You are on page 1of 34

THE JOURNAL OF ENERGY

AND DEVELOPMENT

Paul Adjei Kwakwa and George Adu,

“Effects of Income, Energy Consumption,


and Trade Openness on Carbon Emissions
in Sub-Saharan Africa”
Volume 41, Number 1

Copyright 2016
EFFECTS OF INCOME, ENERGY CONSUMPTION,
AND TRADE OPENNESS ON CARBON EMISSIONS
IN SUB-SAHARAN AFRICA

Paul Adjei Kwakwa and George Adu*

Introduction

T he attention of environmentalists and policy makers to climate change and


global warming in recent years has increased due to their threatening effects.
The estimated effects of climate change and global warming are numerous and
diverse, ranging from the deterioration of the environment to the health impli-
cations for human population.1 However, the effects of climate change and global
warming are found not to be the same for all regions in the world as some
countries, especially those in Sub-Saharan Africa, are considered to be at greater
risk from these events.2 A major contributor to climate change and global warming
is carbon dioxide (CO2) emissions. The concentration of CO2 in the atmosphere as

*Paul Adjei Kwakwa is a Ph.D. candidate with the Department of Economics, Kwame Nkrumah
University of Science and Technology, Kumasi, Ghana, and lectures on economics at the
Presbyterian University College, Ghana. The author’s research interests focus on economic growth,
development, and environmental and resource economics. His recent articles have been published in
The International Journal of Energy Economics and Policy, The Journal of Rural and Industrial
Development, and Renewable and Sustainable Energy Reviews.
George Adu is an Assistant Professor in the Department of Economics, Kwame Nkrumah
University of Science and Technology, Kumasi, Ghana, and a Senior Researcher with The Nordic
Africa Institute, Uppsala, Sweden. His current areas of research include economic growth, impacts
of climate change on economic growth, environmental and resource economics, and energy
economics. The author’s publications have appeared in Environmental Economics and Policy
Studies, Review of Development Finance, African Business, The International Journal of Sustainable
Economy, South African Journal of Economics, Journal of Developing Areas, Journal of Economics
and Econometrics, and African Review of Economics and Finance, as a sampling.

The Journal of Energy and Development, Vol. 41, Nos. 1 and 2


Copyright  2016 by the International Research Center for Energy andEconomic Development
(ICEED). All rights reserved.
85
86 THE JOURNAL OF ENERGY AND DEVELOPMENT

observed by climate scientists has been increasing significantly over the past
century. For the past few decades alone, global carbon dioxide emissions in-
creased from 5,612.9 million tons in 1975 to 8,537.8 million tons in 1995. By
2005, CO2 emissions stood at 11,335.5 million tons and climbed to 13,773.1
million tons in 2011.3 Globally, the current high trend of carbon emissions can be
said to be associated with a high level of economic growth, energy consumption,
and international trade.4
Although the contribution of Sub-Saharan Africa to carbon emissions is very
low compared with some nations, especially the Organization for Economic Co-
operation and Development (OECD) members as well as emerging economies like
China, there has been a 12-fold increment from 1950 to 2008 with a current
emission level of 311 million metric tons.5 For the Sub-Saharan Africa (SSA)
region alone, the emissions of CO2 have increased gradually from 302.9 million
tons in 1975 to 669.7 million tons in 1995 and then to 1,058.3 million tons in
2011.6 The spate of carbon emissions from the African continent suggests po-
tentially serious future implications, especially for the SSA, because of its effect
on climate change and global warming to which the continent is very vulnerable.
This is because many people in the SSA sub-region rely on agriculture that is
dependent on rain for their sustenance.
Available data from the World Bank’s Word Development Indicators suggest
that the rise in carbon emissions in the SSA is found to follow a similar trend with
the sub-region’s growth in income, level of trade openness, and energy con-
sumption from 1971 to 2011 (see figures 1 through 4).7 For instance, the sub-
region’s per-capita income has increased from U.S. $217 in 1971 to U.S. $694 in
1981 and then to U.S. $1,661 in 2011. This development could be attributed to
the impressive growth performance the region has experienced as it has grown at
more than 5 percent on the entire continent and 5.6 percent for the SSA since
2000.8 Energy demand or consumption also has been increasing with carbon
emissions in the sub-region from about 16,948 kilotons (kt) of oil equivalent in
1971 to 312,946 kt of oil equivalent in 1991 and further to 511,879 kt of oil
equivalent in 2011. The high level of energy expenses of some SSA countries
also suggests that energy demand has increased.9 Regarding trade openness, the
figure shows that the sum of import and export as a share of gross domestic
product (GDP) has increased from about 44 percent in 1971 to around 74 percent
in 2010 but reduced marginally to 62 percent in 2011. Over the period, the share
of exports (imports) in GDP rose from about 21 (24) percent in 1971 to 27 (28)
percent in 1996 and then to 30 (31) percent in 2011, although the share of imports
is higher than that of exports. This increase in the trend of exports and imports
can be attributed to the trade liberalization policies associated with the structural
adjustment programs (SAP) and economic recovery programs (ERP) of the In-
ternational Monetary Fund and World Bank that many SSA countries adopted in
the 1980s and 1990s as a remedy for the abysmal economic performances they
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 87

were experiencing. Following that, between 2002 and 2009, the average nominal
export growth rate of SSA is reported to have exceeded the world’s average of
9.7 percent by 5.4 percent. Moreover, it outpaced those of developing countries
in Asia (13.7 percent) and Latin America (9.9 percent). Importation to the sub-
region also increased with an average annual growth rate of 16.9 percent, ex-
ceeding that of developing Asia and America that stood at 9.7 percent over the
same period.10
The data from figures 1 through 4 suggest that the path of carbon dioxide
emissions in the SSA is influenced by the paths of income, trade openness, and
energy consumption. Accordingly, this paper examines the effects of income,
energy consumption, and trade openness on CO2 emissions in SSA using panel
cointegration techniques. This article makes four important contributions to the
literature on the determinants of carbon emissions in general and particularly for
the Sub-Saharan Africa region. First, from the review of the available literature,
we identified a dearth of research on Sub-Saharan Africa that controls for income,
trade, and energy consumption in a single regression. However, the potential
positive effect of trade on income and energy consumption on growth and, hence,
income per person as well as a possible feedback from income to energy con-
sumption is well documented in the literature.11 Thus, omitting one of these
closely related variables from the emissions equation can result in upwards bias of
the estimated effect of income on emissions. The existent literature on the subject
for the SSA commonly suffers from this defect as the study by M. Kohler12 appears
to be the only work that examines the emissions effect of energy consumption,
trade, and income in the sub-region. Some earlier studies on the sub-region ac-
count for the effect of either income and energy consumption13 or the effects of
income and trade14 on emissions.
Second, most of these previous studies on SSA, including the work by
M. Kohler,15 are mainly country case studies that employ time-series approaches,
which have some limitations associated with their estimations.16 This means that
a panel data study should be undertaken, which deals with the shortfalls associated
with time-series studies, providing additional power by combining the cross-
section and time-series data while allowing for heterogeneity across countries.17
Furthermore, findings from previous studies on the subject matter have not yielded
conclusive results, thus necessitating further research. Moreover, to the best of the
authors’ knowledge, many studies on the subject topic (B. Aka, G. Akpan and
U. Akpan, N. Apergis and J. Payne, M. Arouri et al., G. Grossman and A. Krueger,
D. Holtz-Eakin and T. Selden, H. Iwata et al., K. Jayanthakumaran et al., M. Jebli
et al., and M. Kohler) have tested for an inverted U-shape relationship between
income and emissions but none exists for trade.18 In this paper, we confront the
theoretical argument of the potential nonlinearities in the relationship between
trade and emissions with SSA data by including the quadratic trade term in our
specification.
88 THE JOURNAL OF ENERGY AND DEVELOPMENT

Figure 1
ENERGY USE FOR SUB-SAHARAN AFRICA, 1971-2010
(in kilotons of oil equivalent)

Source: The World Bank, World Development Indicators (Washington, D.C.: World
Bank, 2014).

Figure 2
EXPORT AND IMPORT SHARE OF GROSS DOMESTIC PRODUCT (GDP) FOR
SUB-SAHARAN AFRICA, 1971-2011
(in percentage)

Source: The World Bank, World Development Indicators (Washington, D.C.: World
Bank, 2014).
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 89

Figure 3
CARBON DIOXIDE EMISSIONS FOR SUB-SAHARAN AFRICA, 1971-2011
(in kilotons)

Source: The World Bank, World Development Indicators (Washington, D.C.: World
Bank, 2014).

Figure 4
GROSS DOMESTIC PRODUCT (GDP) PER CAPITA FOR SUB-SAHARAN AFRICA,
1971-2011

Source: The World Bank, World Development Indicators (Washington, D.C.: World
Bank, 2014).
90 THE JOURNAL OF ENERGY AND DEVELOPMENT

This research finds that emissions of CO2 in SSA are explained by income,
energy consumption, trade openness, urbanization, and industrialization. In par-
ticular, it confirms that an inverted U-shaped relationship exists between income
and carbon emissions implying that, eventually, promoting economic develop-
ment will reduce carbon emissions. This paper, which again establishes a non-
linear relationship between trade and emissions, concludes that both income and
non-income factors account for carbon emissions in SSA; however, income and
energy consumption have the greatest effects.
The remainder of this article is organized as follows. The next section provides
a comprehensive review of the literature on the core drivers of carbon emissions.
This is followed by a presentation of the empirical model and econometric
strategies toward the identification (estimation) of model parameters. We also
provide a description of data sources and sample as well as descriptive analysis
of the data. The subsequent section presents and discusses the estimate results
followed by our paper’s conclusion with our summary of findings and policy
implications.

