- Email: [email protected]
A bootstrap panel Granger causality analysis of energy consumption and economic growth in the G7 countries
Renewable and Sustainable Energy Reviews 63 (2016) 166–171
Contents lists available at ScienceDirect
Renewable and Sustainable Energy Reviews journal homepage: www.elsevier.com/locate/rser
A bootstrap panel Granger causality analysis of energy consumption and economic growth in the G7 countries Mihai Mutascu a,b,n a LEO (Laboratoire d'Economie d'Orléans) UMR7322, Faculté de Droit d'Economie et de Gestion, University of Orléans, Rue de Blois, B.P. 6739, 45067 Orléans, France b East European Centre for Research in Economics and Business, West University of Timisoara, 16 H. Pestalozzi St., 300115 Timisoara, Romania
art ic l e i nf o
a b s t r a c t
Article history: Received 5 February 2015 Received in revised form 4 February 2016 Accepted 13 May 2016
The paper investigates the causality between energy consumption and economic growth in the countries which are members of the Group of Seven (G7), during the period 1970–2012, by following the bootstrap panel Granger causality approach. The main aim of the paper is to identify the right policies to deal with the interaction between energy consumption and economic growth. The findings show a bi-directional causality between energy consumption and GDP in Canada, Japan and United States. GDP causes energy consumption in France and Germany, while no causality is found for the rest of the sample (i.e. Italy and United Kingdom). These outcomes are very sensitive to the crosssectional dependences between countries. Therefore, their environmental and growth policies should be widely reconsidered when the level of interdependence dramatically decreases. & 2016 Elsevier Ltd. All rights reserved.
Keywords: Energy consumption Growth Bootstrap panel causality G7 countries
Contents 1. 2. 3.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data and methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Cross-section dependence tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Slope homogeneity tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Bootstrap panel Granger causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1. Introduction The connection and causality between energy consumption and economic growth have been extensively explored over the last decades, becoming one of the most important topics in the energy economics literature. Starting with February 2005, when the Kyoto Protocol has been implemented, the interaction between energy
n Correspondence address: East European Centre for Research in Economics and Business, West University of Timisoara, 16 H. Pestalozzi St., 300115 Timisoara, Romania. Tel.: þ40 256 592 556; fax: þ40 256 592500. E-mail addresses: [email protected], [email protected]
http://dx.doi.org/10.1016/j.rser.2016.05.055 1364-0321/& 2016 Elsevier Ltd. All rights reserved.
166 167 168 169 169 169 170 171 171 171
consumption and economic growth registered new valences, especially in developed countries. This international treaty is an extension of the United Nations Framework Convention on Climate Change (UNFCCC) from 1992 and has as main objective the reduction of greenhouse gas emissions in developed countries to the level in 1990. There are two main premises of this target: the global warming exists and it has been caused by the man-made carbon-dioxide (CO2) emissions. During the years, many researchers started to revisit the connection and causality between energy consumption and economic growth, for various countries and periods, by following different methodological tools. The results are very heterogeneous and emphasize different research perspectives, from causality to no causality statements.
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
On this ground, the aim of the paper is to analyse the energy consumption-economic growth nexus, in case of the Group of Seven (G7 countries), by using a bootstrap panel Granger causality approach, for the period 1970–2012. The G7 area includes the world's seven most industrialized economies, more precisely: Canada, France, Germany, Italy, Japan, United Kingdom (UK) and United States (US). Under Kyoto Protocol, this group of developed countries arises a special interest, as any influence of energy consumption policy on growth requires climate change response strategies. US is included in the sample, even if this country does not ratify the Kyoto Protocol. US has intensive trade and financial relationships with the rest of G-7 countries, being one of the first energy consumer in the world. Therefore, these characteristics give consistence of cross-sectional dependence property of G7 panel, which is one of the main assumption of the methodological tool followed (i.e. complex and intensive economic relationships run between G7 countries). Moreover, in 2012, during the 18th Conference of the Parties (COP18), all delegates agreed to extend the Kyoto Protocol until 2020, while US assumed the reduction of the carbon-dioxide emissions and greenhouse gases. Between 1970 and 2012 (Fig. 1), all G7 countries registered an increasing tendency for both energy consumption (average of energy use in kt of oil equivalent) and gross domestic product (GDP) (average GDP expressed in constant 2005 billions of US dollars). For the overall period, the average GDP illustrates an emphasized growth tendency, whereas the average energy consumption seems to follow a smooth increase. Significant negative shocks are registered over the 2007–2010 period, for both variables, as effects of world economic crises. The paper extends the literature in the field by offering the first study to investigate the relationship and causality between energy consumption and economic growth in the case of G7 countries, by following the bootstrap panel Granger causality developed by Kónya [1]. This technique is more powerful compared to existing ones, as it takes into account cross-sectional dependence and cross-country heterogeneity. Besides the contribution of Śmiech and Papież [3], to the best of our knowledge, this is one of the first papers which uses the Kónya's [1] tool in energy economics area. In comparison, Śmiech and Papież [3] explore European Union (EU) member states and analyse the period 1993–2011. The second contribution to the literature is given by the panel dimension, which covers an extended period of time, from 1970 to 2012, on annual basis. Finally, the outputs offer an important support for the economic energy policy makers in the G7 countries, helping them to manage, over the long-run, the interaction between energy consumption and economic growth. The rest of the paper is as follows: Section 2 reviews the literature, Section 3 presents the data and methodology, while Section 4 shows the empirical results. Section 5 concludes.
167
2. Literature review The literature about the linkage and causality between energy consumption and economic growth is prolific and offers a diversity of findings, deriving from the targeted countries, periods of analysis and econometric tools used. Relying on these different outcomes, the literature also offers contradictory policy solutions for policymakers regarding the adjustments of the relationship between energy consumption and economic growth. There are four main hypotheses outlining the literature review in respect to the energy consumption-growth nexus: (i) growthenergy consumption hypothesis, (ii) energy consumption-growth hypothesis, (iii) feed-back hypothesis, and (iv) neutrality hypothesis. The growth-energy consumption hypothesis is also called the conservation hypothesis and suggests that any increase in GDP causes a rise in energy consumption. This statement supports the idea that the policy conservation of energy consumption has no significant or no adverse influence on economic output. The seminal work regarding the causality between economic growth and energy consumption is offered by Kraft and Kraft [4]. Investigating the case of US, between 1947 and 1974, they find strong proof for unidirectional causality, which runs from income to energy consumption. This finding is reinforced by Abosedra and Baghestani [5], who claim the same unidirectional causality from gross national product (GNP) to energy, in the US case. Their analysis covers the period 1947–1987 and follows a mixed co-integration and Granger causality tool. The energy consumption-growth hypothesis is the growth hypothesis and reveals there is one-way causality, from energy consumption to economic growth. This statement highlights the importance of energy use in the growth area, as a complementary factor to labor and capital. Any limitation in energy consumption negatively influences growth. One of the first contributions to support the energy consumption-growth hypothesis belongs to Yu and Choi [6]. They explore a set of 5 countries (i.e. Philippines, Poland, Korea, UK and US), over the 1950-1976 period, by applying Granger causalities. One of the main outputs supports the energy consumption-growth hypothesis, but only in the case of Philippines. Several years later, Stern [7] achieves the first countryspecific study which supports the growth hypothesis. The author explores the case of US, over 1947–1990, by performing a multivariate Vector Autoregressive (VAR) model. The results show that there is no Granger causality from growth to energy consumption, while a one-way causality is registered from energy consumption to growth. The feed-back hypothesis or the bi-directional causality hypothesis assumes that there is a biunivoque connection between energy consumption and economic outputs. In this case, the energy consumption
Fig. 1. Average of energy consumption and GDP in G7 countries, for the period 1970–2012. Source of data: [2] World Bank online database (2015, January).
