Energy Economics 34 (2012) 1051–1057

Contents lists available at SciVerse ScienceDirect

Energy Economics
journal homepage: www.elsevier.com/locate/eneco

Why do some emerging economies proactively accelerate the

adoption of renewable energy?
Ruhul A. Salim a,⁎, Shuddhasattwa Rafiq b
a
School of Economics & Finance, Curtin Business School, Curtin University, Perth, WA 6845, Australia
b
Institute of Business Administration, Jahangirnagar University, Savar, Dhaka, Bangladesh

a r t i c l e i n f o a b s t r a c t

Article history: This article analyses the determinants of renewable energy consumption in a panel of six major emerging
Received 29 December 2010 economies, namely Brazil, China, India, Indonesia, Philippines and Turkey that are proactively accelerating
Received in revised form 13 August 2011 the adoption of renewable energy. Using Fully modified ordinary least square (FMOLS), Dynamic ordinary
Accepted 19 August 2011
least square (DOLS), and Granger causality methods this paper finds that in the long-run, renewable ener-
Available online 28 August 2011
gy consumption is significantly determined by income and pollutant emission in Brazil, China, India and
JEL classifications:
Indonesia while mainly by income in Philippines and Turkey. Causal link between renewable energy and
C22 income; and between renewable energy and pollutant emission are found to be bidirectional in the
C32 short-run. These results suggest that the appropriateness of the efforts undertaken by emerging countries
Q21 to reduce the carbon intensity by increasing the energy efficiency and substantially increasing the share of
Q43 renewable in the overall energy mix.
© 2011 Elsevier B.V. All rights reserved.
Keywords:
Autoregressive distributed lag (ARDL)
Emerging economies
Oil prices
Panel cointegration
Pollutant emission
Renewable energy

1. Introduction capacity grew by 70%, wind power grew by 29%, solar hot water in-
creased by 15%, and small hydro expanded by 8% (El-Ashry, 2009).
Fossil fuel (i.e., coal, oil and natural gas) led economic growth, The Renewable Global Status Report (REN21, 2009) further provides a
through its release of carbon dioxide (CO2) into the atmosphere, is con- ranking of top five renewable energy investor countries along with a
sidered to be the main driver behind global warming and climate ranking of top five countries according to their investment and capacity
change (IPCC, 2007; Stern, 2006). Increased concern over issues related size in renewable energy till 2008. This report reveals that emerging
to energy security and global warming suggests that in the future there economies such as Brazil, China, India, Indonesia, Philippines and Tur-
will be a greater reliance on renewable energy sources (i.e., wind, solar, key are investing substantially in different renewable energy sources.
geothermal, hydro, biomass, wave and tidal). Moreover, with soaring Although MacKenzie (2003) argues that the less developed countries
energy prices at the backdrop, more attention has been drawn to the (LCDs) should be permitted to burn fossil fuels for some time to allow
further exploration of renewable energy as an alternative to the fossil them to industrialize, a number of developing economies have already
fuels. Therefore, academics and industries from different parts of the started to commit substantially on non-fossil fuel energy sources.
globe have started to envision renewable energy driven future world Sadorsky (2009a) investigates the determinants of renewable energy
in the quest for a sustainable energy system. in G7 countries and shows that in the long-run, increases in real GDP
Renewable energy industry is growing rapidly across the world. per capita and CO2 per capita are found to be major drivers behind per
During the peak of global financial crisis in 2008, global investment in capita renewable energy consumption. Sadorsky(2009b) further ana-
new renewable capacity has increased from US$104 to US$120 billion lyses the link between renewable energy consumption and income in
(REN21, 2009; P:9). Annual percentage gains for 2008 also shows sig- twenty two emerging countries by using panel cointegration techniques
nificant achievements–all forms of grid-connected solar photovoltaic and concludes that the demand for renewable energy consumption in
emerging economies are small but rapidly growing. Therefore, identify-
ing the determinants of renewable energy use in developing economies
⁎ Corresponding author. Tel.: + 61 892664577. that are accelerating adoption of renewable energy technology is war-
E-mail address: [email protected] (R.A. Salim). ranted. This study aims to analyse the determinants of renewable energy

0140-9883/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.eneco.2011.08.015
1052 R.A. Salim, S. Rafiq / Energy Economics 34 (2012) 1051–1057

