


































DEM_2019_57to85


© 2019 Nicolaus Copernicus University. All rights reserved.  
http://www.dem.umk.pl/dem 

D Y N A M I C  E C O N O M E T R I C  M O D E L S  
DOI: http://dx.doi.org/10.12775/DEM.2019.004  Vol. 19 (2019) 57−84 

Submitted October 24, 2019  ISSN (online) 2450-7067 
Accepted December 22, 2019 ISSN (print) 1234-3862 

Wondatir Atinafu *  

Energy Consumption and Economic Growth in Ethio-
pia: Evidence from ARDL Bound Test Approach 

A b s t r a c t. The present study aims to investigate the dynamic relationship between economic 
growth and energy consumption. Specifically, the study tries to answer the questions whether 
energy consumption has any significance effect on economic growth of the country and it also 
determined the magnitude of the effect. In doing this, the study used an ARDL bound test ap-
proach to analyze Ethiopian data from 1970 to 2017 with real GDP as a function of energy 
consumption, human capital., physical capital., trade openness and policy change dummy. To 
do so, secondary data were obtained from WDI, UNCTAD stat and NBE. Co-integration test 
approves the existence of long-run relationship among the variables. Moreover, the estimation 
result reveals that, energy consumption found statistically insignificant in affecting economic 
growth in the long-run. However, it was positive and statistically significant in short-run. Like-
wise, the dummy variable incorporated to capture the policy change found insignificant in long-
run and with positive significant result in short-run. Also, we applied the Granger causality test 
in linear multivariate models to evaluate how important is the causal impact of energy con-
sumption on economic growth. The results give the evidence of causality running from eco-
nomic growth to energy consumption supporting “conservation hypothesis”, implying that re-
ducing energy consumption may be implemented with little or no adverse effect on economic 
growth. Hence, this study recommended the policy makers to improve the existing policies on 
energy consumption so as to enhance the level of efficiency in the energy sector i.e. energy 
regulation policies supporting the shift from lower-quality to higher-quality energy services.  
K e y w o r d s: economic-growth, energy-consumption, ARDL, Ethiopia, causality. 

 
* Correspondence to: Wondatir Atinafu, Department of Economics, Jimma University,  

e-mail: wondatiratinafu@gmail.com. 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

58 

Introduction 
 Arguably, energy plays a vital role in economic and social development. 
The role of energy in economic growth has long been a controversial topic in 
economics literature. As a result, the ongoing debate among energy econo-
mists about the relationship between energy use and output growth led to the 
emergence of two opposite views. One point of view suggests that energy is 
the prime source of value because other factors of production such as labor 
and capital cannot do without energy. According to this argument, energy use 
is expected to be a limiting factor to economic growth. The other point of view 
suggests that energy is neutral to growth. This is what became to be known in 
the literature as the ‘neutrality hypothesis’. The main reason for the neutral 
impact of energy on growth is that the cost of energy is very small as a pro-
portion of GDP and, thus, it is not likely to have a significant impact on output 
growth. It has also been argued that the possible impact of energy use on 
growth will depend on the structure of the economy and the stage of economic 
growth of the country concerned (Ghali and El-Sakka, 2009). 
 Theoretical disagreement on the role of energy is matched by mixed em-
pirical evidence. That is, whether economic growth leads to energy consump-
tion or that energy consumption is the engine of economic growth. The direc-
tion of causality has significant policy implications. Empirically it has been 
tried to find the direction of causality between energy consumption and eco-
nomic activities for the developing as well as for the developed countries em-
ploying the Granger or Sims techniques.  
 Like other developing countries Ethiopia is energy using growing econ-
omy, with energy production of 14.1 and 30.9 total million metric tons of oil 
equivalent in 1990 and 2016 respectively. The biomass energy use is predom-
inant which accounts 93.9% and 90.2% for the year 1990 and 2016 respec-
tively and the balance goes to the modern energy. This shows that there is 
a gradual shift from traditional to modern energy sources (WDI, 2017). More-
over, it is believed that the modern energy penetration rate has increased as of 
2017 because of the commissioning of the three hydro power plants in the 
country. Ethiopia is non-oil producing countries and its fossil oil energy needs 
are met by large quantities of imports. 
 The preceding facts show the power sector in Ethiopia is underdeveloped 
and hence energy consumption is very low. As a result, Ethiopia is far from 
having satisfied the current energy demand of its people. Cognizant of this 
problem and in line with the millennium development goals, Ethiopia is trying 
to provide energy to its citizens by investing in major modern energy infra-
structures in the country. This show that Ethiopia has recognized that 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

59 

accessibility to affordable energy services is a prerequisite to poverty allevia-
tion, and necessary condition for sustainable economic growth. This policy 
goal implies that increased energy consumption can help achieve social de-
velopment and enhance economic growth. 
 Thus, to meet its growing needs of energy, Ethiopia faces both energy 
constraints from the supply side and demand management policies (EEA, 
2009). The current concerns about global warming also poses a question about 
how can economic growths in Ethiopia, will be reconciled with stabilization 
in the use of both traditional and fossil fuels. However, for any such policy 
making it is essential to determine the causal relationship between energy con-
sumption and general economic activities. 
 It is important, therefore, to ascertain empirically whether there is a causal 
link between energy consumption and economic growth in Ethiopia. This is 
particularly true given the current debate about global warming and the need 
to reduce Greenhouse Gas Emissions by conserving energy consumption, 
since any constraints put on energy consumption to help reduce emissions will 
have an effect on growth and development if causality from energy to GDP 
exists. 
 Moreover, Ethiopia has huge potential of modern energy resources; how-
ever, availability of modern energy per se is not enough for the economic and 
social problems facing the country.  
 The power investment that is currently taking place in Ethiopia is part of 
the process of the recognition that the quality and quantity of modern power 
supply can play a pivotal role in the country’s social and economic develop-
ment. This investment process is implicitly based on the assumption that in-
vestment in modern energy and the drive towards making the modern energy 
sector more efficient can promote economic growth. 
 Although energy use is a reflection of climatic, geographic and economic 
factors (such as the relative prices of energy), it is closely related to the growth 
in the modern sectors (industry, motorized transport and urban areas). “There 
is a strong connection between the energy sector and a national economy. On 
the one hand, energy demand, supply and pricing have significant impact on 
socio-economic development and the overall quality of life of the population. 
On the other hand, the nature of economic structure and the change in that 
structure, the prevailing macro-economic conditions are key factors of energy 
demand and supply” (EEA, 2009). 
 The data compiled by Energy Information Administration for the periods 
1980 to 2014 shows that GDP per capita have strong correlation coefficient of 
0.6 with energy consumption (EEA, 2016). Although the existence of corre-
lation between the two implies the existence of causality, on the other hand it 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

60 

is source of doubt, on the part of many growth theorists. The fact that eco-
nomic growth tends to be very closely correlated with energy consumption, 
does not a priori mean that energy consumption is the cause of the growth. 
Indeed, most economic models assume the opposite: that economic growth is 
responsible for increasing energy consumption. It is also conceivable that both 
consumption and growth are simultaneously caused by some third factor. 
 With this background, (that is growing of different school of thought with 
regard to resource consumption in general and energy consumption in partic-
ular) there are numerous researches which have tried to figure out the casual 
relationship between energy use growth and economic growth.  
 The answers to questions pose in the hypothesis, which are recognized in 
many previous studies, have important implications for policy makers. As 
noted by Wolde-Rufael (2005), amongst others, if causality runs from energy 
consumption to GDP then it implies that an economy is energy dependent and 
hence energy is a stimulus to growth implying that a shortage of energy may 
negatively affect economic growth or may cause poor economic performance, 
leading to a fall in income and employment. In other words, energy is a limit-
ing factor in economic growth (Stern, 2010). Whereas if causality only runs 
from GDP to energy consumption this implies that an economy is not energy 
dependent hence, as noted by Masih and Masih (1997) amongst others, energy 
conservation policies may be implemented with no adverse effect on growth 
and employment. If, on the other hand, there is no causality in either direction 
(referred to as the ‘neutrality hypotheses), it implies that energy consumption 
is not correlated with GDP, so that energy conservation policies may be pur-
sued without adversely affecting the economy. 
 The non-existence of such research work in the country, at least to the 
knowledge of the research worker, shows there is a gap to be filled, so that 
energy policy lesson can be drawn. And the inconclusive empirical results 
which make it difficult to draw a conclusion about Ethiopia and the important 
role energy plays in economic development in country, the purpose of this 
paper is therefore, to fill this gap by attempting to undertake the energy eco-
nomic growth nexus employing multivariate model consisting of GDP, phys-
ical capital., human capital & energy consumption growth. 
 Moreover, previous studies are that most of them are used Johansen co-
integration method of vector autoregressive method as their method of analy-
sis. Even though the Johansen’s co-integration technique is one of the widely 
used methods of time series analysis, its outcome could not be reliable for 
small sample size; that is observations less than 80 years for the time series 
data (Narayan, 2005; Udoh et al., 2012). Relatively, the Autoregressive dis-
tributed lag (ARDL) method has some advantage over the Johansen’s method 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