168
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
and economic growth are jointly determined and influenced at the same time. Erol and Yu [8] are the pioneers in this topic, but performing an investigation for 6 industrialized countries including Canada, France, Germany, Italy, Japan and UK. Their performed Granger causality states that, between 1952 and 1982, the feed-back hypothesis is supported only by Japan. Following the case of Taiwan, for the period 1961–1990, Hwang and Gum [9] offers one of the first country-specific contributions which validate the feed-back hypothesis. The co-integration and error correction model claim the evidence for a two-way causality between energy consumption and economic growth. Finally, the neutrality hypothesis shows that there is no relationship between energy consumption and economic growth. According to this statement, neither conservation hypothesis nor growth hypothesis is supported. By using the Sims's [10] technique, Akarca and Long [11] and Yu and Hwang [12] offer the first empirical evidence of this hypothesis, in the case of US. The same outputs obtain Yu and Choi [6] and Erol and Yu [8], but for several industrialized countries in their multi-country analyses. The literature in the field also offers many empirical studies investigating the causality between energy consumption and economic growth, for only one country or two or more countries. In the case of the second group of contributions, considering more than two countries, focused on G7 countries or on extended samples of developed countries (i.e. they also include G7 area), the results are different (Table 1). This heterogeneity of findings is generated by the sample used and/or methodological tool chosen, which varies from the classical Granger causality to Nonlinear Panel Vector Error Correction Model. On this theoretical and empirical ground, we analyse the causality between energy consumption and economic growth in G7 countries. We also highlight that, to the best of our knowledge, there is one study (Śmiech and Papież [3]), which uses the bootstrap panel Granger causality proposed by Kónya [1], in order to
analyze the energy consumption-growth nexus, by focusing on more than two countries. In this context, our paper is one of the first studies in the literature using this technique.
3. Data and methodology Two variables are considered in order to investigate the energy consumption-growth nexus: energy consumption (EC) and economic growth (GDP). The energy consumption denotes the energy use in kt of oil equivalent, while the economic growth is captured by GDP expressed in constant 2005 billions of US dollars. The source of data is the World Bank online database [2]. Our panel includes 7 crosssections (7 countries) over 43 years (1970–2012), following the bootstrap panel Granger causality proposed by Kónya [1]. Kónya's [1] approach presents several advantages. In this technique, it is not required to test the unit root and cointegration (i.e. the variables are used in their levels, without any stationarity conditions). Additional panel information can also be obtained given the contemporaneous correlations across countries (i.e. the equations denote a Seemingly Unrelated Regressions system – SUR system). The third advantage of this tool is that no other pre-tests are needed, except the specification of the lag structure. Not at least, this method allows us to identify different Granger causality status, for each country considered. The targeted countries are Canada, France, Germany, Italy, Japan, United Kingdom and United States. Two steps should be followed before applying the bootstrap panel Granger causality: testing the panel for cross-sectional dependence and testing for cross-country heterogeneity. The cross-sectional dependence among the countries and the country-specific heterogeneity are two main assumptions in the bootstrap panel Granger causality proposed by Kónya [1].
Table 1 The main multi-countries studies which investigate the connection between energy consumption and economic growth in G7 area. Author/s
Period
Countries
Methods
Outputs
Erol and Yu [8]
1952–1982
6 Industrialized countries
Granger causality
Soytas and Sari [13]
1950–1992
G7 countries and emerging markets
Co-integration and Granger causality
Lee [14]
1960–2001 11 Developed countries
Granger causality
Soytas and Sari [15]
1960–2004 G-7 countries
Multivariate co-integration, ECM, generalized variance decompositions
Zachariadis [16]
1960–2004 G-7 countries
Bivariate VECM, ARDL, Toda and Yamamoto, Granger causality
EC2GDP: Japan; EC-GDP: Canada; GDP-EC: Italy, Germany; EC GDP: France, UK. EC2GDP: Argentina; EC-GDP: Turkey, France, Japan, Germany; GDP-EC: Italy, Korea. EC2GDP: Sweden, US; EC-GDP: Belgium, Netherlands, Canada, Switzerland; GDP-EC: France, Italy, Japan; EC GDP: Germany, UK. EC2GDP: Canada, Italy, Japan, UK; EC-GDP: France, US; GDP-EC: Germany. EC2GDP: France, Germany, Italy, Japan; EC-GDP: Canada, UK; EC GDP: US. EC-GDP. EC GDP.
Narayan and Smyth [17] 1972–2002 G-7 countries Balcilar et al. [18] 1990–2007 G-7 countries Tugcu et al. [19]
1980–2009 G-7 countries
Omay et al. [20]
1977–2007 G-7 countries
Panel co-integration, Granger causality Bootstrap Granger non-causality tests with fixed size rolling subsamples Autoregressive distributed lag, causality test by Hatemi-J Nonlinear Panel Vector Error Correction Model
EC
GDP, for non-renewable energy.
EC-GDP and/or GDP-EC; (under certain conditions).
Note: EC reveals the energy consumption, GDP illustrates the Gross Domestic Product, - denotes unidirectional causality, 2 indicates bidirectional causality, while means no causality between EC and GDP.
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
3.1. Cross-section dependence tests The cross-sectional dependence is the first assumption in our bootstrap panel Granger causality approach and refers to the presence of common shocks and unobserved components. A set of three tests is constructed in order to check this property: the Breusch and Pagan [21] Lagrange Multiplier (LM) test, the Pesaran [22] Cross-sectional Dependence (CD) test, and the Pesaran et al. [23] bias-adjusted LM test. The first test is proposed by Breusch and Pagan [21]. Their LM test is relied on the sum of squared coefficients of correlation among cross-sectional residuals, which are derived from the ordinary least squares (OLS) estimation. The null hypothesis of no cross-sectional correlations is the core of this test, requiring a large period (T) and small number of cross-sections (N). The test statistics is calculated as follows: 0 1 N 1 X N X 2 LM ¼ T @ ð1Þ ρ^ ij A i ¼ 1 j ¼ iþ1
where ρ denotes the pair-wise correlation coefficients obtained based on ordinary least squares (OLS) residuals estimation, for each cross-sections i, with i ¼1, 2,…,N. The second test is performed by Pesaran [22] for large N and small T. The null hypothesis of this test assumes that there is no cross-sectional dependence (T-1 and N-1). The CD test is as follows: 1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi0 1 X N 2T @NX ^ CD ¼ ð2Þ ρ A NðN 1Þ i ¼ 1 j ¼ i þ 1 ij ^ 2ij
Unfortunately, the CD test has low power when the population pair-wise correlations are zero. In order to deal with this issue, Pesaran et al. [23] propose the bias-adjusted test (LMadj), an adjusted version of classical LM test. The test has the following form: 1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi0 2 N 1 X N X ðT kÞρ^ ij μTij 2 @ A ^ LM adj ¼ ð3Þ ρ NðN 1Þ i ¼ 1 j ¼ i þ 1 ij υTij where k represents the number of regressors. The mean and standard deviation of ðT kÞ_ ρ 2ij are μTij and υTij, respectively. LMadj is asymptotically distributed under the null hypothesis, with T-1 and N-1.
