1



+ γ2i ECTit − 1 + ϕ21i EC2t − 1 + ϕ22i CO2t − 1 2



+ ϕ23i GDPt − 1 + ϕ24i GDPt − 1 + ε2it



q



ΔGDPit = α3i +



q



∑ β31ik ΔCO2it − k + ∑ β32ik ΔECit − k k=1



q



+



k=1 q



∑ β33ik ΔGDPit − k + ∑ ψ11ik ΔGDPt − k k=1 q



+



(4b)



k=1 q



∑ ψ12ik ΔCO2t − k + ∑ ψ13ik ΔECt − k k=1



k=1



423



living, we discuss the GDP, energy consumption and CO2 emissions nexus between developed and developing countries, as well as for the 188 countries as a whole. The model is based on eight indicators of the degree of development, namely, HIE OECD, DAC, CIA AE, IMF AE, WB HIE, HDI VH, Qol Top 30 and CDI CD, where the HIE OECD index determines whether the country belongs to the high-income OECD member economies. The DAC index refers to whether the country is a member of the Development Assistance Committee. The CIA AE index refers to whether the Central Intelligence Agency views the country as a high-income economy. The IMF AE index indicates whether the International Monetary Fund regards the country as having a high level of economic development. The WB HIE index refers to whether the World Bank views the country as a high-income economy. The HDI VH index indicates whether the country has a very high human development index (HDI 40.6). The Qol Top 30 index refers to whether the country ranks among the top 30 in terms of the quality of life. The CDI CD index refers to whether the country is defined as a developed country by the Center for Global Development. We have found that 26 countries meet the requirements of these 8 indexes as developed countries and they are Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Japan, South Korea, Luxembourg, the Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sweden, Switzerland, the United Kingdom, and the United States. The other 162 countries are viewed as developing countries. Table 1 displays the descriptive statistics of real GDP per capita, energy consumption per capita, and CO2 emissions per capita for all 188 countries, developed countries, and developing countries. The means of CO2 emissions per capita for all countries, developed countries, and developing countries are 4.886, 9.344, and 3.732 metric tons, respectively, and the variance in the developing countries is the largest among the three panels. The means of energy consumption per capita in all countries, developed countries, and developing countries are 2353.33, 4412.90, and 1708.90 kt of oil equivalent, respectively, and the variance in developing countries is larger than in developed countries. In terms of real GDP, the means of real GDP per capita in all countries, developed countries, and developing countries are $9323, $31269, and $4035 respectively, and the standard deviation in developed countries is much greater than in developing countries. From the basic descriptive statistics, we find that the variation in the data between developed and developing countries is quite different, which implies that it is necessary to classify the panel data set based on the level of economic development when discussing such a relationship nexus.



+ γ3i ECTit − 1 + ϕ31i GDP2t − 1 + ϕ32i CO2t − 1 + ϕ33i ECt − 1 + ε3it



(4c)



where ECT is the error term that is based on the residuals of the long-run equilibrium relationships between variables using the CCEP estimator, Δ is the first-difference of the variables, q is the optimal lag-length, and the εit represent the error terms with i.i.d. The panel dynamic ECM focuses on two causal relationships: the short-run causality through the lagged dynamic terms and the long-run causality through the ECT terms. To implement Eqs. (3a) to (3c) and (3a) to (3c), a global panel data set needs to be collected. Data sets including GDP per capita (calibrated in 2005 US dollars), CO2 emissions per capita (calibrated in metric tons), and energy consumption per capita (calibrated in kgs of oil equivalent) have been collected. These data sets are from 188 countries with time periods covering the years from 1993 to 2010 and are collected from World Development Indicators. In considering different levels of income, the level of industrialization, the amount of infrastructure, and the standard of



Table 1 Descriptive statistics. Mean



Std. Dev.



Min



Max



Panel A: all countries CO2a 4.886 EC 2353.336 Real GDP 9323.292



6.805 2748.986 14146.000



0.010 58.081 50.042



68.535 22903.378 87716.732



Panel B: developed countries CO2 9.344 EC 4412.905 Real GDP 31268.960



4.441 2207.152 15630.380



1.416 1530.141 6821.645



41.058 17982.521 87716.732



Panel C: developing countries CO2 3.732 EC 1708.901 Real GDP 4035.089



6.835 2590.465 6743.217



0.010 58.081 50.042



68.535 22903.378 57558.719



Note: a metric tons per capita for CO2, kt of oil equivalent per capita for EC, 2000 US dollars per capita for Real GDP.



424



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



4. Results



Table 3 Pesaran's CIPS panel unit root tests.



To identity whether the variables are characterized by CSD, we utilize the Pesaran [87] CD test, and the results are shown in Table 2. We could realize that all the series are cross-sectionally correlated while the Pesaran [87] CD statistics for the variables in the three panels all reject the null hypothesis of cross-sectional independence. Table 3 displays the results of Pesaran [89]’s panel unit root test with a lag of 4, which is grounded on the AIC when choosing the optimal lag-length. In terms of the critical values of CIPS presented in Pesaran [89], the 5% (10%) critical values for the CIPS statistics are  2.11(  2.03) with an intercept, and  2.62 (  2.54) with an intercept and a trend. For CO2, EC, Real GDP, and the square of real GDP in the three panels, the unit root hypotheses with an intercept and with an intercept and trend are not rejected. For the first differences of CO2, EC, Real GDP, and the square of real GDP in the three panels, the unit root hypotheses with an intercept and with an intercept and trend are rejected. Hence, we conclude that these variables in the three panels are all integrated of order 1. In addition, the panel cointegration test result can be found in Table 4. Following the critical values provided by Gengenbach et al. [40] for both statistics, we could find that both the t values and the Wald test statistics in the three panels reject the null hypothesis. In other words, there exist cointegration relationships between CO2, EC and GDP in 188 countries, developed countries, and developing countries as shown in Table 4. The long-run estimates of these variables including CO2, EC and GDP from Eqs. (3a) to (3c) will be estimated using common correlated effects pooled (hereafter CCEP) estimators proposed by Pesaran [88] and Kapetanios et al. [60]. Such estimators will be valid for the case where N is larger than T and could provide the unobserved common factors and cross-sectional dependence. The estimation outcomes are shown in Table 5. In considering the long-run effects of energy consumption and economic growth on CO2 emissions in the three panels as shown in the first row of Table 5, the increase in energy consumption and economic development have significantly positive effects on CO2 emissions in all three panels. Using the elasticity formula of ∂CO2 ∂EC



×



EC , CO2



where EC and are the averages of energy consumption



per capita and CO2 emissions per capita for 188 countries during the period from 1993 to 2010, we found that a 1% increase in energy consumption per capita induces 13.5%, 7.6% and 26.1% increases in CO2 emissions per capita in the world, and developed Table 2 Cross-sectional dependence test. Variable



CO2



Panel A: All countries 0.424 ρ^ε a CD-testb p-value



23.084 0.000



Panel B: Developed countries 0.439 ρ^ε CD-test p-value



30.115 0.000



Panel C: Developing countries 0.408 ρ^ε CD-test p-value



32.217 0.000



EC



GDP



GDP2



0.481



0.769



0.772



69.645 0.000



219.946 0.000



221.465 0.000



0.423



0.900



0.897



22.473 0.000



82.741 0.000



82.449 0.000



0.490



0.722



0.727



53.453 0.000



139.842 0.000



141.932 0.000



Note: a



ρ^ε is the average absolute cross-section correlation of errors ( ε^iy ) where xit = α0i + α1i t + α1i xit − 1 + εit . b Pesaran's CD-statistics test the null hypothesis of independent errors.



