Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

The nexus between carbon emissions, energy consumption

and economic growth in Middle East countries: A panel data analysis
Burcu Ozcan n
Department of Economics, Faculty of Economics and Administrative Sciences, University of Firat, Elazig 23119, Turkey

H I G H L I G H T S

The relationship between CO2 emissions, energy consumption, and growth is examined.

Panel data estimation methods are used for 12 Middle East countries.

We obtain a U-shaped curve contrary to the EKC hypothesis.

The causality runs from economic growth to energy consumption in the short-run.

In the long-run, causality runs from energy consumption and growth to CO2 emissions.

art ic l e i nf o a b s t r a c t

Article history: The environmental Kuznets curve (EKC) hypothesis assumes that there is an inverted U-shaped
Received 11 February 2013 relationship between environmental degradation and income per capita. In other words, as a country
Accepted 6 July 2013 grows, it is assumed that its environmental quality improves. In this study, we aim to test the EKC
hypothesis for 12 Middle East countries during the period 1990–2008 by employing recently developed
Keywords: panel data methods. Our results provide evidence contrary to the EKC hypothesis. We found evidence
Environmental Kuznets curve favorable to the U-shaped EKC for 5 Middle East countries, whereas an inverted U-shaped curve was
CO2 emissions identified for only 3 Middle East countries. Furthermore, there appear to be no causal links between
Panel cointegration income and CO2 emissions for the other 4 countries. Regarding the direction of causality, there appears to
be a unidirectional causality from economic growth to energy consumption in the short-run; in the long-
run, however, the unidirectional causality chain runs from energy consumption and economic growth to
CO2 emissions. We also suggest some crucial policy implications depending on these results.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction decrease and control atmospheric concentration of GHGs emis-

sions. For instance, the United Nations Framework Convention on
Global warming and climate change are serious environmental Climate Change (UNFCCC), adopted in 1992 and opened for
problems of our world. Especially, as a dominant contributor to the signatures at the United Nations Conference on Environment and
greenhouse effect, the increasing amount of carbon dioxide (CO2 ) Development in Rio de Janeiro, is one of the best known environ-
appears to be aggravating environmental problems. These increas- mental conventions. The main aim of UNFCCC is the contracting
ing environmental threats have led scholars and policy-makers to and converging of CO2 emissions across countries to combat global
assess the impacts of global warming on the world economy and climate change and greenhouse effects arising from high atmo-
to debate over reducing greenhouse gases (GHGs) emissions to spheric carbon concentration.
alleviate global warming since 1990s. Furthermore, the mitigation In addition, another international agreement, the Kyoto proto-
assessment of greenhouse gas emissions has been an integral part col, has been signed during the UNFCCC. It was adopted on 11
of national and international climate policy agendas (Haggar, 2012, December 1997 in Kyoto and entered into force on 16 February
p. 358). 2005 with the aim of setting binding targets for 37 industrialized
After a while, industrialized countries have started to arrange countries and the European community to reduce their GHGs
important environmental agreements and agendas in order to emissions. According to the Kyoto protocol, the contracting parties
from developed countries are committed to reducing their green-
house gas emissions by at least 5% from 1990 levels between 2008
n
Tel.: +90 424 237 0000. and 2012.
E-mail address: bozcan@firat.edu.tr

0301-4215/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.enpol.2013.07.016

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
2 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

As a result of an increasing environmental consciousness over Saudi Arabia are the 13th and 16th highest CO2 emitters, having
the last 2 decades, scholars have started to analyze the environ- produced 402 and 365 million metric ton in 2004, respectively.
ment–economy nexus in the framework of environmental eco- Furthermore, among the top 20 countries ranked by percentage
nomics. For instance, the relationship between environmental growth in emissions between 1994 and 2004, there are six Middle
degradation and income is one of the most important topics being East countries, namely Iran, Oman, Saudi Arabia, Turkey, Egypt,
debated under the name of the environmental Kuznets curve and UEA. Thus, it is of great interest to examine the nexus of
(EKC). The EKC was derived from the original Kuznets curve, energy consumption, CO2 emissions, and economic growth in
testing an income–inequality nexus, proposed by Kuznets (1955). Middle East countries.
The EKC hypothesis indicates that there is an inverted U-shaped Contributions of our study to the literature are two-fold. First,
relationship between environmental degradation and economic our panel, consisting of 12 Middle East countries, has not been
growth. During a country′s early stage of development, economic studied before. There are three panel studies analyzing the
growth leads to environmental degradation until a turning point is relationship between energy consumption, GHGs emissions, and
reached. However, this situation is reversed beyond this turning real income in Middle East and North African (MENA) countries.
point. In other words, according to the proponents of the EKC However, we differ from them in that we exclude North African
hypothesis, emissions are considered to be a function of income countries. In addition, to the best of our knowledge, there are few
whereby as income increases, emissions increase until a threshold panel studies (see for instance, Arouri et al., 2012; Haggar, 2012;
level of income is reached after which emissions start to decline Jaunky, 2011) that have taken into account both cross-sectional
(Apergis and Payne, 2010, p. 650). dependence and slope heterogeneity issues while testing the EKC
Testing the validity of the EKC hypothesis or exploring the hypothesis. We employ recently developed panel data methods,
causal links between income and CO2 emissions is crucial when i.e. second generation panel unit root tests and panel cointegration
designing appropriate policy tools for protecting the environment, test instead of first generation panel tests. This is the novelty of
fighting against global warming, and ensuring sustainable eco- this study.
nomic development. In addition, as stated by Narayan and Narayan The rest of paper is organized as follows. In Section 2, we
(2010), examining the relationship between economic growth and present the literature review. In Section 3, we present methodol-
environmental quality allows policymakers to judge the response ogy and empirical results, and in Section 4, we conclude the study
of the environment to economic growth. and suggest some policy implications.
Apart from the EKC hypothesis, the second most debated topic
in the framework of environmental economics concerns the
connection between energy consumption and economic growth. 2. Literature review
Economic development is closely related to energy consumption
given that more energy consumption leads to higher economic There are three empirical research strands examining the
development level via productivity enhancement (Ang, 2007). In aforementioned topics in the environmental economics literature.
addition, using energy in a more efficient way requires a higher The first strand concentrates on the environmental pollutants
level of economic development. Thus, as stated by Ang (2007), and output nexus, and tests the validity of the EKC hypothesis. The
energy consumption and economic development may be jointly first empirical paper concerning the EKC is attributed to Grossman
determined and the direction of causality cannot be determined a and Krueger (1991). After that, several researchers have tested the
priori. EKC hypothesis (see, inter alia, Agras and Chapman, 1999; Dinda
The final and third research area concerning environmental and Coondoo, 2006; Friedl and Getzner, 2003; Galeotti et al., 2009;
economics focuses on the relationship between energy consump- Holtz-Eakin and Selden, 1995; Selden and Song, 1994). Also, Dinda
tion, environmental degradation, and economic growth. This is a (2004), He and Richard (2010), and Stern (2004) provide extensive
synthesis of the first (EKC) and the second (energy consumption– literature reviews on the EKC hypothesis.
economic growth nexus) research areas. In this study, we examine The second strand consists of studies analyzing the growth–
the relationship between energy consumption, environmental energy nexus. These studies date back to Kraft and Kraft′s (1978)
degradation, and real per capita income. Additionally, we test seminal study. The earlier studies, given that they mostly applied a
the validity of the EKC hypothesis for 12 Middle East countries by bivariate model, were criticized due to omitted variables bias and
taking CO2 emissions as an environmental quality indicator. More- failed to get unanimous results (see Akarca and Long, 1980;
over, we examine the direction of causality among economic Bentzen and Engsted, 1993; Erol and Yu, 1987; Yu and Hwang,
growth, energy consumption, and CO2 emissions in the framework 1984). However, recent studies have started to examine the nexus
of panel vector error correction (PVEC) model. of energy consumption and economic growth in a multivariate
Our sample of countries consists of Bahrain, United Arab framework (see Gurgul and Lach, 2011, 2012; Altinay and Karagol,
Emirates (UAE), Iran, Israel, Egypt, Syria, Saudi Arabia, Turkey, 2004; Al-Iriani, 2006; Apergis and Payne, 2009b; Narayan and
Oman, Jordan, Lebanon, and Yemen. The Middle East countries Smith, 2008; Oh and Lee, 2004; Soytas and Sari, 2003, 2006; Stern,
attract a special interest for energy economists due to their 2000; Yang, 2000). Ozturk (2010) provides an extensive literature
abundant natural resource reserves, such as crude oil and natural survey on the energy–growth nexus.
gas. For instance, Iran has about 15% of the world′s total reserves of As stated by Saboori and Soleymani (2011), considering the
natural gas (Farhani and Rejeb, 2012). The Middle East′s share of growth–environment nexus and growth–energy nexus in a bivari-
worldwide oil reserves is about 57.5%. Among Middle East coun- ate framework suffers from omitted-variables bias. Ang (2007) and
tries, Saudi Arabia, Iran, Iraq, Kuwait, and United Arab Emirates are Soytas et al. (2007) have gathered both nexus in a single frame-
the major oil producing countries. For instance, according to the work in their seminal studies, and thus, the third research strand
statistics of Energy Information Administration (EIA, 2012, http:// has emerged. Our study is an example of this third research strand.
www.eia.gov), Saudi Arabia was the world′s largest producer and Studies in this strand can be divided into two sub-groups. The first
exporter of total petroleum liquids in 2010, and the world′s second group includes time-series analyses and focuses on individual
largest crude oil producer behind Russia. Its economy remains countries, whereas the second group consists of panel data
heavily dependent on crude oil and oil export revenues have analyses for a group of countries. The results of these studies
accounted for 80–90% of total Saudi revenues. In addition, accord- change due to their samples, time intervals, and estimation
ing to the report published by the World Bank (2007), Iran and techniques. A brief literature review of this strand is given below.

