Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

journal homepage: www.elsevier.com/locate/rser

CO2 emissions, renewable energy and the Environmental Kuznets Curve, a MARK
panel cointegration approach
Zakaria Zoundi
Department of International Development Policies, Graduate School of International Cooperation Studies, Kobe University, 6578501 Kobe, Japan

A R T I C L E I N F O A BS T RAC T

Keywords: This study combines a panel cointegration analysis with a set of robustness tests to assess the short and long-
CO2 emissions run impacts of renewable energy on CO2 emissions, as well as the Kuznets Environmental Curve hypothesis for
Renewable Energy 25 selected african countries, over the period 1980-2012. The results provide no evidence of a total validation of
Panel cointegration EKC predictions. However, CO2 emissions are found to increase with income per capita. The overall estimations
strongly reveal that renewable energy, with a negative effect on CO2 emissions, coupled with an increasing long-
run effect, remains an efficient substitute for the conventional fossil-fuelled energy. Nonetheless, the impact of
renewable energy is outweighed by primary energy consumption in both the short and long run, entailing more
global synergy for outpacing the environmental challenges.

1. Introduction population accessing electricity, the increase in urbanization, combined

The 2015 Paris conference on climate change appears as a
watershed mark in the efforts of the international community to
grapple with global warming through reducing greenhouse gases
emissions. In spite of the mobilization and determination showed by
governments over decades, at the actual rate of CO2 emissions, the
temperature of the planet will still remain above the threshold of 2 °C
set by scientists [47]. The alarming fact is that the temperature is
projected to reach 30C by 2050 [47], increasing the risk of natural
disasters, and creating more pressure on resources [24]. According to
the IPCC [24], if no sustainable adaptation policy is implemented, the
impact of climate change would reduce the annual GDP growth in
developing countries as much as 2–4% by 2040, and 10% by 2100. The
common global agenda putative by nations is to compel developed
countries to implement more environmental friendly energy policies
(through standard settings, nuclear energy, etc.), and help developing
countries to reduce their CO2 emissions, without jeopardizing their
economic progress. Among these energy policies, actions such as
environmental fund transfers to developing countries, renewable
energy, and technology transfer can help developing countries to take
part in the world's environmental challenge [22,48].
Although Africa pollutes less than other continents, it is the most
severely affected by global warming. According to the IPCC [24], with
around 50% of the urban population and nearly 10% of the rural
with the reduction in crop yields due to climate change, and the high
migration from rural to urban areas would constitute one the most
challenging issues for sub-Saharan Africa in the energy sector. This
paper investigates the role of Africa in global warming by analyzing the
efficiency of an environmental policy suitable for developing countries,
which is spreading across Africa and perceived as the path to sustain-
able growth in Africa [24], namely renewable energy. According to the
International Renewable Energy Agency [23], energies generated from
sources such as wind, solar, and hydro can help boost friendly
environmental development in Africa in the sense that they are
inexhaustible and free. Their deployments can create jobs, contribute
to the improvement of local skills, and they are good pathways for
income generating activities [23].
Two main features distinguish this paper from previous research on
the topic, and contribute to filling the gap in the literature. First, the
topic is focused solely on Africa. The majority of papers on CO2
emissions concentrate either on developed countries or on both
developed and developing countries. For example, Baek and Kim [6],
Apergis and Payne [2], and Ibrahim and Law [20] have analyzed,
respectively, G-20 countries, 6 Central American countries and 69
countries (including developed and developing countries). In their
research, they have found that energy consumption increases CO2
emissions. Apergis et al. [3] have analyzed 19 developed and develop-
ing economies, and have found that nuclear energy has a negative effect

Abbreviations: ASEAN, Association of Southeast Asian Nations; CCEEM, Common Correlated Effects Estimator; EKC, Environmental Kuznets Curve; EIA, Energy Information
Administration; IPA, International Energy Agency; IPCC, Intergovernmental Panel on Climate Change; IRENA, International Renewable Energy Agency; MRIO, Multi-Regional Input–
Output; OECD, Organization for Economic Cooperation and Development; PSTR, Panel Smooth Transition Regression; STIRPAT, Stochastic Impacts by Regression on Population,
Affluence and Technology