EC



GDP



GDP2



 2.054  1.889  1.987



 1.867  1.898  1.848



 1.936  1.842  1.922



Variables in levels: intercept and trend Panel A  1.898  2.001 Panel B  1.853  1.863 Panel C  1.845  1.929 Variables dCO2 dEC



 1.788  1.746  1.738 dGDP



 1.776  1.765  1.821 dGDP2



Variables in first differences: intercept Panel A  3.276**  3.361** Panel B  3.129**  3.209** Panel C  3.076**  3.155**



 3.004**  2.552**  2.849**



 2.793**  2.854**  3.185**



Variables in first differences: intercept and trend Panel A  2.666**  2.652** Panel B  2.637**  2.550* Panel C  2.594*  2.753**



 2.673**  2.609*  2.775**



 2.787**  2.658**  2.631**



Variables



CO2



Variables in levels: intercept Panel A  1.971 Panel B  2.004 Panel C  1.950



Note: (1). The 5% (10%) critical values for the CIPS statistics are  2.11(  2.03) with an intercept, and  2.62 (  2.54) with an intercept and a trend. * **



denotes significance at the 10% level, and denotes significance at the 5% level.



Table 4 ECM panel cointegration test. Dependent variable: CO2 Intercept Panel Aa Panel Bb Panel C Intercept and trend Panel A Panel B Panel C



t¯*  3.180**  4.029**  3.202** t¯*  3.203*  3.371**  3.285**



¯* w 21.295** 23.257** 20.394** ¯* w 17.886** 20.220** 19.388**



Avg. lag length 1.9 2.0 2.1 Avg. lag length 1.8 2.0 1.9



Dependent variable: EC Intercept Panel A Panel B Panel C Intercept and trend Panel A Panel B Panel C



t¯*  3.692**  4.251**  3.087** t¯*  3.254**  3.473**  3.239**



¯* w 21.779** 18.015** 18.063** ¯* w 16.867** 16.902* 17.524**



Avg. lag length 2.0 2.0 2.1 Avg. lag length 1.8 2.0 2.0



Dependent variable: GDP Intercept Panel A Panel B Panel C Intercept and trend Panel A Panel B Panel C



t¯*  4.273**  3.396**  3.528** t¯*  3.320**  3.362**  3.228*



¯* w 18.015** 16.611** 17.044** ¯* w 16.809** 17.063** 16.854**



Avg. lag length 2.2 1.9 2.0 Avg. lag length 2.0 2.0 2.0



Note: a The 5% (10%) critical values for the τ¯* and statistics for Panels A and C are  2.886 (  2.835) and 14.150 (13.866) with an intercept, and  3.237 (  3.193) and 16.268 (15.993) with an intercept and a trend. b The 5% (10%) critical values for the τ¯* and statistics for Panel B are  3.000 (  2.925) and 14.871 (14.394) with an intercept, and  3.361 (  3.286) and 17.023 (16.568) with an intercept and a trend. * denotes significance at the 10% level, and ** denotes significance at the 5% level.



and developing countries, respectively. These figures imply that CO2 emissions per capita will be positively affected by energy consumption per capita but the developed countries have higher energy technology such as cleaner energy used than developing



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



Table 5 Panel long-run estimates. Panel



A



B



C



Dependent variable: CO2 EC



0.028*** (3.518)



0.016** (2.276)



0.057*** (3.940)



GDP



0.007*** (2.429)



0.007** (2.406)



0.005** (2.171)



GDP2



 2.92E-07** (  2.334) 0.497



 1.08E-07** (  2.088) 0.701



 2.62E-07** (  2.256) 0.453



CD test



Dependent variable: EC CO2



0.018 (1.392)



0.024 (1.505)



0.075 (1.716)



GDP



0.601** (2.255)



0.638** (2.263)



0.575** (2.187)



GDP2



4.07E-05** (2.169) 0.522



8.31E-06** (2.040) 0.503



1.67E-04** (2.811) 0.953



CD test



Dependent variable: GDP CO2



12843.856*** (3.074)



39738.752*** (3.613)



6578.103** (2.419)



EC



 11.938** (2.592)



 4.915 (0.686)



 9.960** (2.157)



CD test



0.393



0.448



0.404



Note: (1). The average cross correlation coefficient is evaluated by the average of the pair-wise cross section coefficients of the regression residuals. t-statistics are shown in the parentheses. ** ***



denotes significance at the 5% level, and denotes significance at the 1% level.



countries. Furthermore, there exists the EKC relationship in the world, and in developed and developing countries based on the empirical results in Table 5. We use the formula t =



( ) to cal−α1 2α2



culate the turning point of EKC, where α1 and α2 means the estimated coefficient of GDP and GDP2 in the long-run relationship of (3a), respectively. Hence, the turning point occurs at real GDP per capita levels of $11,986, $32,407, and $11,061, respectively. It indicates that CO2 emissions increases with economic growth before reaching the GDP level of turning point, and decreases when the GDP level is higher than the GDP level of turning point. Such findings are consistent with the studies by Holtz-Eakin and Selden [50], Agras and Chapman [3], Richmond and Kaufmann [92], Managi and Jena [69], He and Richard [47], He and Wang [48], López-Menéndez et al. [67] and Song et al. [97]. The elasticities of CO2 emissions based on GDP in the world, and in developed and developing countries are 2.1,  0.6 and 2.6, respectively, which indicates that the developed countries already have sufficient economic and progressive technological capacities to lessen environmental damage, but the world and developing countries still lie in the stage of economic development endangering the environmental quality. For the long-run effects of CO2 emissions per capita and real GDP per capita on energy consumption per capita as shown in the second row of Table 5, CO2 emissions per capita have no significant impacts on energy consumption per capita, which is as expected. However, real GDP per capita positively influences energy consumption per capita in the world, and in developed and developing countries. We found that a 1% increase in real GDP per capita induces 5.4%, 8.2% and 4.5% increases in energy consumption per capita for the world, and for developed and developing countries, respectively. Such findings are consistent with studies by Chen