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

Panel data studies generally employ standard first generation countries, except for Jordan. Finally, in the short-run, a significant
panel unit root tests and cointegration tests (e.g., tests developed positive relationship between energy consumption and CO2 emis-
by Im et al. (2003), Levin et al. (2002), Pedroni (1999, 2004)). sions was obtained, and in the long-run, all variables in the model
However, these tests do not take cross-sectional dependence into appeared to move together. The last study is of Haggar (2012).
account. Representative studies in this strand are Apergis and He took heterogeneity and cross-sectional dependence into
Payne (2009a, 2010) for 6 Central American countries and 11 account by employing panel cointegration and panel unit root
Commonwealth of Independent States, Al-mulali (2011) and tests developed by Westerlund (2005) and Pesaran (2007), respec-
Farhani and Rejeb (2012) for MENA countries, Lean and Smyth tively. He examined the validity of the EKC hypothesis and the
(2010) for 5 ASEAN countries, Narayan and Narayan (2010) for 43 relationships between variables of interest for 21 Canadian indus-
developing countries, Pao and Tsai (2010) for BRIC countries, and trial sectors for the period 1990–2007. The results suggested that
Wang et al. (2011) for 28 provinces in China, to name a few. energy consumption has a positive and statistically significant
However, to our knowledge, there are only three studies that impact on GHGs emissions. Also, a non-linear relationship was
take cross-sectional dependence and parameter heterogeneity attained between emissions and economic growth, consistent
issues into account while analyzing the relationship between with the EKC hypothesis.
energy consumption, economic growth, and CO2 emissions. First, In addition, there are also time series studies in the related
Jaunky (2011) implemented Pesaran′s (2007) unit root test and literature. For instance, Ang (2007) utilized cointegration and
Westerlund′s (2007) cointegration test while examining the EKC vector error-correction modeling techniques for France; Acaravci
hypothesis and the relationships between variables of interest for and Ozturk (2010), Ghosh (2010), Halicioglu (2009), Ozturk and
the panel of 36 high income countries during the period 1980– Acaravci (2010, 2012), Menyah and Rufael (2010), and Saboori and
2005. The results indicated the presence of both short-run and Soleymani (2011) implemented ARDL bounds testing approach for
long-run unidirectional causality running from real per capita GDP 9 European countries, India, Turkey, Turkey, South Africa, and Iran,
to CO2 emissions. An inverted U-shaped curve was also supported respectively; Soytas and Sari (2009) utilized Toda and Yamamoto
for the whole panel. Second, Arouri et al. (2012) employed the (TY, 1995) causality test for Turkey; Zhang and Cheng (2009)
panel unit root tests of Smith et al. (2004), the panel cointegration conducted a TY causality test and generalized impulse response
test of Westerlund and Edgerton (2007), and the cross correlated analysis for China; Soytas et al. (2007) employed a vector auto-
effects (CCEs) estimator of Pesaran (2006). The results provided regressive (VAR) model and a TY test for the United States; Hwang
support for the inverse U-shaped pattern for the EKC. However, and Yoo (2012) utilized cointegration and the Granger causality
their country-level analysis did not support the EKC for 11 MENA tests for Indonesia; Fodha and Zaghdoud (2010) conducted

Fig. 1. Scatterplots of variables of interest.

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
4 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