http://dx.doi.org/10.1016/j.rser.2016.10.018
Received 4 February 2016; Received in revised form 23 July 2016; Accepted 14 October 2016
Available online 03 November 2016
1364-0321/ © 2016 Elsevier Ltd. All rights reserved.
Z. Zoundi
Fig. 1. Development process and pollution: The Environmental Kuznets Curve.
on CO2 emissions while energy consumption has a positive impact.
Although Africa has comparatively less CO2 emissions (which could be
the reason why the majority of the studies have omitted the continent),
it is important to measure, on one hand, the continent's contribution to
global warming; and, on the other hand, the effectiveness of policies
such as renewable energies. Second, the paper analyzes the long-term
effectiveness of renewable energy and its cointegration with population
growth, energy consumption and income per capita, using a panel
cointegration approach. Although research on the mitigation of global
warming has been driven toward nuclear energy as a potential strategy
for reducing the carbon footprint, in Africa, nuclear energy is in not an
ideal option due to the long lead times, governments disapproval and
the high capital cost, as pointed out by Apergis et al. [3]. Furthermore,
the panel cointegration analysis used in this paper appears more
appropriate than the methods commonly used in the literature, such
as VAR, VEC, panel random and fixed effect or Autoregressive
Distributed Lag (ARDL) (Fig. 1).
Unlike traditional time series or panel data analyses, panel coin-
tegration has an advantage of pooling the long run information
included in the panel, by allowing the short run dynamics and fixed
effect to be heterogeneous across the panels [39]. In addition, the t-
statistic allows for a more flexible alternative hypothesis [37].
Additionally, the paper assesses the hypothesis that pollution
increases as income rises up to a threshold point, then starts decreas-
ing. The inverted U-shape of the relationship income-pollution refers to
the EKC, initiated by Kuznets [29] to explain the relationship growth-
income inequality, and adapted by economists since 1991 in the
context of environmental degradation.
According to the EKC, there are three stages in the intensity of
pollution. Each stage corresponds to a step in the growth process. At
the pre-industrial stage where income per capita is low, environment
pollution increases. This increase in pollution is explained by factors
such as unclean technology used in economic activities, lack of
awareness, prioritization of income growth and profits at the early
stage of growth [51]. However, with the increment in income per capita
and economic growth, followed by improved social indicators, more
investments are made for safer technologies, households become more
willing to target their expenditures towards cleaner goods and assets
(water, houses, cars, etc.). This stage marks the turning point to lower
environment pollution. As the economy crosses the pre-industrializa-
tion stage and moves to the post-industrialization phase, environmen-
tal depletion reduces [35]. The folowing graph summarizes the EKC
insight.
1068
Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075
1.1. Environmental Kuznets Curve
Tremendous efforts have been made by researchers to analyze and
understand the potential factors that can explain the CO2 footprint and
its mitigation (with no harm on development). Although the recent
literature tends to move towards the importance of nuclear energy in
mitigating greenhouse gas emissions (such as Baek and Pride [7];
Apergis, et al. [3]; Menyah and Wolde-Rufael [33], to name just a few),
many studies have paid attention to the potential impact of renewable
energy in greenhouse gas emissions as well as other relevant sources of
pollution such as energy consumption, and population growth. Table 1
summarizes some of the most cited studies.
2. Data and methodology
The study covers 33 years (1980–2012) and 25 countries for a total
of 825 observations. The selection of countries is mainly based on the
availability of data, and the existence of significant renewable energy
initiative (reflected in the data) such as wind, solar, hydropower,
geothermal, and heat pumps. The list of variables includes CO2
emissions in metric ton per inhabitant, as a proxy for air quality; real
GDP per capita in USD (constant 2005); per capita primary energy
consumption (converted into kg of oil equivalent); total renewable
electricity net consumption per capita (in kilowatt-hour) as a proxy for
renewable energy. Renewable electricity comes from 5 main sources:
hydro-electric power, geothermal, solar, wind and biomass such as
wood, wastes and biofuels [16]. To test the effect of demography on
CO2 emissions, the variable population growth is added. While the
variables CO2 emissions, renewable energy and energy consumption
are collected from EIA. The African Development Indicators database
(available in the World Bank databank) provides data on population
and real GDP per capita. Table 2 presents a summary statistics of the
variables.
3. Methodology
3.1. Equation specification
The baseline equation for this study is the equation specified by
previous studies such as Iwata, et al. [25]. In their study, the authors
have used an ARDL approach to assess the EKC hypothesis and the role
of nuclear energy in mitigating CO2 emissions in France. The log-linear
equation specified is as follows:
ln (CO2 )t = α0 + α1 ln yt + α2 (ln yt )2 + α3 ln nuct + εt (1)
where CO2 represents CO2 emissions per capita; y real GDP per capita;
Z. Zoundi Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075

Table 1
Selected literature on CO2 emissions.

Authors Regions Period Methodology Results (impact on CO2 emissions)

Environmental Renewable Energy Population growth

KuznetsCurve energy consumption

Apergis and Central America (6 1971–2004 Panel VEC Holds – Positive impact –
Payne [2] countries)
Alam et al. [1] India, Indonesia, 1970–2012 ARDL bounds test Holds for Brazil, China – Positive impact Positive and significant
China, and Brazil and Indonesia impact for India and Brazil
Apergis et al. 19 countries 1984–2007 Panel VEC Does not hold – Positive impact –
[3] (developed and non-
developed)
Baek and Kim G-20 members 1960–2006 VAR (cointegration) Holds for developed – Positive impact –
[6] countries
Bilgili et al. [10] 17 OECD countries 1977–2010 Panel FMOLS and Holds Negative impact – –
DOLS
Baek and Pride 10 countries 1965–2007 VAR (cointegration) Does not hold – Positive impact –
[7]
Begum et al. [9] Malaysia 1970–2009 ARDL Does not hold – Positive impact No impact
David et al. [14] Worldwide 1977–1990 Panel fixed effect Holds for some – – –
pollutants
Dong et al. [15] Worldwide 1995–2011 MRIO model Mixed – – –
Ibrahim et al. 69 countries 2000–2008 Difference and system Holds Positive impact –
[20] (developed and non- GMM
developed)
Iwata et al. [25] France 1960–2003 ARDL Holds – Positive impact –
Menz and 26 OECD countries 1960–2005 Panel regression – – – Positive impact
Welsch [34]
Lantz and Feng Canada (5 regions) 1970–2000 GLS Does not hold Positive impact – Inverted U-shaped
[30]
Liu et al. [32] China 1990–2012 STIRPAT model Does not hold for Positive impact The impact is function of
industrial waste the regions development
level
Robalino-López Venezuela 1980–2025 Panel cointegration Holds Negative imoact
al. [41]
Nasir and Pakistan 1972–2008 Johansen Holds Positive impact –
Rehman cointegration
[36]
Cemil et al. [12] 36 countries 1971–2009 CCEEM method – Negative impact – –
(developed and non-
developed)
Shi [44] 93 countries 1975–1996 GLS – – – Positive impact
Soytas and Sari Turkey 1960–2000 VAR, Granger Does not hold – No impact –
[45] Causality
Soytas et al. U.S 1960–2004 VAR, Granger Does not hold – Positive impact –
[46] Causality
Wesseh and Lin 34 African countries 1980–2011 Translog production – Negative impact – Inverted U-shaped
[49] function
Heidari et al. ASEAN countries 1980–2008 PSTR model Holds – Positive impact –
[19]
Gene and Alan Worldwide 1977–1990 Panel Random effect Holds – – –
[17]
Jebil et al. [26] 24 sub-Saharan Africa 1980–2010 Panel cointegration Does not hold Mixed –
countries
Shafiei and OECD countries 1980–2011 STIRPAT model Holds in Urban Negative impact Positive impact –
Salim [43]
Atici [5] Bulgaria, Hungary, 1980–2002 Panel regression Holds – Positive impact –
Romania and Turkey
Apergis and 14 Asian countries 1990–2011 GMM Holds – – U-shaped
Ozturk [4]