consumption in six major renewable energy investor emerging econo- Although the drivers behind different types of non-renewable en-
mies within a demand-side framework. Furthermore, in contrast to ergy consumption (i.e. oil, gas and coal) have been well studied, rela-
most of the earlier papers in this field, such as Sadorsky (2009a and tively little is known about the drivers behind renewable energy
2009b); this study employs both panel data and time series econometric consumption. Two studies that identify the drivers for renewable en-
models with renewable energy consumption, income, pollutant emis- ergy in G7 countries and twenty two emerging countries are Sadorsky
sion and oil price. This study adopts time series technique, namely, auto- (2009a) and Sadorsky (2009b), respectively. Both these studies em-
regressive distributed lag (ARDL) approach as it gives more robust result ploy panel cointegration technique and their findings are similar as
in small samples. mentioned above. Apergis and Payne (2010) also perform heteroge-
There are several reasons for selecting these six emerging econo- neous panel cointegration technique to investigate the relationship
mies for our analysis. These emerging economies are growing very between renewable energy consumption and economic growth of
rapidly and striving to attain further growth in future, making it likely OECD countries. Hence, there is a small but rapidly growing literature
that incremental emissions will come from these developing coun- on the demand for renewable energy consumption. However, time
tries increasingly unless adopt environmentally friendly energy series techniques could have been of great help as far as individual
source. In order to mitigate the environmental impacts and diversify country level analyses are concerned. Furthermore, in-depth studies
their energy mix these economies are committed to increase in alter- identifying the linkage between renewable energy and its determi-
native power capacity within the shortest possible time. Available nants are almost non-existent in the literature.
statistics documented in the annual report of the Renewable Energy This study employs both panel and time series econometric tech-
Network (REN21, 2010) show that renewable energy investment in- niques to analyse the link between renewable energy consumption
creases tremendously in emerging economies in recent years. For ex- and output of rapidly accelerating emerging economies in renewable
ample, financial investment in emerging economies increased to energy investment. Panel technique is used to overcome the prob-
$65.86 billion in 2009 from a meagre amount of $1.76 billion in lems of time series technique with small sample size. To examine
2001. Brazil, China, India, Indonesia, Philippines and Turkey are the long-run and short- run dynamics in the panel, the study employs
major renewable energy investor countries of all these emerging a relatively new panel cointegration and error correction method pro-
countries. China is the leading country for investment in renewable posed by Westerlund (2006). The study also examines the long-run
energy with a total of 33.7 billion. While Brazil and India ranked and short-run dynamics at individual country levels. Usually Vector
fifth and eighth in the world with a total investment of $10.5. Other error correction model (VECM) is used for this purpose, however,
developing countries outside the big two (China and India) in Asia In- when sample size is small VECM suffers from low power and size
donesia and Philippines invested an amount of $0.3 and $0.2 billion in properties. To overcome this data requirement issue, the authors
clean energy respectively. Turkey's potential for renewable energy have implemented ARDL because it provides more robust results for
sources remains largely untapped. With the EBRD's (European Bank small samples. Moreover, a number of developing nations like Brazil,
for Reconstruction and Development) investment of over 900 million China, India, Indonesia, Philippines and Turkey are proactively accel-
Euros plus government invest together stand about $1.5 billion dol- erating the adoption of renewable energy. Therefore, identifying de-
lars in 2009. Although these countries are at different stages of eco- terminants of renewable energy consumption in these economies
nomic development with different social and economic institutions, would help them to design appropriate emission mitigation strategies
but all of them have common effort of increasing investment in re- in general and renewable energy policies in particular.
newable energy for the past decade or so.
The rest of the article is structured as follows. The next section 3. Theoretical settings
provides a critical review of earlier literature, followed by theoretical
settings of this paper. A description of data sources and methodolo- The usual approach to modelling energy consumption or demand
gies is presented in Section 4. Section 5 presents an analysis of empir- is to hypothesize a model that relates energy consumption to own
ical results. Conclusions and policy implications are given in the final price, income, and the price of a substitute (see, for example, Masih
section. and Masih, 1997; Narayan and Singh, 2007). Likewise Sadorsky
(2009a and 2009b), the variables to be employed in this research
2. Review of literature are selected in accordance with economic theory and data availabili-
ty. Real GDP is included in the model to measure income. Higher in-
There is an impressive body of literature on the relationship be- come should lead to higher energy consumption and thus there
tween energy consumption and economic growth (Apergis and should have positive association between these two. As oil or prod-
Payne, 2009; Ghosh, 2002; Kraft and Kraft, 1978; Ma et al., 2008; ucts derived from oil are considered to be the most likely substitute
Wolde-Rufael, 2009; Zamani, 2007 and so forth). Research on this for renewable energy for most of the countries, oil prices are included
issue has primarily been aimed at providing significant policy guide- in the model to proxy price of a substitute. Higher oil price should re-
lines in designing efficient energy conservation policies. These studies duce its consumption and increase the demand for renewable energy,
primarily take two different approaches, supply-side and demand-side. implying a negative relationship between the demand for renewable
The supply-side approach analyses the contribution of energy con- energy and its substitute. In accordance with societal concern over
sumption in economic activities within the traditional production greenhouse effects, CO2 emissions are included in the model as an im-
function framework (Ghali and El-Sakka, 2004; Oh and Lee, 2004; portant additional explanatory variable. Higher CO2 emission creates
Sari and Ugur, 2007; Stern, 2000). The demand-side approach ana- demand for cleaner environment and encourages usage of alternative
lyses the relationship between energy consumption, gross domestic renewable energy that is free from this evil effect. So a positive rela-
product (GDP) and energy prices (often taking CPI as a proxy) in a tion between renewable energy and CO2 emission is expected. The re-
tri-variate energy demand model (Asafu-Adjaye, 2000; Lee, 2005; newable energy variable used in this study is a composite variable
Masih and Masih, 1997; Narayan and Singh, 2007; Rafiq and Salim, reflecting renewable energy from several different sources (i.e. bio-
2009; 2011). In addition to the above studies, some recent researches mass, solar, wind and hydroelectricity) for which no reliable price
such as Soytas et al. (2007), Ang (2008) and Soytas and Sari (2009) measure is available. Hence, the equation for renewable energy con-
included pollutant emissions in their analyses to investigate the ener- sumption demand would take the following form:
gy consumption-economic activities relationship. However, none of
these studies does not include price variable in their models and
thus, these cannot be termed as complete demand-side models. REit ¼ βoi þ β1i Yit þ β2i CO2it þ β3i ROPit þ eit ðβ1i N 0; β2i N 0; β3i b 0Þ ð1Þ
 R.A. Salim, S. Rafiq / Energy Economics 34 (2012) 1051–1057 1053

where, i = 1, ......., 6 denotes the country , t = 1980, 1981, .........2006 xit ¼ xit−1 þ vit ð3Þ
denotes the time period, eit the vector of serially and mutually uncor-
related structural innovations. Y, CO2 and ROP stands for real GDP, where t and i represent time and space dimensions of data, respec-
carbon emission and real oil price, respectively. tively. In this formulation, the vector xit is modelled as a pure ran-
dom walk and yit is modelled as the sum of the deterministic term
4. Data sources and analytical framework γ1i + γ2it and a stochastic term zit. This term is modelled as follows:
 