cointegration methods for Tunisia; Akbostanci et al. (2009) efficiency (Baltagi, 2005). To investigate the long-run relationship
employed time series and panel data analyses for Turkey; and between CO2 emissions, energy consumption, and real GDP in the
finally Esteve and Tamarit (2012) used a threshold cointegration panel framework, we adopted the methodology proposed by
test developed by Hansen and Seo (2002) for Spain and examined Apergis and Payne (2009a, 2010) and empirically investigated
the validity of the EKC hypothesis, as well as the relationship the following model based on variables in natural logarithms.
between energy consumption, pollutant emissions and economic
CO2it ¼ αit þ β1i ENC it þ β2i GDP it þ β3i GDP 2it þ εit ð1Þ
growth.
where i ¼ 1; 2; :::::::::; N for each country in the panel and
t ¼ 1; 2; ::::::::::; T refers to the time period. CO2it represents the
3. Methodology and empirical results carbon dioxide emissions, ENC it stands for energy consumption,
GDP it is the per capita real GDP, and GDP 2it is the square of per
3.1. Data and model capita real GDP. The coefficients β1 , β2 , and β3 correspond to the
long-run elasticities of CO2 with respect to energy consumption,
We collected annual data from 1990 to 2008 from the World real GDP, and squared real GDP, respectively.1 The sign of β1 is
Bank Development Indicators database (WDI, 2012) for 12 Middle expected to be positive as an increase in energy consumption
East countries. Energy consumption, CO2 emissions, and real GDP results in an increase in emissions. Furthermore, given that the
per capita were measured as the kt (kiloton) of oil equivalent per EKC hypothesis assumes that as income increases, emissions
capita, metric tons per capita, and constant 2000 international increase until a threshold level of income is reached after which
dollars, respectively. The selection of time period and country emissions begin to decline, β2 and β3 are expected to be positive
sample was dictated by data availability. Fig. 1 presents the plots of and negative, respectively.
GDP versus CO2 emission, GDP versus energy consumption, and
energy consumption versus CO2 emissions. As a visual inspection, 3.2. Testing for cross-sectional dependence
there appears to be a positive relationship between GDP and CO2
emissions contrary to the expection of inverted U shaped EKC. At the first stage of analysis, we first needed to test cross-
Also, our panel result in support of the U shaped Kuznets curve sectional dependence to decide which unit root test would be
could not be supported by the graphical analysis. In addition, the appropriate. To this end, we employed the Lagrange multiplier
graphical representation between energy consumption and GDP (CDLM 1 ) test developed by Breusch and Pagan (1980). It is well
indicates a positive relationship as expected. Finally, a positive rela- known that when T is larger than N (T 4 N), this test is favorable to
tionship between energy consumption and CO2 emissions appears. other tests, namely Frees (1995) and Pesaran (2004) CD tests
Our empirical results support these visual inspections as well. (Haggar, 2012). CDLM 1 test has a χ 2 distribution with a cross-
Table 1 presents the summary statistics associated with CO2 sectional independence null hypothesis. It is based on the sum of
emissions per capita, energy consumption per capita and real GDP squared coefficients of correlation among cross sectional residuals
per capita for each country and panel set over the period 1990– obtained through OLS. The Breusch and Pagan (1980) test statistic
2008. The mean CO2 emission per capita ranges from 0.8797 in was 90.781, indicating that the null hypothesis of cross-sectional
Yemen to 28.05 in UAE. Regarding the mean energy consumption independence was rejected at the 5% significance level. Thus, we
per capita, Yemen has the least energy usage (263.74), whereas proceeded by implementing unit root tests that allow for cross-
Bahrain has the highest (9279.22). In respect of mean GDP per sectional dependence.
capita, UAE is the richest country (33225.20), while Yemen is the
poorest one (526.27). In addition, Yemen has the least variation 3.3. Panel unit root tests
(i.e. the least standard deviation) in CO2 emissions per capita
(0.1122), energy consumption (41.96) and in per capita GDP A number of panel unit root tests have been developed in the
(39.02). Besides, UAE has the highest variation in CO2 emissions literature. For instance, the first group includes first generation
per capita (5.0646) and GDP per capita (2487.96) while Oman has unit root tests that do not allow for cross-sectional dependence
the highest in energy consumption per capita (1124.45). such as those proposed by Choi (2001), Hadri (2000), Im et al.
We used CO2 as an environmental pollutant indicator given (2003), Levin et al. (2002), and Maddala and Wu (1999). The
that “anthropogenic carbon dioxide emissions through the com- second group consists of second generation unit root tests that
bustion of fossil fuels appear to be the major contributor of global take cross-sectional dependence into account (see. e.g. Bai and Ng
warming” (Ghosh, 2010, p. 3008). For instance, in a report (2004), Moon and Perron (2004), Pesaran (2007), Phillips and Sul
provided by the World Bank (2007), among several environmental (2003), and Smith et al. (2004)). However, the result favorable to
pollutants causing climate change CO2 is held responsible for cross-sectional dependence led us to implement a second genera-
about 58.8% of the GHG, and thus it deserves special interest. tion unit root test. For that purpose, the bootstrap unit root tests of
However, there are many studies using other environmental
Smith et al. (2004), namely LM, t, Min, Max, and WS tests were
degradation indicators such as nitrogen oxide, sulfur emissions,
suspended particulate matter, and water pollution (see Grossman conducted. LM is the mean of the individual Lagrange Multiplier
and Krueger, 1995; Imen, 2012; Kaufmann et al., 1998; Perman and (LM i ) test statistics, developed by Solo (1984). It is computed as
Stern, 2003; Rothman, 1998; Selden and Song, 1994; Stern, 2005; LM ¼ N 1 ∑N
i ¼ 1 LM i . t test is a bootstrap version of panel unit root
Zaim and Taskin, 2000). test of Im et al. (2003). It is based on the mean of t i : t ¼
In this study, we implemented a panel data model as panel data N1 ∑N
i ¼ 1 t i . Max ¼N
1 N
∑i ¼ 1 Maxi was developed by Leybourne
has many advantages over time series data. For instance, as stated (1995). The Min ¼ N1 ∑Ni ¼ 1 Mini test is a more powerful variant
by Pao and Tsai (2010), using panel data allows us to gain more
of the LM statistics. It depends on forward and reverse ADF
observations by pooling the time series data across sections and
regressions, which yield the statistics LMf i and LM ri , based on
leads to higher power for the Granger causality test in the case of
short time series. In addition, contrary to time series and cross- their minima Mini ¼ MinðLM f i ; LM ri Þ. Finally, the last test WS was
sectional data, panel data controlling individual heterogeneity
allows for more informative data, more variability, less collinearity 1
We tested the cubic polynomial relationship betweenCO2 emissions and
among the variables, more degrees of freedom, and more income, but we obtained statistically insignificant coefficient for GDP it 3 .

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

Table 1
Summary statistics of data.

Countries Mean S.D. Minimum Maximum

Panel A: CO2 emissions per capita (metric tons per capita)

Bahrain 25.7423 2.9844 20.3778 30.1105
UAE 28.0598 5.0646 15.9465 37.1065
Iran 5.3944 1.0710 3.9640 7.4479
Israel 8.8085 1.6565 5.1276 10.8640
Egypt 1.8885 .4341 1.3360 2.6852
Syria 3.6315 .2846 3.0388 4.1683
Saudi Arabia 14.6228 1.9716 10.7645 17.8491
Turkey 3.2389 .3952 2.7565 4.0669
Oman 9.8894 3.9403 5.5415 17.3493
Jordan 3.3573 .2794 2.7639 3.8855
Lebanon 3.9761 .4304 3.0857 4.6300
Yemen 0.8797 .1122 0.6801 1.0903
Panel 9.1241 9.0565 0.6801 37.1065

Panel B: Energy consumption (kt of oil equivalent per capita)

Bahrain 9279.22 718.38 7976.40 10,432.93
UAE 11297.29 905.55 9409.65 12,607.84
Iran 1926.90 510.51 1236.82 2862.74
Israel 2813.12 192.20 2326.41 3052.84
Egypt 682.19 117.53 544.86 902.79
Syria 992.19 147.16 815.81 1291.49
Saudi Arabia 5123.00 695.08 3702.78 6379.73
Turkey 1139.98 142.61 945.88 1428.79
Oman 3691.90 1124.45 2258.01 6235.39
Jordan 1059.45 106.05 950.19 1271.93
Lebanon 1199.31 234.05 662.70 1462.97
Yemen 263.74 41.96 207.79 325.32
Panel 3289.02 3473.19 207.79 12,607.84

Panel C: Real GDP per capita (constant 2000 international dollars)