425



et al. [29] and others. However, such empirical outcomes indicate that the economic development in developed countries uses up more energy than that in developing countries which implies that there is still room for energy technology to improve in developed countries. The long-run effects of CO2 emissions per capita and energy consumption per capita on real GDP per capita are shown in the last row of Table 5. We find that a 1% increase in energy consumption per capita reduces real GDP per capita by 3.0% and 4.2% all over the world and in developing countries, respectively. Nevertheless, energy consumption per capita in developed countries has no significant impacts on real GDP per capita. However, CO2 emissions per capita have a positive influence on real GDP per capita in all three panels. A 1% increase in CO2 emissions per capita results in 6.7%, 11.9% and 6.1% increases in GDP in the world, and in developed and developing countries, respectively. Such a finding is not consistent with the report compiled by the Intergovernmental Panel on Climate Change [54]. The IPCC has estimated that global damage from climate change during the period from 1991 to 2005 was about US$1190 billion, where the most damaged sectors were crops, fishery, water resources, and human health. This is why many studies including Adamantiades and Kessides [2], DeCanio [31], and Reddy and Assenza [90] emphasize that humans need to take action intermediately to solve the problem of climate change due to global warming, or else there will both be economic damage and an environmental disaster. Based on statistical analysis, greenhouse gas emissions have sharply increased following the economic development in both developed and developing countries. In particular, CO2 emissions account for over half of GHG emissions, which is likely to be related to climate change [102,54,98]. But in our research, the negative influence from CO2 emissions per capita to GDP per capita has not happened during 1993–2010. As the IPCC estimate the climate change may damage the crops, fishery, water resources, and human health but the revenue from the manufacture sector may cover the damage during the period. But how to face the threat of climate change has become the primary task worldwide. Therefore, how to formulate corresponding policies to maintain economic growth together with stable economic development and CO2 emissions is still an important issue. 4.1. Dynamic Panel Granger Causality The existence of long-run cointegrated relationships among the variables in the three panels implies that there exists Granger causality in at least one direction (Engel and Granger [75]). To find out the direction of causality among these three variables and also solve the issue of CSD, we expand the framework of Eqs. (4a)–(4c) to capture the short-run variations and long-run equilibrium by the following two steps [34]. The first step involves making an estimate of the long-run model with CCEP estimators to acquire the estimated residual, and we employ the lagged one residual as the error correction term. In considering the endogeneity issue, the weak instruments problem, small T and large N panels, and reliable asymptotic distribution approximations of the estimators, we take account of the one-step system generalized method of moments (hereafter GMM-SYS) estimator [17,22–24] to examine Granger causality. In testing for the short-run causality, we employ the Wald test to examine the null hypothesis of βk . For long-run causality, we examine whether the coefficient of ECT (i.e., γ ) is significant, while the significance of the coefficient of φi could indicate that the dependent variable responds to deviations from the long-run equilibrium. In time series, the AIC or SBC is employed to select the optimal lag-length p, but there is no similar procedure for the panel ECM when choosing the optimal lag-length [52]. Moreover,



426



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



Huang et al. [52] suggest that owing to the advantages of the mj test proposed by Arellano and Bond [16] for preventing the misspecification of serial correlation with the optimal lag-length, the mj test could pick the more appropriate optimal lag-length. The mj statistics (j ¼1, 2) test for the null hypothesis of no j-th order of autocorrelation in the differenced residuals. Doornik et al. [33] demonstrate that if there exist m1 statistics with the rejection of the null hypothesis of no first-order serial correlation in the differenced residuals and m2 statistics without rejection of the null hypothesis of no second-order serial correlation in the differenced residuals, the error terms in our panel dynamic ECM are serially uncorrelated. We have found that the optimal lag-lengths for Eqs. (4a)–(4c) in the three panels are two lags in length (i.e., q¼2). The estimated results of the short-run and long-run Granger causality tests using the GMM-SYS estimator approach are shown in Table 6. For The different results between long-run and short-run causality imply for distinct economies and energy policies. Firstly, in the short-run, the existence of no causal relationship but long run causality relationship exists. That means the policy of the two variables do not affect each other in the short run, but the effect will converge in the long run. Therefore, if the policy maker pays more attention in the long run effect to do his policy decision, he should take the policy according to the long run result. Otherwise, he may pay more attention to the short run result. However, for the energy policy, it may not affect the GDP or GHG emissions immediately in the short run, but actually influence the environment in the long run. That is why we should discuss the result in different time span. The results lead to some serious policy implications for decision-makers as policies thorough rationalizing the resources. In thinking over the short-run dynamics, when considering the panel data set for all 188 countries (i.e., Panel A), there exist shortrun bidirectional causal relationships between CO2 emissions and GDP and between energy consumption and GDP as shown in the first row of Table 6. Nevertheless, energy consumption and CO2 emissions have negative effects on GDP. There exists short-run positive unidirectional Granger causality from energy consumption to CO2 emissions. Such a short-run relationship may be altered as economic development changes. For developed countries, the empirical results show that CO2 emissions are positively affected by energy consumption and GDP in the short-run while GDP is negatively affected by CO2 emissions but not affected by energy consumption which is shown in the second row of Table 6. According to the database of the World Bank, the average CO2 emissions per $2000 of GDP (at 2005 constant prices) in developed countries present decreasing trends from 0.426 kg to 0.283 kg during the period from 1993 to 2010. However, the average energy consumption per $1000 of GDP (at 2005 constant prices) in developed countries is maintained at about the level of 147.195 kg during the same period. Although the environmental protection policy focusing on the reduction in CO2 emissions may damage GDP since such a reduction in CO2 emissions will push up production costs, the energy consumption does not affect the GDP in developed countries. Accordingly, the conservation policy in relation to energy consumption will not harm the economic growth in developed countries. In developing countries, there exists positive unidirectional Granger causality from energy consumption per capita to CO2 emissions per capita as shown in the last row of Table 6. However, for the short-run results, real GDP per capita is negatively affected by energy consumption per capita and CO2 emissions per capita, which means that rising energy consumption per capita and CO2 emissions per capita are harmful to real GDP per capita. It is worthwhile noting that energy consumption negatively affects GDP in the world as a whole and developing countries, but energy consumption does not affect GDP in developed countries.



Table 6 Panel causality tests for the CO2 emissions-energy consumption-economic growth nexus. Independent variable



Dependent variable



△CO



△EC



–



1.736( þ ) [0.307]



2.947(-)** [0.021]



△EC



5.067(þ)** [0.019]



–



3.904(-)** [0.018]



△GDP



5.582( þ)** [0.013]



6.173( þ )** [0.022]



–



ECT



 3.029** [0.025]



 5.330** [0.016]



 2.565** [0.039]



–



1.519( þ ) [0.330]



4.024(-)** [0.012]



△EC



3.496( þ )** [0.034]



–



1.065(-) [0.381]



△GDP



3.818( þ )*** [0.009]



5.002( þ )** [0.013]



–



ECT



 4.985** [0.011]



 7.014*** [0.006]



 4.223** [0.017]



–



1.903( þ) [0.286]



3.488(-)** [0.025]



△EC



9.716( þ )*** [0.000]



–



5.149(-)*** [0.002]



△GDP



5.310( þ)** [0.013]



10.674( þ )*** [0.000]



–



ECT



 3.208** [0.026]



 3.015** [0.031]



 2.876** [0.034]



2



△GDP



Panel A Short-run



Long-run



△CO



2



Panel B Short-run



Long-run



△CO



2



Panel C Short-run



Long-run



△CO



2



Note: (1). F statistics are shown regarding the short-run variations in the independent variables, and p-values are given in brackets. This study uses SBC as the criterion of the optimal lag lengths, and 7 means that the sum of the lagged coefficients on the independent variables are positive or negative. ECT refers to the error correction terms, and t-statistics relate to the long-run. ** ***



denote significance at the 5% level, respectively. denote significance at the 1% level, respectively.