′ ′
4.1. Data sources αi ðLÞΔ zit ¼ αi zit−1 −β i xit−1 þ λi ðLÞ vit þ eit ð4Þ
pi pi
Annual data for Brazil, China, India, Indonesia, Philippines and j j
where, αi ðLÞ ¼ 1−∑αij L and λi ðLÞ ¼ ∑λij L .
Turkey covering the period 1980 to 2006 are used in this study. Dif- j¼1 j¼0

ferent data series are obtained from various sources. Real GDP data Now substituting Eq. (2) into Eq. (4) gives the following error cor-
with base year of 2005 is collected from the International Financial rection model for yit
Statistics (IFS) database. Carbon emission and oil price data are  
′ ′
taken from the British Petroleum Statistical Review of World Energy. αi ðLÞΔyit ¼ δ1i þ δ2i t þ αi yit−1 −β i xit−1 þ λi ðLÞ vit þ eit ð5Þ
Oil prices are measured using spot price on West Texas Intermediate
(WTI) crude oil as this data has a long history of being used as a rep- where, δ1i = αi(1)γ2i − αiλ1i + αiγ2i and δ2i = − αiλ2i.
resentation of world crude oil prices. The WTI oil price is then trans- In Eq. (5) above, the vector βi defines a long run equilibrium or
formed into local real oil prices by the help of exchange rate data cointegrating relationship between the variables x and y. However,
extracted from United Nations Statistics Department (please see the in the short run there might be disequilibrium, which is corrected
link: http://unstats.un.org/unsd/snaama/Introduction.asp) and pro- by a proportion − 2 b αi ≤ 0 each period. Here, αi is called error cor-
ducer price index of individual countries extracted from world devel- rection parameter. If αi b 0,then there is error correction and the vari-
opment indicators (WDI). Renewable energy consumption data is ables are cointegrated and if αi = 0, then there is no error correction
collected from the US Energy Information Administration (EIA). and the variables are not cointegrated. Group test statistics is given
by 1
4.2. Analytical framework
1 N ⌢ αi
Gτ ¼ ∑ ð6:aÞ
This paper performs both panel data and time series analyses to N i¼1 SEð⌢
αiÞ
identify panel and country specific elasticity of renewable energy in
terms of income, pollutant emission and oil prices and dynamic caus-
1 N T⌢ αi
al relationship among the variables. At the outset we examine time Gα ¼ ∑ ð6:bÞ
N i¼1 ⌢
α i ð1Þ
series properties of the variables under consideration. Careful atten-
tion is given to account for cross-sectional dependence among the
And panel statistics:
data, if there is any. We employ the general diagnostic test for
cross-sectional dependence in panels proposed by Pesaran (2004). ⌢
α
Pτ ¼ ð7:aÞ
SEð⌢
Panel unit root tests that are widely used in empirical research,
αÞ
such as Im et al. (2003) [hereafter IPS], Levin et al. (2002) [hereafter
LLC] and Maddala and Wu (1999) [hereafter MW] are not robust in Pα ¼ T ⌢
α ð7:bÞ
case there is cross-sectional dependency among the variables. Both
LLC and IPS require cross-sectional independence. Maddala and Wu
One distinguishing feature of this test is that it is possible from
(1999) find that MW test is more robust than LLC and IPS tests to
panel statistic (7.b) to estimate the magnitude of adjustment of
the violation of this assumption. However, this paper uses all these
short-run deviation from long-run equilibrium relation, that is, the

magnitude of error correction is ⌢
unit root tests to make sure that the results are not biased to any spe-
α ¼ Pα T .
cific test procedure.
Next we proceed to estimate the cointegrating vector to examine if
To identify long-run relationship among the variables under study
the hypothesized effects of the explanatory variables on the depen-
we use a relatively new cointegration test proposed by Westerlund
dent variable hold in the long run. In doing so uses parametric dynamic
(2006).Unlike residual-based cointegration tests, this test is free
OLS (DOLS) proposed by Kao and Chiang (2000) and group-means
from common factor restriction. Common factor restriction is referred
fully modified OLS (FMOLS) proposed by Pedroni (2000) are estimat-
to the requirement that the long-run cointegrating vector for the vari-
ed. The group-means — fully modified OLS (FMOLS) solves the prob-
ables in their levels being equal to the short-run adjustment process
lem arising from endogeneity and serial correlation in the ordinary
for the variables in their first differences (Kremers et al., 1992). This
least squares (OLS) estimator. Pedroni (2001) further proposed a para-
common factor restriction is forwarded as a plausible explanation
metric dynamic OLS (DOLS) estimator that parametrically eliminated
for the failure of null hypothesis in many studies when cointegration
the endonegeity and serial correlation problems 2. This paper performs
is strongly suggested in theory, such as Ho (2002). Another advantage
both of these techniques as these approaches may provide with a more
of this new cointegration test is that it handles the problem of cross-
accurate idea about the possible elasticity of renewable energy in
sectional dependence by bootstrapping the critical values of the test
terms of the exogenous variables.
statistics.
This study also employs ARDL model considering the small sample
In this new cointegration test, four test statistics are proposed;
size. The motivation behind implementing this technique is that, un-
two are designed to test the alternative that the panel is cointegrated
like traditional unit root and cointegration tests which are known to
as a whole, while the other two are designed to test the alternative
have low power and size properties in small samples, the ARDL
that variables in at least one cross-section unit are cointegrated. The
former two statistics are referred to as group statistics, while the latter
two are referred to as panel statistics. The data generating process in
this test is assumed to be as follows: 1
For derivation of these statistics, please see Westerlund (2006).
2
For further clarification regarding the difference between FMOLS and DOLS Harris
yit ¼ γ1i þ γ2i t þ zit ð2Þ and Sollis (2003) can be consulted.
1054 R.A. Salim, S. Rafiq / Energy Economics 34 (2012) 1051–1057