Bahrain 12,387.92 1531.17 9427.85 14,788.89
UAE 33,225.20 2487.96 25,573.73 36,597.67
Iran 1624.13 264.83 1281.08 2148.23
Israel 18,480.70 1922.05 15,282.69 21,941.11
Egypt 1421.83 213.64 1143.81 1858.85
Syria 1229.50 124.67 955.14 1452.25
Saudi Arabia 9258.02 201.94 8839.18 9571.61
Turkey 4168.22 613.29 3413.64 5323.68
Oman 8610.74 1315.36 6773.15 11,385.64
Jordan 1857.40 275.43 1473.12 2497.53
Lebanon 4693.35 580.71 3066.72 5895.08
Yemen 526.27 39.020 464.83 578.32
Panel 8123.60 9286.30 464.83 36,597.67

developed by Pantula et al. (1994). These tests have a unit root null 3.4. Slope homogeneity and panel cointegration tests
hypothesis and allow for heterogeneous autoregressive roots
under the alternative hypothesis. Thus, rejection of null hypothesis Having established that all variables were integrated of order
indicates that stationarity does hold for at least one country. one, we examined the cointegration relationship among variables
Furthermore, as stated by Arouri et al. (2012), they use sieve of interest as a next step. However, first, the homogeneity of slope
sampling schema to account for both the time series and cross- parameters was tested to decide the right cointegration method.
sectional dependencies of the data.2 For that purpose, Swamy′s (1970),3 slope homogeneity test was
As shown in Table 2, we checked the stationarity of variables implemented. It depends on the dispersion of individual slope
employing t, LM, Min, Max, and WS tests. The results presented estimates from a suitable pooled estimator and is feasible for
that the unit root null hypothesis could not be rejected in level panels where N is small relative to T, as in this paper. The Swamy
variables at all significance levels. However, after taking the first ^ has asymptotically χ 2 distribution with k(N  1)
statistic (S)
differences of series, the variables of interest turn out to be stationary. degrees of freedom. The Swamy (1970) test statistic was 158.59,
Thus, we can confirm that all variables are integrated of order one. implying the rejection of slope homogeneity at 1% level of
significance. Thus, we proceeded to testing the cointegration
under the presence of slope heterogeneity and cross-sectional
dependence. With that in mind, we used the Durbin Hausman-
group mean test (DH g ) developed by Westerlund (2008). This test
2
5000 bootstrap replications were used in order to control for cross-sectional allows for cross-sectional dependence modeled by a factor model
dependency as well as finite- sample bias. We used a block size and maximum lag in which the errors of Eq. (1) are obtained by idiosyncratic
order of the individual unit root test regressions equal to 30 and 4, respectively.
In addition, the results weren′t sensitive to different block sizes and lag orders. We
also employed unit root tests for only constant case and obtained similar results as
3
in trend and intercept case. However, we couldn′t report them due to space See Pesaran and Yamagata (2008) for further details about Swamy′s
conservation. The results are available upon request from the author. (1970) test.

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
6 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Table 2 Table 3
The results of Smith et al. (2004) panel unit root tests. The results of Westerlund (2008) panel cointegration test.

Trend and intercept case DH g 1.89b (0.02)

DH p 0.57 (0.28)
Level 1st difference Level 1st difference
Notes: Probabilities were reported in parentheses. The
Test Statistic p-value Statistic p-value Statistic p-value Statistic p-value criterion used in this paper is IC2(K) with maximum
number of factors (K) set equal to two. For the bandwidth
CO2 ENC selection, M was chosen to the largest integer less than
t  1.89 0.31  4.71 0.00a  2.11 0.47  4.83 0.00a 4ðT=100Þ2=9 as suggested by Newey and West (1994).b
LM 4.76 0.59 10.60 0.00a 5.20 0.44 11.47 0.00a indicates the rejection of no cointegration null hypothesis
Min 3.89 0.44 10.41 0.00a 2.72 0.89 10.31 0.00a at 5% level of significance.
Max  1.69 0.13  4.62 0.00a  1.24 0.90  4.12 0.00a
WS  2.26 0.19  5.27 0.00a  1.99 0.74  4.66 0.00a

GDP GDP 2 Thus, the DH g test statistic applied in this study can be
t -2.19 0.32  3.62 0.00a  2.13 0.38  3.57 0.00a formulated as
LM 4.98 0.47 8.57 0.00a 4.84 0.53 8.49 0.00a
0.00a 0.01a N T
∑ S^ i ðϕ~ i ϕ^ i Þ2 ∑ e^ it1
Min 2.64 0.86 5.86 2.62 0.87 5.85 2
Max  1.30 0.75  2.54 0.00a  1.30 0.74 -2.53 0.00a
ð6Þ
i¼1 t¼2
WS 1.96 0.66  3.15 0.00a  1.95 0.67 -3.14 0.00a
where ϕ~ i is the individual instrumental variable (IV) estimator of
Notes: p-values are the values of bootstrap probability. ^ i is the
ϕi and is obtained by simply instrumenting e^ it1 with e^ it . ϕ
a
indicates rejection of unit root null hypothesis at 1% level of significance.
OLS estimator of ϕi .
For the panel test (DHp), the null and alternative hypotheses
are formulated as H 0 : ϕi ¼ 1 for all i¼1,………….,N versus H p1 :
innovations and unobservable factors that are common across ϕi ¼ ϕ and ϕ≺1 for all i. A common autoregressive parameter is
units of the panel (Auteri and Constantini, 2005). Thus, the errors assumed both under the null and alternative hypotheses.
in Eq. (1) are modeled as follows: In contrast, for the DH g test, H 0 is tested versus the alternative
hypothesis defined as H g1 : ϕi ≺1 for at least some i. In this case,
εit ¼ λ′i F t þ eit ; ð2Þ
heterogeneous autoregressive parameters are assumed across
panel members. Thus, the rejection of null hypothesis indicates
F jt ¼ ρj F jt1 þ ujt ; ð3Þ that there is cointegration for at least some panel units. The results
of the DH g and DHp tests are reported in Table 3.
eit ¼ ϕi eit1 þ υit ; ð4Þ As shown in Table 3, the DHg test statistic indicates that the
null hypothesis of no cointegration is rejected at 5% level of
where F t is a k-dimensional vector of common factors F jt with significance. Hence, it can be inferred that there is a cointegration
j ¼ 1; ::::::::; k and λi is a conformable vector of factor loadings. We for at least some panel members.
ensure that F t is stationary by assuming that ρj o 1 for all j. In this
case, the integration order of the composite regression error εit
3.5. The estimation of long-run relationship
depends only on the integratedness of the idiosyncratic distur-
bance eit . Thus, in this data-generating process, testing the null
Given the presence of cointegration, we estimated the long-run
hypothesis of no cointegration is equivalent to testing whether
parameters in the cointegrating vector. For this purpose, we used
ϕi ¼ 1 in an equation. There are two panel cointegration tests
Pedroni′s (2000) heterogeneous FMOLS estimator that allows for
namely, panel and group mean tests. The panel test is constructed
estimating heterogeneous cointegrated vectors for panel mem-
under the maintained assumption that ϕi ¼ ϕ for all i whereas the
bers. The FMOLS estimator has an advantage since it corrects for
group mean test assumed that ϕi ≠ϕ for all i. Both tests are based
both endogeneity bias and serial correlation. As suggested by
on two estimators of ϕi both of which have different probability
Haggar (2012), it is the most suitable technique to be applied in
limits under the cointegration alternative hypothesis while shar-
the presence of heterogeneous cointegrated panels. The panel
ing the property of consistency under the no cointegration null n
hypothesis. The instrumental variable (IV) and OLS estimators can FMOLS estimator can be defined as β^ GFM ¼ N 1 ∑N βn FMi , where
i¼1
be used to attain the Durbin–Hausman tests. Thus, the following βn FMi is obtained from a time series FMOLS estimation of Eq. (1) for
kernel estimator can be defined. each panel member (i.e. for each country). However, the FMOLS
  T estimator does not take cross-sectional dependence into account.
1 Mi j
ω^ 2i ¼ ∑ 1 ∑ υ^ υ^ ð5Þ Thus, to alleviate this problem, we demeaned the data with
T1 j ¼ Mi M i þ 1 t ¼ jþ1 it itj respect to common time effects. Table 4 reports the results of
individual and panel FMOLS estimates.
where υ^ it is the residuals obtained through OLS from Eq. (4). M i is a
As seen in Table 4, panel FMOLS result renders evidence against
bandwidth parameter which defines how many autocovariances of
the EKC hypothesis. There appears to be a U-shaped relationship
^ 2i is a
υ^ it are required to estimate in the kernel. The quantity of ω between income and CO2 emissions, contrary to the expectations
consistent estimate of the long-run variance of υit (ωi ). Depending
2
derived from the EKC hypothesis. In other words, pollution levels
on these estimates, we can define two variance ratios as S^ i ¼ ω^ 2 =s^ 4 i i
decrease as a country develops, but begin to increase as rising
and S^ N ¼ ω
^ 2N =ðs^ 2N Þ2 , where income passes beyond a certain point. Our result is in line with
that of Wang et al. (2011) since they had evidence favorable to the
1 N 2 1 N 2 U-shaped EKC for a panel set consisting of 28 provinces of China.
^ 2N ¼
ω ^ ;
∑ ω s^ 2N ¼ ∑ s^
Ni¼1 i Ni¼1 i However, our study is in contrast to many studies in the EKC
literature (see, inter alia, Ang, 2007; Apergis and Payne, 2009a,
(^s2i denotes the corresponding contemporaneous variance estimate.) 2010; Lean and Smyth, 2010). The reason for this unexpected