Literature Review

Theoretically, the relationship between income and emissions is explained by


the environmental Kuznets curve (EKC) hypothesis. The EKC hypothesis stems
from S. Kuznets’ argument that economic development initially increases income
inequality, but it reaches a peak and then income inequality is reduced.19 This
relationship between economic development and income inequality is seen as an
inverted U-shaped curve. Similarly, the EKC hypothesis, which was first hy-
pothesized by G. Grossman and A. Krueger, argues an inverted U-shaped re-
lationship between income and environmental degradation.20 That is, an initial
level of growth in income will harm the environment until income reaches a level
where the quality of the environment improves with income. According to
G. Grossman and A. Krueger21 and T. Panayotou,22 three factors explain the inverted
U-shape relationship between income and the quality of the environment: the scale
effect, the composition effect, and the technique effect.23 The scale effect explains
that growth in income occurs with an increase in the scale of economic activities.
Thus, there will be more consumption and production that will increase the level
of environmental pollution than before. The second explanation, known as the
composition effect, is that the output composition of countries changes drastically
as they grow from a more agrarian economy to an industrialized one, which in-
creases pollution initially. Later, the composition shifts from industry to services
resulting in decreasing pollution. The third factor is the technique effect; and the
explanation is that at high levels of income, pollution reduction measures are put in
place since individuals and policy makers become more environmentally conscious.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 91

Given the inverted U-shaped relationship between income and emissions, the
rising portion of the curve indicates the scale effect, while the falling part is due
to the composition and the technical effects.
However, the reality is that as an economy grows and people demand and
produce more goods, both the composition of production and the technique of
production changes and, thus, the effect of an increase in income will depend upon
the direction of these changes. A higher scale effect than the composition and/or
technique effect will increase pollution as the economy grows.24 Also, although it
is clear and easy to identify the location of all the effects on the curve, economists
and researchers have debated the level of income at which the turning point should
take place. For instance, G. Grossman and A. Krueger25 suggest a per-capita in-
come of $8,000; while D. Holtz-Eakin and T. Selden26 provide $35,418 as their
figure. E. Neumayer27 gives a range between $55,000 and $90,000, M. Cole28
mentions $62,700 and $25,100, and B. Aka29 suggests a value of $825.
On the one hand, the environmental damaging effect of trade can be explained
by the pollution haven hypothesis, which argues that trade puts pressure on host
nations, especially poor ones, to ease the environmental standards in order to
protect the hosted foreign firms or entice other firms to come and operate in such
countries. When environmental laws are relaxed, negative externalities increase
since they are not borne by the producers.30 As a result, trade will benefit the
developed countries relative to the developing ones. Another argument to support
the negative effect of trade on the environment is that it encourages countries to
extract more resources that do not possess well-defined property rights. In the
process, according to G. Chichilnisky,31 there is a lower marginal cost of ex-
traction compared with the true value of the resources.32
On the other hand, the effect of trade on pollution can be linked to the EKC
hypothesis. Since trade can promote growth, an increase in trading activities,
which leads to growth in income, will lead to a higher consumption and production
level associated with high pollution. This is the scale effect. However, trade may
cause an economy to move from agriculturally-based to industrial-based and,
lastly, to the service sector, where pollution is low. This is the composition effect.
Finally, there is the technique effect when trade enables countries to export more
and invest in newer and lower pollution production techniques.33 Energy plays an
important role in the growth and development of economies. However, energy
usage also is argued to harm the environment through emissions.34 The environ-
mental impact of energy is explained by the fact that the combustion of fossil
fuels, e.g., gasoline, coal, and diesel, releases emissions such as nitrogen oxides,
sulfur dioxide, and carbon dioxide. As a result, inefficient usage of energy in-
creases the rate of emissions.35 It is in this light that energy-efficient technologies
and clean energy are encouraged to help reduce carbon emissions globally.36
Consequently, energy consumption will increase carbon emissions if it is generated
with less clean energy sources and fewer energy-saving technologies are used,
92 THE JOURNAL OF ENERGY AND DEVELOPMENT

while the use of cleaner energy and more efficient energy technologies can help
reduce carbon emissions.
These arguments indicate clearly that the effect of income, energy consump-
tion, and trade is inconclusive. To test these arguments, empirical studies have
been carried out using different data sets, regions, and methodological ap-
proaches with mixed results. For instance, the pioneering work of G. Grossman
and A. Krueger, which assessed the relationship between income growth and
emissions for 42 countries covering the period of 1977–1984, confirmed the EKC
hypothesis for two out of three emissions.37 D. Holtz-Eakin and T. Selden also
examined the relationship between per-capita income and carbon dioxide emis-
sions for 130 countries over the period 1951–1986 and the findings confirmed the
EKC hypothesis.38 T. Selden and D. Song used a cross-sectional panel data and
found an inverted U-shaped relationship between emissions and income.39 Later,
G. Grossman and A. Krueger estimated the effect of growth on environmental
pollution and confirmed the EKC hypothesis for the United States.40 A study by
B. Aka employed the autoregressive distributed lag (ARDL) model to analyze
both the short- and long-run impacts of trade and growth on carbon emissions on
Sub-Saharan Africa using data covering the 1961–2003 period.41 The author found
that in the short run, a 1-percent increase in economic growth leads to a 1.04-
percent increase in carbon emissions but, in the long run, a 1-percent increase in
GDP per capita reduces emissions by 1.8 percent. Moreover, a 1-percent increase
in trade intensity reduced carbon emissions by 0.15 percent in the short run and
0.57 percent in the long run. Very recently, M. Arouri et al. also modeled carbon
emissions in Thailand as a function of income, square of income, energy con-
sumption, trade, and urbanization.42 Employing the ARDL bounds test for annual
data for 1971–2010, a long-run relationship was established among the vari-
ables. The estimation showed that energy consumption increases emissions
while trade reduces emissions. In addition, the EKC hypothesis was confirmed
in the study.
H. Iwata et al. also examined the determinants of carbon emissions in 11 OECD
countries by employing the ARDL to cointegration approach.43 The authors
modeled carbon emissions as a function of trade, electricity production from
nuclear, energy consumption, income, and square of income. The results show an
insignificant effect of trade on emissions, a positive effect of energy consumption
on emissions, an emission-reducing effect of nuclear energy, and a confirmation of
the EKC hypothesis for some of the selected countries. F. Halicioglu studied the
causal relationships between carbon emissions, energy consumption, income, and
foreign trade in the case of Turkey using the time-series data for the period
1960–2005.44 A positive and significant effect of energy consumption, growth, and
trade on carbon emissions was discovered. In terms of magnitude, it was found
that income is the most significant variable followed by energy consumption and
foreign trade. G. Akpan and U. Akpan applied a multivariate vector error-correction
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 93

model (VECM) to examine the long-run and causal relationship between elec-
tricity consumption, carbon emissions, and economic growth in Nigeria for the
period 1970 to 2008.45 The results showed that an increase in economic growth
increases carbon emissions although there was no support for the EKC hypothesis.
Furthermore, a negative relationship was uncovered between electricity con-
sumption and carbon emissions, which the authors attributed to the large deficit in
the supply and surplus demand of electricity that exists in Nigeria. A Granger
causality test showed an unidirectional causality from income to emissions and no
causality between energy and emissions.
N. Apergis and J. Payne studied the causal relationship between carbon
emissions, energy consumption, and GDP within a panel VECM for six Central
American countries over the period of 1971–2004.46 The long-run results con-
firmed the EKC hypothesis and a positive effect of energy consumption on
emissions. Granger causality test results from the authors showed a short-run
unidirectional causality from energy consumption and real output to emissions,
but a long-run bidirectional causality was found between energy consumption and
emissions. In another study, N. Apergis and J. Payne used VECM to study the
effects of carbon dioxide emissions on energy consumption and real output for
11 countries of the Commonwealth of Independent States over the period
1992–2004.47 The authors found energy consumption has a positive effect on
carbon dioxide emissions and a confirmation of the EKC hypothesis.
K. Jayanthakumaran et al. examined the emissions effect of income, energy
consumption, and trade openness for China and India.48 The bounds testing ap-
proach to cointegration indicated a long-run relationship among the variables in
both countries. In the long run, the EKC hypothesis was established for both
nations. In addition, an insignificant effect of trade on emissions and a positive
effect of energy consumption on emissions for China and India were observed.
J. Baek and H. Kim in their study on the interrelationships between trade, income,
energy consumption, and carbon emissions for G-20 economies employed the
cointegrated vector autoregression (CVAR) for their estimation.49 Using data for
the period 1960–2006 the authors found, that in the long run, income and trade
have negative relationships with emissions for the developed G-20 members. For
the developing nations, trade and income were discovered to have positive effects
on emissions. Again, for most of the countries, energy consumption was found to
positively contribute to emissions. The study by M. Jebli et al. employed the panel
cointegration technique to investigate the causal relationship between carbon
emissions, energy consumption, and trade openness for 25 OECD countries over
the period 1980–2009.50 Estimations from FMOLS (fully modified ordinary least
squares) and DOLS (dynamic ordinary least squares) confirmed the EKC hy-
pothesis: there was an emissions reduction effect of renewable energy and exports/
imports, and a positive effect of non-renewable energy on emissions. Granger
causality tests revealed both the square of income and non-renewable energy
94 THE JOURNAL OF ENERGY AND DEVELOPMENT