and nuc electricity production from nuclear source (as percentage % of included to test the possible impact of demography in CO2 emissions.
total electricity production) and ε the error term. In the equation, lnyt The final equation specification for this study takes the form of:
and (lnyt)2 refer to the EKC hypothesis that there is a nonlinear
quadratic relationship between income and environment pollutants ln (COE )it = ψ0 + ψ1 ln INCit + ψ2 (ln INCit )2 + ψ3 ln RECit + ψ4 ln PECit
(such as carbon dioxide (CO2), nitrous oxide (N2O), methane (CH4) + ψ5 POPit + γit (2)
etc.). As this study focuses on renewable energy as a substitute for
where COE represents CO2 emissions per capita; INC, real GDP per
greenhouse gases, and due to the inexistence of nuclear power plants in
capita; REC, renewable energy consumption per capita; PEC, primary
the majority of African countries (only South Africa has the most active
energy consumption per capita; POP, population growth;ψ0 and γit are
and significant nuclear plant), the variable lnnct is dropped and
the intercept and the error term, respectively. The slope parameters are
replaced by a proxy for renewable energy. To avoid the bias generated
captured by ψi (i=1,4 ). Following EKC insight (pollution increases with
by the omission of relevant variables, the primary energy consumption
income, up to a threshold point, then starts decreasing), ψ1 and ψ2 are
is added to the equation. In addition, the variable population growth is
expected to be positive and negative, respectively. Renewable energy

1069
Z. Zoundi Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075

Table 2
Summary statistics.

Country Statistics CO2 per capita Income per capita Renewable energy consumption Primary energy consumption Population growth

Algeria Mean 3.01 2702.46 10.20 1148.52 0.02

Std. Dev. 0.30 286.08 6.66 140.07 0.01
Min 2.45 2271.44 1.70 859.28 0.01
Max 3.58 3227.80 28.32 1504.07 0.03

Angola Mean 0.82 1845.40 88.13 203.92 0.03

Std. Dev. 0.34 567.46 39.39 54.88 0.00
Min 0.36 1103.61 59.97 150.83 0.02
Max 1.40 3033.13 180.79 341.46 0.04

Cameroon Mean 0.44 983.04 207.28 140.29 0.03

Std. Dev. 0.15 170.62 14.24 14.72 0.00
Min 0.25 779.83 151.14 114.41 0.03
Max 0.88 1370.70 241.76 166.12 0.03

Cote d′Ivoire Mean 0.38 1078.00 107.63 170.29 0.03

Std. Dev. 0.07 114.81 30.55 28.63 0.01
Min 0.30 934.31 37.77 117.85 0.02
Max 0.50 1392.03 200.02 237.57 0.04

Dem. Rep. of the Congo Mean 0.08 369.37 137.03 61.73 0.03
Std. Dev. 0.04 164.13 20.31 18.44 0.00
Min 0.03 201.73 102.13 37.73 0.02
Max 0.13 613.64 172.85 86.74 0.04

Egypt Mean 1.75 1068.69 183.21 728.15 0.02

Std. Dev. 0.36 331.72 24.79 170.27 0.00
Min 0.94 537.81 143.50 410.45 0.02
Max 2.41 1642.35 227.34 1041.53 0.03

Ethiopia Mean 0.06 158.60 28.67 27.38 0.03

Std. Dev. 0.01 37.62 14.49 7.47 0.01
Min 0.03 110.68 12.57 16.62 −0.03
Max 0.09 268.41 71.53 47.10 0.04

Gabon Mean 4.74 7856.67 666.40 934.76 0.02

Std. Dev. 1.25 954.92 89.22 160.59 0.00
Min 2.83 6222.40 352.46 700.34 0.02
Max 6.67 9455.52 795.09 1327.53 0.03

Ghana Mean 0.25 705.83 317.67 169.84 0.03

Std. Dev. 0.06 157.12 86.01 24.30 0.00
Min 0.15 507.65 146.13 105.29 0.02
Max 0.37 1149.89 488.43 206.13 0.03

Guinea Mean 0.16 282.97 38.37 65.22 0.03

Std. Dev. 0.03 15.86 11.44 8.97 0.01
Min 0.12 257.87 24.00 50.10 0.02
Max 0.25 306.47 64.62 91.29 0.06

Kenya Mean 0.28 604.64 116.30 130.08 0.03

Std. Dev. 0.03 29.82 18.81 10.09 0.00
Min 0.24 568.57 66.21 114.82 0.02
Max 0.35 686.62 145.43 153.51 0.04

Madagascar Mean 0.11 300.68 30.61 46.25 0.03

Std. Dev. 0.02 33.83 7.19 7.96 0.00
Min 0.08 249.06 16.69 35.58 0.03
Max 0.15 407.06 59.88 60.69 0.03

Malawi Mean 0.07 295.64 85.59 47.98 0.03

Std. Dev. 0.01 26.65 17.94 7.60 0.01
Min 0.05 252.58 59.90 34.03 0.00
Max 0.12 351.62 121.02 72.13 0.06

Mali Mean 0.05 385.60 21.16 24.07 0.03

Std. Dev. 0.01 73.17 6.77 2.68 0.01
Min 0.03 273.13 11.34 14.83 0.01
Max 0.08 499.69 44.93 31.60 0.03

Mauritius Mean 2.34 4089.63 236.79 858.04 0.01

Std. Dev. 1.13 1569.02 177.67 397.65 0.00
Min 0.75 1890.59 35.05 305.80 0.00
Max 4.23 7064.52 498.42 1476.95 0.01
(continued on next page)

1070
Z. Zoundi Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075

Table 2 (continued)

Country Statistics CO2 per capita Income per capita Renewable energy consumption Primary energy consumption Population growth