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

Table 4 2010). However, it must be borne in mind that our results do not
FMOLS estimation results. give any insight about the direction of causality.
In addition, we can compare our country-based results with the
Country ENC GDP GDP2
findings of Arouri et al. (2012) as their sample of countries (MENA)
Bahrain 5.15a  17.05b 0.72b is similar to ours. Arouri et al. (2012) found inverted U-shaped
(4.09) (  2.49) (2.26) EKCs for Egypt and Lebanon as in our study. However, they found
UAE  2.34b 21.21a  1.01b inverted U-shaped EKCs for Bahrain, Saudi Arabia, and Oman. In
(  2.15) (2.56) (  2.45)
Iran 0.41b 1.71  0.08
addition, a U-shaped curve was obtained for UAE, whereas the
(2.23) (0.57) (  0.40) EKC, in any functional form, was not supported for Lebanon. These
Israel 2.20 1.01  0.08 results are in contrast to ours. However, we could not compare our
(1.49) (0.14) (  0.25) study with the studies of Al-mulali (2011) and Farhani and Rejeb
Egypt 1.04a 6.01b  0.38b
(2012), which have similar samples to ours since Al-mulali (2011)
(2.82) (2.49) (  2.06)
Syria 0.50c  6.45a 0.48a did not estimate long-run parameters, while Farhani and Rejeb
(1.84) (  3.69) (4.39) (2012) tested only the monotonic link between CO2 emissions and
Saudi Arabia 1.56c  13.23 0.87 real GDP.
(1.68) (  1.25) (1.32)
Turkey 1.59a  6.44a 0.38a
(11.17) (  3.44) (3.65) 3.6. Panel short-run and long-run causality tests
Oman  0.09  20.11a 1.19a
(  0.30) (  3.11) (3.64) The existence of a long-run cointegrating vector necessitates
Jordan 0.53a 0.84  0.04
the exploration of Granger causality. To define the direction of
(4.34) (0.60) (  0.41)
Lebanon 0.34a 4.72a  0.22b Granger causality among the variables in both the long-run and
(6.36) (2.80) (  2.26) short-run, we employed a panel-based error correction model,
Yemen 1.14a  9.78a 0.41a following the two steps of Engle and Granger (1987). We first
(4.85) (  5.04) (4.79) estimated the long-run parameters in Eq. (1) via the FMOLS
Panel 1.00a  3.13a 0.19a
estimator to obtain the residuals. Subsequently, we defined the
(11.09) (  2.84) (3.53)
first-lagged residuals as the error correction term and estimated
Notes:a, b, and c indicate significance at 1%, 5%, and 10% levels. t-statistics are the following dynamic error correction model (7a–7d).
reported in parentheses . Intercept and a linear trend are included in regressions. m m
ΔCO2it ¼ γ 1i þ ∑ γ 11ip ΔCO2itm þ ∑ γ 12ip ΔENC itm
k¼1 k¼1
m m
result might be explained as follows. The Middle East countries þ ∑ γ 13ip ΔGDP itm þ ∑ γ 14ip ΔGDP 2 itm þ ϕ1i ECT it1 þ ε1it 7a
under study are rich economies. In other words, they have high k¼1 k¼1
economic growth levels thanks to their abundant oil reserves and
m m
high levels of oil export. However, economic growth does not ΔENC it ¼ γ 2i þ ∑ γ 21ip ΔCO2itm þ ∑ γ 22ip ΔENC itm
always equal economic development or sustainable growth that k¼1 k¼1
aims to improve the environmental quality. Hence, developing m m
þ ∑ γ 23ip ΔGDP itm þ ∑ γ 24ip ΔGDP 2 itm þ ϕ2i ECT it1 þ ε2it 7b
countries might ignore using eco-friendly technologies in their k¼1 k¼1
growth processes. They could attain inverted U-shaped EKCs after
evolving into developed countries. Regarding the impact of energy m m
ΔGDP it ¼ γ 3i þ ∑ γ 31ip ΔCO2itm þ ∑ γ 32ip ΔENC itm
consumption on CO2 emissions, we had expected result. The panel k¼1 k¼1
result indicates that a 1% increase in energy consumption m m
increases CO2 emissions by 1%. The long-run elasticity of CO2 þ ∑ γ 33ip ΔGDP itm þ ∑ γ 34ip ΔGDP 2 itm þ ϕ3i ECT it1 þ ε3it 7c
k¼1 k¼1
emissions with respect to real GDP is  3.13+0.38GDP with the
threshold level of income of 8.23 (in logarithms). m m
In addition, under the presence of heterogeneity of long-run ΔGDP 2 it ¼ γ 4i þ ∑ γ 41ip ΔCO2itm þ ∑ γ 42ip ΔENC itm
k¼1 k¼1
parameters, a different long-run equilibrium must be defined for m m
each country. In this case, as seen in Table 4, a U-shaped relation- þ ∑ γ 43ip ΔGDP itm þ ∑ γ 44ip ΔGDP 2 itm þ ϕ4i ECT it1 þ ε4it 7d
k¼1 k¼1
ship between CO2 emissions and real GDP was supported for five
Middle East countries, namely Bahrain, Syria, Turkey, Oman, and where the term Δ denotes the first differences, m is the lag length
Yemen with the logarithmic threshold income levels of 11.84, 6.71, set at three, which is based on Akaike information criterion. The
8.47, 8.44, and 11.92, respectively. As stated above, based on the data variables in the logarithmic differences stand for the growth rates
in Table 1, Yemen is the poorest country with the least CO2 of variables of interest. ECT is the error-correction term, ϕj;i
emission level. However, inverted U-shaped curves were attained (j¼1,2,3,4) is the adjustment coefficient, and εjit is the disturbance
for the UAE, Egypt, and Lebanon with the logarithmic threshold term presumed to be uncorrelated with zero means.
income levels of 10.50, 7.90, and 10.72, respectively. UAE is the In terms of short-run causality represented in Eq. (7a), causality
richest country among countries under study with the highest level runs from ΔENC to ΔCO2 if the null hypothesis γ 12ip ¼ 0∀ip is
of CO2 emissions. This result indicates that beyond a threshold level
rejected, while causality runs from ΔGDP and ΔGDP 2 to ΔCO2 if
of income, an increase in income may lead to improvement, rather
the joint null hypothesis γ 13ip ¼ γ 14ip ¼ 0∀ip is rejected via a Wald test.
than degradation, in environmental quality. Aside from this, for Iran,
In Eq. (7b), causality runs from ΔCO2 to ΔENC if the null hypothesis
Israel, Saudi Arabia, and Jordan, the EKC hypothesis, irrespective of
λ21ip ¼ 0∀ip is rejected, whereas causality runs from ΔGDP and
its functional form, could not be supported as real GDP and squared
real GDP did not significantly affect CO2 emissions. In addition, ΔGDP 2 to ΔENC if the joint null hypothesis γ 23ip ¼ γ 24ip ¼ 0 ¼ ∀ip is
regarding the effect of energy consumption on CO2 emissions, we rejected. In case of Eqs. (7c) and (7d), due to the presence of two
had positive and significant results except for UAE, Israel, and, variables related to real GDP growth (i.e., ΔGDP and ΔGDP 2 ) in the
Oman. This result is also in line with many studies in the literature system, we need to make cross-equation restrictions to determine
(see, among others, Ang, 2007; Arouri et al., 2012; Lean and Smyth, the causality from either ΔCO2 or ΔENC to GDP growth via