Brazil China India

4

Indonesia Philippines Turkey

4

0
1980 1990 2000 2010 1980 1990 2000 2010 1980 1990 2000 2010
Years
LRE LY
LCO2 LOP

Fig. 1. Time series plots of variables (in log scale).

approach, proposed by Pesaran and Shin (1999) and Pesaran et al. visually to get an idea of their evolution. Fig. 1 below plots the vari-
(2001), does not require knowledge of order of integration or cointe- ables (in log form) for six countries.
gration ranks of the variables (Cheung and Lai, 1993; Harris and Sollis, The plots show that all series are upward trending with no sign of
2003; Wolde-Rufael, 2006). Hence, this method avoids the inherent mean-reversion. Another feature of these variables that is evident
limitations in testing for unit roots prior to testing for cointegration from these plots is that none of the series experience any abrupt
and can be conveniently applied to studies that have small sample change either in intercept or in trend. Therefore, traditional panel
size. unit root tests can be used to capture the behaviour of data. However,
The basic idea of the ARDL approach is to test for the existence of it is difficult to gauge the presence of structural breaks in data by just
long-term relationship among the variables in the model. The test in- visual inspection of graphs 3 so we also carry out here unit root tests
volves estimating the following unrestricted error correction model allowing structural breaks by following Carrion-i-Silvestre et al.
(UECM) considering each variable in turn as a dependent variable: (2005) [here after CIS].
Table 1 presents conventional panel unit root test and Table 2 CIS
m n o tests allowing structural breaks for renewable energy, income, oil
Δ REt ¼ β0 þ ∑βit ΔYt−i þ ∑β2j ΔCO2t−j þ ∑β3k Δ ROPt−k price and pollutant emission. All the variables seem to be stationary
i¼1 j¼0 k¼0 ð8Þ in first differences, while non-stationary in their levels. Hence, results
þβ4 Yt−1 þ β5 CO2t−1 þ β6 ROPt−1 þ εt of all four panel unit root tests indicate that the variables involved in
this study are integrated in the level of 1, i.e., they are I(1) variables. Al-
though all tests give identical results, ADF-Fisher and PP-Fisher tests re-
where REt, Yt, CO2t and ROPt are renewable energy, income, carbon sults are more important than other in the case of cross-sectional
emission and real oil price, respectively. εt is assumed to be white dependence (CD). Pesaran (2004) CD test (results are reported in Ap-
noise ‘well behaved’ random disturbance term. pendix Table 3) indicates that there is high dependence among the
The null hypothesis of ‘no-long-run relation’ is tested through F or panel variables. All these tests indicate that the oil price variable is
χ 2-test of the joint significance of the lagged level coefficients. Short- also non-stationary at levels and become stationary at its first differ-
run dynamics or causality can be tested through the significance of ence. Hence, our next move is to conduct panel cointegration test.
differenced terms. Pesaran et al. (2001) demonstrate that the distri- Table 3 provides the results of the panel unit root tests allowing
bution of the test statistics (F or χ 2) is non-standard irrespective of for multiple breaks for all four variables (renewable energy, income,
whether the regressors are I(0) or I(1). oil price and pollutant emission) by following Carrion-i-Silvestre
et al. (2005). The critical values reported are computed by the boot-
5. Analysis of empirical results strapping technique. These results indicate presence of unit root
even when allowing structural breaks in a panel setting at all conven-
Before going to a more rigorous panel and time series analyses, we tional level of significance except for OP series at 5% level of signifi-
work out partial correlation coefficients among the concerned vari- cance 4. That is, these results do not altered our earlier conclusions
ables for different countries (Appendix Table 1). The results show in regard to all level series. Thus, overall evidence shows that vari-
that for most of the countries there exist significant correlation ables are in fact non-stationary at levels.
among the variables. However, correlation coefficient cannot be
used to draw meaningful conclusions about the long-run relationship
between these variables. Therefore, this study proceeds to formalise 3
One of the referees pointed this and suggests us to carry out unit root tests allow-
this relationship by checking the stationarity properties of the vari- ing structural breaks.
ables to avoid the danger of spurious relation among the variables. 4
We identified two breakpoints for each series by using the Bai and Perron (2003)
Before applying econometric tests it is useful to inspect the data procedures and the results are presented in Appendix Table 2.
 R.A. Salim, S. Rafiq / Energy Economics 34 (2012) 1051–1057 1055

Table 1
Panel unit root test.

Method(s) LRE ΔLRE LGDP ΔLY LCO2 ΔLCO2 LOP ΔLOP

Levin, Lin & Chu t* 1.072 [0.858] − 4.291 [0.000] 2.597 [0.995] − 1.492 [0.00] 0.375 [0.646] − 5.677 [0.000] 0.241 [0.849] − 7.053 [0.000]
Breitung t-stat 1.427 [0.923] − 3.592 [0.000] 1.529 [0.937] − 4.707 [0.000] − 0.988 [0.162] − 3.316 [0.001] 0.652 [0.247] − 5.057 [0.002]
Im, Pesaran and Shin W-stat 0.187 [0.574] − 5.035 [0.000] − 1.432 [0.077] − 7.294 [0.000] − 0.094 [0.463] − 6.268 [0.000] 2.536 [0.948] 7.389 [0.000]
ADF-Fisher χ2 10.789 [0.547] 46.891 [0.000] 24.282 [0.019] 64.821 [0.000] 13.232 [0.352] 55.648 [0.000] 4.903 [0.472] 6.283 [0.000]
PP-Fisher χ2 10.519 [0.571] 330.399 [0.000] 9.408 [0.668] 48.259 [0.000] 5.029 [0.957] 53.081 [0.000] 9.037 [0.583] 34.026 [0.000]

Note: all the variables are expressed in natural logarithms. Individual intercept and time trend included in test regressions. Figures in the parenthesis indicate p-values as optimum lag
lengths are automatically determined by Schwarz Information Criteria (SIC). RE, Y, CO2 and OP stands for renewable energy, output, carbon emission and oil prices, respectively.
Variables are expressed in log forms. The null hypothesis for the first test is the presence of a common unit root while for the other three tests the null is the presence of individual unit root.