5. Discussion The 1993–2010 average energy intensity (measured as units of energy per dollar of GDP in 2005) in the world, and in developed and developing countries is 17,979, 7,079, and 12,529 Btu/USD, respectively. Higher energy intensity in developing countries exhibits inefficiency in terms of energy usage, while a lower degree of energy intensity in developed countries implies lower energy production cost. As a result, inefficient energy use results in a waste of energy. Therefore, an increase in energy consumption damages GDP [15,100]. Such empirical outcomes with respect to the level of economic development also reflect an inequality issue in relation to climate change. The developed countries together with the industrialized regions of China account for approximately 23% of the population, but emit two-thirds of greenhouse gases globally. Such increases in greenhouse gas emissions have induced climate change and the impact of this behavior is overwhelmingly apparent in its effect on the rest of the world, and particularly among the poor and vulnerable populations living in the poor countries. Finally, in terms of the long-run causality, regardless of the level of economic development, energy consumption, CO2 emissions and GDP affect each other as shown in the row for the long run for each



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



Table 7 Comparison within the literature.



Table 7 (continued ) Literature



Literature



Region (Period) Method



Ang [12]



In the long run: GDP-CO2 GDP-EC In the short run: EC-GDP USA (1960– Toda-Yamamo- In the long run: 2004) to procedure EC-CO2 In the long run: Pedroni coinSix central tegration tests; GDP ↔ EC American panel FMOLS; countries EC ↔ CO2 panel VECM (1971–2004) In the short run: GDP-CO2 EC ↔ GDP EC-CO2 Turkey (1960– ARDL bound In the long run: 2005) test EC ↔ CO2 GDP-CO2 EC-GDP In the short run: GDP ↔ CO2 EC-CO2 China (1975– ARDL bound In the long run and 2005) test short run: GDP-CO2 EC-CO2 Turkey (1960– Toda-Yamamo- In the long run: 2000) to procedure CO2-EC China (1960– Toda-Yamamo- In the long run: 2007) to procedure GDP-EC EC-CO2 In the long run: Denmark, Ger- ARDL bound test EC-CO2: Denmark, many, Greece, Iceland, Italy, Greece, Iceland, Italy, Portugal, SwitPortugal zerland (1960– GDP-CO2: Denmark, 2005) Greece, Iceland, Italy, Portugal EC ↔ CO2: Switzerland GDP↔ CO2: Switzerland In the short run: GDP-CO2: Denmark and Italy GDP-EC: Greece and Italy GDP ↔ EC: Switzerland In the long run: Pedroni coin11 integration tests; EC ↔ CO2 dependent panel FMOLS; states (1992– In the short run: panel VECM 2004) GDP ↔ EC GDP-CO2 EC-CO2 In the long run: 26 OECD coun- VECM GDP-EC tries and 45 In the short run: non-OECD GDP ↔ EC countries China (1981– Johansen coin- In the long run: GDP-CO2 2006) tegration test; VECM GDP-Coal CO2 ↔ Coal Natural gas ↔ GDP Natural gas-CO2 Electricity ↔ GDP Electricity-CO2 Johansen Fisher In the long run: Five ASEAN EC-GDP panel coincountries CO2-GDP tegration test; (1980–2006) panel DOLS; In the short run: panel VECM CO2-EC USA (1960– Toda-Yamamo- In the long run: 2007) to procedure Nuclear EC-CO2 GDP ↔ CO2 Renewable EC- Nuclear EC GDP -Renewable EC South Africa ARDL bound In the long run:



Soytas et al. [99] Apergis and Payne [14]



Halicioglu [44]



Jalil and Mahmud [57]



Soytas and Sari [98] Zhang and Cheng [108] Acaravci and Ozturk [1]



Apergis and Payne [15]



Costantini and Martini [25]



Chang [27]



Lean and Smyth [63]



Menyah and Wolde-Rufael [71]



Menyah and



France (1960– 2000)



427



Johansen cointegration test; ARDL bound test; VECM



Region (Period) Method



Results



(1965–2006)



test



Ozturk and Acaravci [76]



Turkey



ARDL bound test



Pao and Tsai [79]



BRIC countries (1971–2005)



Pedroni cointegration tests; Johansen Fisher panel cointegration test; panel VECM



Alam et al. [7]



India (1971– 2006) Greece (1977– 2007)



Toda-Yamamoto procedure Johansen cointegration test; VECM



Hossain [51]



10 newly-industrialized countries (1971–2007)



Johansen Fisher panel cointegration test; panel VECM



Iwata et al. [56]



OECD countries ARDL regresand non-OECD sion with PMG estimator countries



Pao and Tsai [80]



BRIC countries (1992–2007)



Pedroni cointegration tests; Kao Test; Johansen Fisher panel; panel VECM



Pao and Tsai [81]



Brazil



Johansen Fisher panel cointegration test; panel VECM



Pao et al. [82]



Russia (1990– 2007)



Johansen Fisher panel cointegration test; panel VECM



Wang et al. [103]



28 provinces in Pedroni cointegration tests; China (1995– panel VECM 2007)



Al-mulali and Sab [10]



19 selected countries (1980–2008)



Pedroni cointegration tests; panel VECM



Al-mulali and Sab [11]



Sub Saharan African countries (1980– 2008)



Pedroni cointegration tests; panel VECM



EC-GDP CO2-GDP EC-CO2 In the short run: EC-GDP CO2-GDP In the long run: CO2-GDP EC-CO2 In the long run: GDP ↔ EC CO2-GDP CO2-EC In the short run: CO2 ↔ EC CO2-GDP EC-GDP In the long run: CO2 ↔ EC In the long run: GDP-EI (energy intensity) GDP-CO2 CO2 ↔ EI In the short run: GDP-CO2 EI-CO2 No long-run causal relationships. In the short run: GDP-CO2 GDP-EC In the long run: Nuclear energy-CO2 (OECD and no OECD) GDP-CO2 (OECD and no OECD) In the short run: GDP-CO2 (OECD) In the long run: EC-CO2 GDP-CO2 In the short run: GDP ↔ CO2 GDP ↔ EC EC-CO2 In the long run: GDP ↔ CO2 EC-CO2 EC-GDP In the short run: GDP -CO2 In the long run: CO2 ↔ EC EC ↔ GDP GDP ↔ CO2 In the short run: CO2-GDP CO2-EC In the long run: CO2 ↔ EC In the short run: GDP ↔ EC CO2 ↔ EC In the long run: CO2 ↔ EC GDP ↔ EC CO2 ↔ GDP In the short run: EC-GDP CO2 ↔ EC GDP ↔ CO2 In the long run: CO2 ↔ EC GDP ↔ EC CO2 ↔ GDP



Results Wolde-Rufael [72]



Hatzigeorgiou et al. [46]



428



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



Table 7 (continued ) Literature



Alam et al. [6]



Table 7 (continued ) Region (Period) Method



Bangladesh (1972–2006)



Johansen cointegration test; ARDL bound test; VECM



Bloch et al. [21]



China (1977– 2008)



Hamit-Haggar [45]



Pedroni coinCanadian industrial sectors tegration tests; panel FMOLS; (1990–2007) panel VECM



Jayanthakumaran et al. [58]



China and India (1971– 2007)