61 

(Pesaran et al., 1999). These advantages are it can be applied irrespective of 
whether the regressors are I(1) and I(0). It can also provide valid and statis-
tically significant result or avoids the problem of biasness in small sample 
sizes (Pesaran et al., 1999; Narayan, 2005; Chaudhry et al., 2006; Udoh et al., 
2012). This ARDL procedure can provide unbiased and valid estimates of 
the long run model even when some of the regressors are endogenous (Harris 
et al., 2003, Pesaran et al., 1999; Ang, 2007). Furthermore, in using this Ap-
proach, a dummy variable can be included in the co-integration test process, 
which is not permitted in Johansen’s method (Rahimi et al., 2011). Hence in 
this paper, ARDL model is adopted so as to provide valid empirical evidence 
on the main target of this study which is assessing the nexus between energy 
consumption and economic growth in Ethiopia. 
 The overall objective of this paper is to empirically investigate the nexus 
between energy consumption and economic growth in Ethiopia. More specif-
ically: to empirically examine the effect of energy consumption on the aggre-
gate economic growth of Ethiopia in both short run and long run and to inves-
tigate the possible causal relationship between economic growth and energy 
consumption in Ethiopia, the ARDL bound test approach and Granger's cau-
sality test were used.  
 The reminder of the paper organized as follows. Section 2 includes review 
literature, section 3 presents methodology applied in the study, Section 4 re-
ports the findings and discussion of our analysis and conclusion follows in 
section 5. 

1. Review of Related Literature 

1.1. Theoretical Literature 
 The laws of thermodynamics and the conservation of matter describe the 
immutable constraints within which the economic system must operate. The 
mass-balance principle means that, in order to obtain a given material output, 
greater or equal quantities of matter must be used as inputs with the residual 
a pollutant or waste product. Therefore, there are minimal material input re-
quirements for any production process producing material outputs. The sec-
ond law of thermodynamics (the efficiency law) implies that a minimum quan-
tity of energy is required to carry out the transformation or movement of mat-
ter or, more generally, perform physical work. Carrying out transformations 
in finite time requires more energy than these minima. All production involves 
work. Therefore, all economic activities must require energy, and there must 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

62 

be limits to the substitution of other factors of production for energy so that 
energy is always an essential factor of production. 
 Primary factors of production are defined as inputs that exist at the begin-
ning of the period under consideration and are not directly used up in produc-
tion (though they can be degraded or accumulated from period to period), 
while intermediate inputs are those created during the production period under 
consideration and are used up entirely in production. Mainstream economists 
usually think of capital., labor and land as the primary factors of production, 
and goods (such as fuels and materials) as intermediate inputs. The prices paid 
for the various intermediate inputs are seen as eventually being payments to 
the owners of the primary inputs for the services provided directly or embod-
ied in the produced intermediate inputs. This approach has led to a focus in 
mainstream growth theory on the primary inputs, and in particular, capital and 
labor. The classical factor of land, including all-natural resource inputs, grad-
ually diminished in importance in economic theory as its value share of GDP 
fell steadily and is usually subsumed as a subcategory of capital. 

Growth Models with Resources and no Technical Change 
 Adding non-renewable natural resources that are essential in production 
to the basic mainstream growth models means that capital also needs to be 
accumulated to compensate for resource depletion. When there is more than 
one input – both capital and natural resources – there are many alternative 
paths that economic growth can take, determined by both the nature of tech-
nology and institutional arrangements. Solow showed that sustainability is 
achievable in a model with a non-renewable natural resource with no extrac-
tion costs and non-depreciating capital when the elasticity of substitution be-
tween the two inputs is unity, and when certain other technical conditions are 
met. Sustainability, and even indefinite growth in consumption, can occur 
when the utility of individuals is given equal weight without regard to when 
they happen to live. However, under competition the same model economy 
results in exhaustion of the resource and consumption and social welfare even-
tually fall to zero. With any constant discount rate, the efficient growth path 
also leads to eventual depletion of the natural resource and the collapse of the 
economy. The Hartwick rule shows that if sustainability is technically feasi-
ble, a constant level of consumption can be achieved by reinvesting the re-
source rents in other forms of capital., which in turn can substitute for re-
sources. 
 A common interpretation of this body of work is that substitution and tech-
nical change can effectively decouple economic growth from the use of energy 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

63 

and other resources. Depleted resources can be replaced by more abundant 
substitutes, or by ‘equivalent’ forms of human-made capital (people, ma-
chines, factories, etc.). 

Growth Models with Resources and Technical Change 
 In addition to substitution of capital for resources, technological change 
might permit growth or at least constant consumption in the face of a finite 
resource base. When the elasticity of substitution between capital and re-
sources is unity, exogenous technical progress will allow consumption to 
grow over time if the rate of technological change divided by the discount rate 
is greater than the output elasticity of resources. Technological change might 
enable sustainability even with an elasticity of substitution of less than one. 
Once again, technical feasibility does not guarantee sustainability. Depending 
on preferences for current versus future consumption, technological change 
might instead result in faster depletion of the resource. Therefore, mainstream 
economic growth theory assumes that resource consumption is a consequence, 
not a cause, of growth. 

Synthesis: Unified Models of Energy and Growth 
 The mainstream growth models ignore energy in the economic growth, by 
contrast, the ecological economics literature posits a central role for energy in 
driving growth but argues that limits to substitutability and/or technological 
change might limit or reverse growth in the future. But none of the models 
and theories reviewed so far really provides a satisfactory explanation of the 
long-run history of the economy. Until the industrial revolution, output per 
capita was generally low and economic growth was not sustained. Ecological 
economists point to the invention of methods to use fossil fuels as the cause 
of the industrial revolution. But the mainstream growth models that ignore 
energy resources can at least partly explain economic growth over the last half 
a century. 
There are currently two principal mainstream theories that explain the growth 
regimes of both the preindustrial and modern economies and the cause of the 
industrial revolution, which formed the transition between them. These are 
endogenous technical change approach, and the second approach is repre-
sented by two sectors – Malthusian Sector and Solow Sector. 
 To integrate the different approaches, Stern (2011) proposed to modify 
Solow’s growth model. In the model Stern added an energy input that has low 
substitutability with capital and labor, while allowing the elasticity of substi-
tution between capital and labor to remain at unity. In this model, depending 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

64 

on the availability of energy and the nature of technological change, energy 
can be either a constraint on growth or an enabler of growth. Omitting time 
indexes for simplicity, the model consists of two equations: 

Y = [(1-g ) (ALbBb K1–b + g(AEE)F )]F  (1) 
D K  =  s(Y – PEE ) - dK (2) 

Equation (1) embeds a Cobb–Douglas production function of capital (K) and 
labor (L) in a constant elasticity of substitution (CES) function of value added 
and energy (E) that produces gross output Y. 

f = (d - 1)/ d;  
Where d is the elasticity of substitution between energy and the value-added 
aggregate; PE the price of energy; and g is a parameter reflecting the relative 
importance of energy and value added. AL and AE are the augmentation in-
dexes of labor and energy, which can be interpreted as reflecting both changes 
in technology that augment the effective supply of the factor in question and 
changes in the quality of the respective factors. 
Equation (2) is the equation of motion for capital that assumes like Solow that 
the proportion of gross output that is saved is fixed at s and that capital depre-
ciates at a constant rate d. As d -> 1and g -> 0 we have the Solow model as 
a special case, where in the steady state, K and Y grows at the rate of labor 
augmentation. Additionally, depending on the scarcity of energy, the model 
displays either Solow-style or energy constrained behaviour. 