169
test. As long as the errors are normally distributed, it is assumed that there are no restrictions on relative expansion. The test is calculated as follows: ! pffiffiffiffi N 1 S~ k pffiffiffiffiffiffi Δ~ ¼ N ð5Þ 2k Aside of this, an improved version of test can be performed, having this form ! pffiffiffiffi N 1 S~ Eðz^ iT Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Δ~ adj ¼ N ð6Þ varðz^ iT Þ where Eðz^ iT Þ ¼ k; while the variance is given by 2k(T k 1)/T þ1. 3.3. Bootstrap panel Granger causality The Granger's [26] contribution is considered the seminal work in the causality topic and reveals how past values (lags) of one variable help predicting another one. Such approach seems to be more complex when we deal with a multiple-country time-series. In this situation, it is required the deal with two issues: the cross-sectional dependence and the cross-country heteroskedasticity. The first issue is characterized by the transmission of shocks from one variable to others. This means that in our study all the countries considered are influenced by globalization and have common economic characteristics. The second issue is related to the cross-country heteroskedasticity, which illustrates that a significant economic connection in one country is not necessarily replicated by the others. The Kónya's [1] approach deals with both issues, based on SUR systems estimation and identification of Wald tests with country specific bootstrap critical values. This procedure allows us to consider all variables in their levels and perform causality output for each country. The SUR system form is as follows: EC 1;t ¼ α1;1 þ
mlE X1
λ1;1;l EC 1;t 1 þ
l¼1
EC 2;t ¼ α1;2 þ
mlE X1
mlR X1
δ1;1;l GDP 1;t 1 þ ε1;1;t
l¼1
λ1;2;l EC 2;t 1 þ
l¼1
mlR X1
δ1;2;l GDP 2;t 1 þ ε1;2;t
l¼1
:::::::::::::::::::: EC N;t ¼ α1;N þ
mlE X1
λ1;N;l EC 2Nt 1 þ
l¼1
mlR X1
δ1;N;l GDP N;t 1 þ ε1;N;t
ð7Þ
l¼1
and 3.2. Slope homogeneity tests
GDP 1;t ¼ α2;1 þ
The slope heterogeneity is the second assumption in the bootstrap panel causality approach. For small N and large T panels, the Wald test principle seems to be a good choice to check this characteristic. Wald test is a parametric statistical test which allows us to verify the true value of a coefficient relied on a sample estimation. The null hypothesis (H0) assumes that all coefficients are equal to each other, while the alternative one (Ha) considers that at least one coefficient is different from the rest. Swamy [24] offers the first slope homogeneity test needed to identify the cross-country heterogeneity, with this form: S~ ¼
N X i¼1
β^ i β~ WFE
0 x0 M x i τ i ^
σ
^ 2i
β i β~ WFE
ð4Þ
where β^ i is the estimator of pooled OLS, β~ WFE reveals the estimator calculated based on the weighted fixed-effect pooled esti2 mation, M τ shows an identity matrix, whereas σ^ i denotes the estimator of variance corresponding to the error term. Having as starting point the contribution of Swamy [24], Pesaran and Yamagata [25] proposed a standardized version of S~
mlE X2
λ2;1;l EC 1;t 1 þ
l¼1
GDP 2;t ¼ α2;2 þ
mlE X2
mlR X2
δ2;1;l GDP 1;t 1 þ ε2;1;t
l¼1
λ2;2;l EC 2;t 1 þ
l¼1
mlR X2
δ2;2;l GDP 2;t 1 þ ε2;2;t
l¼1
:::::::::::::::::::: GDP N;t ¼ α2;N þ
mlE X2 l¼1
λ2;N;l EC 2Nt 1 þ
mlR X2
δ2;N;l GDP N;t 1 þ ε2;N;t
ð8Þ
l¼1
where GDP and EC denote the log of GDP and log of energy consumption, respectively. N is the cross-section dimension (in our case, N ¼7), t denotes the time period (in our analysis, t¼43), and l represents the lag length. The common coefficient is α, the slopes are λ and δ, while ε is the error term. For each system there are maximal lags for GDP and EC, which are the same across equations. The optimal joint lag represents the lag for which the Akaike Information Criterion (AIC) and Schwartz Bayesian Criterion (SBC) have minimal levels. We also note that, following Kónya's ([1], p. 981) contribution, for any i country,” (i) there is one-way Granger-causality from GDP to EC if not all δ1,i are zero, but all λ2,i are zero, (ii) there is one-way Granger
170
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
Table 2 Cross-sectional dependence and slope homogeneity test outputs
Table 3 The bootstrap panel Granger causality results.
Method
Test statistics
p-Value
Cross-sectional dependence tests LM test CD test LMadj test
352nnn 183.7nnn 17.45nnn
0.0000 0.0000 0.0000
Slope homogeneity tests Δ~ test Δ~ adj test
32.018nnn 33.168nnn
0.000 0.000
(1) nnn The significance for 0.01 levels. (2) LM test, CD test and LMadj test show the cross-sectional dependence tests of Breusch and Pagan [21], Pesaran [22], and Pesaran et al. [23], respectively. (3) Δ~ test and Δ~ adj test illustrate the slope homogeneity tests proposed by Pesaran and Yamagata [25].
causality running from EC to GDP if all δ1,i are zero, but not all λ2,i are zero, (iii) there is two-way Granger causality between EC and GDP if neither δ1,i nor λ2,i are zero, and (iv) there is no Granger causality between EC and GDP if all δ1,i and λ2,i are zero.” For sensitivity, the classical Granger causality test is employed for each country. In this case, no cross-sectional dependence between countries is considered. We assume Vector Autoregressive (VAR) models, with variables treated in their firstdifference. Augmented Dickey–Fuller (ADF), Phillips–Perron (PP) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) tests are performed to examine the variables for the unit root property.
Country
Canada France Germany Italy Japan United Kingdom United States
Two points are crucial in order to follow the bootstrap panel Granger causality between energy consumption and GDP, in the case of G7 countries, over the period 1970–2012: the crosssectional dependence and the slope homogeneity. The results of cross-sectional dependence tests (LM test, CD ~ test and Δ ~ test and LMadj test) and slope homogeneity tests (Δ adj test) are presented in Table 2. The first set of tests, for cross-sectional dependence, clearly shows that the null hypothesis of no cross-sectional dependence is rejected for all significance levels. More precisely, this means that there is a cross-sectional dependence in the case of G7 countries. Any shock in one G7 country is transmitted to others, the SUR system estimator being more appropriate than country-bycountry pooled OLS estimator. The second part of outputs reveals that the null hypothesis of slope homogeneity is rejected for both tests and for all significance levels. In this case, the economic relationship in one G7 country is not replicated by the others. As there are both cross-sectional dependence and slope heterogeneity, the bootstrap panel Granger causality approach can be followed. Table 3 shows the main results of this technique.1 The outputs illustrate that in three cases there is a two-way Granger causality, for two situations there is one-way causality, whereas for two other cases no Granger causality is registered. More specifically, we find biunivoque causality between energy consumption and GDP in Canada, Japan and United States. Oneway causality, which runs from GDP to energy consumption, is registered for France and Germany, while no Granger causality is found for the rest of the sample (i.e. Italy and United Kingdom). 1 The TSP codes used in the bootstrap panel Granger causality approach is offered by the courtesy of Laszlo Kónya.