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
8 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Table 5 FMOLS estimator provided evidence in support of a U-shaped

The results of long-run and short-run causality tests. relationship between CO2 emissions and real income, contrary to
the EKC hypothesis. Hence, it could be stated that CO2 emissions
Dependent variables Sources of causation (independent variables)
decrease with economic growth, stabilize, and then increase.
ΔGDP, ΔGDP 2 Short-run ΔCO2 Long-run Furthermore, we also reported the country-based FMOLS results
ΔENC ECT under the presence of slope heterogeneity. The individual results
provided evidence favorable to the U-shaped curves for 5 Middle
ΔGDP, ΔGDP 2 – 3.64 (0.72) 8.17 (0.22) 1.05 [0.58]
East countries, namely Bahrain, Syria, Turkey, Oman, and Yemen.
ΔENC 3.41a (0.00) – 0.31 (0.81) 0.02 [0.61]
In addition, there appear to be inverted U-shaped EKCs for UAE,
ΔCO2 1.17 (0.32) 1.29 (0.27) –  0.90a[  8.94]
Egypt, and Lebanon whereas there is no significant relationship
Notes: The p-values are given in parentheses. χ 2 statistic and F statistic are provided between CO2 emissions and real output for the other 4 Middle East
for the likelihood ratio test and Wald test, respectively for the short-run causality. countries.
t-statistics are reported in brackets for the long-run causality. a indicates 1% level of To define the direction of long-run and short-run causalities,
significance.
we estimated a panel error correction model. The results from
error correction model indicated the presence of a short-run
a likelihood ratio (LR) test. Causality from ΔCO2 (ΔENC) to ΔGDP and unidirectional causality running from economic growth to energy
ΔGDP 2 is supported if the null hypothesis γ 31ip ¼ 0∀ip and γ 41ip ¼ 0∀ip consumption. In other words, economic growth leads to increases
(γ 32ip ¼ 0∀ip and γ 42ip ¼ 0∀ip ) is rejected. in energy consumption. However, there appears no feedback effect
With respect to long-run causality represented in Eq. (7a), if the from energy consumption to economic growth. Thus, this finding
null hypothesis ϕ1i ¼ 0∀i is rejected, then ΔCO2 responds to indicates that energy conservation policies, such as rationing
deviations from long-run equilibrium. In Eq. (7b), if the null energy consumption and controlling CO2 emissions, are likely to
hypothesis ϕ2i ¼ 0∀i is rejected, then ΔENC responds to deviations have no adverse effect on the real output growth of Middle East
from long-run equilibrium. Finally, in terms of Eqs. (7c) and (7d), if countries considered in this study (Acaravci and Ozturk, 2010). In
the null hypothesis ϕ3i ¼ ϕ4i ¼ 0∀i is rejected, then ΔGDP and addition, in the long-run, there seems only a unidirectional causal
ΔGDP 2 jointly respond to deviations from long-run equilibrium. link from energy consumption and real output growth to CO2
The short-run and long-run Granger causality test results are emissions. This result implies that energy conservation policies
reported in Table 5. may be efficient in reducing the rate of pollutant emissions in the
As shown in Table 5, in the short-run, there is a unidirectional long-run. Also, as a long-term development policy, Middle East
Granger causality running from ΔGDP and ΔGDP 2 to ΔENC. This countries need to arrange sustainable growth policies aiming to
result is consistent with the findings of Haggar (2012) for the contract the emissions rate of pollutants.
Canadian industrial sectors, Hwang and Yoo (2012) for Indonesia,
Yuan et al. (2008), and Zhang and Cheng (2009) for China.
Depending on this unidirectional causality, it could be asserted Acknowledgment
that real output growth has a positive impact on energy consump-
tion. There is no feedback effect between energy consumption and I would like to adress special thanks to L. Vanessa Smith and
real output growth. Economic growth leads to higher levels of Joakim Westerlund for providing Gauss codes of their tests, and to
energy consumption. The absence of a short-run causality running Saban Nazlioglu for his endless support and patience.
from energy consumption to real output growth indicates that
policy measures improving energy efficiency can be implemented
without jeopardizing economic growth (Haggar, 2012). For the References
presence of a long-run causality, the estimated coefficient for one
period of lagged ECT needs to be statistically significant at the 10% Acaravci, A., Ozturk, I., 2010. On the relationship between energy consumption, CO2
level or better. As seen, in the long-run, there appears only one emissions and economic growth in Europe. Energy 35, 5412–5420.
Agras, J., Chapman, D., 1999. A dynamic approach to the environmental Kuznets
unidirectional panel Granger causality running from real output
curve hypothesis. Ecological Economics 28 (2), 267–277.
growth and energy consumption to carbon dioxide emissions. Akarca, A.T., Long, T.V., 1980. On the relationship between energy and GNP: a
Thus, it could be suggested that ΔCO2 responds to deviations from reexamination. Journal of Energy Development 5, 326–331.
Akbostanci, E., Asik, S., Tunc, G.I., 2009. The relationship between income and
long-run equilibrium, whereas ΔGDP/ΔGDP 2 and ΔENC do not.
environment in Turkey: is there an environmental Kuznets curve. Energy Policy
This result is in line with the results of Apergis and Payne (2009a) 37, 861–867.
for Central America and Halicioglu (2009) for Turkey. Al-Iriani, M.A., 2006. Energy–GDP relationship revisited: an example from GCC
countries using panel causality. Energy Policy 34 (17), 3342–3350.
Al-mulali, U., 2011. Oil consumption,CO2 emission and economic growth in MENA
countries. Energy 36, 6165–6171.
4. Conclusion and policy implications Altinay, G., Karagol, E., 2004. Structural break, unit root and causality between
energy consumption and GDP in Turkey. Energy Economics 26, 985–994.
In this study, we aimed to test the EKC hypothesis for 12 Middle Ang, J.B., 2007. CO2 emissions, energy consumption, and output in France. Energy
Policy 35, 4772–4778.
East countries over the period 1990–2008 by employing recently Apergis, N., Payne, J.E., 2009a. CO2 emissions, energy usage, and output in Central
developed panel data methods. We took cross-sectional depen- America. Energy Policy 37, 3282–3286.
dence and slope homogeneity issues into account in our estima- Apergis, N., Payne, J.E., 2009b. Energy consumption and economic growth:
evidence from the Commonwealth of Independent States. Energy Economics
tion procedure. For that purpose, we first analyzed the integration 31, 641–647.
orders of variables of interest using Smith et al.'s (2004) panel unit Apergis, N., Payne, J.E., 2010. The emissions, energy consumption, and growth
root tests that allow for cross-sectional dependence and slope nexus: evidence from the commonwealth of independent states. Energy Policy
38, 650–655.
heterogeneity. Having established that all variables are integrated
Arouri, M.E.H., Youssef, A.B., M′henni, H., Rault, C., 2012. Energy consumption,
of order one, we examined whether CO2 emissions, energy economic growth and CO2 emissions in Middle East and North African
consumption, real GDP, and squared real GDP are cointegrated or countries. Energy Policy 45, 342–349.
not. Next, based on the presence of a long-run relationship, we Auteri, M., Constantini, M., 2005. Intratemporal Substitution and Government
Spending: Unit Root and Cointegration Tests in a Cross Section Correlated
estimated cointegrating vector (long-run parameters) employing Panel. XVII Conference Paper, Societa Italiana di economia pubblica, Pavia,
the FMOLS estimator of Pedroni (2000). The results of the panel Universita, URL:〈http://www-3.unipv.it/websiep/wp/419.pdf〉.