1.2. Empirical Literature Review 
Over the past few years, the relationship between energy consumption and 
economic growth has been extensively researched. Yet, there seems to be no 
consensus regarding the direction of causality between energy consumption 
and economic growth. 
 In a study of over more than hundred countries, Chontanawat et al. (2008) 
find that the causal relationship between energy consumption and economic 
growth is more pronounced in developed than in developing countries. Cau-
sality running from energy consumption to economic growth. Ethiopia was 
included in the study and the result shows there is Granger causality running 
from economic growth to energy consumption. 
Stern (1993) examined the causal relationship between energy use and GDP 
in the USA. He employed a multivariate vector autoregressive (VAR) analysis 
and used a weighting index of energy quality, where content of energy use 
shifts from lower quality energy such as coal to high quality energy such as 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

65 

electricity, rather than using a measure of total energy use. Also, found that 
total energy use does not Granger cause GDP. 
 Masih and Masih (1996) used cointegration analysis to study this relation-
ship in a group of six Asian countries and found cointegration between energy 
use and GDP in India, Pakistan, and Indonesia. No cointegration is found in 
the case of Malaysia, Singapore and the Philippines. The flow of causality is 
found to be running from energy to GDP in India and from GDP to energy in 
Pakistan and Indonesia. 
 Nondo and Mulugeta (2009) applied panel data techniques to investigate 
the long-run relationship between energy consumption and GDP for a panel 
of 19 African countries (COMESA) based on annual data for the period 1980–
–2005. They have estimated the long-run relationship and test for causality 
using panel-based error correction models. The results indicate that long-run 
and short-run causality is unidirectional., running from energy consumption 
to GDP. The paper did not elaborate county specific result, it simply indicated 
the result in its aggregate form, and the study did not include Ethiopia. 
 Using a bivariate analysis Ebohon (1996), examines the causal directions 
between energy consumption and economic growth for Nigeria and Tanzania. 
The results show a simultaneous causal relationship between energy and eco-
nomic growth for both countries. In a bivariate relationship between energy 
consumption and economic growth in African countries, Wolde-Rufael 
(2005) also found conflicting evidence with the neutrality hypothesis sup-
ported in a substantial number of countries, with little support for the hypoth-
esis that energy consumption causes economic growth. 
 Bi-directional causality was detected for two countries, Gabon and Zam-
bia. For the remaining nine countries where there was no causality in any di-
rection between economic growth and energy consumption, energy consump-
tion seems neither to promote nor to retard economic growth.  
 The most striking result of the empirical evidence is that the introduction 
of both gross capital formation and labor has altered the direction of causality 
in thirteen countries that were previously investigated by Wolde-Rufael 
(2005). In seven of the countries where Wolde-Rufael (2005) found no evi-
dence of causality in any direction between energy consumption and eco-
nomic growth, he now found evidence of Granger causality for seven of these 
countries, Benin, Senegal., South Africa, Sudan, Togo, Tunisia and Zimba-
bwe. In Benin and South Africa causality runs now from energy consumption 
to economic growth; in Senegal., Sudan and Tunisia causality runs now from 
economic growth to energy consumption, and in Togo and Zimbabwe we find 
now that energy and economic growth were mutually causal. 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

66 

 Causality was also reversed in another six counties: Algeria, Cameroon, 
Gambia, Ghana, Morocco and Nigeria. In Algeria causality was reversed from 
economic growth to energy consumption, to the opposite causality running 
from energy consumption to economic growth contrary to the no causality 
found by Chontanawat et al. (2008).  
 Amirat and Bouri (2010) undertook analyses of the causal relationship be-
tween the per capita energy consumption and the per capita GDP in Algeria 
by using annual data from 1980 to 2007. They include capital and labor as 
additional variables to the energy growth nexus. They used Granger causality 
test and the variance decomposition analysis. The results give the evidence of 
causality running from energy consumption to economic growth.  
 Similarly, using a multivariate causality test, Akinlo (2008) found also 
conflicting results for eleven African countries. The result shows that energy 
consumption is co-integrated with economic growth in seven of the countries. 
In addition, in few of the countries, the result suggests that energy consump-
tion has a significant long run impact on economic growth.  
 Olatunji Adeniran (undated) tested for causal relationship between energy 
consumption and GDP in Nigeria using systematic econometric techniques. 
The study found that there is a unidirectional causality that runs from GDP to 
electricity consumption.  
 Jumbe (2004) examined cointegration and causality between electricity 
consumption (kWh) and, respectively, overall GDP (GDP), agricultural GDP 
(AGDP) and non-agricultural GDP (NGDP) using Malawi data for 1970– 
–1999 periods. The results show that kWh is, respectively, cointegrated with 
GDP and NGDP, but not with AGDP. The granger causality results show a bi-
directional causality between kWh and GDP, but a unidirectional causality 
running from NGDP to kWh.  
 Yohannes (2010) has conducted causal relationship between economic 
growth and energy consumption in Ethiopia and he found energy consumption 
Granger cause economic growth. 

2. Methodology of the Study  

Autoregressive Distributive Lag (ARDL) Approach to Co-integration 
So as to capture the nexus between energy consumption and economic growth, 
time series secondary data was employed. Data for all variables was taken 
from only two sources so as to keep its consistency and avoid possible biases 
due to difference in measurement techniques. The data sources for this study 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

67 

were World Bank (WB) and UNCTAD. The study considers annual data of 
Ethiopia for the years from 1970 to 2017. 
 Most of the time series studies in this area previously conducted are used 
the Engle-Granger approach following Engle and Granger (1987) and the Jo-
hansen’s co-integration technique following Johansen (1988) and Johansen 
and Juselius (1990). But its outcome could not be reliable for small sample 
size (Narayan, 2005; Udoh et al., 2012). Relatively, the Autoregressive dis-
tributed lag method of co-integration (ARDL) has more advantage over the 
Johansen’s method (Pesaran et al., 1999). Johansen’s co-integration technique 
requires that all the variables in the system have equal order of integration, 
that is the application of the Johansen technique will fail when the underlying 
regressors have different order of integration, especially when some of the 
variables are I(0) (Pesaran et al., 2001). That means the trace and maximum 
eigen value tests may lead to erroneous co-integrating relations with other 
variables in the model when I(0) variables are present in the system (Harris, 
1999). 
 Fortunately, to overcome this problem a new Autoregressive Distributed 
Lag (ARDL) model is developed by Pasaran, Shin and Smith (2001) which 
have more advantages than the Johansen co-integration approach. First, the 
ARDL approach can be applied irrespective of whether the regressors are I(1) 
and I(0) or have a mix of these integration orders. The only exception is that 
none of the variables in the model is integrated of order 2 or higher. Second, 
while the Johansen co-integration techniques require large data samples for 
validity, the ARDL procedure provides statistically significant result in small 
samples (Pesaran et al., 1999; Narayan, 2005; Udoh et al., 2012). That means, 
it avoids the problem of biasness that arise from small sample size (Chaudhry 
et al., 2006). Third, the ARDL procedure provides unbiased and valid esti-
mates of the long run model even when some of the regressors are endogenous 
(Harris et al., 2003; Pesaran et al., 1999; Ang, 2007).  
 Moreover, the ARDL procedure employs only a single reduced form equa-
tion, while the other co-integration procedures estimate the long-run relation-
ships within a context of system equations. Further, in using the ARDL ap-
proach, a dummy variable can be included in the co-integration test process, 
which is not permitted in Johansen’s method (Rahimi et al., 2011). Therefore, 
in order to achieve the targeted objectives of the study, the model of economic 
growth equation is estimated using ARDL model of econometric technique. 
 The above advantages of the ARDL technique over other standard co-in-
tegration techniques justify the application of ARDL approach in the present 
study to investigate the link between economic growth and energy consump-
tion.  