Morocco Mean 1.03 1692.72 46.72 375.15 0.02

Std. Dev. 0.18 380.58 24.81 83.03 0.01
Min 0.75 1214.57 16.38 261.80 0.01
Max 1.48 2503.15 127.45 573.72 0.02

Mozambique Mean 0.11 267.37 333.70 114.16 0.02

Std. Dev. 0.04 106.62 319.74 80.24 0.01
Min 0.07 153.55 1.83 33.15 0.00
Max 0.22 493.56 1123.37 370.43 0.04

Nigeria Mean 0.74 1032.47 44.18 173.52 0.03

Std. Dev. 0.13 272.72 10.12 19.05 0.00
Min 0.50 742.49 23.28 127.23 0.02
Max 0.94 1617.45 63.07 197.67 0.03

South Africa Mean 8.76 5074.58 37.88 2605.09 0.02

Std. Dev. 0.44 477.59 18.69 137.96 0.01
Min 8.05 4413.38 3.69 2337.24 0.01
Max 9.73 6047.27 89.84 2938.28 0.03

Sudan Mean 0.21 686.42 52.40 87.55 0.03

Std. Dev. 0.08 170.47 26.25 30.32 0.00
Min 0.13 481.79 25.89 56.46 0.03
Max 0.41 1027.00 145.74 163.03 0.03

Tanzania Mean 0.11 402.52 55.38 52.99 0.03

Std. Dev. 0.03 85.20 9.60 9.16 0.00
Min 0.08 318.95 34.37 38.32 0.02
Max 0.19 596.80 75.11 77.55 0.03

Uganda Mean 0.06 336.90 50.74 32.87 0.03

Std. Dev. 0.02 105.10 10.12 7.89 0.00
Min 0.03 220.76 35.80 22.02 0.03
Max 0.10 538.15 71.33 54.91 0.04

Zambia Mean 0.29 670.43 954.31 327.24 0.03

Std. Dev. 0.13 80.14 294.50 108.22 0.00
Min 0.14 557.29 714.25 230.68 0.02
Max 0.61 814.00 1576.39 564.09 0.03

Zimbabwe Mean 1.08 649.05 328.46 421.47 0.02

Std. Dev. 0.31 96.10 109.76 81.29 0.01
Min 0.58 432.29 141.85 278.54 0.01
Max 1.61 863.05 544.10 571.27 0.04
Mean 1.12 1398.38 174.27 373.13 0.03

Full sample Std. Dev. 1.98 1858.91 242.11 571.52 0.01

Min 0.03 110.68 1.70 14.83 −0.03
Max 9.73 9455.52 1576.39 2938.28 0.06

Note: Data for Sudan excludes South Sudan for the years 2011 and 2012, mainly due to the unavailability of recent statistics for south Sudan (South Sudan was separated from Sudan in
2011).

consumption is supposed to lead to a reduction in greenhouse
pollutants. Therefore ψ3 is expected to be negative. Primary energy
consumption is to increase the emission of CO2; ψ4 is expected to be
positive. Finally, the increase in population growth creates more
demand for transportation, goods and services; which can accelerate
the depletion of the environment. Thus, ψ5 is expected to be positive.
3.2. Panel cointegration approach
The application of panel cointegration analysis is justified by many
factors such as the dimension and characteristic of the data. With small
T and large N usually found in microeconomic datasets such as surveys,
the traditional panel methods (random effect, fixed effect, etc.) remains
suitable. However, the analysis of panel data with T > N generates
spurious results, since the feature of the data behavior tends to be close
to time series. The spuriousness increases when analyzing macroeco-
nomic variables (which is the case for this study), as series in macro
data are usually non-stationary.
To handle the problem generated by the accumulation of observa-
tions over time, Baltagi [8] suggests two possible options: 1) hetero-
geneous regressions for each individual to avoid the homogeneity of
coefficients that would be obtained with a single regression; and 2) the
application of time series processes to panels to deal with non-
stationarity and cointegrations among series. The panel cointegration
is an extension of time series analysis to panel data with large T. In
addition to its capacity to pool long run information included in panels,
by allowing the short run dynamics and fixed effect to be heterogeneous
across the panel. [39], the panel cointegration approach provides short
and long run estimates. The process can be summarized as follows: the
preliminary investigation is a unit root test. If a series were found to be
integrated, one would check the possible cointegration among variables
by running a cointegration test. Finnaly, if variables are cointegrated,
in other words if there is a long-run relationship among variables, one
would estimate the long-run coefficients.

1071
Z. Zoundi Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075

4. Empirical results although series are stationary, there are still some series that are
non-stationary, which need to be appropriately addressed.
4.1. Panel unit root