H0: GDP does not Granger causes EC
Wald test
Wald test
n
2.855 0.442 2.186 0.726 7.254nnn 0.655 3.362n
P-value 0.091 0.505 0.139 0.393 0.007 0.418 0.066
n
2.856 9.789nnn 5.325nn 1.407 11.076nnn 0.217 8.641nnn
p-Value 0.091 0.001 0.021 0.235 0.000 0.882 0.003
n
The significance for 0.1 levels. The significance for 0.05 levels. nnn The significance for 0.01 levels. nn
Table 4 The classical Granger causality results. Country
Canada France Germany Italy Japan United Kingdom United States n
4. Results
H0: EC does not Granger causes GDP
H0: EC does not Granger causes GDP
H0: EC does not Granger causes GDP
F-statistic
p-Value
F-statistic
p-Value
5.606nnn 0.335 0.508 0.859 0.623 3.258n 0.881
0.007 0.717 0.605 0.432 0.541 0.051 0.423
0.071 0.495 1.267 0.311 0.798 1.254 0.024
0.931 0.613 0.294 0.734 0.458 0.297 0.976
The significance for 0.1 levels. The significance for 0.01 levels.
nnn
Our results obtained for Canada and Japan fit the outputs of Erol and Yu [8], Soytas and Sari [15], and Zachariadis [16], respectively. The finding for US confirms the contributions of Lee [14]. For France and Germany, our outputs also support the results of Lee [14], Erol and Yu [8] and Soytas and Sari [15]. For Italy, our findings confirm the output of Balcilar et al. [18] and Tugcu et al. [19]. In the case of UK, we reinforce the results of Erol and Yu [8], Lee [14], Balcilar et al. [18] and Tugcu et al. [19]. Nonetheless, our findings do not confirm the rest of literature referring to G7 countries. This heterogeneity of results derives from the period of time covered and the econometric tool used. Table 4 shows the sensitivity of these outcomes, analyzed based on the results offered by the classical Granger tests, the variables being considered in their first difference.2 Based on the classical Granger causality approach, the tests show different findings. Only two countries, Canada and United Kingdom, reveal a Granger causality running from energy consumption to economic growth. For the rest of the countries no causality was found between the variables considered. The results emphasize that the energy consumption and growth nexus is very sensitive to the cross-sectional dependences between countries. In other words, the Granger causality between energy consumption and growth differs as economic transmission shocks from one G7 country to another are either considered or not.
2
Upon request, the unit root tests of variables are available for each country.
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
171
5. Conclusions
References
The findings state that both cross-sectional dependence and slope heterogeneity are present in the case of G7 countries. This strongly supports the idea that all G7 countries are influenced by globalization, having common economic features, as the most developed world countries. Any economic structural connection of one country is not replicated by the others. Canada, Japan and United States support the feed-back hypothesis, the energy consumption and economic growth being jointly determined and influenced at the same time. The conservation hypothesis is validated in the case of France and Germany. In these countries, the economic growth influences the energy consumption. The neutrality hypothesis is demonstrated by Italy and United Kingdom. Here, neither conservation hypothesis nor growth hypothesis is supported. Regarding the policy implications, in Canada, Japan and United States, the policymakers should jointly manage the energy consumption and economic growth, following at the same time energy and economic growth adjustments. They should take into account that the energy conservation policy for reducing energy consumption can damage the growth performance and, furthermore, affect the energy consumption through growth compression. France and Germany show that the policy conservation of energy consumption is not the right way for political decisions. In this situation, only the independent economic growth adjustments are recommended while the interest should be especially oriented towards the labor and capital factors. For Italy and United Kingdom, neither conservation hypothesis nor growth hypothesis is supported. In these countries, the policymakers should take into account that the energy conservation policies are neutral in respect to economic growth. Conversely, the economic growth has no impact on energy consumption. All these recommendations are valid only if the considered countries have a strong'economic openness' to each other. If such dependencies are attenuated, their policies should be thoroughly reconsidered. Canada and United Kingdom follow the growth hypothesis trend, assuming that energy consumption is a good incentive for growth. In this case the decision should be carefully made, as the energy conservation policies which reduce energy consumption will have a negative influence on economic growth. For the rest of the countries, the neutral assumption should be the main signal for policymakers.
[1] Kónya L. Exports and growth: Granger causality analysis on OECD countries with a panel data approach. Econ Model 2006;23(6):978–92. [2] World Bank. World Bank online database; January 2015. [3] Śmiech S, Papież 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. [4] Kraft J, Kraft A. On the relationship between energy and GNP. J Energy Dev 1978;3:401–3. [5] Abosedra S, Baghestani H. New evidence on the causal relationship between United States energy consumption and gross national product. J Energy Dev 1989;14:285–92. [6] Yu ESH, Choi JY. The causal relationship between energy and GNP: an international comparison. J Energy Dev 1985;10:249–72. [7] Stern DI. Energy and economic growth in the USA: a multivariate approach. Energy Econ 1993;15:137–50. [8] Erol U, Yu ESH. On the causal relationship between energy and income for industrializing countries. J Energy Dev 1987;13:113–22. [9] Hwang D, Gum B. The causal relationship between energy and GNP: the case of Taiwan. J Energy Dev 1991;16:219–26. [10] Sims CA. Money, income and causality. Am Econ Rev 1972;62:540–52. [11] Akarca AT, Long TV. On the relationship between energy and GNP: a reexamination. J Energ Dev 1980;5:326–31. [12] Yu ESH, Hwang BK. The relationship between energy and GNP, further results. Energy Econ 1984;6:186–90. [13] Soytas U, Sari R. Energy consumption and GDP: causality relationship in G7 countries and emerging markets. Energy Econ 2003;25:33–7. [14] Lee CC. Energy consumption and GDP in developing countries: a cointegrated panel analysis. Energy Econ 2005;27:415–27. [15] Soytas U, Sari R. Energy consumption and income in G7 countries. J Policy Model 2006;28:739–50. [16] Zachariadis T. Exploring the relationship between energy use and economic growth with bivariate models: new evidence from G-7 countries. Energy Econ 2007;29(6):1233–53. [17] Narayan PK, Smyth R. Energy consumption and real GDP in G7 countries: new evidence from panel cointegration with structural breaks. Energy Econ 2008;30:2331–41. [18] Balcilar M, Ozdemir ZA, Arslanturk Y. Economic growth and energy consumption causal nexus viewed through a bootstrap rolling window. Energy Econ 2010;32(6):1398–410. [19] Tugcu CT, Ozturk I, Aslan A. Renewable and nonrenewable energy consumption and economic growth relationship revisited: evidence from G7 countries. Energy Econ 2012;34(6):1942–50. [20] Omay T, Hasanov M, Ucar N. Energy consumption and economic growth: evidence from nonlinear panel cointegration and causality tests. Appl Econ 2014;34(2):36–55. [21] Breusch TS, Pagan AR. The Lagrange multiplier test and its applications to model specification in econometrics. Rev Econ Stud 1980;47:239–53. [22] Pesaran MH. General diagnostic tests for cross section dependence in panels. Cambridge Working Papers in Economics No. 0435. Faculty of Economics. University of Cambridge; 2004. [23] Pesaran MH, Ullah A, Yamagata T. A bias-adjusted LM test of error crosssection independence. Economet J 2008;11:105–27. [24] Swamy PAVB. Efficient inference in a random coefficient regression model. Econometrica 1970;38:311–23. [25] Pesaran MH, Yamagata T. Testing slope homogeneity in large panels. J Econometr 2008;142:50–93. [26] Granger CWJ. Investigating causal relations by econometric models and crossspectral methods. Econometrica 1969;37:424–38.
Acknowledgments The author would like to thank the anonymous reviewers for their helpful comments. Special thanks go to Lázsló Kónya for his support and suggestions offered for the empirical part of this research. The author also thanks Suleyman Bolat, Aviral Tiwari and Stefano Fachin. Any error or omission in this paper belongs to the author.