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

68 

The Empirical Model in ARDL Framework 
 According to Pesaran and Pesaran (1997), the ARDL approach requires 
the following two steps. In the first step, the existence of any long-term rela-
tionship among the variables of interest is determined using an F-test. The 
second step of the analysis is to estimate the coefficients of the long-run rela-
tionship and determine their values, followed by the estimation of the short-
run elasticity of the variables with the error correction representation of the 
ARDL model. By applying the ECM version of ARDL, the speed of adjust-
ment to equilibrium will be determined. According to Pesaran and Pesaran 
(1997), the ARDL model is represented by the following equation: 
 After checking for the order of integration of all variables in the model, 
the Autoregressive Distributed Lag (ARDL) model involves two steps for es-
timating the long-run relationship (Pesaran et al., 2001). In the first step the 
existence of long-run relationship among all variables in an equation should 
be examined and then in the second step the long-run and short-run coeffi-
cients of the variables can be estimated in the model. One can run the second 
step only if we find along run co-integration relationship among the variables 
in the first step. 
 In order to examine the long-run relationship and dynamic interaction be-
tween economic growth and energy consumption, this study employs an 
ARDL model. In general., there are three steps in estimating the model. The 
first step is to estimate the long-run relationship among the variables. This is 
done by testing the significance of the lagged levels of the variables in the 
error correction form of the underlying ARDL model. Our ARDL model can 
be written as follows: 
∆𝐿𝑁𝑅𝐺𝐷𝑃! = 𝛼" + 𝛽#𝐿𝑁𝑅𝐺𝐷𝑃!$# + 𝛽%𝐿𝑁𝐸𝑁𝐸𝑅𝐺𝑌!$# + 𝛽&𝐿𝑁𝑇𝑂!$#

+ 𝛽'𝐿𝑁𝑃𝐶!$# + 𝛽(𝐿𝑁𝐻𝐶!$# +2𝛿#∆𝐿𝑁𝑅𝐺𝐷𝑃!$)

*

)+#

+2𝛿%∆𝐿𝑁𝐸𝑁𝐸𝑅𝐺𝑌!$)

*

)+#

+2𝛿&∆𝐿𝑁𝑇𝑂!$)

*

)+#

+2𝛿'∆𝐿𝑁𝑃𝐶!$)

*

)+#

+2𝛿(∆𝐿𝑁𝐻𝐶!$)

*

)+#

+ 𝛾𝐷_𝑒𝑛𝑒𝑟𝑔𝑦	 + 𝜀!	 

where, LNRGDP is log of real GDP, LNENERGY is log of energy consump-
tion, LNTO is log of trade openness, LNPC is log of physical capital., 𝐿𝑁𝐻𝐶 
is log of human capital. The selection of the optimum lagged orders of the 
ARDL models is based on Akaike Information criteria (AIC). In order to test 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

69 

co-integration among the variables, the Wald F-statistics for testing the joint 
hypotheses has to be compared with the critical values as tabulated by Pesaran 
et al. (2001). 
The joint hypotheses to be tested are: 
  𝐻":	𝛽# = 𝛽% = 𝛽& = 𝛽' = 𝛽( = 𝛽, = 𝛽- = 0 
  𝐻#:	𝛽) ≠ 0	, 𝑖 = 1,2… .7 
If the F-statistics is higher than the upper bound critical value, the null hy-
pothesis (𝐻") is rejected, indicating that there is a long run relationship be-
tween the lagged level variables in the model. In contrast, if the F-statistic falls 
below the lower bound, then the 𝐻" cannot be rejected and no long run rela-
tionship exists. However, if the F-statistics falls in between the upper bound 
and lower bound critical values, the inference is inconclusive. At this condi-
tion, the order of integration of each variable should be determined before any 
inference can be made. 
 In the second step, once the co-integration is established, the conditional 
ARDL (p,q,r,s,t,) long-run model of the economic growth and energy con-
sumption can be estimated as below: 

𝐿𝑁𝑅𝐺𝐷𝑃! = 𝛼" +2𝛽#𝐿𝑁𝑅𝐺𝐷𝑃!$#

*

)+#

+2𝛽%𝐿𝑁𝐸𝑁𝐸𝑅𝐺𝑌!$#

.

)+"

+2𝛽&𝐿𝑁𝑇𝑂!$#

/

)+"

+2𝛽'𝐿𝑁𝑃𝐶!$#

0

)+"

+2𝛽(𝐿𝑁𝐻𝐶!$#

!

)+"

+ 𝐷

+ 𝜀! 
 In the final step, we obtain the short-run dynamic parameters by estimat-
ing an error correction model (ECM) associated with the long-run estimates. 
This is specified as follows: 
∆𝐿𝑁𝑅𝐺𝐷𝑃! = 𝛼" + ∑ 𝛽#∆𝐿𝑁𝑅𝐺𝐷𝑃!$)1

2+" +
												∑ 𝛽%∆𝐿𝑁𝐸𝑁𝐸𝑅𝐺𝑌!$)1

2+" + ∑ 𝛽&∆𝐿𝑁𝑇𝑂!$)1
2+" +∑ 𝛽'∆𝐿𝑁𝑃𝐶!$)1

2+" +
												∑ 𝛽(∆𝐿𝑁𝐻𝐶!$)1

2+" + 𝛾𝐷 + 	𝛿𝐸𝐶𝑇𝑡 − 1 + 𝜀!  

where,	𝛿#, 𝛿%, 𝛿&, 𝛿', 𝛿(, 𝛿, ,	𝛿- are the short-run dynamic coefficients of the 
model’s convergence to equilibrium, and 𝛿 is the speed of adjustment.  
 The theoretical foundation of the study is based on the augmented Solow 
model and endogenous growth model for economic growth equation which 
aims to show the impact of energy consumption on economic growth of Ethi-
opia. It is constructed based on the theoretical framework of the augmented 
Solow Model and endogenous growth model with a modification that extends 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

70 

the basic production function framework to permit human capital as an addi-
tional input in to the production function following Romer (1996) and energy 
following Stern and Cleveland (2004). As implied by Solow’s formulation, 
economic growth is a function of capital accumulation, an expansion of labor 
force and exogenous factor, technological progress which makes physical cap-
ital and labor more productive. 
 The presence of co-integration alone does not indicate the direction of 
causality. Hence, we need to test whether the relationship between the varia-
bles is unidirectional or bidirectional. Since the underlying series (LNRGDP 
and LNENERGY) are integrated of the same order, the ordinary Granger cau-
sality test can be applied to perform causality tests. The test proceeds in esti-
mating the following two equations.  
𝐿𝑁𝑅𝐺𝐷𝑃! = 𝛼" + ∑ 𝛼#∆𝐿𝑁𝑅𝐺𝐷𝑃!$)3

2+" +∑ 𝛼%∆𝐿𝑁𝐸𝑁𝐸𝑅𝐺𝑌!$)3
2+" + 𝜀#! 	

𝐿𝑁𝐸𝑁𝐸𝑅𝐺𝑌! = 𝛽" + ∑ 𝛽#∆𝐿𝑁𝐸𝑁𝐸𝑅𝐺𝑌!$)3
2+" + ∑ 𝛽%∆𝐿𝑁𝑅𝐺𝐷𝑃!$)3

2+" +
																										𝜀%!  
The null hypothesis is that: 

 H0:	𝛽11 = β12 =.... = β1j = 0  

Implying LENERGY does not Granger Cause LRGDP 
 H1: β11 ≠ 	𝛽12 ≠ ....  ≠ β1j	≠	0 
Implying LNENERGY does Granger Cause LNRGDP 
The null hypothesis can be stated as: 

 H0:	𝛼11 = α12 = .... = α1j  = 0  

Implying LRGDP does not Granger Cause LNENERGY 
 H1: α11 ≠ 	𝛼12 ≠ .... ≠ 	𝛼1j  ≠	0  
Implying LNRGDP does Granger Cause LNENERGY 
 The decision is that there is causality from energy consumption 
(LNENERGY) to economic growth (LNRGDP) if the null hypothesis H0:	𝛽11 
= β12 = .... = β1j = 0 can be rejected at least at 10% level of significance. 
Similarly, there is causality from economic growth to energy consumption if 
the null hypothesis H0:	𝛼11 = α12 = .... = α1j = 0 can be rejected at least at 5% 
level of significance. 
  