Pao et al. [78]



China



Johansen cointegration test; VECM



ARDL bound test



Johansen cointegration test



Al-mulali et al. [9] NEBA countries Pedroni coin(1980–2009) tegration tests; panel VECM



Chandran and Tang [26]



Govindaraju and Tang [41]



ASEAN-5 countries (1971–2008)



China and India (1965– 2009)



Johansen cointegration test; VECM



Combined cointegraion test (Bayer and Hanck [20]); VAR (India); VECM (China)



Results In the short run: CO2 ↔ EC In the long run: CO2 ↔ EC EC-GDP CO2-GDP In the short run: EC-GDP CO2-GDP EC-CO2 In the long run: GDP-CO2 GDP-coal consumption CO2 ↔ coal consumption In the short run: CO2 ↔ GDP GDP-coal consumption In the long run: EC-GHG (greenhouse gas) GDP-GHG In the short run: GDP-GHG GDP-EC GHG ↔ EC In the long run: GDP-EC: China and India EC-GDP: China and India In the short run: EC-GDP: China In the long run: CO2 ↔ EC GDP ↔ EC CO2 ↔ GDP In the long run: EC ↔ CO2 Urbanization ↔ EC CO2 ↔ Urbanization In the short run: EC ↔ CO2 Urbanization-EC CO2 ↔ Urbanization In the long run: GDP- CO2: Malaysia GDP ↔ CO2: Indonesia, Thailand EC- CO2: Indonesia CO2 ↔ EC: Malaysia, Thailand EC-GDP: Indonesia, Malaysia EC ↔ GDP: Thailand In the short run: GDP- CO2: Indonesia, Malaysia, Thailand CO2-GDP: Philippines GDP ↔ CO2: Singapore CO2-EC: Indonesia, Malaysia CO2 ↔ EC: Philippines, Thailand EC-GDP: Indonesia, Singapore GDP-EC: Philippines EC ↔ GDP: Malaysia In the long and short run (China): GDP- CO2 CO2 ↔ Coal Coal ↔ GDP In the short run (India):



Literature



Region (Period) Method



Results



Khan et al. [61]



Pakistan (1975–2011)



GDP ↔ CO2 CO2 ↔ Coal Coal ↔ GDP No casual relationships in the long run.



Saboori and Sulaiman [92]



Indonesia; Malaysia; Philippines; Singapore; Thailand (1971– 2009)



Johansen cointegration test; VAR ARDL bound test



Shahbaz et al. [95] Indonesia (1975–2011)



ARDL bound test



Akhmat et al. [5].



Conditional ARDL-VECM



López-Menéndez et al. [67]



Šmiech and Papiez [96]



Liddle and Lung [66]



Özbugday and Erbas [77] This study



South Asia, Middle East and North Africa (MENAregion), Sub-Saharan Africa, East Asia and Pacific and the aggregate data of the world. (1975–2011) 27 EU countries(1996– 2010)



Panel regression



In the long run: EC-GDP: Indonesia, Malaysia, Philippines CO2 ↔ EC: Indonesia, Malaysia, Philippines, Singapore, Thailand GDP ↔ CO2: Indonesia, Malaysia, Philippines GDP-CO2: Singapore, Thailand GDP-EC: Singapore, Thailand In the short run: GDP ↔ CO2: Indonesia, Singapore, Thailand CO2 ↔ EC: Malaysia, Singapore EC-GDP: Malaysia, Thailand CO2-GDP: Philippines EC-CO2: Philippines In the long run: CO2 ↔ EC GDP ↔ EC CO2 ↔ GDP In the short run: CO2 ↔ EC GDP ↔ CO2 In the long run: EC-CO2: all regions GDP ↔ EC: all regions but not exists in SubSaharan Africa.



Environmental Kuznets Curve exists because of the impacts of renewable energies on CO2 emissions 25 EU member Bootstrap panel EC-GDP: Greece, Poland GDP-EC: Cyprus, Granger causstates (1993– France, Romania, Sloality test 2011) vakia GDP ↔ EC: Bulgaria, Latvia, In the long run: OECD countries Dynamic ECM GDP-EC: OECD coun(1960–2006) tries and and non-OECD non-OECD groups, but countries not include Middle(1971–2005) high energy intensity group and Low energy intensity group 36 countries Heterogeneous In the long run: Energy efficiency-CO2 (1971–2009) panel CCE approach In the long run and 128 countries; Gengenbach et al. [40] panel short run: developed GDP ↔ CO2: all, develcointegration countries; developing coun- test; CCEP esti- oped, and developing mator; panel tries (1993– countries VECM 2008) GDP ↔ EC: all and developing countries GDP-EC: developed countries EC-CO2: all, developed, and developing countries



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



of these three panels in Table 6. Such a finding could be compared with the findings of another 42 studies as shown in Table 7. There exist three types of directional causality between energy consumption and CO2 emissions. The occurrence of directional causality running from energy consumption to CO2 emissions is observed by Soytas et al. [99], Halicioglu [44], Jalil and Mahmud [57], Zhang and Cheng [108], Acaravci and Ozturk [1], Apergis and Payne [15], Menyah and Wolde-Rufael [71], Menyah and WoldeRufael [72], Ozturk and Acaravci [76], Hatzigeorgiou et al. [46], Pao and Tsai [80], Pao and Tsai [81], Hamit-Haggar [45], and Chandran and Tang [25]. However, there are 15 studies that have found that a bi-directional causality between energy consumption and CO2 emissions exists as a result of these 42 studies. Such a bi-directional relationship is found to exist in developing countries such as China, Indonesia and Bangladesh. This empirical finding indicates that the positive relationship between CO2 emissions and energy consumption in developing countries may imply that the energy use efficiency needs to be improved in order to stop the increase in greenhouse gas emissions. This concept is also consistent with recent research from Özbugday and Erbas [77] who indicate that positive significant effect of energy efficiency on CO2 emissions in the long-run and roles of energy efficiency and renewable energy in curbing CO2 emissions. Based on the summary of the literature in Table 7, many studies, including Ang [12], Soytas et al. [99], Halicioglu [44], Jalil and Mahmud [57], Soytas and Sari [99], Zhang and Cheng [108], Chang [27], Menyah and Wolde-Rufael [71], Menyah and Wolde-Rufael [73], Ozturk and Acaravci [76], Alam et al. [7], Hatzigeorgiou et al. [46], Pao and Tsai [81], Pao et al. [82], Alam et al. [6], Bloch et al. [21], Pao et al. [78], Khan et al. [61], and Shahbaz et al., [95] have focused on the issue of relationships among energy consumption, GDP, and CO2 emissions in a particular country which may only provide part of the truth regarding the relationship between energy consumption and economic growth or the relationship between energy consumption and CO2 emissions. The relationship between GDP and CO2 emissions needs to be examined from a global perspective due to the climate change being induced by the increased global greenhouse gas emissions with their concentration. This explains why there is no causality running from CO2 emissions to GDP based on the observations in Table 7. This is also why we apply global data sets from 188 countries to examine such relationships among energy consumption, economic growth and CO2 emissions in this study. Three major findings could be addressed after comparing the previous studies with our findings on the issue of the relationships among energy consumption, GDP, and CO2 emissions as shown in Table 7.