Contents lists available at ScienceDirect
Renewable and Sustainable Energy Reviews journal homepage: www.elsevier.com/locate/rser
A bootstrap panel Granger causality analysis of energy consumption and economic growth in the G7 countries Mihai Mutascu a,b,n a LEO (Laboratoire d'Economie d'Orléans) UMR7322, Faculté de Droit d'Economie et de Gestion, University of Orléans, Rue de Blois, B.P. 6739, 45067 Orléans, France b East European Centre for Research in Economics and Business, West University of Timisoara, 16 H. Pestalozzi St., 300115 Timisoara, Romania
art ic l e i nf o
a b s t r a c t
Article history: Received 5 February 2015 Received in revised form 4 February 2016 Accepted 13 May 2016
The paper investigates the causality between energy consumption and economic growth in the countries which are members of the Group of Seven (G7), during the period 1970–2012, by following the bootstrap panel Granger causality approach. The main aim of the paper is to identify the right policies to deal with the interaction between energy consumption and economic growth. The findings show a bi-directional causality between energy consumption and GDP in Canada, Japan and United States. GDP causes energy consumption in France and Germany, while no causality is found for the rest of the sample (i.e. Italy and United Kingdom). These outcomes are very sensitive to the crosssectional dependences between countries. Therefore, their environmental and growth policies should be widely reconsidered when the level of interdependence dramatically decreases. & 2016 Elsevier Ltd. All rights reserved.
Keywords: Energy consumption Growth Bootstrap panel causality G7 countries
Contents 1. 2. 3.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data and methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Cross-section dependence tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Slope homogeneity tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Bootstrap panel Granger causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1. Introduction The connection and causality between energy consumption and economic growth have been extensively explored over the last decades, becoming one of the most important topics in the energy economics literature. Starting with February 2005, when the Kyoto Protocol has been implemented, the interaction between energy
n Correspondence address: East European Centre for Research in Economics and Business, West University of Timisoara, 16 H. Pestalozzi St., 300115 Timisoara, Romania. Tel.: þ40 256 592 556; fax: þ40 256 592500. E-mail addresses: [email protected], [email protected]
http://dx.doi.org/10.1016/j.rser.2016.05.055 1364-0321/& 2016 Elsevier Ltd. All rights reserved.
166 167 168 169 169 169 170 171 171 171
consumption and economic growth registered new valences, especially in developed countries. This international treaty is an extension of the United Nations Framework Convention on Climate Change (UNFCCC) from 1992 and has as main objective the reduction of greenhouse gas emissions in developed countries to the level in 1990. There are two main premises of this target: the global warming exists and it has been caused by the man-made carbon-dioxide (CO2) emissions. During the years, many researchers started to revisit the connection and causality between energy consumption and economic growth, for various countries and periods, by following different methodological tools. The results are very heterogeneous and emphasize different research perspectives, from causality to no causality statements.
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
On this ground, the aim of the paper is to analyse the energy consumption-economic growth nexus, in case of the Group of Seven (G7 countries), by using a bootstrap panel Granger causality approach, for the period 1970–2012. The G7 area includes the world's seven most industrialized economies, more precisely: Canada, France, Germany, Italy, Japan, United Kingdom (UK) and United States (US). Under Kyoto Protocol, this group of developed countries arises a special interest, as any influence of energy consumption policy on growth requires climate change response strategies. US is included in the sample, even if this country does not ratify the Kyoto Protocol. US has intensive trade and financial relationships with the rest of G-7 countries, being one of the first energy consumer in the world. Therefore, these characteristics give consistence of cross-sectional dependence property of G7 panel, which is one of the main assumption of the methodological tool followed (i.e. complex and intensive economic relationships run between G7 countries). Moreover, in 2012, during the 18th Conference of the Parties (COP18), all delegates agreed to extend the Kyoto Protocol until 2020, while US assumed the reduction of the carbon-dioxide emissions and greenhouse gases. Between 1970 and 2012 (Fig. 1), all G7 countries registered an increasing tendency for both energy consumption (average of energy use in kt of oil equivalent) and gross domestic product (GDP) (average GDP expressed in constant 2005 billions of US dollars). For the overall period, the average GDP illustrates an emphasized growth tendency, whereas the average energy consumption seems to follow a smooth increase. Significant negative shocks are registered over the 2007–2010 period, for both variables, as effects of world economic crises. The paper extends the literature in the field by offering the first study to investigate the relationship and causality between energy consumption and economic growth in the case of G7 countries, by following the bootstrap panel Granger causality developed by Kónya [1]. This technique is more powerful compared to existing ones, as it takes into account cross-sectional dependence and cross-country heterogeneity. Besides the contribution of Śmiech and Papież [3], to the best of our knowledge, this is one of the first papers which uses the Kónya's [1] tool in energy economics area. In comparison, Śmiech and Papież [3] explore European Union (EU) member states and analyse the period 1993–2011. The second contribution to the literature is given by the panel dimension, which covers an extended period of time, from 1970 to 2012, on annual basis. Finally, the outputs offer an important support for the economic energy policy makers in the G7 countries, helping them to manage, over the long-run, the interaction between energy consumption and economic growth. The rest of the paper is as follows: Section 2 reviews the literature, Section 3 presents the data and methodology, while Section 4 shows the empirical results. Section 5 concludes.
167
2. Literature review The literature about the linkage and causality between energy consumption and economic growth is prolific and offers a diversity of findings, deriving from the targeted countries, periods of analysis and econometric tools used. Relying on these different outcomes, the literature also offers contradictory policy solutions for policymakers regarding the adjustments of the relationship between energy consumption and economic growth. There are four main hypotheses outlining the literature review in respect to the energy consumption-growth nexus: (i) growthenergy consumption hypothesis, (ii) energy consumption-growth hypothesis, (iii) feed-back hypothesis, and (iv) neutrality hypothesis. The growth-energy consumption hypothesis is also called the conservation hypothesis and suggests that any increase in GDP causes a rise in energy consumption. This statement supports the idea that the policy conservation of energy consumption has no significant or no adverse influence on economic output. The seminal work regarding the causality between economic growth and energy consumption is offered by Kraft and Kraft [4]. Investigating the case of US, between 1947 and 1974, they find strong proof for unidirectional causality, which runs from income to energy consumption. This finding is reinforced by Abosedra and Baghestani [5], who claim the same unidirectional causality from gross national product (GNP) to energy, in the US case. Their analysis covers the period 1947–1987 and follows a mixed co-integration and Granger causality tool. The energy consumption-growth hypothesis is the growth hypothesis and reveals there is one-way causality, from energy consumption to economic growth. This statement highlights the importance of energy use in the growth area, as a complementary factor to labor and capital. Any limitation in energy consumption negatively influences growth. One of the first contributions to support the energy consumption-growth hypothesis belongs to Yu and Choi [6]. They explore a set of 5 countries (i.e. Philippines, Poland, Korea, UK and US), over the 1950-1976 period, by applying Granger causalities. One of the main outputs supports the energy consumption-growth hypothesis, but only in the case of Philippines. Several years later, Stern [7] achieves the first countryspecific study which supports the growth hypothesis. The author explores the case of US, over 1947–1990, by performing a multivariate Vector Autoregressive (VAR) model. The results show that there is no Granger causality from growth to energy consumption, while a one-way causality is registered from energy consumption to growth. The feed-back hypothesis or the bi-directional causality hypothesis assumes that there is a biunivoque connection between energy consumption and economic outputs. In this case, the energy consumption
Fig. 1. Average of energy consumption and GDP in G7 countries, for the period 1970–2012. Source of data: [2] World Bank online database (2015, January).