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

71 

Description of the Macroeconomic Variables  
The descriptions of the dependent and the explanatory variables that are in-
cluded in the study model are explained as follows: 
Real gross domestic product (RGDP): It is the total market value of all final 
domestically produced products at constant price. It is a dependent variable of 
the model. Here RGDP has been transformed into log so as to keep the line-
arity of the variable vis-á-vis the other variables. 
Energy consumption (EC): Energy consumption is proxied by (GDP per 
unit of energy use) which measured by the PPP GDP per kilogram of oil 
equivalent of energy use.  
Physical capital (PC): Capital stock is defined as the value of the existing 
supply of physical goods that are used in the production process at a given 
point of time and includes buildings, machinery, equipment and inventory. 
There are points of view that capital stock is generally believed to be of critical 
importance, not only as a component of final aggregate demand, but also in 
terms of the impact of capital stock on the economy’s growth and employment 
opportunities (Ghali, 1999). Hence, we expect that gross capital formation 
should have a positive coefficient in explaining economic growth. 
Human capital (HC): In this study human capital is proxied by secondary 
school enrolments (% gross). Romer (1996) and Gungor (1997) notes that hu-
man capital which describes the knowledge and skills embodied in individuals 
are an important source of economic growth. Human capital accumulation that 
is the ability of individuals to solve problems and to think critically is believed 
to promote higher growth by improving labor force which will be more pro-
ductive. Therefore, human capital variable is expected to have positive impact 
on the production and economic growth of the country. 
Trade Openness (TO): trade openness is the sum of export and import di-
vided by two divided by GDP and expected to affect economic growth posi-
tively. Romer, (1993) claimed that the countries have higher possibility to im-
plement leading technologies from other countries if countries are more open 
to trade. In addition, Chang et.al (2005) emphasized trade openness promotes 
the efficient comparative advantage which allows the dissemination of 
knowledge and technological progress and encourages competition in the in-
ternational market. 
Policy dummy (D): Changes in political and economic policies (the dummy 
variable D in the model) can influence the performance of the economy 
through investment on human capital and infrastructure, improvement in po-
litical and legal institutions and so on (Easterly, 1993). 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

72 

3. Results and Discussion 

3.1. Empirical Results for Unit Root Testing 
It is vital and must to test the nature of stationarity of the variables before 
running ARDL model, a model used to determine the existence of long run 
relationship among the variables. Doing so avoids the possibility of running 
a spurious regression, which makes the result to be unreliable and incon-
sistent. The null hypothesis of no stationarity cannot be rejected for all varia-
bles in level. However, every variable become stationary with trend once they 
are first differenced. This indicates that none of the above variables are inte-
grated of order two I(2), which is a precondition to use ARDL model (see 
Appendix 1) 
 As a result, Autoregressive Distributed Lag approach to Co-integration is 
the right technique to apply in this scenario. Therefore, ARDL or bound test-
ing approach to co-integration is the preferred and appropriate method of re-
gression in this case. 

3.2. Bounds Test for Long Run Relationship 
 In the ARDL approach to Co-integration, the first step is to test the pres-
ence of co-integration or long run relationship among the variables. This test 
for the long run relationship is done using the F-statistic. Given the annual 
nature of the data; it is recommended that the optimal lag length for the ARDL 
model is maximum two lags. Moreover, AIC is used to determine the optimal 
lag because of small sample size at hand. 
 The F statistic will then be compared with the lower and upper bounds of 
Kripfganz and Schneider (2018) critical values, based on the rational men-
tioned in chapter three. The calculated F-statistics is 5.416 and this value is 
higher than the upper bound critical values at 5% level of significance. The 
results indicate that there is strong evidence of long-run relationship or co-
integration between log of RGDP and the remaining variables. This represents 
a co-integrated RGDP equation in Ethiopia. Thus, the null hypothesis of no 
co-integration between log of RGDP and its fundamentals is rejected (see Ap-
pendix 2). 

A. Dynamic Long-run ARDL Estimates 
 Based on the confirmation obtained from the unit root test about the ab-
sence of a variable which is I(2) and given the F statistic result which indicated 
the existence of long run cointegration among the variables, it is now possible 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

73 

to proceed to the estimation of the long run coefficients of the model. The 
following table presents the results found after running the appropriate ARDL 
model to find out the long run coefficients.  

Table 1.  Estimated Long Run Coefficients using the ARDL Approach (Dependent 
variable is LNRGDP; 44 observations used for estimation from 1974 to 
2017) 

Regressors  Coefficient  ST. Error T-Ratio 
LNENERGY 3.84 9.27 0.41 

LNTO 0.544* 0.287 1.90 
LNHC 0.099 0.101 0.98 
LNPC 0.455*** 0.138 3.3 

D –0.274 0.176 –1.56 
Constant  –5.299 15.675 –0.34 

Note: The signs ***, ** and * indicate the significance of the coefficients at 1%, 5% and 10% level of 
significance respectively; ARDL (4, 1, 1, 0, 4, 4) selected based on Akaike Information Criterion (AIC). 

 The real GDP equation or growth model is specified in a log-linear form; 
hence, the coefficient of the dependent variable can be interpreted as elasticity 
with respect to economic growth. As we observe from the long-run ARDL 
regression result, log of energy consumption has an insignificant impact on 
log of real GDP. Additionally, human capital found to be statistically insig-
nificant in the long-run. The result is inconsistent with the outcome found by 
Driffield and Jones (2013), and Fayissa and Nsiah (2008) where human capital 
is found to positively and significantly affecting output. Moreover, the dummy 
for policy change found statistically insignificant to affect economic growth 
(i.e. other things remain constant, policy change from Derg regime to post 
Derg regime of the country doesn’t significantly affect the performance of the 
economy in the long-run).  
 Apart from these, both trade openness and physical capital found to be 
positively and statistically significant to affect economic growth in Ethiopia 
(see Appendix 3).  

B. Short-run Error Correction Model 
 The short run model results are different from the long run. For instance, 
energy consumption is significantly and positively affecting output which is 
dissimilar to the long run result. Also, even though trade openness has statis-
tically significant in both long run and short run estimate, it has negative sign 
in short run, however. The result also suggests that, openness can be pain for 
an economy and invoke a call for protectionism. This may arise in line with 
poor quality of institutions and weak exporting capacity of the country or large 
share of import content of the countries international trade participation. 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

74 

Table 2. Error Correction Representation for the Selected ARDL Model (Dependent 
variable is dLNRGDP; 44 observations used for estimation from 1974 to 
2017) 

Regressors  Coefficient  ST. Error T-Ratio 
dLNRGDP 0.3527* 0.194 1.82 

dLNENERGY 6.035** 2.486 2.43 
dLNTO –0.00971* 0.0506 –1.92 
dLNPC –0.0321 0.0593 –0.54 

dD 0.206*** 0.0708 2.92 
ECM (–1) –0.2779** 0.12564 -2.21 

R-squared = 0.8575 
R-adjusted = 0.7448 

Note: The signs ***, ** and * indicate the significance of the coefficients at 1%, 5% and 10% level of 
significance respectively. 

 More interestingly, the dummy of policy change found positive and sta-
tistically significant. That means the policy transition during 1991 (departure 
from the previous socialist system) had significant effect on economic growth 
of Ethiopia in the short run.  
 The speed of adjustment of any disequilibrium towards long-run equilib-
rium or the equilibrium error correction coefficient (ECM), estimated  
(–0.2779) is highly significant and has the correct sign. It implies a high speed 
of adjustment to equilibrium after a shock. Approximately 27.79 % of the dis-
equilibrium from the previous year’s shock converges back to the long-run 
equilibrium in the current year and such significant error correction term is 
another proof for the existence of a stable a long-run equilibrium relationship 
among the variables.  
 Regarding the short run model’s goodness of fit, the regression result im-
ply that real gross domestic product is moderately explained by the explana-
tory variables incorporated in the model. The adjusted R-squared reveals that 
74.48% of the short-run variation in real gross domestic product is explained 
by the explanatory variable (see Appendix 2). 