6. Conclusions and policy implications The major objective in this study is to investigate the long-run relationships and direction of causal relationships among CO2 emissions per capita, energy consumption per capita, and real GDP per capita for developed countries, developing countries, and the world during the period from 1993 to 2010. Global data sets for real GDP per capita, energy consumption per capita, and CO2 emissions per capita are collected from 188 countries of which there are 26 developed countries, with the others all being developing countries. In considering the characteristics of large N and small T panel data sets, we employ the recently developed panel unit root test [89] and panel cointegration test [40] to confirm whether there exist long-run cointegrated relationships among CO2 emissions, energy consumption, and GDP. The long-run equilibrium relationships among these three variables with the CCE estimator are subsequently explored based



429



on the studies by Pesaran [88] and Kapetanios et al. [60]. The empirical results indicate that energy consumption per capita has significant impacts on CO2 emissions. A 1% increase in energy consumption per capita augments CO2 emissions per capita by 7.6%, 26.1%, and 13.5% in developed countries, developing countries, and the world, respectively, which implies a positive relationship between these two variables in the long-run. We have found that energy consumption per capita in developing countries induces more CO2 emissions per capita than in developed countries. On the other hand, real GDP per capita has been found to positively affect energy consumption per capita both in developed countries and developing countries where developing countries lead to more energy consumption than developed countries as the economy grows. Augmenting energy consumption causes damage to GDP for the world as a whole in the long-run. For the long-run, A 1% increase in CO2 emissions per capita leads to a 6.7% increases in real GDP per capita, however, a 1% increase in energy consumption per capita results in a 3.0% losses of real GDP per capita. Therefore, the related environmental regulations for preventing environmental degradation need to be considered for each country as the economy grows. These empirical results also raise the issue of inequality in climate change and economic development. As developed countries emitted two thirds of global greenhouse gas emissions due to the economic development, they have enjoyed higher living quality but have simultaneously given rise to negative external effects of global warming. Such increases in greenhouse gas emissions have induced climate change and have significantly affected the rest of the world, especially for the people living in the poor countries. Such an inequality issue needs to be resolved through multiple strategies including energy use efficiency, tools to mitigate greenhouse gas emissions such as a carbon tax or emissions trading, and a compensating mechanism. After testing the long-run equilibrium relationships among GDP, energy consumption and CO2 emissions, we employ the expanded framework of Gengenbach et al. [40] conditional ECM model to examine the causal relationships among these three variables in both developed and developing countries. As for the short-run causal relationship between GDP and energy consumption, the developed countries indicate that GDP has positive effects on energy consumption, but energy consumption does not affect GDP. In developing countries, there exists a bidirectional causal relationship between GDP and energy consumption, but energy consumption has a negative effect on GDP, which is the opposite of the effect of GDP on energy consumption. In addition, the empirical results also indicate that the energy consumption has positive effects on CO2 emissions both in developed and developing countries. The relationship between GDP and CO2 emissions in developed countries is similar to that in the developing countries, with higher GDP bringing about higher CO2 emissions, and higher CO2 emissions giving rise to lower GDP. CO2 emissions are likely to cause climate change and this has led to a decline in global environmental quality which has caused much damage around the world. In terms of the long-run causality, CO2 emissions, energy consumption and GDP could respond to deviations from the long-run equilibrium. The major finding of this paper is that the relationship between energy consumption and GDP varies with different levels of economic development. Therefore, it is necessary to take account of the level of economic development when setting the environmental policy. However, both the developed and developing countries indicates that higher GDP gives rise to more energy consumption and more CO2 emissions, and such increases in CO2 emissions in turn harm the economy in the short-run, however, increases in CO2 emissions lead to the rising of economy growth in the long-run. That is, currently economic development is high dependence on consumption of fossil fuel energy in the long-run.



430



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



Furthermore, high dependence on consumption of fossil fuel energy causes more and more CO2 emissions which will lead to the rising of damage risk for climate change. Therefore, all countries should consider enforcing relevant environmental regulations to prevent environmental degradation as the economy grows. Besides cutting down on CO2 emissions, developed countries should also serve as the leaders in mitigating climate change, including developing more efficient energy conversion technologies and clean energy technologies such as natural gas, hydropower, nuclear power, and renewable energy. In developing countries, the external cost of energy use such as the inefficiency of energy use and pollutants is greater than the resulting benefits, and so it is necessary to enhance the efficiency of energy usage with energyrelated equipment and the process of energy conservation.



References [1] Acaravci A, Ozturk I. On the relationship between energy consumption, CO2 emissions and economic growth in Europe. Energy 2010;35:5412–20. [2] Adamantiades A, Kessides I. Nuclear power for sustainable development: Current status and future prospects. Energy Policy 2009;37:5149–66. [3] Agras J, Chapman D. A dynamic approach to the environmental Kuznets curve hypothesis. Ecol Econ 1999;28:267–77. [4] Akbostanci E, Turut-Asik S, Tunc GI. The relationship between income and environment in Turkey: Is there an environmental Kuznets curve? Energy Policy 2009;37:861–7. [5] Akhmat G, Zaman K, Shukui T, Sajjad F. Does energy consumption contribute to climate change? Evidence from major regions of the world Renew Sustain Energy Rev 2014;36:123–34. [6] Alam MJ, Begum IA, Buysse J, Huylenbroeck GV. Energy consumption, carbon emissions and economic growth nexus in Bangladesh: Cointegration and dynamic causality analysis. Energy Policy 2012;45:217–25. [7] Alam MJ, Begum IA, Buysse J, Rahman S. Dynamic modeling of causal relationship between energy consumption, CO2 emissions and economic growth in India. Renew Sustain Energy Rev 2011;15:3243–51. [8] Al-Iriani MA. Energy-GDP relationship revisited: An example from GCC countries using panel causality. Energy Policy 2006;34:3342–50. [9] Al-mulali U, Fereidouni HG, Lee JYM, Sab CNBC. Exploring the relationship between urbanization, energy consumption, and CO2 emission in MENA countries. Renew Sustain Energy Rev 2013;23:107–12. [10] Al-mulali U, Sab CNBC. The impact of energy consumption and CO2 emission on the economic and financial development in 19 selected countries. Renew Sustain Energy Rev 2012;16:4365–9. [11] Al-mulali U, Sab CNBC. The impact of energy consumption and CO2 emission on the economic and financial development in the Sub Saharan African countries. Energy 2012;39:180–6. [12] Ang JB. CO2 emissions, energy consumption, and output in France. Energy Policy 2007;35:4772–8. [13] Ang JB. Economic development, pollutant emissions and energy consumption in Malaysia. J Policy Modeling 2008;30:271–8. [14] Apergis N, Payne JE. CO2 emissions, energy usage, and output in Central America. Energy Policy 2009;37:3282–6. [15] Apergis N, Payne JE. The emissions, energy consumption, and growth nexus: Evidence from the commonwealth of independent states. Energy Policy 2010;38:650–5. [16] Arellano M, Bond S. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 1991;58:277–97. [17] Arellano M, Bover O. Another look at the instrumental-variable estimation of error components models. J Econometrics 1995;68:29–52. [18] Bai J, Kao C. On the estimation and inference of a panel cointegration model with cross-sectional dependence. In: Baltagi B, editor. Panel Data Econometrics: Theoretical Contributions and Empirical Applications. In: Contributions to Economic Analysis, vol. 274. Elsevier; 2006. [19] A. Banerjee, J.L. Carrion-i-Silvestre, Cointegration in panel data with breaks and cross-section dependence. In: ECB Working Paper Series, vol. 591. European Central Bank, 2006. [20] Bayer C, Hanck C. Combining non-cointegration tests. J Time Ser Anal 2013;34:83–95. [21] Bloch H, Rafiq S, Salim R. Coal consumption, CO2 emission and economic growth in China: Empirical evidence and policy responses. Energy Econ 2012;34:518–28. [22] Blundell R, Bond S. Initial conditions and moment restrictions in dynamic panel data models. J Econ 1998;87:115–43. [23] Blundell RW, Bond SR, Windmeijer F. Estimation in dynamic data models: Improving on the performance of the standard GMM estimator. In: Baltagi B, editor. Nonstationary Panels, Panel Cointegration and Dynamic Panels. Advances in Econometrics, vol. 15. JAI Elsevier Science; 2000. [24] Bond SR. Dynamic panel data models: a guide to micro data methods and



