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...