Diagnostic Testing and Model Stability 
In this study Akaike information criterion is used to determine the optimal lag 
length of each variable automatically because it is a better choice for small 
sample size data. Moreover, according to Pesaran and Shin (1999), for the 
annual data a maximum of two lag length is recommended to choose the op-
timal lag for each variable. Therefore, in this paper a maximum lag length of 
2 was chosen for the conditional ARDL model. Finally, in this model, AIC 
selects the optimal lag length of each variable (LNRGDP, LNENERGY, 
LNTO, LNHC, LNPC, D), respectively and it is ARDL(4, 1, 1, 0, 4, 4). This 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

75 

automatically determination of the lag length is to get the valid result and in-
ferences (see Appendix 4) 
 To check the reliability and verifiability of the estimated long-run and 
short-run models, diagnostic tests are undertaken. These tests include serial 
correlation (Breusch and Godfrey LM test), Functional form (Ramsey’s RE-
SET test), Normality (Jarque-Bera test), Hetroscedasticity (Breusch-Pagan-
Godfrey test) and also CUMSUM recursive residuals and CUMSUM square 
recursive residuals tests are applied to check the overall stability of the long-
run and short-run coefficients which are recommended by Pesaran et al. 
(2001).  
 The results indicate that both the LM version and the F version of the sta-
tistics are unable to reject the null hypothesis specified for each test. Hence, 
there is no serial correlation problem and the Ramsey functional form test 
confirms that the model is specified well. Likewise, the errors are normally 
distributed and the model doesn’t suffer from heteroskedasticity problem (see 
Appendix 6). 
A. The null hypothesis of no serial correlation (Bruesch-Godfrey LM test) is 
failed to reject for the reason that that the p-value associated with test statistic 
is greater than the standard significant level (0.234> 0.05). Since the lagged 
dependent variable appear as a regressor in the model, LM test avoid the use 
of the traditional Durbin Watson test statistic. 
B. For Ramsey’s RESET test, which tests whether the model suffers from 
omitted variable bias or not we failed to reject the null hypothesis of this test 
which says that the model is correctly specified, because the p-value is larger 
than the conventional significance value (0.716> 0.05). 
C. Similarly, we could not reject the null hypothesis for the Jarque-Bera nor-
mality test which says that the residuals are normally distributed, for the rea-
son that the p-value associated is larger than the standard significance level 
(0.627>0.05). Therefore, the error term is normally distributed. 
D. The last diagnostic test is hetroscedasticity test and as we can understand 
from the result, the null hypothesis of no heteroscedasticity is failed to be re-
jected at 5% significant level due to its p-value associated is greater than the 
standard significance level (0.301> 0.05).  
 Pesaran and Shin (1997) further suggested that structural stability or pres-
ence of structural break of the long run and short run relationships for the 
sample period can be better examined by cumulative sum (CUMSUM) and 
the cumulative sum of squares (CUMSUMSQ) of the recursive residual test. 
The test is based the first set of n observations and is updated recursively 
which will then be plotted against the break points to assess the given param-
eter consistency. In this study the plot of CUMSUM and CUMSUMSQ starts 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

76 

from 1994/95, implying that the test is based on the recursive residuals from 
observations before 1994/95. The test chooses the first n observation by itself. 
 For the stability test the graph plots both the cumulative sum and the 5% 
critical lines. And, if the cumulative sum remains inside between the two crit-
ical lines or bounds back after it is out of the boundary lines, the null hypoth-
esis of correct specification of the model cannot be rejected. But, if the 
CUSUM goes outside (never returns back) between the two critical bounds 
there exists series parameter instability problem. 

 

Figure 1.  Cumulative Sum of Recursive Residuals 

 

Figure 2.  Cumulative Sum of Square of Recursive Residuals (ARDL(4, 1, 1, 0, 4, 4) 
result). 

As the two plots above clearly reveal the plots of CUMSUM and 
CUMSUMSQ stay within the lines, and, therefore, this confirms the equation 
is correctly specified and the model is stable. Furthermore, the result shows 
that there is no structural instability in the model during the sample period. 
From this, the model appears to be robust in estimating short run and long run 
relationship between real gross domestic product and the included regressor.  
  



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

77 

Granger Causality Test Results 
Granger causality test provides important information of the causal direction 
between the variables and knowing the direction of causality between the var-
iables. In this study, Granger causality Wald test after VAR model was em-
ployed to look at the causal linkages between economic growth and energy 
consumption in Ethiopia.  

C. Pairwise Granger Causality Test 
 This section is concerned with tests of Granger causality between GDP 
and energy. The estimated F-statistics of the causality test are reported in the 
Tables below. From the result we fail to accept the null hypothesis that 
LNGDP does not Granger causes LNENERGY, but we fail to reject the null 
hypothesis that LNENERGY does not Granger cause LNGDP. Therefore, it 
appears that Granger causality runs one-way from GDP to ENERGY and not 
the other way (See Appendix 5). 
 From the above pairwise granger causality we fail to reject the null hy-
pothesis for LNENERGY does not granger cause LNGDP because the p-value 
is 0.93326 which much higher than 0.05. However, in the second case we can 
reject the Ho and accept the alternative which states LNRGDP can granger 
cause LNENERGY.  

D. Vector Error Correction Granger Causality (Wald Test) 
 After undertaking pair wise granger causality test, Error Correction Model 
is also used and the result are shown in below. 

Table 3. Granger causality test results for LNRGDP equation 
Dependent Variable: LNRGDP (log of real GDP) 

Excluded  Chi 2 P-value  
LNENERGY  0.02524 0.874 

LNTO 2.8082 0.094* 
LNHC 0.5262 0.468 
LNPC 8.2672 0.004*** 

D 3.7145 0.054* 
All 53.656 0.000*** 

 In Table 3 where GDP is dependent variable the null hypothesis energy 
consumption does not Granger cause economic growth and the alterative hy-
pothesis is energy consumption Granger cause economic growth. From the 
Table 3 it shown that the p-value is 0.874 and based on the ‘p-value’ we tend 
to accept HO. That is, energy does not Granger cause economic growth. 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

78 

Table 4. Granger causality test results for LNENERGY equation 
Dependent Variable: LNENERGY (log of energy consumption) 

Excluded  Chi 2 p-value  
LNRGDP 8.344 0.004*** 

LNTO 5.5232 0.019** 
LNHC 1.1691 0.28 
LNPC 4.413 0.036** 

D 16.982 0.000*** 
All 38.881 0.000*** 

Note: The signs ***, ** and * indicate the significance of the coefficients at 1%, 5% and 10% level of 
significance respectively. 

 In the Table 4 where energy is dependent variable and with hypothesis 
GDP does not Granger cause energy consumption and the alternative hypoth-
esis that GDP does Granger cause energy consumption. The ‘p-value’ is 0.004 
and accordingly we have to reject the null hypothesis and hence we tend to 
accept the alterative hypothesis. Therefore, the evidence of multi-variate anal-
ysis is in line with the growth-led energy consumption hypothesis where cau-
sality running from economic growth to energy consumption. 
 The above two results that is the pair wise Granger causality (which is bi-
variate analysis) and the vector error correction model Granger causality test 
(which is multivariate analysis including physical & human capital., trade 
openness and policy dummy) are consistent with each other. Both evidences 
are in line with the growth-led energy consumption hypothesis where causal-
ity running from economic growth to energy consumption, implying that eco-
nomic development seems to take precedence over energy consumption and 
that economic growth caused greater demand for energy. The economy of 
Ethiopia is heavily dominated by the agricultural sector. However, the energy 
use of the sector is insignificant. And the results show that shortage of energy 
may not adversely affect GDP growth or cause a fall in the GDP in the short 
run. This is because the agricultural sector does not depend on energy. 
 The above result of Granger causality running from economic growth to 
energy consumption in Ethiopia goes in line with the finding of Chontanawat 
et al. (2008) who found economic growth Granger cause energy consumption 
using bivariate analysis for Ethiopia. 
 GDP is generally less in the developing world than the developed world 
(or alternatively causality from energy to GDP generally increases at higher 
stages of development). Hence the results support the view that energy is gen-
erally neutral with respect to its effect on economic growth in the developing 
world, implying that the effect of energy conservation policies to help combat 
global warning would have a greater detrimental effect on the overall growth 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