Granger cause emissions. H. Chebbi et al. modeled carbon emissions as a function


of trade and growth in Tunisia for the period 1961–2004.51 A co-integration
technique employed showed that in the long run there is a positive linkage between
trade openness and per-capita emissions and a negative linkage between economic
growth and per-capita pollution emissions. In the short run a positive effect of
trade on emissions was established while growth was found to have a negative
effect on emissions. The Granger causality test found bidirectional relationships
between emissions and output growth in the long run.
S. Farhani et al. examined the effects of GDP, energy consumption, trade
openness, and urbanization on carbon emissions for Middle East and North Af-
rican countries over the period 1980–2009.52 The results from FMOLS and DOLS
show that income, trade openness, and energy consumption positively affect
carbon emissions, while squared income and urbanization have negative effects on
emissions. A short-run causality from GDP and energy consumption to emissions
was established. In the long run, unidirectional causality from trade, GDP, and
energy consumption was obtained. In his study, U. Al-mulali examined the effects
of energy consumption, foreign direct investment, GDP, and total trade on carbon
emissions for 12 Middle Eastern nations.53 Using a data set for the 1990–2009
period, the research established that all the variables have positive effects on
carbon emissions.
S. Amin et al. relied on time-series data from 1976–2007 to examine the causal
relationship among energy use, growth, and carbon emissions in Bangladesh.54 A
Johansen cointegration test confirmed a long-run relationship among all the var-
iables and a Granger causality test found a one-way causality from energy use to
emissions. Z. Gu et al. also examined the effects of trade on carbon emissions in
China.55 The authors used annual data from 1981 to 2010 and the Johansen
cointegration test showed there is a long-term equilibrium among carbon emis-
sions, foreign trade dependency, and foreign direct investment dependency. The
estimation revealed that an increase in foreign trade dependency and foreign direct
investment dependency will increase emissions. K.-M. Lim et al. studied the
causality issues among oil consumption, carbon emissions, and economic growth
in the Philippines.56 The authors relied on annual time-series data for the period
1965–2012 and by employing the Granger causality test, a bidirectional causality
between oil consumption and carbon emissions and a unidirectional causality
running from carbon emissions to economic growth were detected. A. Rahman and
A. Porna used data spanning 1970–2008 to investigate the relationship between
environmental parameters and economic growth in six South Asian countries and
found a long-run relationship between growth and emissions and a one-way Granger
causality from carbon dioxide to growth.57
S. Sharma examined the effect of energy, growth, urbanization, and trade on
emissions for 69 countries.58 The results from the generalized method of moments
estimator for three separate panels generally indicated a positive effect of income
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 95

and energy consumption on emissions. A study by J. Ang looked at the re-


lationship between emissions, energy consumption, and output for France.59 The
author employed the VECM for data covering the period 1960–2000 and found
that GDP Granger caused emissions. I. Ozturk and G. Uddin examined the cau-
sality relationship between energy consumption, carbon dioxide emissions, and
economic growth in India over the period 1971–2007.60 Granger causality tests
showed a causal flow from energy consumption to carbon emissions. G. Liu ex-
amined the causal relationship between growth and carbon emissions among other
variables using annual data for the period 1973–2003.61 The test results indicate a
unidirectional causal relationship exists between GDP and air emissions. P. Adom
et al. investigated causal relationships among carbon emissions, economic growth,
technical efficiency, and industrial structure for Ghana, Senegal, and Morocco.62
Using the bounds cointegration approach for data covering the period of 1971 to
2007, a long-run equilibrium was established for two countries while the Toda and
Yomamoto Granger causality test showed a blend of unidirectional, bidirectional,
and neutral relationships for all countries.
The general conclusion from the above review shows a diversity of opinions on
the effect of income, energy consumption, and trade on CO2 emissions. Thus,
there is no consensus yet on how these factors affect the environment through
carbon emissions and this may be due to the differences in the estimation tech-
niques, data sources, and time periods used for the various studies. Again, very few
studies have incorporated all three variables—energy consumption, income, and
trade—in the estimation process. These then call for further research to deepen the
understanding on the subject matter.

Methods

Theoretical and Empirical Specifications: Theoretically, it is argued through


the EKC hypothesis that income has a quadratic relationship with carbon emis-
sions, energy has a linear relationship with emissions, and a non-linear relation-
ship for trade and emissions through the EKC hypothesis. Based on this, we model
the emissions effects of income, trade openness, and energy consumption as
follows:
CPCit = a + bXit + lZit + jit ð1Þ
where CPC is carbon emissions, i represents countries used in the study, t rep-
resents time, and j consists of the unobservable country-specific effect and the
remainder disturbance. X represents vector of explanatory variables, namely, in-
come, square of income, energy consumption, trade openness, and square of trade
openness. Z consists of control variables, namely, urbanization and industrialization
96 THE JOURNAL OF ENERGY AND DEVELOPMENT

added to avoid biasedness in the results. The inclusion of these two variables was
based on the fact that they are known to affect carbon emissions.63
In our estimation, we advance six separate models where two of them serve as
our base-line models, which examine the emission effects of energy consumption,
income, and square of income as well as the emission effect of income, square of
income, and trade openness. The other two models are with income or energy
consumption tests for the nonlinearity relationship between trade and emissions.
The rest of the models add the control variables for the estimations. The natural
logarithmic forms of all these variables are used for estimation, which gives us
their growth rate and, thus, affords us the opportunity to interpret the coefficients
bs and ls as elasticities. If an inverted U-shaped relationship is found between
income and emissions (i.e., a positive sign for b1 and a negative for b2), the
turning point (K) will be estimated as:
k = ex½ – ðb1 =2b2 Þ ð2Þ
where b1 is the coefficient of the non-squared income while b2 represents the
coefficient of the square of income.

Empirical Strategy—Testing for Unit Roots: In estimating equation (1), the


following steps are taken into consideration. First, all the stationarity tests for the
variables are conducted. This test is important since most panel data are known to
contain unit roots and, as such, estimation with such variables will produce spu-
rious results. Though quite a number of methodologies are available to conduct
unit root tests, this study utilized the popular Levin-Lin-Chu (LLC), Breitung, and
Im, Pesaran and Shin (IPS) tests. The LLC test by A. Levin et al.64 is seen as
a panel extension of the augmented Dickey-Fuller (ADF) test and it is based on the
following equation:
Xri
Dxiy = ai + bxit – 1 + b Dxit – 1 + mit
j = 1 ij
ð3Þ

where i = 1,…I denotes the country; t = 1,…,T is the time period; and xit represents
the series for country i over the time period t. The number of lags is represented by
ri and mit is the stationary error term. In carrying out this test A. Levin et al.
suggest a three-step procedure: (i) carrying out separate ADF regressions for each
individual variable and generating two orthogonalized residuals; (ii) estimating
the ratio of long-run to short-run innovation standard deviations for each individual
variable; and (iii) computing the pooled t-statistics, with the average number of
observations per individual variable and average lag length.65 The LLC works with
a null hypothesis Ho: b=r – 1 = 0 and an alternate hypothesis of H1 = b < 0.
An extended version of the LLC test is the IPS test by K. Im et al.66 The IPS test
allows for heterogeneity on the autoregressive (AR) coefficient, b, for all the panel
unit, a feature that makes IPS superior to the LLC. As a result, according to IPS,
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 97

the null hypothesis is stated as Ho : bi = 0 "I; while the alternate is bi < 0 "I.
J. Breitung identifies that the LLC and IPS tests suffer from a dramatic loss of power
based on how the deterministic components are specified and, consequently,
proposed a test statistic to overcome this difficulty.67 Breitung’s test statistic as-
sumes a common unit root process and also is shown to be asymptotically dis-
tributed as a standard normal.68

Empirical Strategy—Testing for Cointegration: Next, the panel cointegra-


tion is tested to determine if there is a long-run relationship among the variables.
The Pedroni cointegration test was used in the study.69 Pedroni’s test is an ex-
tended version of the Engle-Granger70 cointegration test, which explains that there
is a cointegration for I(1) series if the residual from their spurious regression is
found to be I(0).71 By extending this approach, Pedroni developed a test that
took into consideration the individual heterogeneous and fixed-effect cointe-
grated panels by proposing seven test statistics for testing panel cointegration.72
The seven tests fall into two groups: those that test for heterogeneous cointegration
based on the within dimension technique and those that are based on the between
means techniques. The former group consists of the panel Phillips-Perron (PP),
panel augmented Dickey-Fuller (ADF), panel v-statistic, and panel rho-statistic; the
latter consists of the group PP, group ADF, and group rho-statistic.
The Pedroni conitegration test is based on the regression in equation (4):73
yit = xit bi + di + qi + mit ð4Þ

Examining the existence of cointegration requires the test to be done on the


residual from the static long-run regression term expressed in equation (5):
mit = ri mit – 1 + git ð5Þ
with the null hypothesis of no cointegration rI = 1.