Having identified I(0) at levels and I(1) in first differenced series, validity of the error correction specification. The long-run elasticities of
next step of analysis involves identification of cointegrating relation- income are significant for all countries. According to the ARDL results,
ship among the variables. In most of the cases, researchers use panel the long run elasticities of income with respect to renewable energy
cointegration technique proposed by Pedroni (1999, 2004). However, are found to be 9.531%, 0.543%, 1.999%, 2.422%, 2.805%, and 3.779% for
Pedroni's cross-sectional independence assumption is highly restric- Brazil, China, India, Indonesia, Philippines, and Turkey respectively. Car-
tive (Banarjee and Carrion -i- Silvestre (2006). Given cross-section bon emission has positive, significant elasticity for most of the countries
dependence as well as presence of structural breaks in data in the except for Philippines and Turkey. The long run elasticities for pollutant
present paper, cointegration technique proposed by Westerlund emission are turn out to be 2.981%, 0.336%, 6.920%, and 2.664% for Brazil,
(2006) is used 5. The results are given in Table 3 below. China, India, and Indonesia, respectively.
After making allowance for cross-section dependence through This result conforms to the robust findings of the panel study previ-
bootstrapping, null of no cointegration is rejected both by group ously employed. Hence, it can be inferred through both panel and indi-
and panel statistics. These results indicate that demand for renewable vidual country analyses that for all the countries except Philippines and
energy bears long-run equilibrium relationship with income, CO2 Turkey carbon emission has been a significant determinant of renew-
emission and real oil price. However, this long-run relationship may able energy consumption in the long run. However, for Philippines
be disturbed in the short run for various reasons. The speed of adjust-
ment or the error correction parameter in Eq. (5) can be calculated by
putting the value of panel statistic Tα in Eq. (7.b) as follows ⌢ α¼
 − Table 2
Pα
T
¼ 6:825=27 ¼ −0:2528 that is, around 26% discrepancy be- Panel unit root tests with structural breaks.
tween renewable energy demand and its determinants are corrected
each year. Panel of 6 countries Bartlett Kernel Bootstrap critical values

Westerlund (2006) cointegration test confirms that there is long- 5% 2.5% 1%

run equilibrium relation between renewable energy demand and its GDP
determinants, namely income, CO2 emission and real oil price. How- Homogenous breaks 4.03 9.24 11.70 13.33
ever, this finding has no policy implication unless the effects of the in- Heterogeneous breaks 4.10 19.86 24.35 31.59
dependent variable on the dependent variable are known. It is
RE
therefore needed to identify the cointegrating vectors. Panel FMOLS Homogenous breaks 5.86 15.74 21.75 27.39
and panel DOLS are estimated to serve this purpose and the results Heterogeneous breaks 19.23 19.40 23.23 27.28
are reported in Table 4 6.The panel cointegrating vector under both
FMOLS and DOLS supports our hypothesized relationship of income CO2
Homogenous Breaks 3.18 27.17 35.81 46.66
and CO2 with renewable energy demand. Both income and CO2 elas-
Heterogeneous breaks 3.29 24.04 29.52 38.95
ticities are positive and significant at 1% levels, that means, with an
increase in income and CO2 emission, demand for renewable energy OP
increases. To be specific long-run income elasticities under FMOLS Homogenous Breaks 29.91 19.30 27.29 39.70
Heterogeneous breaks 29.91 19.30 27.29 39.70
and DOLS are 1.228% and 0.203% respectively. In case of CO2, these
elasticities are 0.033% and 0.127% respectively. Quadratic Kernel Bootstrap critical values
Under both these methods, oil price seem to have least impact on
5% 2.5% 1%
renewable energy consumption. These results conform to Sadorsky's
(2009b) findings. This exogeneity of oil price variable may be due to GDP
Homogenous breaks 4.12 9.54 11.59 14.68
the fact that real oil prices were falling for much of the estimation pe-
Heterogeneous breaks 4.18 20.02 24.61 32.90
riod. Furthermore, another reason might be that for most of these
countries oil prices have been subsidized to avoid any adverse effect RE
on the economy. Homogenous breaks 6.24 16.45 21.32 26.01
Using ARDL approach this study next performs country specific time Heterogeneous breaks 18.60 19.68 22.73 29.79

series analysis to identify long-run elasticity and short-run causality CO2

among the variables for individual countries. The long-run elasticities Homogenous breaks 3.52 27.43 35.01 47.78
for individual countries are presented in Table 5. The coefficients of all Heterogeneous breaks 3.45 22.84 27.88 34.37
the error correction terms are negative and significant confirming the
OP
Homogenous breaks 29.36 16.89 23.25 33.47
Heterogeneous breaks 29.36 16.89 23.25 33.47
5
As Pedroni's cointegration test is widely used in empirical research, this test is per- Note: all tests results allows for up to 5 structural breaks. The long-run variance is
formed and the results are presented in Table 3 in the Appendix. The results show that estimated using the Bartlett kernel with automatic spectral window bandwidth
the variables are cointegrated. selection as in Sul et al. (2005). In addition, all bootstrap critical values allow for
6
Individual country specific FMOLS and DOLS are reported in Appendix Table-5. cross-sectional dependence.
1056 R.A. Salim, S. Rafiq / Energy Economics 34 (2012) 1051–1057

Table 3 Table 5
Westerlund (2006) panel cointegration test. Results of cointegration test and long-run coefficients in the ARLD approach.