practice. Portuguese Econ J 2002:141–62. [25] Costantini V, Martini C. The causality between energy consumption and economic growth: a multi-sectoral analysis using non-stationary cointegrated panel data. Energy Econ 2010;32(3):591–603. [26] Chandran VGR, Tang CF. The impacts of transport energy consumption, foreign direct investment and income on CO2 emissions in ASEAN-5 economies. Renew Sustain Energy Rev 2013;24:445–53. [27] Chang CC. A multivariate causality test of carbon dioxide emissions, energy consumption and economic growth in China. Appl Energy 2010;87:3533–7. [28] Chang Y. Residual based tests for cointegration in dependent panels. Mimeo, Rice University; 2005. [29] Chen PY, Chen ST, Chen CC. Energy consumption and economic growth New evidence from meta analysis. Energy Policy 2012;44:245–55. [30] Coondoo D, Dinda S. Causality between income and emissions: a country group-specific econometric analysis. Ecol Econ 2002;40:351–67. [31] DeCanio SJ. The political economy of global carbon emissions reductions. Ecol Econ 2009;68:915–24. [32] Dinda S, Coondoo D. Income and emission: A panel data-based cointegration analysis. Ecol Econ 2006;57:167–81. [33] J.A., Doornik, M., Arellano, S., Bond, 2006. Panel Data Estimation using DPD for Ox. mimeo. [34] Engle RF, Granger CWJ. Co-integration and error correction: Representation, estimation and testing. Econometrica 1987;55:251–76. [35] Esteve V, Tamarit C. Threshold cointegration and nonlinear adjustment between CO2 and income: the environmental Kuznets Curve in Spain, 1857– 2007. Energy Econ 2012;34:2148–56. [36] Fodhaa M, Zaghdoud O. Economic growth and pollutant emissions in Tunisia: an empirical analysis of the environmental Kuznets curve. Energy Policy 2010;38:1150–6. [37] Friedl B, Getzner M. Determinants of CO2 emissions in a small open economy. Ecol Econ 2003;45:133–48. [38] Galeotti M, Lanza A. Desperately seeking environmental Kuznets. Environ Model Software 2005;20:1379–88. [39] Gengenbach C, Palm FC, Urbain JP. Panel unit root tests in the presence of cross-sectional dependencies: Comparison and implications for modeling. Econom Rev 2010;29:111–45. [40] Gengenbach C, Urbain JP, Westerlund J. Error correction testing in panels with global stochastic trends. Maastricht University; 2009 (METEOR). [41] Govindaraju VGRC, Tang CF. The dynamic links between CO2 emissions, economic growth and coal consumption in China and India. Appl Energy 2013;104:310–8. [42] Groen JJJ, Kleibergen F. Likelihood-based cointegration analysis in panels of vector error-correction models. J Business Econ Stat 2003;21:295–318. [43] G.M. Grossman, Krueger, Environmental impacts of a North American Free Trade Agreement. NBER Working Paper, vol. 3914, Cambridge, MA, 1991. [44] Halicioglu F. An econometric study of CO2 emissions, energy consumption, income and foreign trade in Turkey. Energy Policy 2009;37:1156–64. [45] Hamit-Haggar M. Greenhouse gas emissions, energy consumption and economic growth: a panel cointegration analysis from Canadian industrial sector perspective. Energy Econ 2012;34:358–64. [46] Hatzigeorgiou E, Polatidis H, Haralambopoulos D. CO2 emissions, GDP and energy intensity: a multivariate cointegration and causality analysis for Greece, 1977–2007. Appl Energy 2011;88:1377–85. [47] He J, Richard P. Environmental Kuznets curve for CO2 in Canada. Ecol Econ 2010;69:1083–93. [48] He J, Wang H. Economic structure, development policy and environmental quality: an empirical analysis of environmental Kuznets curves with Chinese municipal data. Ecol Econ 2012;76:49–59. [49] Holly S, Pesaran MH, Yamagata T. A spatio-temporal model of house prices in the USA. J Econom 2010;158:160–73. [50] Holtz-Eakin D, Selden TM. Stoking the fires? CO2 emissions and economic growth J Public Econom 1995;57:85–101. [51] Hossain SM. Panel estimation for CO2 emissions, energy consumption, economic growth, trade openness and urbanisation of newly industrialized countries. Energy Policy 2011;39:6991–9. [52] Huang BN, Hwang MJ, Yang CW. Causal relationship between energy consumption and GDP growth revisited: a dynamic panel data approach. Ecol Econom 2008;67:41–54. [53] Im K, Pesaran MH, Shin Y. Testing for unit roots in heterogeneous panels. J Econom 2003;115:53–74. [54] Intergovernmental Panel on Climate Change (IPCC). Climate Change 2007: Impacts, adaptation and vulnerability.Cambridge: Cambridge University Press,; 2007. [55] International Energy Agency (IEA). World Energy Outlook 2011.Paris: OECD/ IEA,; 2011. [56] Iwata H, Okada K, Samreth S. A note on the environmental Kuznets curve for CO2: A pooled mean group approach. Appl Energy 2011;88:1986–96. [57] Jalil A, Mahmud SF. Environment Kuznets curve for CO2 emissions: a cointegration analysis for China. Energy Policy 2009;37:5167–72. [58] Jayanthakumaran K, Verma R, Liu Y. CO2 emissions, energy consumption, trade and income: A comparative analysis of China and India. Energy Policy 2012;42:450–60. [59] Kao C. Spurious regression and residual-based tests for cointegration in panel data. J Econom 1999;90:1–44. [60] Kapetanios G, Pesaran MH, Yamagata T. Panels with nonstationary multifactor error structures. J Econom 2011;160:326–48.