79 

of OECD/developed countries than that of the non-OECD/developing coun-
tries”. 
 And it also supports the finding of Wolde-Rufael (2005) for five countries 
(Algeria, Democratic Republic of Congo, Egypt, Ghana and Ivory Coast) who 
found economic growth Granger cause energy consumption using bivariate 
analysis. And similarly, it goes in line with what found by Wolde-Rufael 
(2009) for Sudan and Zimbabwe which Granger causality test shows that eco-
nomic growth Granger cause energy consumption using multi-variate analysis 
consisting GDP, capital., energy and labor. However, it contradicts with the 
results for Cameroon, Gambia, Ghana, Morocco and Nigeria. And the result 
goes in line with the result of Akinlo (2008) who found for Sudan and Zim-
babwe Granger causality running from economic growth to energy consump-
tion. The result is also consistent with Masih and Masih (1996) for Pakistan 
and Indonesia, Olatunji Adeniran(undated) for Nigeria, Jumbe (2004) for Ma-
lawi. 
 This finding is contrary with the result of Yohannes (2010) in Ethiopia 
and Amirat and Bouri (2010) who undertook analyses of the causal relation-
ship between the per capita energy consumption and the per capita GDP in 
Algeria adding capital and labor to the economic growth and energy consump-
tion nexus and found Granger causality running from energy consumption to 
economic growth which reverse the result of Chontanawat et al. (2008) for 
Algeria. It is also inconsistent with Nondo et.al (2009) for 19 CEMESA mem-
ber countries. 
 The implication of the uni-directional causality running from economic 
development to energy consumption result is that, the result may statistically 
suggest that energy conservation measures may be taken without jeopardizing 
economic development. In practice however, to suggest measures that can 
lead to the reduction of energy consumption to the end-user in order to halt 
any conservation problem arising out energy consumption may not be a viable 
option for Ethiopia particularly given the magnitude of the energy problems 
and the fact that the current energy infrastructure of the country is still inade-
quate to support the quest for rapid economic growth that is required to erad-
icate poverty and to raise the living standards of the people. Reducing energy 
consumption while the overwhelmingly majority of the population is still de-
nied access to the use of modern form of energy may not be a viable option. 
Ethiopia has not yet reached the energy ladder that may warrant such a sug-
gestion but it can still substantially improve the detrimental consequences of 
energy consumption (example the loss of natural resource for energy and the 
subsequent loss of soil fertility and erosion) without reducing its use. By mak-
ing its energy sector more efficient and by making it available to a larger part 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

80 

of the population (especially electricity) energy used per unit of output can be 
raised. 

Conclusions and Policy Implications 
 This study aimed to examine the dynamic relationship between economic 
growth and energy consumption in Ethiopia. In order to achieve objectives, 
data from different relevant source were collected over the years 1970–2017 
and the parameters of the model were estimated using ARDL system of data 
estimation technique. The estimation result reveals that, energy consumption 
found statistically insignificant in affecting economic growth in the long-run. 
However, it was positive and statistically significant in short-run. Similarly, 
the dummy variable incorporated to capture the policy change effect found 
insignificant in long-run and with positive significant result in short-run. 
 Regarding the causality, the evidence is in line with the growth-led energy 
consumption hypothesis where causality running from economic growth to 
energy consumption, implying that reducing energy consumption may be im-
plemented with little or no adverse effect on economic growth. In practice 
however any conservation measures taken to reduce energy consumption may 
not be a viable option for Ethiopia particularly given the magnitude of its en-
ergy problems and the fact that the current energy infrastructure of the country 
is still inadequate to support its quest for rapid economic growth and for erad-
icating poverty. 
 The option therefore might be for Ethiopia to enhance the level of effi-
ciency in the energy sector. Increasing energy efficiency can cut down growth 
of energy demand that can mitigate conservation and health problem. As noted 
by IEA (2002), finding ways of expanding the quality and quantity of energy 
services while simultaneously addressing the environmental impacts associ-
ated with energy use represents one of the critical challenges Africa is facing. 
This means that energy regulation policies supporting the shift from lower-
quality (typically less efficient and more polluting) to higher-quality energy 
services could provide impulse to economic growth rather than be detrimental 
to the development process (Costantini and Martini, 2010). 
 Since short run energy shortages may have significant impacts on the long 
run economic performances, the country needs to attract new capital for its 
energy industries. However, expanding energy production is not the one and 
only solution to the growth problems of the country. Promoting energy effi-
ciency and focusing on decreasing energy intensity may also have positive 
impacts on economic growth rates without putting considerable pressure on 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

81 

the environment. Developing energy sources that are renewable and that have 
low or no carbon content seem to be essential for this purpose. 
 Irrespective of the strength of the causal relationship between energy con-
sumption and economic growth, the energy challenge facing Ethiopia is 
daunting. Unfortunately, in Africa, it is not energy lack that is the basic prob-
lem but the lack of institutions, rules, financing mechanisms, and regulations 
needed to make markets work in support of energy for sustainable develop-
ment .Until these elementary limitations that are restraining the development 
of an efficient and accessible energy sector are fully solved, energy supply 
will still persist to be a major obstacle for the economic and social develop-
ment Ethiopia.  

Reference 
Akinlo, A.E. (2008). Energy Consumption and Economic Growth: Evidence from 11 African 

Countries. Energy Economics, 30, 2391–2400.   
http://dx.doi.org/10.1016/j.eneco.2008.01.008 

Amirat, A. & Bouri, A. (2010). Energy and Economic Growth: The Algerian Case, typewritten. 
Chontanawat, J., Hunt, L. & Pierce, R. (2008). Does Energy Consumption Cause Economic 

Growth? Evidence from Systematic Study of over 100 Countries. Journal of Policy 
Modelling, 30, 209–220. http://dx.doi.org/10.1016/j.jpolmod.2006.10.003 

Easterly, W. (1993). How Much Do Distortions Affect Growth? Journal of Monetary Econom-
ics, 32(2), 187–212. 

Ebohon, O. J. (1996). Energy, Economic Growth and Causality in Developing Countries: 
A Case Study of Tanzania and Nigeria. Energy Policy, 24, 447–453. 

EEA (2009). Problems and Prospects of Energy Sector in Ethiopia. Bulletin of the Ethiopian 
Economic Association, 3(5), 7–11. 

Enders,W. (1996), Applied Econometric Time Series, Lowa State University: John Wiley & 
Sons INC. 

Engle, R.F. & Granger, C.W.J. (1987). Cointegration, Error Correction Representation, Esti-
mation and Testing. Economertica, 55, 251–276. 

Ghali, K.H. & El-Sakka (2004). Energy Use and Output Growth in Canada: a Multivariate 
Cointegration Analysis. Energy Economics, 26, 225–238.   
http://dx.doi.org/10.1016/S0140-9883(03)00056-2 

Granger, C.W.J. (1988). Causality, Co-integration, and Control. Journal of Economic Dynam-
ics and Control, 12, 551–559. 

Harris, R. (1999), Using Co-integration Analysis in Econometric Modeling, London: Prentice 
Hall. 

Jumbe, C.B.L. (2004). Co-integration and Causality between Electricity Consumption and 
GDP: Empirical Evidence from Malawi. Journal of Energy Economics, 26(1), 26–68. 

Johansen, S. and K. Juselius (1990), Maximum Likelihood Estimation and Inference of Co-
integration: with Applications to the Demand for Money. Oxford Bulletin of Economics 
and Statistics, 52, 169–210. 

Masih, A. M. M., Masih, R. (1998). A Multivariate Co-integrated Modeling Approach in Test-
ing Temporal Causality between Energy Consumption, Real Income and Prices with an 
Application to Two Asian LDCs. Applied Economics, 30, 1287–1298. 