Statistic Value Pvalue Robust Pvalue (1000 replications) Country Cointegration test Long-run coefficients

Gτ − 3.180 0.000* 0.014** F-statistics Error correction LY LCO2 LOP

Gα − 5.326 0.835 0.018** term ecm(− 1)
Tτ − 4.817 0.112 0.082***
Brazil 24.807*** − 2.573* 9.531*** 2.981** − 0.470
Tα − 6.825 0.161 0.020**
China 7.936*** − 5.148*** 0.543*** 0.336*** − .152**
Note: *, ** and *** indicate 1%, 5%, and 10% significance levels respectively. Following India 5.851*** − 3.031*** 1.999* 6.920*** 0.476
Westerlund (2006) maximum lag length is selected according to 4(T/100)2/9. The Indonesia 5.742*** − 4.575*** 2.422*** 2.664*** − 0.417**
null hypothesis of the test is H0: No cointegration. Philippines 21.173*** − 5.317*** 2.805** 0.558 − 0.473
Turkey 7.743*** − 0.438** 3.779*** − 0.879 − 0.438

Note: *, **, and **** indicate significant at 10%, 5%, and 1% level, respectively. The F-statistics
is non-standard and is tabulated in Pesaran et al. (2001). Due to the small size, a maximum
and Turkey income has been the only long-run determinant of renew- lag structure of three (m = 3) was considered for the UECM in Eq. (1). The appropriate
able energy consumption. Likewise panel analysis, real oil price seems lag structure was selected according to Schwarz information criteria. A battery of
to play a very limited role in the model. Apparently, results vary across tests according to Pesaran and Pesaran (1997) was undertaken for check for model
the selected countries since these countries differ in their stages of de- misspecification.

velopment as well as their social and economic institutions and poli-

cies. Moreover, these countries vary in their renewable consumption
of energy mix. For example, renewable energy consumption accounts
energy investor emerging economies. Both panel data and time series
for about 8.3% of China's primary energy consumption, 80% of which
econometric techniques are employed in this regard. The empirical re-
comes from hydroelectric, followed by wind power while in India, re-
sults show that in the long-run, renewable energy consumption is signif-
newable energy comprised 15 percent of consumption of which 7 per-
icantly determined by income and pollutant emission in Brazil, China,
cent and 6 percent comes from hydroelectric and solar power
India and Indonesia. However, income turns out to be the main driver
respectively.
for renewable energy consumption in Philippines and Turkey. With re-
Country wise short-run and long-run causality are examined next
spect to all of these countries, oil price seems to have lesser, negative im-
and the results are reported in Table 6. It is apparent from Table 6 that
pact on renewable energy consumption. The elasticities from both panel
in case of Brazil, there are bi-directional causalities between renew-
methods (FMOLS and DOLS) and the time series method (ARDL) are
able energy and income, renewable energy and pollutant emission,
found to be pretty consistent. The FMOLS method indicates that, taking
and income and pollutant emission. Causality results are turned out
the countries altogether, 1% increase in GDP causes 1.228% increase in re-
to be similar for China. In case of India, there is a uni-directional cau-
newable energy consumption, while 1% increase in pollutant emission
sality running from income to renewable energy, while bi-directional
lead to 0.033% increase in renewable energy. In the short-run, causalities
causality exists between pollutant emission and renewable energy.
between renewable energy and income and between renewable energy
Causality results for Indonesia are similar to those of India. Causalities
and pollutant emission are turn out to be bi-directional in Brazil and
between income and renewable energy are found to be bi-directional
China.
both in Philippines and Turkey. However, there is no causal link be-
In cases of Brazil, China, India and Indonesia our results are in line
tween pollutant emission and renewable energy in these countries.
with those of Sadorsky (2009a) who found that both income and
For most of these countries, oil price seems to be very much inactive
within the model even in the short term. This result can be inter-
preted as oil price does not significantly contribute to renewable en-
Table 6
ergy adoption and pollutant emission per se. This is well justified in Granger causality test result in ARDL approach.
the light of the fact that for most of these countries oil price is
Country Wald χ2-statistics
subsidized.
ΔLRE ΔLY ΔLCO2 ΔLOP
6. Conclusions and policy implications Brazil ΔLRE – 19.821*** 92.514*** 0.221
ΔLY 13.561*** – 3.172* 4.488**
In response to the environmental and economic threats posed by fos- ΔLCO2 6.391** 4.359** – 0.148
ΔLOP 0.298 0.712 3.769* –
sil fuels, countries need to transform their addictions of non-renewable
China ΔLRE – 27.551*** 23.449*** 20.689***
energy to an increased consumption of renewable energy. Surprisingly ΔLY 41.245*** – 3.71** 0.538
enough, some of the developing nations like Brazil, China, India, Indone- ΔLCO2 9.114*** 9.609*** – 4.736**
sia, Philippines and Turkey are taking the initiative to restructure their ΔLOP 1.753 2.683 2.107 –
energy consumption profile through investing heavily in renewable en- India ΔLRE – 3.778* 9.033*** 0.222
ΔLY 2.307 – 4.255** 0.034
ergy industry. This study performs the unique task of identifying the de- ΔLCO2 2.716* 0.424 – 0.579
terminants of renewable energy consumption in six leading renewable ΔLOP 4.281** 0.012 0.435 –
Indonesia ΔLRE – 5.661** 8.616*** 5.095**
ΔLY 0.312 – 0.784 0.411
ΔLCO2 6.537** 0.242 – 8.558***
Table 4 ΔLOP 0.922 0.013 0.008 –
Renewable energy consumption long run elasticities. Philippines ΔLRE – 11.199*** 0.286 9.739***
ΔLY 12.187*** – 0.022 1.419
Panel results
ΔLCO2 28.162*** 12.869*** – 0.934
Variables FMOLS DOLS ΔLOP 0.046 2.680 0.289 –
Turkey ΔLRE – 27.701*** 1.031 1.869
Coefficient t statistic Coefficient t statistic
ΔLY 9.588*** – 4.606** 0.018
LY 1.228*** 12.451 0.203*** 5.793 ΔLCO2 2.459 10.082*** – 2.285**
LCO2 0.033*** 5.789 0.127*** 7.499 ΔLOP 0.023 0.599 0.922 –
LP 0.101 − 1.354 − 0.766 − 1.233
Note: the ARDL model is based on an optimally determined (Schwarz information
Note: t-statistics are given in the parenthesis. ***, ** and * represent significant at 1%, 5% criteria) lag structure and a constant. *, **, and **** indicate significant at 10%, 5%,
and 10% levels, respectively. and 1% level, respectively.
 R.A. Salim, S. Rafiq / Energy Economics 34 (2012) 1051–1057 1057