168
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
and economic growth are jointly determined and influenced at the same time. Erol and Yu [8] are the pioneers in this topic, but performing an investigation for 6 industrialized countries including Canada, France, Germany, Italy, Japan and UK. Their performed Granger causality states that, between 1952 and 1982, the feed-back hypothesis is supported only by Japan. Following the case of Taiwan, for the period 1961–1990, Hwang and Gum [9] offers one of the first country-specific contributions which validate the feed-back hypothesis. The co-integration and error correction model claim the evidence for a two-way causality between energy consumption and economic growth. Finally, the neutrality hypothesis shows that there is no relationship between energy consumption and economic growth. According to this statement, neither conservation hypothesis nor growth hypothesis is supported. By using the Sims's [10] technique, Akarca and Long [11] and Yu and Hwang [12] offer the first empirical evidence of this hypothesis, in the case of US. The same outputs obtain Yu and Choi [6] and Erol and Yu [8], but for several industrialized countries in their multi-country analyses. The literature in the field also offers many empirical studies investigating the causality between energy consumption and economic growth, for only one country or two or more countries. In the case of the second group of contributions, considering more than two countries, focused on G7 countries or on extended samples of developed countries (i.e. they also include G7 area), the results are different (Table 1). This heterogeneity of findings is generated by the sample used and/or methodological tool chosen, which varies from the classical Granger causality to Nonlinear Panel Vector Error Correction Model. On this theoretical and empirical ground, we analyse the causality between energy consumption and economic growth in G7 countries. We also highlight that, to the best of our knowledge, there is one study (Śmiech and Papież [3]), which uses the bootstrap panel Granger causality proposed by Kónya [1], in order to
analyze the energy consumption-growth nexus, by focusing on more than two countries. In this context, our paper is one of the first studies in the literature using this technique.
3. Data and methodology Two variables are considered in order to investigate the energy consumption-growth nexus: energy consumption (EC) and economic growth (GDP). The energy consumption denotes the energy use in kt of oil equivalent, while the economic growth is captured by GDP expressed in constant 2005 billions of US dollars. The source of data is the World Bank online database [2]. Our panel includes 7 crosssections (7 countries) over 43 years (1970–2012), following the bootstrap panel Granger causality proposed by Kónya [1]. Kónya's [1] approach presents several advantages. In this technique, it is not required to test the unit root and cointegration (i.e. the variables are used in their levels, without any stationarity conditions). Additional panel information can also be obtained given the contemporaneous correlations across countries (i.e. the equations denote a Seemingly Unrelated Regressions system – SUR system). The third advantage of this tool is that no other pre-tests are needed, except the specification of the lag structure. Not at least, this method allows us to identify different Granger causality status, for each country considered. The targeted countries are Canada, France, Germany, Italy, Japan, United Kingdom and United States. Two steps should be followed before applying the bootstrap panel Granger causality: testing the panel for cross-sectional dependence and testing for cross-country heterogeneity. The cross-sectional dependence among the countries and the country-specific heterogeneity are two main assumptions in the bootstrap panel Granger causality proposed by Kónya [1].
Table 1 The main multi-countries studies which investigate the connection between energy consumption and economic growth in G7 area. Author/s
Period
Countries
Methods
Outputs
Erol and Yu [8]
1952–1982
6 Industrialized countries
Granger causality
Soytas and Sari [13]
1950–1992
G7 countries and emerging markets
Co-integration and Granger causality
Lee [14]
1960–2001 11 Developed countries
Granger causality
Soytas and Sari [15]
1960–2004 G-7 countries
Multivariate co-integration, ECM, generalized variance decompositions
Zachariadis [16]
1960–2004 G-7 countries
Bivariate VECM, ARDL, Toda and Yamamoto, Granger causality
EC2GDP: Japan; EC-GDP: Canada; GDP-EC: Italy, Germany; EC GDP: France, UK. EC2GDP: Argentina; EC-GDP: Turkey, France, Japan, Germany; GDP-EC: Italy, Korea. EC2GDP: Sweden, US; EC-GDP: Belgium, Netherlands, Canada, Switzerland; GDP-EC: France, Italy, Japan; EC GDP: Germany, UK. EC2GDP: Canada, Italy, Japan, UK; EC-GDP: France, US; GDP-EC: Germany. EC2GDP: France, Germany, Italy, Japan; EC-GDP: Canada, UK; EC GDP: US. EC-GDP. EC GDP.
Narayan and Smyth [17] 1972–2002 G-7 countries Balcilar et al. [18] 1990–2007 G-7 countries Tugcu et al. [19]
1980–2009 G-7 countries
Omay et al. [20]
1977–2007 G-7 countries
Panel co-integration, Granger causality Bootstrap Granger non-causality tests with fixed size rolling subsamples Autoregressive distributed lag, causality test by Hatemi-J Nonlinear Panel Vector Error Correction Model
EC
GDP, for non-renewable energy.
EC-GDP and/or GDP-EC; (under certain conditions).
Note: EC reveals the energy consumption, GDP illustrates the Gross Domestic Product, - denotes unidirectional causality, 2 indicates bidirectional causality, while means no causality between EC and GDP.
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
3.1. Cross-section dependence tests The cross-sectional dependence is the first assumption in our bootstrap panel Granger causality approach and refers to the presence of common shocks and unobserved components. A set of three tests is constructed in order to check this property: the Breusch and Pagan [21] Lagrange Multiplier (LM) test, the Pesaran [22] Cross-sectional Dependence (CD) test, and the Pesaran et al. [23] bias-adjusted LM test. The first test is proposed by Breusch and Pagan [21]. Their LM test is relied on the sum of squared coefficients of correlation among cross-sectional residuals, which are derived from the ordinary least squares (OLS) estimation. The null hypothesis of no cross-sectional correlations is the core of this test, requiring a large period (T) and small number of cross-sections (N). The test statistics is calculated as follows: 0 1 N 1 X N X 2 LM ¼ T @ ð1Þ ρ^ ij A i ¼ 1 j ¼ iþ1
where ρ denotes the pair-wise correlation coefficients obtained based on ordinary least squares (OLS) residuals estimation, for each cross-sections i, with i ¼1, 2,…,N. The second test is performed by Pesaran [22] for large N and small T. The null hypothesis of this test assumes that there is no cross-sectional dependence (T-1 and N-1). The CD test is as follows: 1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi0 1 X N 2T @NX ^ CD ¼ ð2Þ ρ A NðN 1Þ i ¼ 1 j ¼ i þ 1 ij ^ 2ij
Unfortunately, the CD test has low power when the population pair-wise correlations are zero. In order to deal with this issue, Pesaran et al. [23] propose the bias-adjusted test (LMadj), an adjusted version of classical LM test. The test has the following form: 1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi0 2 N 1 X N X ðT kÞρ^ ij μTij 2 @ A ^ LM adj ¼ ð3Þ ρ NðN 1Þ i ¼ 1 j ¼ i þ 1 ij υTij where k represents the number of regressors. The mean and standard deviation of ðT kÞ_ ρ 2ij are μTij and υTij, respectively. LMadj is asymptotically distributed under the null hypothesis, with T-1 and N-1.