Many techniques have been developed to estimate the order of

integration of series in panel data. Levin et al. [31] suggest a panel unit
root test (LLC) as an extension of the augmented Dickey-Fuller (ADF): 4.2. Cointegration tests
n
Δyit = ϕit ψi + ρyit −1 + ∑ ji=1 φij Δyi,t −j + ξit (3) The existence of I (1) in the series prompts analyzing the presence
of long-run equilibrium among the variables. Three types of tests are
where ϕit includes individual deterministic components (such as fixed applied to the variables of interest: Pedroni [38] panel and group
effect, trend, or a mixture of fixed effects and trend); ρ the autore- statistics, Kao [27] and Westerlund [50]. Each test uses different
gressive coefficient; ξit the error terms; and n the lag order. techniques and assumptions to compute the statistics, and is based
However, the LLC test assumes ρ constant across panels; which on the null hypothesis of no cointegration among series, against the
may suffer from loss of power [11]. Im et al. [21] extend the LLC test by alternative that series are cointegrated. Pedroni panel tests are based
allowing ρ to vary across panels: on the within dimension analysis, and statistics are computed by
n summing numerators and denominators along series, independently.
Δyit = ϕit ψi + ρi yit −1 + ∑ ji=1 φij Δyi,t −j + ξit (4) The panel statistics have four components: a panel-v statistic (which is
non-parametric, based on variances ratio), a panel-rho, panel-PP and
Breitung [11] proposes a test that corrects for bias generated in the
panel ADF statistics, similar to Phillips-Perron ρ statistics, Phillips-
application of LLC or IPS. The bias generally comes from the difference
Peron t-statistics and augmented Dickey-Fuller t-statistics in univariate
in size between N and T (LLC and IPS appear stronger when T is larger
time series, respectively. Pedroni group statistics are based on the
than N), or from the inclusion of an individual deterministic trend in
between dimension approach. Statistics are computed by dividing the
the tests. Besides, the Fisher tests (ADF and Phillips-Perron) suggested
numerator and denominator prior to summing over cross sections. The
by Choi [13] use the time series ADF and PP tests as a framework and
three computable statistics are group-rho, group-PP and group-ADF
application to panel data. The most distinctive feature is that the tests
statistics. The statistics are analogous to Phillips-Perron ρ statistics,
combine each series p-value resulting from their unit root tests, instead
Phillips-Peron t-statistics and augmented Dickey-Fuller (ADF) t-sta-
of averaging individual test statistics as suggested by IPS [21]. The
tistics in univariate time series.
Hadri [18] unit root test is based on the Lagrangian multiplier and uses
Kao t-statistic uses ADF as a framework, and assumes homogeneity
residuals derived from individual ordinary least squared regression on
in the panels. The statistic is derived from panel least squared dummy
deterministic components to compute the statistics. While LLC,
variable (LSDV) analysis. The Westerlund test provides four alterna-
Breitung IPS and Fisher test the null hypothesis that each series is
tives, Gt, Ga, Pt, and Pa. The test use an error correction model to
non-stationary across individuals (H0: ρi =0) against the alternative that
analyze the existence of cointegration in individual panels or in the
at least one individual in the series is stationary (H1: ρi < 0), the Hadri
whole panel. The rejection of Ga and Gt null hypothesis implies the
test assumes the opposite (null hypothesis: no unit root against the
existence of cointegration in at least one of the cross sections, as these
alternative that some or all series are non-stationary). In addition, the
two statistics use individual weighted average process and individual t-
LLC, Breitung and Hadri tests are based on homogeneity in the unit
statistic, respectively. On the contrary, Pa and Pt use a pooling process
root process ( ρi = ρ across panels), while the IPS and Fisher tests assume
over cross-section and the rejection null hypothesis implies the
the autoregressive coefficient to be heterogeneous.
existence of cointegration among variables in the panel, as a whole.
The panel unit root tests specified in this study include individual
Table 4 presents the results of the cointegration tests. In the
effects and the deterministic time trend. The Akaike information
Pedroni and Kao tests, the Schwarz Information Criterion (SIC)
criterion (AIC) is used to determine the country-specific lag length,
determines lag length, and the spectral estimation is based on
with a maximum lag set to 7. The spectral estimation is based on
Bartlett kernel with the bandwidth being determined by Newey and
Bartlett kernel and the bandwidth is determined by Newey and West
West automatic lag selection. The Pedroni test includes individual
automatic lag selection. As reported in Table 3, results provide fair
intercept and trend. In the Westerlund test, the number of lags is set to
evidence of stationarity of the variable. However, the LLC and Breitung
one. The results provide evidence of a long-run relationship among
tests do not reject the null hypothesis of non-stationarity of renewable
variables. Except rho (panel and group) and Westerlund Ga statistics,
energy consumption, although IPS, and the Fisher-type two tests (ADF
all the tests reject the null hypothesis of no cointegration. In other
and PP) reject the null hypothesis. In addition, the Phillips-Perron
words, variables are cointegrated. This suggests that the series tend to
Fisher-type test does not reject the null hypothesis of non-stationarity
move together in the long run, and the difference among them is
of the variable per capita CO2. Hadri z-statistic provides a challenging
stationary. The panel tests seem to have a higher power than the group
alternative. As the null hypothesis is the stationarity of the series
tests, confirming that pooling increases efficiency.
(rather than non-stationarity), the significant results suggest that

Table 3
Panel unit root.

LLC Breitung IPS Fisher Hadri

Variables t* stat t-stat W-stat ADF. Chi-sq PP. Chi-sq z-stat

Log per capita CO2 −15.05*** −3.21*** −17.23*** 379.96*** 1584.96 1.87**
Log per capita income −8.20*** −6.911*** −11.47*** 220.41*** 410.91*** 5.27***
(Log income per capita)^2 −8.51*** −7.13*** −11.13*** 214.01*** 413.21*** 4.59***
Log renewable energy consumption per capita −15.93 −6.77 −17.85*** 355.92*** 1245.28*** 7.73***
Log primary energy consumption per capita −15.23*** −4.53*** −17.63*** 365.59*** 1277.92*** 7.19***
Population growth 13.85 −2.83*** −7.00*** 142.54*** 318.08** 2.53***

** significane at 5%, *** significane at 1%.

1072
Z. Zoundi Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075

Table 4 Table 5
Cointegration tests. Estimations/robustness.