pollutant emission are the long run drivers of renewable energy in G7 Group III to the Fourth Assessment Report of the Intergovernmental Panel on Cli-
mate Change, 2007. Cambridge University Press, United Kingdom.
countries. However, for Philippines and Turkey it is only income that Kao, C., Chiang, M.H., 2000. On the estimation and inference of a cointegrated regres-
leads to increased consumption of renewable energy. This difference sion in panel data. Adv. Econ. 15, 179–222.
may arise from increasing concern over environmental sustainability Kraft, J., Kraft, A., 1978. On the relationship between energy and GNP. J. Energy Dev. 3,
401–403.
from countries where the economic growth is more rapid. This may Kremers, J., Ericsson, N., Dolado, J., 1992. The power of cointegration tests. Oxf. Bull.
also be due to greater global political, civil society and media pressure Econ. Stat. 54, 325–348.
that create awareness campaign that is happening globally. Neverthe- Lee, Chien-Chiang, 2005. Energy consumption and GDP in developing countries: a coin-
tegrated panel analysis. Energy Econ. 27, 415–427.
less, while Sadorsky (2009b) cannot find any short-run causality
Levin, A., Lin, C.F., Chu, J., 2002. Unit root test in panel data: asymptotic and finite sam-
other than significant negative error terms, this study finds that ple properties. J. Econ. 108, 1–24.
there are significant short-run causal relationship between renewable Ma, Hengyun, Oxley, Les, Gibson, John, Kim, Bonggeun, 2008. China's energy economy:
energy consumption and income and pollutant emission, especially in technical change, factor demand and interfactor/interfuel substitution. Energy
Econ. 30 (5), 2167–2183.
Brazil and China. This research suggests that the appropriateness of MacKenzie, James J., 2003. Technology growth curves: a new approach to reducing
the efforts undertaken by emerging countries to reduce the carbon in- global CO2 emissions. Energy Policy 31, 1183–1187.
tensity by increasing the energy efficiency and substantially increas- Maddala, G.S., Wu, S., 1999. A comparative study of unit root tests with panel data and
a new simple test. Oxf. Bull. Econ. Stat. 61, 631–652.
ing the share of renewable in the overall energy mix. These findings Masih, Abul M.M., Masih, Rumi, 1997. On the temporal causal relationship between en-
reinforce the proposition that despite its competing uses, investment ergy consumption, real income, and prices: Some new evidence from Asian-energy
in renewable energy should continue to grow to achieve renewable dependent NICs based on a multivariate cointegration/vector error-correction ap-
proach. J. Policy Model. 19 (4), 417–440.
energy targets.
Narayan, Paresh Kumar, Singh, Baljeet, 2007. The electricity consumption and GDP
nexus for the Fiji Island. Energy Econ. 29 (6), 1141–1150.
Acknowledgement Oh, Wankeun, Lee, Kihoon, 2004. Causal relationship between energy consumption
and GDP revisited: the case of Korea 1970–1999. Energy Econ. 26 (1), 51–59.
Pedroni, P., 1999. Critical values for cointegration tests in heterogeneous panels with
Helpful comments from anonymous referees and the editor of this multiple regressors. Oxf. Bull. Econ. Stat. 61, 653–670.
journal and useful computing guidance from Dr. Vinod Mishra, Monash Pedroni, P., 2000. Fully modified OLS for heterogeneous cointegrated panels. In: Baltagi,
University, Australia, A/Professors AFM Kamrul Hassan, University of B.H., Fomby, T.B., Hill, R.C. (Eds.), Advances in Econometrics. JAI Press, Elsevier Sci-
ences, Amsterdam.
Rajshahi, Bangladesh and Felix Chan of Curtin University, Australia are Pedroni, P., 2001. Purchasing power parity tests in cointegrated panels. Rev. Econ. Stat.
gratefully acknowledged without implicating them for any error re- 83 (4), 727–731.
mains. We are also grateful to Joseph L. Carrion-i-Silvestre from the Uni- Pedroni, P., 2004. Panel cointegration: asymptotic and finite sampleproperties of
pooled time series tests with an application to the PPP hypothesis. Econometric
versity of Barcelona, for providing us GAUSS codes for structural break Theory 20, 597–625.
testing in panel setting. Pesaran, M.H., 2004. General Diagnostic Tests for Cross Section Dependence in Panels;
CESifo Working Paper No. 1229. Cambridge University, United Kingdom.
Pesaran, H.M., Pesaran, B., 1997. Working with Microfit 4.0: Interactive Economteric
Appendix A. Supplementary data Analysis. Oxford University Press, London.
Pesaran, M.H., Shin, Y., 1999. An autoregressive distributed lag modeling approach to
Supplementary data to this article can be found online at doi:10. cointegration analysis. In: Storm, S. (Ed.), Econometric and Economic Theory in