Empirical Strategy—Estimations Technique: Having tested for cointe-


gration, the final step is to estimate the effects of energy consumption, trade
openness, income, industrialization, and urbanization on carbon emissions in
equation (1) by using the extended version of Phillip and Hansen74 fully modified
ordinary least squares (FMOLS) estimator for time-series data as proposed by
Pedroni75 for heterogeneous panel data. The choice of FMOLS estimator over
other estimators is determined by the fact that it is appropriate for estimating
equations with I(1) variables and endogenous regressors. Another advantage of this
estimator is that it does not require testing for the order of cointegration rank. Ad-
ditionally, it is found to be robust for both non-stationary and endogenous variables.
The panel FMOLS estimator, generally, is given by equation (6):
98 THE JOURNAL OF ENERGY AND DEVELOPMENT

hXN XT i– 1 hXN XT i
^ fmol =
b ðx – xi Þ0 ðx – x Þ^
y +
+ T ^
D +
em ð6Þ
i=1 t = 1 it i=1 t = 1 it i it
+
where Dem is the serial correlation correction term and yit+ is the transformed
variable of yit to achieve the endogeneity correction.
Again, the panel extension of Stock and Watson’s76 dynamic ordinary least
squares (DOLS) is estimated to check for the robustness of the results. The esti-
mated coefficient of DOLS is given by:
XN XT – 1  XT 
^ dol =
b z it z it
0
zit ^
y +
ð7Þ
i=1 t=1 t=1 it

where zit = ½xit – xi ; Dxi;t – q ; …; Dxi;t + q  is 2(q + 1) x 1 vector of regressors.

The Data—Sources and Descriptive Analysis: The study used an unbalanced


cross-country panel data for 19 SSA countries (Benin, Botswana, Burkina Faso,
Cameroon, Congo Democratic Republic, Congo Republic of Brazzaville, Côte
d’Ivoire, Ethiopia, Ghana, Kenya, Mozambique, Nigeria, Senegal, Sudan, South
Africa, Tanzania, Togo, Zambia, and Zimbabwe) for the period 1977–2012. The
initial plan of using all the SSA countries was not possible because a number of the
countries were missing data for the variables of interest. Ultimately, we focused on
the 19 countries that had most of the data available for the 1977–2012 time span.
The use of the panel model also is informed by the fact that nations differ in many
respects and it is important to control for the heterogeneity differences, which, if
not accounted for, would lead to misspecifications of the results. Again, panel data
are known to be more informative, have greater variability, less collinearity, possess
more degrees of freedom, and the estimates are more efficient.77 All variables were
from the 2014 version of the World Bank’s World Development Indicators.78
The dependent variable, CPC, is measured by carbon dioxide emissions in
metric tons per capita. Income (Y) is measured by the per-capita GDP and Y2
denotes the square of income. The EKC hypothesis is expected to hold in this
region and, as such, income is expected to have a positive coefficient and a neg-
ative coefficient for income squared. Our trade openness (TO) variable is mea-
sured as the sum of exports and imports as a percentage of GDP. It, too, is expected
to have a positive effect on emissions since the region under study is a developing
one and the pollution haven hypothesis is likely to hold in this case. We introduce
in our estimation the square of trade openness (TO2) based on the grounds that the
literature has established the emission effect of trade through the EKC hypothesis.
This implies trade also may have an inverted U-shape relationship with emissions;
consequently, this study tests for such a relationship. Therefore, the square of trade
openness is expected to have a negative effect on emissions. Energy consumption
per capita (EC) is expected to have a positive coefficient. Urbanization (LUB)
measured as the total urban population is expected to positively increase emissions,
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 99

and the level of industrialization (INV) measured by the industrial value added as
a share of GDP is expected to increase carbon emissions.
The summary statistics of carbon emissions, income, industrialization, energy
consumption, trade, and urbanization are shown in table 1. The mean for income is
U.S. $1,481.69, and this is low compared with the developed and other devel-
oping Asian countries. The value of the standard deviation tells us that there is
little variation in the level of income across countries and time. Trade as a share of
income has a mean of 0.695, which suggests that within the period of study, trade’s
share of real GDP for the average country in the sample is in excess of 60 percent.
This indicates that, on average, the sub-region over the period under study was
modestly open to international trade. The minimum and maximum values of 0.109
and 2.153, respectively, are recorded for trade. The mean energy consumption is
639.511 kilograms of oil equivalent per capita, and when compared with the
minimum and maximum values, it is clear that there is not much variability in the
energy consumption across countries and time. Carbon emissions in terms of
metric tons per capita vary greatly across space and time as the standard deviation,
mean, minimum, and maximum values suggest.
As a precursor to preview the relationship between carbon dioxide emissions
and the variables of interest, the study employed scatter plots for the variables in
their natural logarithmic form. Figure 5 plots the GDP per capita against carbon

Table 1
a
SUMMARY OF DESCRIPTIVE STATISTICS

2 2
CPC Y Y TO TO EC LUB INV

Statistics
Mean
1.195 1481.69 6092981 0.695 0.615 639.51 4.31 29.42
Median
0.319 643.97 414711 0.646 0.417 412.43 4.11 25.21
Maximum
10.918 10647.12 1.13E+08 2.153 4.635 2961.35 15.34 77.41
Minimum
0.031 111.88 12518 0.109 0.012 207.76 1.12 7.18
Standard deviation
2.390 1975.96 14035977 0.363 0.640 557.05 1.72 14.41

a
CPC = carbon dioxide emissions in metric tons per capita; Y = income measured by the per-
2
capita gross domestic product; Y = income squared; TO = trade openness measured as the sum of
2
exports and imports as a percentage of gross domestic product; TO = trade openness squared; EC =
mean energy consumption measured in kilograms of oil equivalent per capita; LUB = urbanization
measured as the total urban population; and INV = industrialization measured by the industrial value
added as a share of gross domestic product.
100 THE JOURNAL OF ENERGY AND DEVELOPMENT

dioxide emissions and the line of best fit indicates a positive relationship sug-
gesting that GDP per capita is detrimental to carbon emissions. In figure 6, where
we plot the square of income and carbon dioxide emissions, the line of best fit also
depicts a positive relationship between the two. Figure 7 plots carbon emissions
and energy consumption; a positive relationship is seen, although it is concen-
trated between 5.5 and 6.5 kilograms per capita of energy consumption. In figure
8, a positive relationship is seen between trade and emissions. This suggests that
opening up the region to trade with other countries has a detrimental effect on the
environment. However, figure 9 indicates a negative relationship between the
square of trade and emissions of carbon dioxide. The relationship between
emissions and urbanization (figure 10) demonstrates a negative effect; however,
the relationship between emissions and industrialization (figure 11) is positive.
Thus, all the variables from the scatter plots are expected to harm the environment
with the exceptions of the square of trade and urbanization.

Results and Discussion

This section presents and discusses the results of econometric techniques


employed to establish the long-run effect of income, trade, energy consumption,

Figure 5
PLOT OF CARBON DIOXIDE EMISSIONS (LCPC) AND INCOME (LY)
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 101

Figure 6
2
PLOT OF CARBON DIOXIDE EMISSIONS (LCPC) AND SQUARE OF INCOME (LY )

Figure 7
PLOT OF CARBON DIOXIDE EMISSIONS (LCPC) AND ENERGY CONSUMPTION (LEC)
102 THE JOURNAL OF ENERGY AND DEVELOPMENT

Figure 8
PLOT OF CARBON DIOXIDE EMISSIONS (LCPC) AND TRADE OPENESS (LTO)

Figure 9
PLOT OF CARBON DIOXIDE EMISSIONS (LCPC) AND SQUARE OF TRADE OPENESS
2
(LTO )
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 103

Figure 10
PLOT OF CARBON DIOXIDE EMISSIONS (LCPC) AND URBANIZATION (LUB)

Figure 11
PLOT OF CARBON DIOXIDE EMISSIONS (LCPC) AND INDUSTRIALIZATION (LINV)
104 THE JOURNAL OF ENERGY AND DEVELOPMENT

urbanization, and level of industrialization on the rate of carbon dioxide emissions


in the SSA. The analysis presented in this section follows three sequential steps.
First, panel unit root tests are conducted on the individual series in our sample.
Second, panel cointegration tests are performed to ascertain the existence of long-
run equilibrium relationships among the variables in the specified equation. Last,
but not least, a different specification of the model is estimated using panel
FMOLS and DOLS.