Cointegration test Statistics Dynamic System Dynamic Mean Pooled

OLS GMM Fixed Group Mean
Pedroni Panel-v 1.39* Effect Group
Panel-rho 0.84
Panel-PP −8.87*** Variables Log per Log per D. (Log D. (Log D. (Log
Panel-ADF −9.57*** capita capita per capita per capita per capita
Group-rho 3.86 CO2 CO2 CO2) CO2) CO2)
Group-PP −19.30***
Group-ADF −11.98*** Short-
Kao t −7.58*** run
Westerlund Gt −2.935*** Constant −1.466 −1.259*** −23.575 −2.055***
Ga −6.672 (−1.211) (−0.415) (−38.989) (−0.416)
Pt −13.138*** Log income per −0.527*
Pa −11.207** capita
(−0.309)
Note: Pedroni and Kao tests are generated on Eviews, while Westerlund test is generated (Log income 0.045*
on Stata with the command “xtwest”. The number of bootstrap replicates in the per capita)
Westerlund test has been reduced from the maximum of 800 to 300 to obtain the result. ^2
*
Significance at 10%. (−0.025)
**
significance at 5%.; Log renewable −0.100***
***
significance at 1%. energy
consump-
tion per
4.3. Estimation/robustness capita
(−0.03)
This section constitutes the core of the study. Having found that Log primary 0.558***
variables are cointegrated, the final step is to estimate the relationship. energy
consump-
To get more acumen on the findings, the analysis includes a short-run
tion per
estimation. In addition, a series of robustness would help ascertain the capita
relevance of the outcome. Two classical long-run estimations are (−0.129)
commonly used in the panel cointegration analysis: the dynamic OLS Population −0.008
growth
(DOLS) and the fully modified OLS (FMOLS) developed by Pedroni
(−0.014)
[39,40] and Kao and Chiang [28]. However, as Kao and Chiang [28] use L. (Log per 0.445*** −0.273*** −0.801*** −0.347***
a Monte Carlo simulation to posit that the finite-sample properties of capita CO2)
the DOLS estimator are higher than the FMOLS properties, the (−0.079) (−0.024) (−0.063) (−0.07)
analysis for this study favors the DOLS over FMOLS. D. (Log per −0.519 5.335 7.242*
capita
The DOLS uses leads and lags of the first difference regressors to
income)
estimate the long-run relation among variables. It assumes indepen- (−0.525) (−6.312) (−3.797)
dence between cross sections. However, the test does not provide any D. (Log income 0.049 −0.355 −0.530*
information on the short-run estimates. To obtain the short-run per capita)
^2
estimates, as well to ascertain the relevance of the findings, a set of
(−0.04) (−0.475) (−0.308)
robustness tests are added to the baseline estimator (DOLS). The list of D. (Log −0.033** −0.110* −0.292***
tests includes: the system GMM, the Dynamic fixed effect (DFE), the renewable
Pooled mean Group (PMG) and the Mean Group (MG) estimators. energy
Each of the tests possesses a particular advantage and proceeds consump-
tion per
differently to compute the parameters.
capita)
The system GMM deals with the issue of endogeneity by treating (−0.014) (−0.06) (−0.096)
each variable as endogenous, and instruments the variables by their D. (Log primary 0.452*** 0.095 0.718***
own lag (with the possibility to add external instruments). However, energy
consump-
the estimator is more efficient when T < N [42]. The DFE, MG and
tion per
PMG are closely related. The DFE assumes the slope of the cointegrat- capita)
ing vectors and error variances to be the same across panels. In other (−0.045) (−0.098) (−0.17)
words, all series are pooled, making the slope parameter homogeneous. D. (Population −0.011 0.321 0.055
Only intercepts are specific to each panel. The MG averages slopes growth)
(−0.011) (−0.33) (−0.118)
across panels and the intercept. Error variances are allowed to vary
Observations 743 743 743 743
across panels. Finally, the PMG is an intermediate between the MG and R-squared
DFE in the sense that it averages and pools the coefficients across Instruments 473
panels. The short run coefficients are specific to each cross section, AR1 0.014
while the long run effect of the coefficients are constrained to be AR2 0.277
Hausman 0.019
homogeneous across panels. Long-run
Results of the estimation as well as the robustness tests are Log per capita 0.21 −0.95* −0.519 6.147 −0.117
presented in Table 5. The tests report also the short-run estimations. income
The DOLS provides evidence that renewable energy is an efficient (−0.461) (−0.525) (−11.478) (−0.173)
(Log income 0.017 0.08* 0.049 −0.453 0.007
substitute for conventional fossil-fuelled energy. Renewable energy
per capita)
allows meeting households’ need for energy and contributes to the ^2
improvement of air quality. A 1% increase in the consumption of (−0.034) (−0.04) (−0.816) (−0.01)
renewable energy reduces CO2 emissions by 0.13%, everything held Log renewable −0.130*** −0.18*** −0.130*** −0.122 −0.015**
constant elsewhere. The effect of renewable energy on CO2 emissions (continued on next page)

1073
Z. Zoundi Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075

Table 5 (continued) The EKC hypothesis seems to be confirmed for only the PMG in the
short run. The coefficient is positive (but overestimated) for income per
Dynamic System Dynamic Mean Pooled
capita, and negative the square of income per capita. Population
OLS GMM Fixed Group Mean
Effect Group growth however, has no effect on CO2 emissions (GMM, DFE and
PMG).
Variables Log per Log per D. (Log D. (Log D. (Log
capita capita per capita per capita per capita 5. Concluding remarks
CO2 CO2 CO2) CO2) CO2)

energy This paper attempted to assess the relevance of possible factors that
consump- deplete environment quality, by assessing the EKC hypothesis and
tion per analyzing the effectiveness of renewable energy in Africa. For this
capita
(−0.024) (−0.014) (−0.144) (−0.006)
purpose, the analysis used a panel cointegration approach and robust-
Log primary 0.855*** 1.01*** 0.855*** 1.279*** 1.093*** ness tests to estimate the short and long-run relationship among
energy variables. Results provide clear evidence that renewable energy, with
consump- a negative effect on CO2 emissions, and increasing impact in the long
tion per
run, remains an efficient substitute for conventional energy. However,
capita
(−0.063) (−0.045) (−0.136) (−0.028) this impact is undermined by the higher effect of primary energy
Population 0.026* −0.014 −0.011 −0.063 0.008 consumption on CO2 emissions, implying that tremendous effort
growth should be made at both the national and international level for a
(−0.014) (−0.011) (−0.04) (−0.007) sustainable impact of renewable energy. Additionally, although the
EKC is not completely confirmed for the sample of countries, income
Note: The GMM long-run coefficient is given by β /(1 − α ) where β is the short-term
coefficient (associated with the variable of interest) and α the coefficient of the lagged
per capita is found to contribute to CO2 emissions, confirming that at
dependent variable; the system GMM generates 473 instruments for 743 observations; the earlier stage of development the need for higher income, and the
the Dynamic OLS includes 1 leads and 1 lag. The test is generated on Stata with the lack of awareness of environmental issues lead to higher depletion of
command “xtdolshm” (prior to running this command, the 6 Mata functions should be air quality (first half of EKC). The unequal repartition of income across
installed with the command “ssc install ltimbimata”; the Mean Group, Pooled Mean Africa could explain the unconvincing result for the EKC hypothesis. A
Group and Dynamic Fixed Effect are all obtained with the command “xtpmg”. In
country case study, or a region-based approach would bring different
addition, for all the three estimates, the coefficient of the lagged dependent variable is
given by the error correction term.
insights. Finally, demography, with its perverse effect on air quality,
*
Significance at 10%. constitutes an additional challenge for Africa.
**
significance at 5%.
***
significance at 1%. Appendix A. Supporting information