169
test. As long as the errors are normally distributed, it is assumed that there are no restrictions on relative expansion. The test is calculated as follows: ! pffiffiffiffi N 1 S~ k pffiffiffiffiffiffi Δ~ ¼ N ð5Þ 2k Aside of this, an improved version of test can be performed, having this form ! pffiffiffiffi N 1 S~ Eðz^ iT Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Δ~ adj ¼ N ð6Þ varðz^ iT Þ where Eðz^ iT Þ ¼ k; while the variance is given by 2k(T k 1)/T þ1. 3.3. Bootstrap panel Granger causality The Granger's [26] contribution is considered the seminal work in the causality topic and reveals how past values (lags) of one variable help predicting another one. Such approach seems to be more complex when we deal with a multiple-country time-series. In this situation, it is required the deal with two issues: the cross-sectional dependence and the cross-country heteroskedasticity. The first issue is characterized by the transmission of shocks from one variable to others. This means that in our study all the countries considered are influenced by globalization and have common economic characteristics. The second issue is related to the cross-country heteroskedasticity, which illustrates that a significant economic connection in one country is not necessarily replicated by the others. The Kónya's [1] approach deals with both issues, based on SUR systems estimation and identification of Wald tests with country specific bootstrap critical values. This procedure allows us to consider all variables in their levels and perform causality output for each country. The SUR system form is as follows: EC 1;t ¼ α1;1 þ
mlE X1
λ1;1;l EC 1;t 1 þ
l¼1
EC 2;t ¼ α1;2 þ
mlE X1
mlR X1
δ1;1;l GDP 1;t 1 þ ε1;1;t
l¼1
λ1;2;l EC 2;t 1 þ
l¼1
mlR X1
δ1;2;l GDP 2;t 1 þ ε1;2;t
l¼1
:::::::::::::::::::: EC N;t ¼ α1;N þ
mlE X1
λ1;N;l EC 2Nt 1 þ
l¼1
mlR X1
δ1;N;l GDP N;t 1 þ ε1;N;t
ð7Þ
l¼1
and 3.2. Slope homogeneity tests
GDP 1;t ¼ α2;1 þ
The slope heterogeneity is the second assumption in the bootstrap panel causality approach. For small N and large T panels, the Wald test principle seems to be a good choice to check this characteristic. Wald test is a parametric statistical test which allows us to verify the true value of a coefficient relied on a sample estimation. The null hypothesis (H0) assumes that all coefficients are equal to each other, while the alternative one (Ha) considers that at least one coefficient is different from the rest. Swamy [24] offers the first slope homogeneity test needed to identify the cross-country heterogeneity, with this form: S~ ¼
N X i¼1
β^ i β~ WFE
0 x0 M x i τ i ^
σ
^ 2i
β i β~ WFE
ð4Þ
where β^ i is the estimator of pooled OLS, β~ WFE reveals the estimator calculated based on the weighted fixed-effect pooled esti2 mation, M τ shows an identity matrix, whereas σ^ i denotes the estimator of variance corresponding to the error term. Having as starting point the contribution of Swamy [24], Pesaran and Yamagata [25] proposed a standardized version of S~
mlE X2
λ2;1;l EC 1;t 1 þ
l¼1
GDP 2;t ¼ α2;2 þ
mlE X2
mlR X2
δ2;1;l GDP 1;t 1 þ ε2;1;t
l¼1
λ2;2;l EC 2;t 1 þ
l¼1
mlR X2
δ2;2;l GDP 2;t 1 þ ε2;2;t
l¼1
:::::::::::::::::::: GDP N;t ¼ α2;N þ
mlE X2 l¼1
λ2;N;l EC 2Nt 1 þ
mlR X2
δ2;N;l GDP N;t 1 þ ε2;N;t
ð8Þ
l¼1
where GDP and EC denote the log of GDP and log of energy consumption, respectively. N is the cross-section dimension (in our case, N ¼7), t denotes the time period (in our analysis, t¼43), and l represents the lag length. The common coefficient is α, the slopes are λ and δ, while ε is the error term. For each system there are maximal lags for GDP and EC, which are the same across equations. The optimal joint lag represents the lag for which the Akaike Information Criterion (AIC) and Schwartz Bayesian Criterion (SBC) have minimal levels. We also note that, following Kónya's ([1], p. 981) contribution, for any i country,” (i) there is one-way Granger-causality from GDP to EC if not all δ1,i are zero, but all λ2,i are zero, (ii) there is one-way Granger
170
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
Table 2 Cross-sectional dependence and slope homogeneity test outputs
Table 3 The bootstrap panel Granger causality results.
Method
Test statistics
p-Value
Cross-sectional dependence tests LM test CD test LMadj test
352nnn 183.7nnn 17.45nnn
0.0000 0.0000 0.0000
Slope homogeneity tests Δ~ test Δ~ adj test
32.018nnn 33.168nnn
0.000 0.000
(1) nnn The significance for 0.01 levels. (2) LM test, CD test and LMadj test show the cross-sectional dependence tests of Breusch and Pagan [21], Pesaran [22], and Pesaran et al. [23], respectively. (3) Δ~ test and Δ~ adj test illustrate the slope homogeneity tests proposed by Pesaran and Yamagata [25].
causality running from EC to GDP if all δ1,i are zero, but not all λ2,i are zero, (iii) there is two-way Granger causality between EC and GDP if neither δ1,i nor λ2,i are zero, and (iv) there is no Granger causality between EC and GDP if all δ1,i and λ2,i are zero.” For sensitivity, the classical Granger causality test is employed for each country. In this case, no cross-sectional dependence between countries is considered. We assume Vector Autoregressive (VAR) models, with variables treated in their firstdifference. Augmented Dickey–Fuller (ADF), Phillips–Perron (PP) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) tests are performed to examine the variables for the unit root property.
Country
Canada France Germany Italy Japan United Kingdom United States
Two points are crucial in order to follow the bootstrap panel Granger causality between energy consumption and GDP, in the case of G7 countries, over the period 1970–2012: the crosssectional dependence and the slope homogeneity. The results of cross-sectional dependence tests (LM test, CD ~ test and Δ ~ test and LMadj test) and slope homogeneity tests (Δ adj test) are presented in Table 2. The first set of tests, for cross-sectional dependence, clearly shows that the null hypothesis of no cross-sectional dependence is rejected for all significance levels. More precisely, this means that there is a cross-sectional dependence in the case of G7 countries. Any shock in one G7 country is transmitted to others, the SUR system estimator being more appropriate than country-bycountry pooled OLS estimator. The second part of outputs reveals that the null hypothesis of slope homogeneity is rejected for both tests and for all significance levels. In this case, the economic relationship in one G7 country is not replicated by the others. As there are both cross-sectional dependence and slope heterogeneity, the bootstrap panel Granger causality approach can be followed. Table 3 shows the main results of this technique.1 The outputs illustrate that in three cases there is a two-way Granger causality, for two situations there is one-way causality, whereas for two other cases no Granger causality is registered. More specifically, we find biunivoque causality between energy consumption and GDP in Canada, Japan and United States. Oneway causality, which runs from GDP to energy consumption, is registered for France and Germany, while no Granger causality is found for the rest of the sample (i.e. Italy and United Kingdom). 1 The TSP codes used in the bootstrap panel Granger causality approach is offered by the courtesy of Laszlo Kónya.