Unit Root Tests: The results of the panel unit root tests for the series of in-
come, trade openness, energy consumption, carbon emissions, urbanization, and
industrialization are reported in table 2. The study relied on the LLC, Breitung,
and IPS tests to examine stationarity of the series in their levels; the results show
Table 2
a
RESULTS OF THE PANEL UNIT ROOT TESTS

LLC Breitung IPS


Series name t-stat t-stat t-stat

Variables in levels
LCPC –0.8349 –1.2286 –2.4636
LY –0.5971 4.9320 1.9042
2
LY –0.3152 5.2455 2.1650
LEC –0.3441 3.0848 2.0285
LTO 0.7375 1.0915 1.8271
2
LTO 0.9339 2.7386 3.4001
LUB 3.5142 8.1651 1.9093
LINV 1.0305 –2.9399* –1.2168
Variables at first difference
DLCPC –22.4471*** –12.7463*** –16.1760***
DLY –7.0980*** –7.3512*** –10.0329***
2
DLY –7.2194*** –7.1592*** –10.1161***
DLEC –6.2299*** –3.7970*** –11.5805***
DLTO –11.0145*** –7.3780*** –12.3626***
2
DLTO –7.4612*** –1.8427** –2.2030**
DLUB –5.1597*** –8.1384*** –6.7774***
DLINV –11.4190*** –12.5671*** –12.9091***

a
LLC = Levin-Lin-Chu unit root test; IPS = Im, Pesaran, and Shin unit root test; CPC = carbon
dioxide emissions in metric tons per capita; Y = income measured by the per-capita gross domestic
2
product; Y = income squared; EC = mean energy consumption measured in kilograms of oil
equivalent per capita; TO = trade openness measured as the sum of exports and imports as a percentage
2
of gross domestic product; TO = trade openness squared; UB = urbanization measured as the total
urban population; INV = industrialization measured by the industrial value added as a share of gross
domestic product; *** = rejection of the null hypothesis at the 1-percent level; ** = rejection of the null
hypothesis at the 5-percent level; and * = rejection of null hypothesis at the 10-percent level.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 105

all the tests could not reject the presence of unit root for the variables with the
exception of the Breitung that rejects the presence of unit root for industrialization.
At first difference of the variables, all three tests reject the presence of unit root at
the 1-percent level of significance. Thus, the variables become stationary at first
difference and, in light of this, we can then conclude that all variables are in-
tegrated of order one, I(1). With the individual series integrated of order one and,
thus, considered nonstationary, we proceed to the panel cointegration test.

Test for Long-Run Equilibrium: The study employs the Pedroni79 test to
check for cointegration among the series with the results presented in table 3. The
results for each group of series shows some tests accepting the null hypothesis of

Table 3
a
RESULTS OF THE PANEL COINTEGRATION TESTS

2
LCPC, LEC, LTO, LTO , LCPC, LTO, LCPC, LEC,
2 2
LY, LINV, and LUB LY, and LY LY, and LY
Test statistic Statistic Statistic Statistics

Panel v-statistic 1.5162* –1.3228 –1.0900


Panel rho-statistic –1.0742 –3.1609*** –4.0797***
Panel pp-statistic –9.3082*** –5.3478*** –6.9340***
ADF statistic –2.4289*** –1.5714* –1.6470**
Group rho-statistic 2.6012 0.1548 –0.2620
Group pp-statistic –7.6105*** –2.9981** –4.6563***
Group ADF-statistic –2.1594** –0.6547 –2.4549***
LCPC, LEC, LCPC, LEC, LTO, LINV,
2 2 2
LTO, LTO , and LY LTO, and LTO LUB, LY, and LY
Test statistic Statistic Statistic Statistic
Panel v-statistic 1.1286 –0.4681 0.5406
Panel rho-statistic –3.3258*** –2.1791** –1.1853
Panel pp-statistic –5.5323*** –5.1011*** –9.2670***
ADF statistic –1.3147** –0.0728 –1.9426**
Group rho-statistic 2.3910 –0.0652 2.7943
Group pp-statistic –2.8827*** –3.7308*** –10.5969***
Group ADF-statistic –1.3114* –2.1154** –2.7243***

a
CPC = carbon dioxide emissions in metric tons per capita; Y = income measured by the per-
2
capita gross domestic product; Y = income squared; EC = mean energy consumption measured in
kilograms of oil equivalent per capita; TO = trade openness measured as the sum of exports and imports
2
as a percentage of gross domestic product; TO = trade openness squared; UB = urbanization measured
as the total urban population; INV = industrialization measured by the industrial value added as
a share of gross domestic product; *** = rejection of the null hypothesis at the 1-percent level;
** = rejection of the null hypothesis at the 5-percent level; and * = rejection of null hypothesis at
the 10-percent level.
106 THE JOURNAL OF ENERGY AND DEVELOPMENT

no cointegration while others reject it. Thus, mixed results are provided from the
cointegration tests for each group of series. Notwithstanding this, the conclusion
that there is a stable long-run relationship among the variables for each group can
be established on two grounds. First, we have the PP-statistics, which J. Gao and
L. Zhang80 have noted to be more powerful than rho-statistics in confirming the
presence of long-run relationships for each group in the series. Second, the ma-
jority of the statistics indicates the presence of cointegration among the variables.

Long-Run Analysis: Having tested and confirmed that the variables are
cointegrated, the fully modified OLS (FMOLS) by Pedroni81 and an extension of
Stock and Watson82 DOLS estimation techniques for heterogeneous cointe-
grated panels are estimated; the results are reported in table 4. As observable,

Table 4
a
LONG-RUN PANEL ESTIMATES, SUB-SAHARAN AFRICA, 1977–2012

FMOLS Estimation
MODEL 1 MODEL 2 MODEL 3 MODEL 4 MODEL 5 MODEL 6

LEC 1.1151*** 1.2087*** 0.9633*** 1.1201***


LTO 0.0642* 0.0238 0.0636** 0.2245*** –0.0566
2
LTO –0.0832** –0.1113*** –0.1851***
LY 1.2321*** 1.0064*** 1.3745*** 0.8892*** 0.2714***
2
LY –0.0729** –0.0384 –0.0975**
LINV 0.1374*** 0.1365***
LUB 0.0598*** 0.0679*
DOLS Estimation
MODEL 1 MODEL 2 MODEL 3 MODEL 4 MODEL 5 MODEL 6
LEC 1.4257*** 1.9864*** 1.2504*** 1.5560***
LTO –0.0964 0.0716*** –0.1608 2.1565** –0.1170
2
LTO –0.1092* 0.0107 –0.1915***
LY 2.3680*** 2.9500** 4.0938*** 1.248*** 0.8524***
2
LY –0.1656*** –0.1706** –0.2907***
LINV 0.3207*** 0.1574**
LUB 0.0300* 0.0260***

a
Dependent variable = LCPC (carbon dioxide emissions in metric tons per capita); FMOLS =
fully modified ordinary least squares; DOLS = dynamic ordinary least squares; EC = mean energy
consumption measured in kilograms of oil equivalent per capita; TO = trade openness measured as
2
the sum of exports and imports as a percentage of gross domestic product; TO = trade openness
2
squared; Y = income measured by the per-capita gross domestic product; Y = income squared;
INV = industrialization measured by the industrial value added as a share of gross domestic product;
UB = urbanization measured as the total urban population; *** = rejection of the null hypothesis at
the 1-percent level; ** = rejection of the null hypothesis at the 5-percent level; and * = rejection of
null hypothesis at the 10-percent level.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 107

both FMOLS and DOLS estimations under the six models yield similar results.
The general conclusion from the results is that all the variables have a positive and
significant effect on CO2 emissions with the exceptions of the square of income
and square of trade, which have negative coefficients. In addition, we find that the
inclusion of the variable terms did not affect the magnitude of the estimated co-
efficient that much.
The result for energy consumption is found to positively contribute to CO2
emissions in the sub-region with a coefficient ranging from 0.96 to 1.99. The
interpretation is that a 1-percent increase in energy consumption will lead to be-
tween a 0.96-percent to 1.99-percent increase in the sub-region’s carbon emissions.
Thus, over time, higher energy consumption in the sub-region gives rise to more
carbon emissions. The reason why energy consumption increases emissions can be
explained by the fact that the sub-region is mainly rural where a high number of the
population relies on unclean energy sources, which emit more carbon, for their
cooking and other domestic activities.83 Even in the urban areas, the high price of
cleaner energy coupled with its relatively scarcity compel many to use less clean
energy sources, thereby increasing carbon emissions in the sub-region. The findings
here are consistent with J. Gao and L. Zhang, who report a coefficient of 0.27,
N. Apergis and J. Payne with a coefficient of 0.42, S. Hossain with a coefficient
of 1.09, and S. Farhani et al. with a coefficient lying between 0.82 and 0.92.84
The income elasticity of carbon emissions is positive and generally elastic but
negative and generally inelastic for the square of income. Specifically, a 1-percent
increase in real per-capita GDP leads to between a 0.3-percent to 4.1-percent
increase in carbon dioxide emissions; however, a 1-percent increase in the square
of income will reduce carbon emissions by almost 0.03 percent to 0.29 percent.
The positive sign for income and the negative sign for income squared in models 1
(FMOLS and DOLS), 2 (DOLS), and 3 (FMOLS and DOLS) support the EKC
hypothesis that pollution emissions initially increase with income and then de-
crease after income reaches a certain level. From the results, it can be said that the
reduction impact of growth in income from the composition and techniques effects
outweighs the negative scale effect of income growth. This implies that if eco-
nomic growth and development continue to increase in the sub-continent, it is
expected that carbon emissions will fall. The outcome in this study corroborates
S. Farhani et al. and J. Gao and L. Zhang.85 From the results we estimated the
turning point k = ex[–(b1/2b2)] of the EKC and found it to be between about U.S.
$1,142.85 to U.S. $5,687.09, which was higher than B. Aka’s value of U.S. $824
for the SSA.86 However, our results are lower than G. Grossman and A. Krueger’s
value of $8,000, D. Holtz-Eakin and T. Selden’s value of $35,418, and E. Neumayer’s
values of $55,000–$90,000.87 The variances in the values of the turning points
clearly indicates why, although it is clear and easy to identify the location of all the
three effects on the EKC, economists and researchers have debated the level of
income at which the turning point should take place.
108 THE JOURNAL OF ENERGY AND DEVELOPMENT