P.-Y. Chen et al. / Renewable and Sustainable Energy Reviews 65 (2016) 420–431



[61] Khan MA, Khan MZ, Zaman K, Khan MM, Zahoor H. Causal links between greenhouse gas emissions, economic growth and energy consumption in Pakistan: a fatal disorder of society. Renew Sustain Energy Rev 2013;25:166– 76. [62] Kraft J, Kraft A. On the relationship between energy and GNP. J Energy Develop 1978;3:401–3. [63] Lean HH, Smyth R. CO2 emissions, electricity consumption and output in ASEAN. Appl Energy 2010;87:1858–64. [64] Lee CC, Lee JD. Income and CO2 emissions: evidence from panel unit root and cointegration tests. Energy Policy 2009;37:413–23. [65] Levin A, Lin F, Chu C. Unit root tests in panel data: asymptotic and finite sample properties. J Econom 2002;108:1–24. [66] Liddle B, Lung S. Revisiting energy consumption and GDP causality: importance of a priori hypothesis testing, disaggregated data, and heterogeneous panels. Appl Energy 2015;142:44–55. [67] López-Menéndez AJ, Pérez R, Moreno B. Environmental costs and renewable energy: re-visiting the environmental Kuznets Curve. J Environm Manag 2014;145:368–73. [68] Maddala GS, Wu S. A comparative study of unit root tests with panel data and a new simple test. Oxford Bull Econom Statistics 1999;61:631–52. [69] Managi S, Jena PR. Environmental productivity and Kuznets curve in India. Ecol Econ 2008;65:432–40. [70] Martinez-Zarzoso I, Bengochea-Morancho A. Pooled mean group estimation of an environmental Kuznets curve for CO2. Econ Lett 2004;82:121–6. [71] Menyah K, Wolde-Rufael Y. CO2 emissions, nuclear energy, renewable energy and economic growth in the US. Energy Policy 2010;38:2911–5. [72] Menyah K, Wolde-Rufael Y. Energy consumption, pollutant emissions and economic growth in South Africa. Energy Econom 2010;32:1374–82. [73] Nelson M, Ogaki M, Sul D. Dynamic seemingly unrelated cointegrating regressions. Rev Econ Studies 2005;72:797–820. [74] Orubu CO, Omotor DG. Environmental quality and economic growth: Searching for environmental Kuznets curves for air and water pollutants in Africa. Energy Policy 2011;39:4178–88. [75] Oxley L, Greasley D. Vector autoregression, cointegration and causality: testing for causes of the British Industrial Revolution. Appl Econ 1998;30:1387–97. [76] Ozturk I, Acaravic A. CO2 emissions, energy consumption and economic growth in Turkey. Renew Sustain Energy Rev 2010;14:3220–5. [77] Özbugday FC, Erbas BC. How effective are energy efficiency and renewable energy in curbing CO2 emissions in the long run? A heterogeneous panel data analysis Energy 2015;82:734–45. [78] Pao HT, Fu HC, Tseng CL. Forecasting of CO2 emissions, energy consumption and economic growth in China using an improved grey model. Energy 2012;40:400–9. [79] Pao HT, Tsai CM. CO2 emissions, energy consumption and economic growth in BRIC countries. Energy Policy 2010;38:7850–60. [80] Pao HT, Tsai CM. Multivariate Granger causality between CO2 emissions, energy consumption, FDI foreign direct investment and GDP (gross domestic product): evidence from a panel of BRIC (Brazil, Russian Federation, India, and China) countries. Energy 2011;36:685–93. [81] Pao HT, Tsai CM. Modelling and forescasting the CO2 emissions, energy consumption, and economic growth in Brazil. Energy 2011;36:2450–8. [82] Pao HT, Yu HC, Yang YH. Modeling the CO2 emissions, energy use, and economic growth in Russia. Energy 2011;36:5094–100. [83] Payne JE. Survey of the international evidence on the causal relationship between energy consumption and growth. J Econ Studies 2010;37:53–95. [84] Pedroni P. Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxford Bull Econ Stat 1999;61:653–69.



431



[85] Pedroni P. Panel cointegration: Asymptotic and finite sample properties of pooled time series tests with an application to the PPP hypothesis. Econom Theory 2004;20:597–625. [86] Pedroni P, Vogelsang T. Robust unit root and cointegration rank tests for panels and large systems. Mimeo. Williams College,; 2005. [87] M.H., Pesaran, General diagnostic tests for cross section dependence in panels. CESifo Working Papers. No. 1233, 2004. [88] Pesaran MH. Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica 2006;74:967–1012. [89] Pesaran MH. A simple panel unit root test in the presence of cross section dependence. J Appl Econom 2007;22:265–312. [90] Reddy BS, Assenza GB. The great climate debate. Energy Policy 2009;37:2997–3008. [91] Richmond AK, Kaufmann RK. Is there a turning point in the relationship between income and energy use and/or carbon emissions? Ecol Econom 2006;56:176–89. [92] B. Saboori, J. Sulaiman, CO2 emissions, energy consumption and economic growth in Association of Southeast Asian Nations (ASEAN) countries: a cointegration approach. Energy. In Press, 2013. [93] Selden T, Song D. Environmental quality and development: is there a Kuznets curve for air pollution emissions? J Environ Econom Manag 1994;27:147–62. [94] Sephtona P, Mann J. Further evidence of an Environmental Kuznets Curve in Spain. Energy Econom 2013;36:177–81. [95] Shahbaz M, Hye QMA, Tiwari AK, Leitão NC. Economic growth, energy consumption, financial development, international trade and CO2 emissions in Indonesia. Renew Sustain Energy Rev 2013;25:109–21. [96] Šmiech S, Papiez M. Energy consumption and economic growth in the light of meeting the targets of energy policy in the EU: the bootstrap panel Granger causality approach. Energy Policy 2014;71:118–29. [97] Song ML, Zhang W, Wang SH. Inflection point of environmental Kuznets curve in Mainland China. Energy Policy 2013;57:14–20. [98] Soytas U, Sari R. Energy consumption, economic growth, and carbon emissions: challenges faced by an EU candidate member. Ecol Econ 2009;68:1667–75. [99] Soytas U, Sari R, Ewing BT. Energy consumption, income, and carbon emissions in the United States. Ecol Econ 2007;62:482–9. [100] Squalli J. Electricity consumption and economic growth: bounds and causality analyses of OPEC countries. Energy Econ 2007;29:1192–205. [101] Tamazian A, Chousa JP, Vadlamannati KC. Does higher economic and financial development lead to environmental degradation: Evidence from BRIC countries. Energy Policy 2009;37:246–53. [102] The World Bank, Growth and CO2 emissions: How do different countries fare? Environment Department, Washington, DC, 2007. [103] Wang SS, Zhou DQ, Zhou P, Wang QW. CO2 emissions, energy consumption and economic growth in China: a panel data analysis. Energy Policy 2011;39:4870–5. [104] Westerlund J. New simple tests for panel cointegration. Econom Rev 2005;24:297–316. [105] Westerlund J. Panel cointegration tests of the Fisher effect. J Appl Econom 2008;23:193–233. [106] Westerlund J, Costantini M. Panel cointegration and the neutrality of money. Empir Econom 2009;36:1–26. [107] Westerlund J, Edgerton D. A panel bootstrap cointegration test. Econom Lett 2007;97:185–90. [108] Zhang XP, Cheng XM. Energy consumption, carbon emissions, and economic growth in China. Ecol Econom 2009;68:2706–12.