H0: GDP does not Granger causes EC
Wald test
Wald test
n
2.855 0.442 2.186 0.726 7.254nnn 0.655 3.362n
P-value 0.091 0.505 0.139 0.393 0.007 0.418 0.066
n
2.856 9.789nnn 5.325nn 1.407 11.076nnn 0.217 8.641nnn
p-Value 0.091 0.001 0.021 0.235 0.000 0.882 0.003
n
The significance for 0.1 levels. The significance for 0.05 levels. nnn The significance for 0.01 levels. nn
Table 4 The classical Granger causality results. Country
Canada France Germany Italy Japan United Kingdom United States n
4. Results
H0: EC does not Granger causes GDP
H0: EC does not Granger causes GDP
H0: EC does not Granger causes GDP
F-statistic
p-Value
F-statistic
p-Value
5.606nnn 0.335 0.508 0.859 0.623 3.258n 0.881
0.007 0.717 0.605 0.432 0.541 0.051 0.423
0.071 0.495 1.267 0.311 0.798 1.254 0.024
0.931 0.613 0.294 0.734 0.458 0.297 0.976
The significance for 0.1 levels. The significance for 0.01 levels.
nnn
Our results obtained for Canada and Japan fit the outputs of Erol and Yu [8], Soytas and Sari [15], and Zachariadis [16], respectively. The finding for US confirms the contributions of Lee [14]. For France and Germany, our outputs also support the results of Lee [14], Erol and Yu [8] and Soytas and Sari [15]. For Italy, our findings confirm the output of Balcilar et al. [18] and Tugcu et al. [19]. In the case of UK, we reinforce the results of Erol and Yu [8], Lee [14], Balcilar et al. [18] and Tugcu et al. [19]. Nonetheless, our findings do not confirm the rest of literature referring to G7 countries. This heterogeneity of results derives from the period of time covered and the econometric tool used. Table 4 shows the sensitivity of these outcomes, analyzed based on the results offered by the classical Granger tests, the variables being considered in their first difference.2 Based on the classical Granger causality approach, the tests show different findings. Only two countries, Canada and United Kingdom, reveal a Granger causality running from energy consumption to economic growth. For the rest of the countries no causality was found between the variables considered. The results emphasize that the energy consumption and growth nexus is very sensitive to the cross-sectional dependences between countries. In other words, the Granger causality between energy consumption and growth differs as economic transmission shocks from one G7 country to another are either considered or not.
2
Upon request, the unit root tests of variables are available for each country.
M. Mutascu / Renewable and Sustainable Energy Reviews 63 (2016) 166–171
171
5. Conclusions
References
The findings state that both cross-sectional dependence and slope heterogeneity are present in the case of G7 countries. This strongly supports the idea that all G7 countries are influenced by globalization, having common economic features, as the most developed world countries. Any economic structural connection of one country is not replicated by the others. Canada, Japan and United States support the feed-back hypothesis, the energy consumption and economic growth being jointly determined and influenced at the same time. The conservation hypothesis is validated in the case of France and Germany. In these countries, the economic growth influences the energy consumption. The neutrality hypothesis is demonstrated by Italy and United Kingdom. Here, neither conservation hypothesis nor growth hypothesis is supported. Regarding the policy implications, in Canada, Japan and United States, the policymakers should jointly manage the energy consumption and economic growth, following at the same time energy and economic growth adjustments. They should take into account that the energy conservation policy for reducing energy consumption can damage the growth performance and, furthermore, affect the energy consumption through growth compression. France and Germany show that the policy conservation of energy consumption is not the right way for political decisions. In this situation, only the independent economic growth adjustments are recommended while the interest should be especially oriented towards the labor and capital factors. For Italy and United Kingdom, neither conservation hypothesis nor growth hypothesis is supported. In these countries, the policymakers should take into account that the energy conservation policies are neutral in respect to economic growth. Conversely, the economic growth has no impact on energy consumption. All these recommendations are valid only if the considered countries have a strong'economic openness' to each other. If such dependencies are attenuated, their policies should be thoroughly reconsidered. Canada and United Kingdom follow the growth hypothesis trend, assuming that energy consumption is a good incentive for growth. In this case the decision should be carefully made, as the energy conservation policies which reduce energy consumption will have a negative influence on economic growth. For the rest of the countries, the neutral assumption should be the main signal for policymakers.
[1] Kónya L. Exports and growth: Granger causality analysis on OECD countries with a panel data approach. Econ Model 2006;23(6):978–92. [2] World Bank. World Bank online database; January 2015. [3] Śmiech S, Papież 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. [4] Kraft J, Kraft A. On the relationship between energy and GNP. J Energy Dev 1978;3:401–3. [5] Abosedra S, Baghestani H. New evidence on the causal relationship between United States energy consumption and gross national product. J Energy Dev 1989;14:285–92. [6] Yu ESH, Choi JY. The causal relationship between energy and GNP: an international comparison. J Energy Dev 1985;10:249–72. [7] Stern DI. Energy and economic growth in the USA: a multivariate approach. Energy Econ 1993;15:137–50. [8] Erol U, Yu ESH. On the causal relationship between energy and income for industrializing countries. J Energy Dev 1987;13:113–22. [9] Hwang D, Gum B. The causal relationship between energy and GNP: the case of Taiwan. J Energy Dev 1991;16:219–26. [10] Sims CA. Money, income and causality. Am Econ Rev 1972;62:540–52. [11] Akarca AT, Long TV. On the relationship between energy and GNP: a reexamination. J Energ Dev 1980;5:326–31. [12] Yu ESH, Hwang BK. The relationship between energy and GNP, further results. Energy Econ 1984;6:186–90. [13] Soytas U, Sari R. Energy consumption and GDP: causality relationship in G7 countries and emerging markets. Energy Econ 2003;25:33–7. [14] Lee CC. Energy consumption and GDP in developing countries: a cointegrated panel analysis. Energy Econ 2005;27:415–27. [15] Soytas U, Sari R. Energy consumption and income in G7 countries. J Policy Model 2006;28:739–50. [16] Zachariadis T. Exploring the relationship between energy use and economic growth with bivariate models: new evidence from G-7 countries. Energy Econ 2007;29(6):1233–53. [17] Narayan PK, Smyth R. Energy consumption and real GDP in G7 countries: new evidence from panel cointegration with structural breaks. Energy Econ 2008;30:2331–41. [18] Balcilar M, Ozdemir ZA, Arslanturk Y. Economic growth and energy consumption causal nexus viewed through a bootstrap rolling window. Energy Econ 2010;32(6):1398–410. [19] Tugcu CT, Ozturk I, Aslan A. Renewable and nonrenewable energy consumption and economic growth relationship revisited: evidence from G7 countries. Energy Econ 2012;34(6):1942–50. [20] Omay T, Hasanov M, Ucar N. Energy consumption and economic growth: evidence from nonlinear panel cointegration and causality tests. Appl Econ 2014;34(2):36–55. [21] Breusch TS, Pagan AR. The Lagrange multiplier test and its applications to model specification in econometrics. Rev Econ Stud 1980;47:239–53. [22] Pesaran MH. General diagnostic tests for cross section dependence in panels. Cambridge Working Papers in Economics No. 0435. Faculty of Economics. University of Cambridge; 2004. [23] Pesaran MH, Ullah A, Yamagata T. A bias-adjusted LM test of error crosssection independence. Economet J 2008;11:105–27. [24] Swamy PAVB. Efficient inference in a random coefficient regression model. Econometrica 1970;38:311–23. [25] Pesaran MH, Yamagata T. Testing slope homogeneity in large panels. J Econometr 2008;142:50–93. [26] Granger CWJ. Investigating causal relations by econometric models and crossspectral methods. Econometrica 1969;37:424–38.
Acknowledgments The author would like to thank the anonymous reviewers for their helpful comments. Special thanks go to Lázsló Kónya for his support and suggestions offered for the empirical part of this research. The author also thanks Suleyman Bolat, Aviral Tiwari and Stefano Fachin. Any error or omission in this paper belongs to the author.