The effect of trade openness is found to be positive and generally inelastic,


meaning that a 1-percent increase in trade openness will increase carbon emissions
by a rate of less than 1 percent. On the other hand, the square of trade openness is
negative and significant. Specifically, a 1-percent increase in the square of trade
openness will reduce carbon emissions by about 0.08 percent to 0.19 percent.
Thus, the results strongly confirm nonlinearity in the relationship between trade
and emissions with SSA data. Therefore, this indicates that the composition and
the technique effects of trade exceed the scale effect of trade for the sub-region. As
a result, the pollution haven hypothesis can be concluded to be relatively a short-
run phenomenon for the SSA sub-region since in the long run trade will reduce
carbon emissions despite the fact that the sub-region seems to have weaker en-
vironmental laws for multinational firms from advanced nations that contribute to
regional emissions.88
With respect to urbanization, it is found that a 1-percent increase in the urban
population will cause carbon emissions to grow. The positive coefficient of ur-
banization can be explained by the fact that urban centers in the sub-region receive
the best services from utility companies, which have led to the fast rate of in-
dustrialization that requires less clean energy, such as gasoline, to operate well.
Again, heavy vehicular traffic congestion is a common thing to see in many urban
settings, the effect of which is more fuel usage, most of which is regarded as
unclean energy thereby leading to an increase in carbon emissions in the sub-
region. The finding here confirms the results of S. Sharma.89 Industrialization is
found to positively affect the emissions of carbon dioxide. The estimated in-
dustrialization coefficient implies that for every 1-percent increase in the in-
dustrial share of the GDP of the sub-region, carbon emissions will increase by
between 0.13 to 0.32 percent and vice versa. Thus, as the economy of the SSA
region becomes more industrialized, there is the associated negative effect of
carbon emissions from the machines used.

Diagnostic Test: To ensure reliability of the results, the estimated FMOLS and
DOLS models are tested against correlation problems because of the way some of
the variables are measured in this study. The variance inflation factor (VIF) was
used here to test for multicollinearity. The literature suggests a VIF value of
1 means there is no correlation among a predictor and the other remaining predictor
variables while a VIF exceeding 4 warrants further investigation. However, if the
VIF exceeds 10, then there are clear signs of serious multicollinearity that will
require correction. The test results are shown in table 5 for FMOLS models 3, 4, 5,
and 6, which indicate that the variables are likely to be correlated. As can be seen,
there is no problem of collinearity among the variables as they all have VIF values
of less than 1.07. It is to be noted that similar results were obtained for all the other
models (though not reported here). Testing for serial correlation was not much of
a concern since the main estimator—FMOLS—accounts for that.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 109

Table 5
a
VARIANCE INFLATION FACTOR (VIF) MULTICOLLINEARITY TEST RESULTS

FMOLS Model 3 FMOLS Model 6


Coefficient variance Uncentered VIF Coefficient variance Uncentered VIF

LTO 0.001284 1.021551 0.000998 1.036232


2
LTO 0.000687 1.055739
LY 0.000281 1.014262 0.001161 1.006565
2
LY 0.000714 1.056409
LEC 0.001121 1.048953 0.000297 1.028194
LUB 0.000653 1.013261 0.001147 1.010850
LINV 0.000281 1.014262 0.001081 1.006889
FMOLS Model 5 FMOLS Model 4
Coefficient variance Uncentered VIF Coefficient variance Uncentered VIF
LTO 0.000804 1.004857 0.003174 1.016380
2
LTO 0.001211 1.004244 0.000724 1.016380
LY 0.001427 1.018790
2
LY
LEC 0.000659 1.005372
LUB
LINV

a
FMOLS = fully modified ordinary least squares; TO = trade openness measured as the sum of exports
2
and imports as a percentage of gross domestic product; TO = trade openness squared; Y = income
2
measured by the per-capita gross domestic product; Y = income squared; EC = mean energy
consumption measured in kilograms of oil equivalent per capita; INV = industrialization measured
by the industrial value added as a share of gross domestic product; and UB = urbanization
measured as the total urban population.

Conclusions and Policy Implications

This paper is the result of four motivations. First, existing studies on the de-
terminants of carbon dioxide emissions have yet to yield conclusive results.
Second, there are limited studies on carbon dioxide emissions that account for the
effects of income, energy consumption, and trade openness for Sub-Saharan
Africa in a single regression despite the fact that there is the possibility of trade
openness positively affecting income, on the one hand, and energy consumption
affecting growth and, hence, income per person as well as a possible feedback
from income to energy consumption on the other hand. This suggests that omitting
one of these closely related variables from the emissions equation can result in
upwards bias of the estimated effect of income on emissions for the sub-region.
Furthermore, most existing research on the subject for the SSA area has employed
time-series techniques for individual countries with little evidence from panel
110 THE JOURNAL OF ENERGY AND DEVELOPMENT

data. The last motivation for the paper stemmed from the fact that studies ex-
amining the non-linearity in the relationship between income and emissions
abound but, to the best of the authors’ knowledge, no study exists that tests for
such a relationship between trade and emissions (since it can be argued that trade
affects emissions through the EKC hypothesis). Consequently, this work sought to
examine the effects of trade openness, income, and energy consumption on carbon
dioxide emissions for 19 Sub-Saharan African countries using data covering the
period 1977–2012.
The study employed the FMOLS model, which is found to be more robust in
dealing with the problem of endogeneity and serial correlations that are more or
less very prevalent in panel data. The DOLS estimation also was employed to
check for the robustness of the results. Unit root tests using the LLC, Breitung, and
IPS test statistics indicated all variables to be integrated of the order one, I(1), with
the exception of the Breitung test that showed industrialization as an I(0). How-
ever, since the other two tests showed industrialization as I(1) variable, all the
variables were considered to be integrated of the order one. Pedroni’s heterogeneous
panel cointegration test reveals a long-run relationship among trade openness, in-
come, energy consumption, carbon emissions, industrialization, and urbanization
variables. The FMOLS and DOLS estimations indicated that omitting either trade
openness or energy consumption in the estimation procedure did not affect the
magnitudes of the estimated coefficients. Again, the results confirmed that the EKC
hypothesis holds for the sub-region with an estimated turning point values between
U.S. $1,142.85 to U.S. $5,687.09. The study also confirms a nonlinear relationship
between trade and emissions for the sub-region. The results actually suggest an
inverted U-shape relationship between trade and emissions.
These findings suggest that economic expansion as predicted by the EKC
hypothesis will eventually reduce carbon dioxide emissions in Sub-Saharan
Africa. Therefore, as the economy in the sub-region grows and develops,
carbon emissions will be reduced. This implies the need to promote economic
growth and development as a means of reducing carbon emissions. Conse-
quently, any growth and development impediment in the sub-region needs to be
addressed. Additionally, it is inferred from the results that opening up the sub-
region for international trade, in the long run, will help to reduce emissions.
Moreover, an increase in the level of energy consumption is expected to in-
crease the level of emissions in the sub-region. The implication of this is that,
since energy has been recognized as a necessity for growth and development in
the sub-region by previous studies and that a reduction may be harmful to the
economy, policies like investment grants, which will ensure the availability of
cleaner energy for economic activities, will help reduce emissions. It also
implies it is imperative for countries in the region to embrace more energy
conservation policies—e.g., pricing, education, and training—in order to re-
duce emissions.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 111