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

82 

Nondo and Mulugeta. (2009). Energy Consumption and Economic Growth: Evidence from 
COMESA Countries, Annual Meeting, January 31-February 3, Atlanta, Georgia. 

Pesaran, M.H. & Shin, Y. (1998). Generalised Impulse Response Analysis in Linear Multivar-
iate Models. Economics Letters, 58, 17–29. 

Pesaran, H. and Y. Shin (1999), An Autoregressive Distributed Lag Modeling Approach to 
Cointegration Analysis, in: Econometrics and Economic Theory in the 20th Century: 
The Ragnar Frisch Centennial Symposium, Strom, S. (ed.) Cambridge University Press. 

Pesaran M.H., Shin, Y. and Smith, R. (2001), Bound Testing Approach to the Analysis of Level 
Relationship. Journal of Applied Econometrics, 16(3), 289–326.     
https://doi.org/10.1002/jae.616 

Rahimi, M. and A. Shahabadi (2011), Trade Liberalization and Economic Growth in Iranian 
Economy, Bu-Ali Sina University, Hamedan, Iran.  
http://dx.doi.org/10.2139/ssrn.1976299  

Romer, D. (1996), Advanced Macroeconomics. New-York: McGraw-Hill. 
Stern, D.I. (1993). Energy Use and Economic Growth in the USA, a Multivariate Approach. 

Energy Economics, 15, 137–150. 
Stern, D.I. (2010). Energy Quality. Ecological Economics, 69(7), 1471–1478.   

http://dx.doi.org/10.1016/j.ecolecon.2010.02.005 
Stern, D. I. (2011). The Role of Energy in Economic Growth. Ecological Economics Reviews, 

1219, 26–51. 
Wolde-Rufael, Y. (2004). Electricity Consumption and Economic Growth: a Time Series Ex-

perience for 17 African Countries. Energy Policy, 34, 1106–1114. 
Wolde-Rufael, Y. (2005). Energy Demand and Economic Growth: the African Experience. 

Journal of Policy Modelling, 27(8), 891–903.  
http://dx.doi.org/10.1016/j.jpolmod.2005.06.003 

World Bank (2017). World Development Indicators. 
Yohannes, H. (2010). Energy, Growth, and Environmental Interaction in the Ethiopian Econ-

omy. Journal of Economic & Financial Modelling, 2(2), 35–47. 

Websites Accessed 
World Bank, world development indicator accessed at http//www.world bank.org/data 
UNCTAD, UNCTADSTAT, accessed at http//unctadstat.unctad.org/EN/ 

Appendix 

Appendix 1. Results of Augmented Dickey Fuller Test 
Variables At level I(0) At 1st difference I (1) Order of Integration  

Ln Real GDP Intercept 3.264 -4.643**  
  I (1) Trend 0.565 -5.633** 

Ln Energy cons. Intercept 0.783 -6.427**   
  I (1) Trend -1.534 -6.991** 

Ln Trade Openness Intercept -1.209 -5.448**   
  I (1) Trend -2.001 -5.406** 

Ln Physical capital Intercept 1.821 -6.852**  
  I (1) Trend -0.570 -7.883** 

Ln Human capital Intercept -1.298 -4.026*   
  I (1) Trend -1.188 -4.067** 



Energy Consumption and Economic Growth in Ethiopia… 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 
 

83 

Appendix 2. Bound Test Result 

 

Appendix 3. ARDL regression result (using AIC lag selection criteria)  

 
  

      (if p-values < desired level for I(1) variables)
    both F and t are more extreme than critical values for I(1) variables
reject H0 if
      (if p-values > desired level for I(0) variables)
    both F and t are closer to zero than critical values for I(0) variables
do not reject H0 if

 t    -2.518   -3.826    -2.867   -4.238    -3.572   -5.065     0.037    0.299
 F     2.428    3.691     2.911    4.327     4.043    5.806     0.002    0.015
                                                                              
        I(0)     I(1)      I(0)     I(1)      I(0)     I(1)      I(0)     I(1)
     10%                5%                 1%                 p-value         

Kripfganz and Schneider (2018) critical values and approximate p-values

Finite sample (5 variables, 46 observations, 4 short-run coefficients)

Case 3                                                           t =    -3.009
H0: no level relationship                                        F =     5.416

Pesaran, Shin, and Smith (2001) bounds test

. estat ectest

                                                                              
       _cons     -5.29906   15.67556    -0.34   0.738    -37.65184    27.05371
              
        L3D.     .0888911   .0534729     1.66   0.109    -.0214716    .1992538
        L2D.     .1624917   .0663116     2.45   0.022     .0256313    .2993522
         LD.    -.0235592   .0707011    -0.33   0.742    -.1694792    .1223607
         D1.     .2069438   .0708412     2.92   0.007     .0607347    .3531528
           D  
              
        L3D.    -.0716564   .0520453    -1.38   0.181    -.1790727    .0357599
        L2D.    -.1246635   .0604452    -2.06   0.050    -.2494161    .0000892
         LD.    -.0298997   .0594317    -0.50   0.619    -.1525608    .0927614
         D1.    -.0321597   .0593711    -0.54   0.593    -.1546955    .0903762
        lnpc  
              
         D1.    -.0971763   .0506137    -1.92   0.067    -.2016378    .0072852
        lnto  
              
         D1.     6.035592    2.48643     2.43   0.023     .9038526    11.16733
    lnenergy  
              
        L3D.      .334598   .1949743     1.72   0.099    -.0678092    .7370053
        L2D.    -.0992182   .1940987    -0.51   0.614    -.4998183    .3013818
         LD.     .3527601   .1940001     1.82   0.082    -.0476366    .7531567
      lnrgdp  
SR            
                                                                              
           D    -.2748932   .1761806    -1.56   0.132    -.6385121    .0887256
        lnpc     .4555515   .1382543     3.30   0.003     .1702086    .7408945
        lnhc     .0992804   .1013946     0.98   0.337    -.1099879    .3085486
        lnto     .5444754   .2870811     1.90   0.070    -.0480309    1.136982
    lnenergy     3.843454   9.277611     0.41   0.682    -15.30459     22.9915
LR            
                                                                              
         L1.    -.2779594   .1256454    -2.21   0.037    -.5372788     -.01864
      lnrgdp  
ADJ           
                                                                              
    D.lnrgdp        Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
                                                                              

Log likelihood =  98.496424                       Root MSE        =     0.0349
                                                  Adj R-squared   =     0.7448
                                                  R-squared       =     0.8575
Sample: 1974 - 2017                               Number of obs   =         44

ARDL(4,1,1,0,4,4) regression

. ardl lnrgdp lnenergy lnto lnhc lnpc D, aic ec



Wondatir Atinafu 

DYNAMIC ECONOMETRIC MODELS 19 (2019) 57–84 

84 

Appendix 4. Optimal lag length for each variable (Akaike information criterion 

 

Appendix 5. Pair wise granger causality test  

 
Appendix 6: Diagnostic tests of the model 

Test statistics LM version  F version 
Serial Correlation  CHSQ(1)= 1.2024[.234]**   F(4, 41)= .62469[.423]** 
Functional Form CHSQ(1)= .011370[.716]**  F(4, 39)= .0053317[.943]** 

Normality CHSQ(2)= 1.5745[.627]** Not applicable 
Heteroscedasticity CHSQ(1)= 1.3321[.301]** F(4, 38)= 1.3031[.263]** 

A: Lagrange multiplier test of residual serial correlation 
B: Ramsey's RESET test using the square of the fitted values 

C: Based on a test of skewness and kurtosis of residuals 
D: Based on the regression of squared residuals on squared fitted values 

 
 

 

r1         4         1         1         0         4         4
      lnrgdp  lnenergy      lnto      lnhc      lnpc         D
e(lags)[1,6]

. matrix list e(lags)

Pairwise Granger Causality Tests
Date: 04/06/19   Time: 22:25
Sample: 1970 2017
Lags: 2

 Null Hypothesis: Obs F-Statistic Prob. 

 LNENERGY does not Granger Cause LNRGDP 46 0.069179... 0.93326...
 LNRGDP does not Granger Cause LNENERGY 7.031050... 0.00236...