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

Bai, J., Ng, S., 2004. A panic attack on unit roots and cointegration. Econometrica 7 Menyah, K., Rufael, Y.W., 2010. Energy consumption, pollutant emissions and
(4), 1127–1177. economic growth in South Africa. Energy Economics 32, 1374–1382.
Baltagi, B.H., 2005. Econometric Analysis of Panel Data, third ed. John Wiley & Sons Moon, H.R., Perron, B., 2004. Testing for a unit root in panels with dynamic factors.
Ltd, England. Journal of Econometrics 122, 81–126.
Bentzen, J., Engsted, T., 1993. Short-and long-run elasticities in energy demand: a Narayan, P.K., Smith, R., 2008. Energy consumption and real GDP in G7 countries:
cointegration approach. Energy Economics 15, 9–16. new evidence from panel co-integration with structural Breaks. Energy
Breusch, T.S., Pagan, A.R., 1980. The Lagrange multiplier test and its applications to Economics 30 (5), 2331–2341.
model specification in econometrics. Review of Economic Studies 47 (1), Narayan, P.K., Narayan, S., 2010. Carbon dioxide emissions and economic growth:
239–253. panel data evidence from developing countries. Energy Policy 38, 661–666.
Choi, I., 2001. Unit roots tests for panel data. Journal of International Money and Newey, W.K., West, K.D., 1994. Automatic lag selection in covariance matrix
Finance 20 (2), 229–272. estimation. Review of Economic Studies 61 (4), 631–654.
Dinda, S., 2004. Environmental Kuznets curve hypothesis: a survey. Ecological Oh, W., Lee, K., 2004. Causal relationship between energy consumption and GDP
Economics 49, 431–455. revisited: the case of Korea 1970–1999. Energy Economics 26, 51–59.
Dinda, S., Coondoo, D., 2006. Income and emission: a panel data based cointegra- Ozturk, I., 2010. A literature survey on energy–growth nexus. Energy Policy 38 (1),
tion analysis. Ecological Economics 57, 167–181. 340–349.
Energy Information Administration, URL: 〈http://www.eia.gov〉 (accessed 10.11.12.). Ozturk, I., Acaravci, A., 2010. Renewable and Sustainable Energy Reviews 14 (9),
Engle, R.F., Granger, C.W.J., 1987. Cointegration and error correction: representa- 3220–3225.
tion, estimation, and testing. Econometrica 55 (2), 251–276. Ozturk, I., Acaravci, A., 2012. The long-run and causal analysis of energy, growth,
Erol, U., Yu, E.S.H., 1987. Time-series analysis of the causal relationships between US openness and financial development on carbon emissions in Turkey. Energy
energy and employment. Resources Energy 9, 75–89. Economics. (URL): http://dx.doi.org/10.1016/j.eneco.2012.08.025.
Esteve, V., Tamarit, C., 2012. Threshold cointegration and nonlinear adjustment Pao, H.T., Tsai, C.M., 2010. CO2 , emissions, energy consumption and economic
between CO2 and income: the environmental Kuznets curve in Spain, 1857– growth in BRIC countries. Energy Policy 38, 7850–7860.
2007. Energy Economics 34, 2148–2156. Pantula, S.G., Gonzalez, F.G., Fuller, W.A., 1994. A comparison of unit-root test
Farhani, S., Rejeb, J.B., 2012. Energy consumption, economic growth and CO2 criteria. Journal of Business and Economic Statistics 12, 449–459.
emissions: evidence from panel data for MENA region. International Journal Pedroni, P., 1999. Critical values for cointegration tests in heterogeneous panels
of Energy Economics and Policy 2 (2), 71–81. with multiple regressors. Oxford Bulletin of Economics and Statistics 61,
Fodha, M., Zaghdoud, O., 2010. Economic growth and pollutant emissions in 653–670.
Tunisia: an empirical analysis of the environmental Kuznets curve. Energy Pedroni, P., 2000. Fully modified OLS for heterogeneous cointegrated panels.
Policy 38, 1150–1156. Advanced in Econometrics 15, 93–130.
Friedl, B., Getzner, M., 2003. Determinants of CO2 emissions in a small open Pedroni, P., 2004. Panel cointegration: asymptotic and finite sample properties of
economy. Ecological Economics 45 (1), 133–148. pooled time series tests with an application to the PPP hypothesis. Econometric
Frees, E.W., 1995. Assessing cross-sectional correlation in panel data. Journal of Theory 20 (3), 597–625.
Econometrics 69, 393–414. Perman, R., Stern, D.I., 2003. Evidence from panel unit root and cointegration tests
Galeotti, M., Manera, M., Lanza, A., 2009. On the robustness of robustness checks of that the environmental Kuznets curve does not exist. Australian Journal of
the environmental Kuznets curve hypothesis. Environmental and Resource Agricultural and Resource Economics 47, 325–347.
Economics 42 (4), 551–574. Pesaran, M.H., 2004, General diagnostic tests for cross section dependence in
Ghosh, S., 2010. Examining carbon emissions economic growth nexus for India: a panels. Working Paper 0435, Cambridge Working Papers in Economics. URL:
multivariate cointegration approach. Energy Policy 38, 3008–3014. 〈http://www.econ.cam.ac.uk/dae/repec/cam/pdf/cwpe0435.pdf〉.
Grossman, G., Krueger, A., 1991. Environmental impacts of a North American free Pesaran, M.H., 2006. Estimation and inference in large heterogeneous panels with
trade agreement. Working Paper 3194, National Bureau of Economics Research multifactor error structure. Econometrica: Journal of the Econometric Society
Cambridge, November. URL: 〈http://www.nber.org/papers/w3914.pdf〉. 74, 967–1012.
Grossman, G., Krueger, A., 1995. Economic growth and the environment. The Pesaran, H.M., 2007. A simple panel unit root test in the presence of cross-section
Quarterly Journal of Economics 110 (2), 353–377. dependence. Journal of Applied Econometrics 22, 265–312.
Gurgul, H., Lach, Ł., 2012. The electricity consumption versus economic growth of Pesaran, M.H., Yamagata, T., 2008. Testing slope homogeneity in large panels.
the Polish economy. Energy Economics 34 (2), 500–510. Journal of Econometrics 142 (1), 50–93.
Gurgul, H., Lach, Ł., 2011. The role of coal consumption in the economic growth of Phillips, P.C.B., Sul, D., 2003. Dynamic panel estimation and homogeneity testing
the Polish economy in transition. Energy Policy 39, 2088–2099. under cross section dependence. Econometrics Journal 6, 217–259.
Hadri, K., 2000. Testing for stationarity in heterogeneous panel data. Econometrics Rothman, D.S., 1998. Environmental Kuznets curves-real progress or passing the buck?
Journal 3 (2), 148–161. A case for consumption-based approaches. Ecological Economics 25, 177–194.
Haggar, M.H., 2012. Greenhouse gas emissions, energy consumption and economic Saboori, B., Soleymani, A., 2011. CO2 emissions, economic growth and energy
growth: a panel cointegration analysis from Canadian industrial sector per- consumption in Iran: a co-integration approach. International Journal of
spective. Energy Economics 34, 358–364. Environmental Sciences 2 (1), 44–53.
Halicioglu, F., 2009. An econometric study of CO2 emissions, energy consumption, Selden, T.M., Song, D., 1994. Environmental quality and development: is there a
income and foreign trade in Turkey. Energy Policy 37, 1156–1164. Kuznets curve for air pollution emissions? Journal of Environmental Economics
Hansen, B.E., Seo, B., 2002. Testing for two-regime threshold cointegration in vector and Management 27 (2), 147–162.
error-correction models. Journal of Econometrics 110, 293–318. Smith, V., Leybourne, S., Kim, T.H., 2004. More powerful panel unit root tests with
He, J., Richard, P., 2010. Environmental Kuznets curve for CO2 in Canada. Ecological an application to the mean reversion in real exchange rates. Journal of Applied
Economics 69, 1083–1093. Econometrics 19, 147–170.
Holtz-Eakin, D., Selden, T.M., 1995. Stroking the fires: CO2 emissions and economic Solo, V., 1984. The order of differencing in ARIMA models. Journal of the American
growth. Journal Public Economics 57 (1), 85–101. Statistical Association 79, 916–921.
Hwang, J.H., Yoo, S.H., 2012. Energy consumption, CO2 emissions, and economic Soytas, U., Sari, R., 2003. Energy consumption and GDP: causality relationship in
growth: evidence from Indonesia. Quality & Quantity 46 (5), 1–16, http://dx. G-7 countries and emerging markets. Energy Economics 25 (1), 33–37.
doi.org/10.1007/s11135-012-9749-9. Soytas, U., Sari, R., 2006. Energy consumption and income in G-7 countries. Journal
Im, K.S., Pesaran, M.H., Shin, Y., 2003. Testing for unit roots in heterogeneous of Policy Modeling 28 (7), 739–750.
panels. Journal of Econometrics 115, 53–74. Soytas, U., Sari, R., Ewing, T., 2007. Energy consumption, income, and carbon
Imen, T., 2012. Is there an EKC relevant to the industrial emission of water pollution emissions in the United States. Ecological Economics 62, 482–489.
for SEMC and EU countries? Journal of Arts Science & Commerce 3 (2), 68–75. Soytas, U., Sari, R., 2009. Energy consumption, economic growth, and carbon
Jaunky, V.C., 2011. The CO2 emissions-income nexus: evidence from rich countries. emissions: challenges faced by an EU candidate member. Ecological Economics
Energy Policy 39, 1228–1240. 68, 1667–1675.
Kaufmann, R.K., Davidsdottir, B., Garnham, S., Pauly, P., 1998. The determinants of Stern, D.I., 2000. Multivariate cointegration analysis of the role of energy in the U.S.
atmospheric SO2 concentrations: reconsidering the environmental Kuznets macroeconomy. Energy Economics 22, 267–283.
curve. Ecological Economics 25, 209–220. Stern, D.I., 2004. The rise and fall of the environmental Kuznets curve. World
Kraft, J., Kraft, A., 1978. On the relationship between energy and GNP. Journal of Development 32, 1419–1439.
Energy and Development 3 (2), 401–403. Stern, D.I., 2005. Beyond the environmental Kuznets curves: diffusion of sulfur-
Kuznets, S., 1955. Economic growth and income inequality. American Economic emissions-abating technology. Journal of Environment and Development 14
Review 45 (1), 1–28. (1), 101–124.
Lean, H.H., Smyth, R., 2010. CO2 emissions, electricity consumption and output in Swamy, P., 1970. Efficient inference in a random coefficient regression model.
ASEAN. Applied Energy 87, 1858–1864. Econometrica 38, 311–323.
Levin, A., Lin, C.F., Chu, C., 2002. Unit root tests in panel data: asymptotic and finite- Toda, H.Y., Yamamoto, T., 1995. Statistical inference in vector autoregression with
sample properties. Journal of Econometrics 108, 1–24. possibly integrated processes. Journal of Econometrics 66, 225–250.
Leybourne, S., 1995. Testing for unit roots using forward and reverse Dickey–Fuller Wang, S.S., Zhou, D.Q., Zhou, P., Wang, Q.W., 2011. CO2 emissions, energy
regressions. Oxford Bulletin of Economics and Statistics 57 (4), 559–571. consumption and economic growth in China: a panel data analysis. Energy
Maddala, G.S., Wu, S., 1999. A comparative study of unit root tests with panel data Policy 39, 4870–4875.
and a new simple test. Oxford Bulletin of Economics and Statistics 61 (S1), Westerlund, J., 2005. New simple tests for panel cointegration. Econometrics
631–652. Reviews 24, 297–316.