In addition, putting measures in place to decentralize growth in the country can


help reduce pressures in the urban areas and, hence, reduce urbanization and
eventually carbon emissions. A possible way to achieve this is for governments in
the sub-region to promote rural development. Last, the results suggest the pro-
motion of low carbon emissions technologies in the industrial sector to help reduce
emissions. Generally, the results from the study conclude that both income and
non-income variables explain carbon emissions in SSA, although income and
energy consumption have a greater effect.
NOTES
1
O. Fernández-Amador, J. F. Francois, and P. Tomberger, “Carbon Dioxide Emissions and
International Trade in the Turn of the Millennium,” Ecological Economics, vol. 125, issue C
(2016), pp. 14–26.
2
United Nations, United Nations Fact Sheet on Climate Change: Africa is Particularly Vul-
nerable to the Expected Impacts of Global Warming (Nairobi, Kenya: United Nations Climate
Change Conference, 2006).
3
Organisation for Economic Co-operation and Development (OECD) and International Energy
Agency (IEA), CO2 Emissions from Fuel Combustion: 2012 Highlights (Paris: OECD, 2013).
4
Ibid.; S. Sharma, “Determinants of Carbon Dioxide Emissions: Empirical Evidence from 69
Countries,” Applied Energy, vol. 88, no. 1 (2013), pp. 376–82; United Nations Environment Pro-
gramme (UNEP), Green Economy and Trade – Trends, Challenges and Opportunities (New York:
UNEP, 2013); and World Trade Organization (WTO) and United Nations Environment Programme
(UNEP), Trade and Climate Change (New York: WTO and UNEP, 2009).
5
T. A. Boden, G. Marland, and R. J. Andres, “Global, Regional, and National Fossil-Fuel CO2
Emissions,” Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S.
Department of Energy, Oak Ridge, Tennessee, 2011.
6
Organisation for Economic Co-operation and Development (OECD) and International Energy
Agency (IEA), CO2 Emissions from Fuel Combustion: 2012 Highlights.
7
The World Bank, World Development Indicators (Washington, D.C.: World Bank, 2014).
8
African Development Bank Group, African Development Report — Towards Green Growth in
Africa (Tunis: African Development Bank Group, 2012).
9
B. Stambuli, “Price and Income Elasticities of Oil Demand in Tanzania: An Autoregressive
Approach,” Business Management Dynamics, vol. 3, no. 1 (2013), pp. 75–83, and Institute of
Economic Affairs (IEA), “Energy in Kenya Energy Scenarios,” paper delivered at the IEA
Workshop, Stanley Hotel, Nairobi, Kenya, July 24, 2013.
10
A. Chea, “Sub-Saharan Africa and Global Trade: What Sub-Saharan Africa Needs to Do to
Maximize the Benefits from Global Trade Integration, Increase Economic Growth and Reduce
Poverty?” International Journal of Academic Research in Business and Social Sciences, vol. 2, no.
4 (2012), pp. 358–84.
112 THE JOURNAL OF ENERGY AND DEVELOPMENT
11
D. Gómez, G. Álvarez-Ude, and I. Gálvez, “Trade-led Growth Hypothesis in Cuban Com-
mercial Regimes since 1960,” Economics of Transition, vol. 19, no. 2 (2012), pp. 287–304; P. K.
Narayan, S. Narayan, B. C. Prasad, and A. Prasad, “Export-led Growth Hypothesis: Evidence from
Papua New Guinea and Fiji,” Journal of Economic Studies, vol. 34, no. 4 (2007), pp. 341–51; P. A.
Kwakwa, “Disaggregated Energy Consumption and Economic Growth in Ghana,” International
Journal of Energy Economics and Policy, vol. 4, no. 1 (2012), pp. 34–40; S.-H. Yoo, “The Causal
Relationship between Electricity Consumption and Economic Growth in ASEAN Countries,”
Energy Policy, vol. 34, no. 18 (2006), pp. 3573–582; and S. Solarin and M. Shahbaz, “Trivariate
Causality between Economic Growth, Urbanization and Electricity Consumption in Angola: Coin-
tegration and Causality Analysis,” Energy Policy, vol. 60, issue C (2013), pp. 876–84.
12
M. Kohler, “CO2 Emissions, Energy Consumption, Income and Foreign Trade: A South Af-
rican Perspective,” ERSA working paper no. 356, Cape Town, South Africa, Economic Research
South Africa (ERSA), 2013.
13
G. E. Akpan and U. F. Akpan, “Electricity Consumption, Carbon Emissions and Economic
Growth in Nigeria,” International Journal of Energy Economics and Policy, vol. 2, no. 4 (2012),
pp. 292–306, and P. A. Kwakwa, F. S. Arku, and S. Aboagye, “Environmental Degradation Effect
of Agricultural and Industrial Growth in Ghana,” Journal of Rural and Industrial Development, vol.
2, no. 2 (2014), pp. 22–9.
14
B. F. Aka, “Effects of Trade and Growth on Air Pollution in the Aggregated Sub-Saharan
Africa,” International Journal of Applied Econometrics and Quantitative Studies, vol. 5, no. 1
(2008), pp. 5–14.
15
M. Kohler, op. cit.
16
B. H. Baltagi, Econometric Analysis of Panel Data (Chichester, West Sussex, United King-
dom: John Wiley & Sons Ltd., 2005).
17
N. Apergis and J. E. Payne, “Energy Consumption and Growth in South America: Evidence
from a Panel Error Correction Model,” Energy Economics, vol. 32, no. 6 (2010), pp. 1421–426.
18
B. F. Aka, op. cit.; G. E. Akpan and U. F. Akpan, op. cit.; N. Apergis and J. E. Payne, “CO2
Emissions, Energy Usage, and Output in Central America,” Energy Policy, vol. 37, no. 8 (2009),
pp. 3282–286; M. Arouri, M. Shahbaz, R. Onchang, F. Islam, and F. Teulon, “Environmental
Kuznets Curve in Thailand: Cointegration and Causality Analysis,” The Journal of Energy and
Development, vol. 39, no. 2 (2014), pp. 149–70; G. M. Grossman and A. B. Krueger, “Environ-
mental Impacts of a North American Free Trade Agreement,” NBER Working Paper no. 3914,
Cambridge, Massachusetts, National Bureau of Economic Research (NBER), 1991, and “Economic
Growth and the Environment,” Quarterly Journal of Economics, vol. 110, no. 2 (1995), pp. 353–77;
D. Holtz-Eakin and T. M. Selden, “Stoking the Fires? CO2 Emissions and Economic Growth,”
Journal of Public Economics, vol. 57, no. 1 (1995), pp. 85–101; H. Iwata, K. Okada, and
S. Samreth, “Empirical Study on the Determinants of CO2 Emissions: Evidence from OECD
Countries,” Applied Economics, vol. 44, no. 27 (2011), pp. 3513–519; K. Jayanthakumaran,
R. Verma, and Y. Liu, “CO2 Emissions, Energy Consumption, Trade and Income: A Comparative
Analysis of China and India,” Energy Policy, vol. 42, issue C (2012), pp. 450–60; M. Ben Jebli,
S. Ben Youssef, and I. Ozturk, “The Environmental Kuznets Curve: The Role of Renewable and
Non-Renewable Energy Consumption and Trade Openness,” MPRA paper no. 51672, Munich,
Germany, Munich Personal RePEc Archive (MPRA), 2013; and M. Kohler, op. cit.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 113
19
S. Kuznets, “Economic Growth and Income Inequality,” American Economic Review, vol. 49,
no. 1 (1955), pp. 1–28.
20
G. M. Grossman and A. B. Krueger, “Economic Growth and the Environment.”
21
Ibid.
22
T. Panayotou, “Demystifying the Environmental Kuznets Curve: Turning a Black Box into
a Policy Tool,” Environment and Development Economics, vol. 2, no. 4 (1997), pp. 465–84.
23
G. Mohapatra and A. K. Giri, “Economic Development and Environmental Quality: An
Econometric Study in India,” Management of Environmental Quality: An International Journal,
vol. 20, no. 2 (2009), pp. 175–91.
24
R. A. Bouvier, “Air Pollution and Per Capita Income: A Disaggregation of the Effects of
Scale, Sectoral Composition, and Technological Change,” PERI working paper series no. 84,
Amherst, Massachusetts, Political Economy Research Institute (PERI), University of Massachu-
setts, Amherst, 2004.
25
G. M. Grossman and A. B. Krueger, “Economic Growth and the Environment.”
26
D. Holtz-Eakin and T. M. Selden, op. cit.
27
E. Neumayer, “National Carbon Dioxide Emissions,” Geography Matters, vol. 36, no. 1
(2004), pp. 33–40.
28
M. Cole, “Trade, the Pollution Haven Hypothesis and the Environmental Kuznets Curve:
Examining the Linkages,” Ecological Economics, vol. 48, no. 1 (2004), pp. 71–81.
29
B. F. Aka, op. cit.
30
M. Cole, op. cit.
31
G. Chichilnisky, “North-South Trade and the Global Environment,” The American Economic
Review, vol. 84, no. 4 (1994), pp. 851–74.
32
M. Aklin, “Trade, Development & The Environment: The Diffusion of Pollution,” SSRN
Working Paper Series, Social Science Research Network (SSRN), 2011.
33
P. Erickson, H. van Asselt, E. Kemp-Benedict, and M. Lazarus, “International Trade and
Global Greenhouse Gas Emissions: Could Shifting the Location of Production Bring GHG Bene-
fits?” Project Report 2013-02, Stockholm Environment Institute, Stockholm, Sweden, 2013.
34
C. Bozkurt and Y. Akan, “Economic Growth, CO2 Emissions and Energy Consumption: The
Turkish Case,” International Journal of Energy Economics and Policy, vol. 4, no. 3 (2014), pp.
484–94.
35
Organisation for Economic Co-operation and Development (OECD) and International Energy
Agency (IEA), CO2 Emissions from Fuel Combustion: 2012 Highlights; United Nations Environ-
ment Programme (UNEP), Green Economy and Trade – Trends, Challenges and Opportunities; and
World Trade Organization (WTO) and United Nations Environment Programme (UNEP), Trade
and Climate Change.
114 THE JOURNAL OF ENERGY AND DEVELOPMENT
36
International Energy Agency (IEA), Energy Technology Perspectives 2012: Pathways to
a Clean Energy System (Paris: IEA, 2012).
37
G. M. Grossman and A. B. Krueger, “Environmental Impacts of a North American Free Trade
Agreement.”
38
D. Holtz-Eakin and T. M. Selden, op. cit.
39
T. M. Selden and D. Song, “Environmental Quality and Development: Is There a Kuznets
Curve for Air Pollution Emissions?” Journal of Environmental Economics and Management, vol.
27, no. 2 (1994), pp. 147–62.
40
G. M. Grossman and A. B. Krueger, “Economic Growth and the Environment.”
41
B. F. Aka, op. cit.
42
M. Arouri et al., op. cit.
43
H. Iwata et al., op. cit.
44
F. Halicioglu, “An Econometric Study of CO2 Emissions, Energy Consumption, Income and
Foreign Trade in Turkey,” MPRA paper no. 11457, Munich, Germany, Munich Personal RePEc
Archive (MPRA), 2008.
45
G. E. Akpan and U. F. Akpan, op. cit.
46
N. Apergis and J. E. Payne, “CO2 Emissions, Energy Usage, and Output in Central America.”
47
N. Apergis and J. E. Payne, “The Emissions, Energy Consumption, and Growth Nexus: Evidence
from the Commonwealth of Independent States,” Energy Policy, vol. 38, no. 1 (2010), pp. 650–55.
48
K. Jayanthakumaran et al., op. cit.
49
J. Baek and H. S. Kim, “Is Economic Growth Good or Bad for the Environment? Empirical
Evidence from Korea,” Energy Economics, vol. 36, no. 6 (2013), pp. 744–49.
50
M. Ben Jebli et al., op. cit.
51
H. E. Chebbi, M. Olarreaga, and H. Zitouna, “Trade Openness and CO2 Emissions in Tunisia,”
in proceedings of the 16th Economic Research Forum Annual Conference, Cairo, Egypt, November
7–9, 2009.
52
S. Farhani, M. Shahbaz, and M. Arouri, “Panel Analysis of CO2 Emissions, GDP, Energy
Consumption, Trade Openness and Urbanization for MENA Countries,” MPRA paper no. 49258,
Munich, Germany, Munich Personal RePEc Archive (MPRA), 2013.
53
Usama Al-mulali, “Factors Affecting CO2 Emission in the Middle East: A Panel Data
Analysis,” Energy, vol. 44, no. 1 (2012), pp. 564–69.
54
S. B. Amin, S. S. Ferdaus, and A. K. Porna, “Causal Relationship among Energy Use, CO2
Emissions and Economic Growth in Bangladesh: An Empirical Study,” World Journal of Social
Sciences, vol. 2, no. 4 (2012), pp. 273–90.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 115
55
Z. Gu, Y Gao, and C. Li, “An Empirical Research on Trade Liberalization and CO2 Emissions
in China,” in Proceedings of the International Conference on Education Technology and In-
formation System (ICETIS), Hainan Sanya, China, June 21–22, 2013, pp. 243–46.
56
K.-M. Lim, S.-Y. Lim, and S.-H. Yoo, “Oil Consumption, CO2 Emission, and Economic
Growth: Evidence from the Philippines,” Sustainability, vol. 6, no. 2 (2014), pp. 967–79.
57
A. F. M. Ataur Rahman and A. K. Porna, “Growth Environment Relationship: Evidence from
Data on South Asia,” Journal of Accounting, Finance and Economics, vol. 4, no. 1 (2014), pp. 86–96.
58
S. Sharma, op. cit.
59
J. B. Ang, “CO2 Emissions, Energy Consumption, and Output in France,” Energy Policy, vol.
35, no. 10 (2007), pp. 4772–778.
60
I. Ozturk and G. S. Uddin, “Causality among Carbon Emissions, Energy Consumption and
Growth in India,” Economic Research—Ekonomska Istrazivanja, vol. 25, no. 3 (2012), pp.
752–75.
61
G. Liu, “A Causality Analysis on GDP and Air Emissions in Norway,” Statistics Norway
discussion paper no. 447, Oslo, Norway, Statistics Norway, Research Department, 2006.
62
P. K. Adom, W. Bekoe, F. Amuakwa-Mensah, J. T. Mensah, and E. Botchway, “Carbon
Dioxide Emissions, Economic Growth, Industrial Structure, and Technical Efficiency: Empirical
Evidence from Ghana, Senegal, and Morocco on the Causal Dynamics,” Energy, vol. 47, no. 1
(2012), pp. 314–25.
63
S. Sharma, op. cit., and Z. Wang, C. Shi, Q. Li, and G. Wang, “Impact of Heavy In-
dustrialization on the Carbon Emissions: An Empirical Study of China,” Energy Procedia, vol. 5
(2011), pp. 2610–616.
64
A. Levin, C.-F. Lin, and C.-S. J. Chu, “Unit Root Tests in Panel Data: Asymptotic and Finite-
Sample Properties,” Journal of Econometrics, vol. 108, no. 1 (2002), pp. 1–24.
65
Ibid.
66
K. Im, M. Pesaran, and Y. Shin, “Testing for Unit Roots in Heterogeneous Panels,” Journal of
Econometrics, vol. 115, no. 1 (2003), pp. 53–74.
67
J. Breitung, “The Local Power of Some Unit Root Tests for Panel Data,” in Advances in
Econometrics, Volume 15: Nonstationary Panels, Panel Cointegration, and Dynamic Panels, ed. B.
H. Baltagi (Amsterdam: JAY Press, 2000), pp. 161–78.
68
O. I. Kareem, “A Dynamic Panel Analysis of the Effects of International Tourism Exports on
African Economic Growth,” presentation at the 14th African Econometrics Society (AES)
Conference, Abuja, Nigeria, July 2009, and P. M. G. Martins, “Aid Absorption and Spending in
Africa: A Panel Cointegration Approach,” Journal of Development Studies, vol. 47, no. 12 (2011),
pp. 1925–953.
69
P. Pedroni, “Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple
Regressors,” Oxford Bulletin of Economics and Statistics, vol. 61, no. 0 (1999), pp. 653–70, and
“Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time Series Tests with
116 THE JOURNAL OF ENERGY AND DEVELOPMENT