the Twentieth Century: The Ragnar Frisch Centennial Symposium. Cambridge Uni-
1016/j.eneco.2011.08.015. versity Press, Cambridge.
Pesaran, M. Hashem, Shin, Yongcheol, Smith, Richard J., 2001. Bounds testing ap-
proaches to the analysis of level relationships. J. Appl. Econ. 16 (3), 289–326.
References Rafiq, Shuddhasattwa, Salim, Ruhul A., 2009. Temporal causality between energy con-
sumption and income in six Asian emerging countries. Appl. Econ. Q. 55 (4), 1–16.
Ang, J.B., 2008. Economic development, pollutant emissions and energy consumption Rafiq, Shuddhasattwa, Salim, Ruhul A., 2011. The linkage between energy consumption
in Malaysia. J. Policy Modell. 30, 271–278. and income: a multivariate cointegration analysis in Six Emerging economies of
Apergis, Nicholas, Payne, James E., 2009. Energy consumption and economic growth in Asia. J. Emerg. Markets 6 (1), 50–73.
Central America: Evidence from a panel cointegration and error correction model. REN21, 2009. Renewables Global Status Report Paris: REN21 Secretariate.
Energ. Econ. 31 (2), 211–216. REN21, 2010. Global Trends in Sustainable Energy Investment 2010: REN21 Secretariate.
Apergis, Nicholas, Payne, James E., 2010. Reneable energy consumption and economic Sadorsky, Perry, 2009a. Renewable energy consumption, CO2 emissions and oil prices
growth: evidence from panel of OECD countries. Energ. Policy 38, 656–660. in the G7 Countries. Energy Economis 31 (456–62).
Asafu-Adjaye, John, 2000. The relationship between energy consumption, energy Sadorsky, Perry, 2009b. Renewable energy consumption and income in emerging econ-
prices and economic growth: time series evidence from Asian developing coun- omies. Energy Policy 37, 4021–4028.
tries. Energy Econ. 22 (6), 615–625. Sari, Ramazan, Ugur, Soytas, 2007. The growth of income and energy consumption in
Bai, J., Perron, P., 2003. Computation and analysis of multiple structural change model. six developing countries. Energy Policy 35 (2), 889–898.
J. Appl. Econ. 18, 1–9. Soytas, U., Sari, R., 2009. Energy consumption, economic growth, and carbon emissions:
Banarjee, A., Carrion -i- Silvestre, J., 2006. Cointegration in panel data and with breaks challenges faced by an EU candidate member. Ecol. Econ. 68 (6), 1667–1675.
and cross-section dependence. European Central Bank. Working paper no. 591. Soytas, U., Sari, R., Ewing, B.T., 2007. Energy consumption, income, and carbon emis-
Carrion-i-Silvestre, J.L., del Barrio-Castro, T., Lopez-Bazo, E., 2005. Breaking the panels: sions in the United States. Ecol. Econ. 62, 482–489.
an application to GDP per capita. Econ. J. 8, 159–175. Stern, David I., 2000. A multivariate cointegration analysis of the role of energy in the
Cheung, Y.-W., Lai, K.S., 1993. Finite-sample sizes of Johansen's likelihood ratio tests for US macroeconomy. Energy Economics 22 (2), 267–283.
cointegration. Oxf. Bull. Econ. Stat. 55 (3), 313–328. Stern, Nicholas, 2006. Stern review on the economics of climate change. Stern Review
El-Ashry, Mohamed, 2009. Renewable Energy-A Global Perspective In 3rd European Re- on the Economics of Climate Change, UK Treasury.
newable Energy Policy Conference. Brussels. Sul, D., Phillips, P.C.B., Choi, C.Y., 2005. Prewhitening bias in HAC estimation. Oxf. Bull.
Ghali, Khalifa H., El-Sakka, M.I.T., 2004. Energy use and output growth in Canada: a Econ. Stat. 67, 517–546.
multivariate cointegration analysis. Energy Econ. 26 (2), 225–238. Westerlund, Joakim, 2006. Testing for panel cointegration with multiple structural
Ghosh, Sajal, 2002. Electricity consumption and economic growth in India. Energy Pol- breaks. Oxf. Bull. Econ. Stat. 68, 101–132.
icy 30 (2), 125–129. Wolde-Rufael, Yemane, 2006. Electricity consumption and economic growth: a time
Harris, R., Sollis, R., 2003. Applied time series modelling and forecasting. John Wiley & series experience for 17 African countries. Energy Policy 34 (10), 1106–1114.
Sons, Chichester, England. Wolde-Rufael, Yemane, 2009. Energy consumption and economic growth: the experi-
Ho, T., 2002. A panel cointegration approach to the saving-investment correlation. ence of African countries revisited. Energy Economics 31 (2), 217–224.
Empir. Econ. 27, 91–100. Zamani, Mehrzad, 2007. Energy consumption and economic activities in Iran. Energy
Im, K.S., Pesaran, M.H., Shin, Y., 2003. Testing for unit roots in heterogeneous panels. J. Economics 29 (6), 1135–1140.
Econ. 115, 53–74.
IPCC, 2007. Climate change 2007: Mitigation of climate change. In: Metz, O.R.D.B.,
Bosch, P.R., Dave, R., Meyer Cambridge, L.A. (Eds.), Contribution of Working