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i
10 B. Ozcan / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Westerlund, J., Edgerton, D., 2007. A panel bootstrap cointegration test. Economics Yu, E.S.H., Hwang, B.K., 1984. The relationship between energy and GNP: further
Letters 97, 185–190. results. Energy Economics 6, 186–190.
Westerlund, J., 2008. Panel cointegration tests of the fisher effect. Journal of Applied Yuan, J.H., Kang, J.G., Zhao, C.H., Hu, Z.G., 2008. Energy consumption and economic
Econometrics 23, 193–233. growth: evidence from China at both aggregated and disaggregated levels.
World Bank, 2007. Growth and CO2 emissions: how do different countries fare. Energy Economics 30, 3077–3094.
Environment Department, Washington, DC, URL:〈http://siteresources.world Zaim, O., Taskin, F., 2000. A Kuznets curve in environmental efficiency: an
bank.org/INTCC/214574-1192124923600/21511758/CO2Decompositionfina application on OECD countries. Environmental and Resource Economics 7,
lOct2007.pdf〉, (accessed 19.10.12.). 21–36.
Word Development Indicators, URL:〈http://data.worldbank.org/data-catalog/ Zhang, X.P., Cheng, X.M., 2009. Energy consumption, carbon emissions, and
world-development-indicators〉, (accessed 12.10.12.). economic growth in China. Ecological Economics 68, 2706–2712.
Yang, H.Y., 2000. A note on the causal relationship between energy and GDP in
Taiwan. Energy Economics 22, 309–317.

Please cite this article as: Ozcan, B., The nexus between carbon emissions, energy consumption and economic growth in Middle East
countries: A panel data analysis. Energy Policy (2013), http://dx.doi.org/10.1016/j.enpol.2013.07.016i