an Application to the PPP Hypothesis: New Results,” Econometric Theory, vol. 20, no. 3 (2004),
pp. 597–625.
70
R. F. Engle and C. W. J. Granger, “Co-Integration and Error Correction: Representation,
Estimation, and Testing,” Econometrica, vol. 55, no. 2 (1987), pp. 251–76.
71
O. I. Kareem, op. cit.
72
P. Pedroni, “Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple
Regressors,” and “Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time
Series Tests with an Application to the PPP Hypothesis: New Results.”
73
Ibid.
74
B. Hansen and P. C. B. Phillips, “Statistical Inference in Instrumental Variables Regression
with I(1) Processes,” Review of Economic Studies, vol. 57, no. 1 (1990), pp. 99–125.
75
P. Pedroni, “Fully Modified OLS for the Heterogeneous Cointegrated Panels,” Advances in
Econometrics, vol. 15 (2000), pp. 93–130.
76
J. H. Stock and M. W. Watson, “A Simple Estimator of Cointegrating Vectors in Higher Order
Integrated Systems,” Econometrica, vol. 61, no. 4 (1993), pp. 783–820.
77
N. A. Klevmarken, “Panel Studies: What Can We Learn from Them? Introduction,” European
Economic Review, vol. 33, nos. 2–3 (1989), pp. 523–29.
78
The World Bank, op. cit.
79
P. Pedroni, “Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple
Regressors,” and “Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time
Series Tests with an Application to the PPP Hypothesis: New Results.”
80
J. Gao and L. Zhang, “Electricity Consumption–Economic Growth–CO2 Emissions Nexus in
Sub-Saharan Africa: Evidence from Panel Cointegration,” African Development Review, vol. 26,
no. 2 (2014), pp. 359–71.
81
P. Pedroni, “Fully Modified OLS for the Heterogeneous Cointegrated Panels.”
82
J. H. Stock and M. W. Watson, op. cit.
83
International Energy Agency (IEA), World Energy Outlook 2014 Factsheet: Energy in Sub-
Saharan Africa Today (Paris: IEA, 2014).
84
J. Gao and L. Zhang, op. cit.; N. Apergis and J. E. Payne, “Energy Consumption and Growth
in South America: Evidence from a Panel Error Correction Model;” S. Hossain, “An Econo-
metric Analysis for CO2 Emissions, Energy Consumption, Economic Growth, Foreign Trade
and Urbanization of Japan,” Low Carbon Economy, vol. 3, no. 3A (2012), pp. 92–105; and S.
Farhani et al., op. cit.
85
S. Farhani et al., op. cit., and J. Gao and L. Zhang, op. cit.
86
B. F. Aka, op. cit.
CARBON EMISSIONS IN SUB-SAHARAN AFRICA 117
87
G. M. Grossman and A. B. Krueger, “Economic Growth and the Environment;” D. Holtz-
Eakin and T. M. Selden, op. cit.; and E. Neumayer, op. cit.
88
S. O. Aghalino and B. Eyinla, “Oil Exploitation and Marine Pollution: Evidence from the
Niger Delta, Nigeria,” Journal of Human Ecology, vol. 28, no. 3 (2009), pp. 177–82, and A.
Shinsato, “Increasing the Accountability of Transactional Corporations for Environmental Harms:
The Petroleum Industry in Nigeria,” Northwestern Journal of International Human Rights, vol. 4,
no. 1 (2005), pp. 185–209.
89
S. Sharma, op. cit.