seems to be outweighed by the preserve effects of usual energy
(primary energy consumption). In fact, a 1% increment in primary
energy consumption increases CO2 emissions by 0.85%, implying that
the consumption of renewable energy along with primary energy
depletes air quality by around 0.72% (0.85–0.13), for a 1% consump-
tion of each of these sources. The lower impact of renewable energy on
CO2 emissions compared with primary energy implies that to obtain a
considerable reduction in CO2 emissions, more efforts should be done
to spread renewable energy policies across Africa, and efforts should be
made by households to switch to this new technology.
Besides, the DOLS shows that there is no evidence of the confirma-
tion of the EKC prediction. Both income per capita and the square of
income per capita do not significantly affect CO2 emissions, although
the impact of income per capita appears to be positive (0.21).
Population growth remains one of the environmental challenges in
Africa. The results indicate that 1% population growth leads to an
additional 0.026% CO2 emissions (with a 10% significance level). The
positive impact of population on CO2 is explained by the surge in
transportation, and demand for goods and services, which creates more
pressure on resources and the environment, the more so as Africa is
behind in terms of infrastructure and green technology.
The overall estimators used in the robustness tests provide strong
evidence that renewable energy contributes to the reduction of CO2
emissions. The average impact is around a 0.1% decrease in CO2
emissions for a 1% increment in per capita renewable energy con-
sumption in the short-run. In the long-run, the impact increases and
reaches around 0.13%, similar to the result for the DOLS. The impact
of primary energy consumption remains high, and increases in the
long-run for all the estimators. A 1% percent increment in primary
energy consumption increases CO2 emissions by around 0.5% in the
short-run, and about 1% in the long-run. Note that the Hausman test
on the difference in estimates between the MG and PMG shows a
significant result at the 5% level (0.019). Therefore, the interpretation
is based on the PMG results, as pooling has more power than grouping.
Supplementary data associated with this article can be found in the
online version at doi:10.1016/j.rser.2016.10.018.
References
[1] Alam MM, Murad MW, Noman AHM, Ozturk I. Relationships among carbon
emissions, economic growth, energy consumption and population growth: Testing
Environmental Kuznets Curve hypothesis for Brazil, China, India and Indonesia.
Ecol Indic 2016;70:466–79.
[2] Apergis N, Payne JE. CO2 emissions, energy usage, and output in Central America.
Energy Policy 2009;37:3282–6.
[3] Apergis N, Payne JE, Menyah K, Wolde-Rufael Y. On the causal dynamics between
emissions, nuclear energy, renewable energy, and economic growth. Ecol Econ
2010;69:2255–60.
[4] Apergis N, Ozturk I. Testing environmental Kuznets curve hypothesis in Asian
countries. Ecol Ind 2015;52:16–22.
[5] Atici C. Carbon emissions in Central and Eastern Europe: environmental Kuznets
curve and implications for sustainable development. Sustain Dev 2009;17:155–60.
[6] Baek J, Kim HS. Trade liberalization, economic growth, energy consumption and
the environment: time series evidence from G-20 countries. J East Asia Econ Integr
2011;15:3–32.
[7] Baek J, Pride D. On the income-nuclear energy-CO2 emissions nexus revisited.
Energy Econ 2014;43:6–10.
[8] Baltagi B. Econometric analysis of panel data, fourth edition. Chichester: John
Wiley & Sons; 2008.
[9] Begum RA, Sohag K, Abdullah SMS, Jaafar M. CO2 emissions, energy consumption,
economic and population growth in Malaysia. Renew Sustain Energy Rev
2015;41:594–601.
[10] Bilgili F, Koçak E, Bulut Ü. The dynamic impact of renewable energy consumption
on CO2 emissions: a revisited Environmental Kuznets Curve approach. Renew
Sustain Energy Rev 2016;54:838–45.
[11] Breitung J. The local power of some unit root tests for panel data. In: Baltagi B,
editor. Nonstationary panels, panel cointegration, and dynamic panels, advances in
econometrics. Vol. 15; 2000p. 161–78.
[12] Cemil F, Gday O, Erbas BC. How effective are energy efficiency and renewable
energy in curbing CO2 emissions in the long run? A heterogeneous panel data
analysis. Energy 2015;82:734–45.
[13] Choi I. Unit root tests for panel data. J Int Money Financ 2001;20:249–72.
[14] David F. Bradford, Rebecca Schlieckert and SHS. The Environmental Kuznets
Curve: exploring a fresh specification. Natl Bur Econ Res Work Pap Ser 2000:8001.
[15] Dong B, Wang F, Guo Y. The global EKCs. Int Rev Econ Financ 2016;43:210–21.
[16] Energy Information Administration. Monthly Energy Review. 10. Renewable

1074
Z. Zoundi
energy. 〈https://www.eia.gov/totalenergy/data/monthly/pdf/mer.pdf〉; 2015.
[17] Gene MG, Alan B. Economic Growth and the Environment. J. TQ
2010;110:353–77, The MIT Press.
[18] Hadri K. Testing for stationarity in heterogeneous panel data. Econ J
2000;3:148–61.
[19] Heidari H, Turan Katircioğlu S, Saeidpour L. Economic growth, CO2 emissions, and
energy consumption in the five ASEAN countries. Int J Electr Power Energy Syst
2015;64:785–91.
[20] Ibrahim MH, Law SH. Social capital and CO2 emission–output relations: a panel
analysis. Renew Sustain Energy Rev 2014;29:528–34.
[21] Im KS, Pesaran MH, Shin Y. Testing for unit roots in heterogeneous panels. J Econ
2003;115(53):74.
[22] International Energy Agency. World Energy Outlook, Paris, France. 〈http://www.
worldenergyoutlook.org/media/weowebsite/2009/WEO2009.pdf〉; 2009a.
[23] International Renewable Energy Agency. Africa’s Renewable Future. 〈http://www.
irena.org/documentdownloads/publications/africa_renewable_future.pdf〉; 2009.
[24] IPCC . Climate Change 2007: the Physical Science Basis. Contribution of Working
Group I to the Fourth. In: Solomon S, Qin D, Manning M, Chen Z, Marquis M,
Averyt KB, Tignor M, Miller HL, editors. Assessment report of the intergovern-
mental panel on climate change. Cambridge, United Kingdom and New York, NY,
USA: Cambridge University Press; 2007.
[25] Iwata H, Okada K, Samreth S. Empirical study on the determinants of CO2
emissions: evidence from OECD countries. Appl Econ 2012;44:3513–9.
[26] Jebli MB, Youssef MB, Ozturk I. The Role of renewable energy consumption and
trade: Environmental Kuznets Curve Analysis for Sub-Saharan. Afr Ctries Afr Dev
Rev 2015;27:288–300.
[27] Kao C. Spurious regression and residual-based tests for cointegration in panel data.
J Econ 1999;90:1–44.
[28] Kao C, Chiang MH. On the estimation and inference of a cointegrated regression in
panel data. In: Baltagi BH, Fomby TB, Hill RC, editors. . Nonstationary panels,
panel cointegration, and dynamic panels, 15. Advanced in Econometrics; 2000. p.
170–222.
[29] Kuznets S. Economic growth and income inequality. Am Econ Rev 1955;45:1–28.
[30] Lantz V, Feng Q. Assessing income, population, and technology impacts on CO2
emissions in Canada: Where's the EKC?. Ecol Econ 2006;57:229–38.
[31] Levin M, Lin CF, Chu CS. Unit root tests in panel data: asymptotic and finite sample
properties. J Econ 2002;108:1–24.
[32] Liu Y, Zhou Y, Wu W. Assessing the impact of population, income and technology
on energy consumption and industrial pollutant emissions in China. Appl Energy
2015;155:904–17.
[33] Menyah K, Wolde-Rufael Y. CO2 emissions, nuclear energy, renewable energy and
economic growth in the US. Energy Policy 2010;38:2911–5.
[34] Menz T, Welsch H. Population aging and carbon emissions in OECD countries:
accounting for life-cycle and cohort effects. Energy Econ 2012;34:842–9.
1075
Renewable and Sustainable Energy Reviews 72 (2017) 1067–1075
[35] Munasinghe M. Is environmental degradation an inevitable consequence of
economic growth: tunneling through the environmental Kuznets curve. Ecol Econ
1999;29:89–109.
[36] Nasir M, Ur Rehman F. Environmental Kuznets Curve for carbon emissions in
Pakistan: an empirical investigation. Energy Policy 2011;39:1857–64.
[37] Pedroni P. Fully Modified OLS for Heterogeneous Cointegrated Panels and the Case
of Purchasing Power Parity. Department of Economics, Indiana University; 1996,
[WP 96–020].
[38] Pedroni P. Critical values for cointegration tests in heterogeneous panels with
multiple regressors. Oxf Bull Econ Stat 1999;61:653–70.
[39] Pedroni P. Fully modified OLS for heterogeneous cointegrated panels. Adv Econ
2000;15:93–130.
[40] Pedroni P. Purchasing power parity tests in cointegrated panels. Rev Econ Stat
2001;83:727–31.
[41] Robalino-López A, Mena-Nieto Á, García-Ramos J-E, Golpe AA. Studying the
relationship between economic growth, CO2 emissions, and the environmental
Kuznets curve in Venezuela (1980–2025). Renew Sustain Energy Rev
2015;41:602–14.
[42] Roodman D. How to do xtabond2: an introduction to difference and system GMM
in Stata. Stata J 2009;9:86–139.
[43] Shafiei S, Salim RA. Non-renewable and renewable energy consumption and CO2
emissions in OECD countries: a comparative analysis. Energy Policy
2014;66:547–56.
[44] Shi A. The impact of population pressure on global carbon dioxide emissions,
1975–1996: evidence from pooled cross- country data. Ecol Econ 2003;44:29–42.
[45] Soytas U, Sari R, Ewing BT. Energy consumption, income, and carbon emissions in
the United States. Ecol Econ 2007;62:482–9.
[46] Soytas U, Sari R, Ewing BT. Energy consumption, income, and carbon emissions in
the United States. Ecol Econ 2007;62:482–9. http://dx.doi.org/10.1016/j.ecole-
con.2006.07.009.
[47] United Nations Framework Convention on Climate Change. Adoption of The Paris
Agreement., Conference of the Parties Twenty-first session Paris. Agenda item 4(b);
2015, p. 1–32.
[48] United Nations Framework Convention on Climate Change. Synthesis Report on
the Aggregate Effect of the Intended Nationally Determined Contributions. 〈http://
www.cop21.gouv.fr/wpcontent/uploads/2015/10/rapport_de_synthese_de_la_
ccnucc.pdf〉; 2007.
[49] Wesseh PK, Lin B. Can African countries efficiently build their economies on
renewable energy?. Renew Sustain Energy Rev 2016;54:161–73.
[50] Westerlund J. Testing for error correction in panel data. Oxf Bull Econ Stat
2007;69:709–48.
[51] Yandle B, Vijayaraghavan M, Bhattarai M. The Environmental Kuznets Curve a
primer. Prop Environ Res Cent Res Study 2007, [02-01].