مصرف انرژی و رشد اقتصادی شواهد جدیدی از متا آنالیز

The causal relationships between energy consumption and economic growth have given rise to much discussion but remain controversial. Alternative data sets based on different time spans, countries, energy policies and econometric approaches result in diverse outcomes. A meta analysis using a multinomial logit model with 174 samples governing the relationships between GDP and energy consumption is applied here to investigate the major factors that affect these controversial outcomes. The empirical results have demonstrated how the time spans, subject selections including GDP and energy consumption, econometric models, and tools for greenhouse gases emission reduction characteristics significantly affect these controversial outcomes.

The relationship between GDP and energy consumption has given rise to considerable debate since the pioneering research of Kraft and Kraft (1978) who used annual data for the United States over the period 1947–1974 to explore the causal relationship between gross national product and energy consumption. Their research suggests that there is a uni-directional causal relationship from economic growth to energy consumption. However, in subsequent research, Akarca and Long (1980) used the same data set of Kraft and Kraft but changed the time periods of the sample and obtained a different result, which provided evidence that different sample intervals may change the empirical results on the issue of the relationship between gross domestic product (hereafter GDP) and energy consumption. Estimation methodologies may also play an important role in this issue. Yu and Hwang (1984) adopted the E–G two-step method (Engle and Granger, 1987) using US quarterly data between 1974 and 1990, and found that there is no long-term cointegration equilibrium relationship between GDP and energy consumption. However, Stern (1993) applied a vector autoregressive model (VAR) to US data sets and found a uni-directional Granger causal relationship to exist from energy consumption to GDP if fuel composition consumption was replaced by energy consumption. Later, Stern (2000) used a single static cointegration analysis and multiple dynamic cointegration analysis to extend his previous research (Stern, 1993) and found energy consumption had a significant effect on GDP, which confirms that there is an obvious long-term equilibrium cointegration relationship between GDP and energy consumption. Furthermore, some literatures have provided more comprehensive evidences to support the relationship between energy consumption and economic growth through panel data set. For instance, Belke et al. (2011) found the bi-directional relationship between energy consumption and real GDP for 25 OECD countries from 1981 to 2007. The newly trivariate panel VECM model has also widely applied to examine the relationship between energy consumption and economic growth by Lee et al. (2008), Costantini and Martini (2010) and Lee and Lee (2010). Such evidences have shown that the improved or updated econometric models may change the findings of the relationship between energy consumption and GDP. Such research has also proceeded in different countries including the U.K., Germany, Italy, France, Japan, and other industrial countries, while the major econometric models applied have included the standard Granger causality test model, the comprehensive error correction model, and the vector autoregressive model. For example, Hwang and Gum (1992), Asafu-Adjaye (2000) and Paul and Bhattacharya (2004) found that there exists a bi-directional relationship between the energy and economic variables, while Masih and Masih, 1996 and Masih and Masih, 1997 found that for Malaysia, Singapore and the Philippines a natural structural compliance relationship was found to exist between energy consumption and income, India was found to exhibit uni-directional causality from energy consumption to GDP, Indonesia to exhibit a reverse causal relationship from GDP to energy consumption, and Pakistan and Taiwan to exhibit a bi-directional causal relationship between energy and GDP. Cheng and Lai (1997) also used a Granger causality test to test Taiwan's data over the period 1955–1993 and found evidence of only a uni-directional causal relationship from energy consumption to GDP. However, Asafu-Adjaye (2000) applied cointegration and error correction models to estimate the relationship between energy consumption and economic growth in India, Indonesia, Thailand, and the Philippines. The empirical results for the Philippines and Indonesia are different from those of Masih and Masih (1997). Furthermore, Yang (2000) updated the above sample interval to cover the period 1954–1997 and also obtained a controversial outcome. Oh and Lee's (2004) research on South Korea found that there is no causal relationship between energy consumption and GDP in the short term, but that there is a uni-directional causal relationship from energy consumption to GDP in the long run. Lee (2005) analyzed the cointegration relationship between energy consumption and economic growth in 18 developing countries and the evidence showed that there is a long-run cointegration relationship from the perspective of the heterogeneous country effect. The results also indicated that there exist uni-directional causalities from energy consumption to GDP in both the short run and long run. Wolde-Rufael (2006) examined the long-term causal relationship between GDP per capita and electricity consumption per capita in 17 African countries over the period 1971–2001. The results indicated that there is long-term stable relationship between GDP per capita and energy consumption per capita in nine African countries, and that there is Granger causality between GDP per capita and energy consumption per capita in 12 African countries. Chen et al. (2007) conducted a similar study in 10 newly industrializing and developing Asian countries using both single data sets and panel data procedures. The empirical results based on single data sets indicated that the causality directions in the 10 Asian countries are mixed while there is a uni-directional short-run causality running from economic growth to electricity consumption and a bi-directional long-run causality between electricity consumption and economic growth if the panel data procedure is implemented. The newly developing countries also get lots of attention to discuss the relationship between energy consumption and economic variables recently. For example, Wang et al. (2011) applied 28 provinces in China over the period 1995–2007 and found the bi-directional relationship between energy consumption and economic variables. Zhang (2011) and Sadorsky (in press) found the bi-directional relationship between energy consumption and economic variables in Russia and seven South America countries, respectively. And the bi-directional relationship is found by Eggoh et al. (2011) who estimated the relationship between energy consumption and economic growth for 21 African countries over the period from 1970 to 2006. Such relationship will be more diverse as developed countries are taken into consideration. For example, the research from Narayan and Smyth (2008) found the uni-directional causality from energy consumption to economic growth when G7 countries are considered. However, Belke et al. (2011) and Costantini and Martini (2010) found the bi-directional relationship between energy consumption and real GDP when OECD countries are applied. This indicates that the similar subject may come to different empirical results. In this study, we refer the energy consumption to electricity consumption since electricity usage plays an essential role in the development of the global economy. Ferguson et al. (2000) examined the relationship between electricity consumption per capita and GDP per capita in more than 100 countries and found that the rich countries had more significant correlation between electricity usage and economic growth than poor countries. However, such relationships between electricity consumption and GDP have been found to be inconsistent among studies and depend on the economic development of a country, its energy usage, the econometric approach adopted, the time frame, and so on (Murray and Nan, 1994, Yang, 2000, Fatai et al., 2004, Shiu and Lam, 2004, Wolde-Rufael, 2004, Altinay and Karagol, 2005, Yoo, 2005, Chen et al., 2007, Ho and Siu, 2007, Narayan and Singh, 2007, Yuan et al., 2007, Yuan et al., 2008, Hu and Lin, 2008, Narayan and Prasad, 2008 and Akinlo, 2009). Although the related conclusions of these studies on the relationships between energy and economic growth are still controversial, such outcomes could be briefly categorized into four types based on the studies by Ozturk (2010) and Payne (2010). In other words, the directions of the causal relationships between energy consumption and economic growth are listed as follows: no causality, uni-directional causality running from economic growth to energy consumption, uni-directional causality running from energy consumption to economic growth, and bi-directional causality between energy consumption and economic growth. Although the controversial and inconsistent conclusions from the literature have not been resolved, the relationship between income and energy consumption may be described as one of these four types of relationships. We also observe that such alternative conclusions or these four types of relationships within the literature may be due to the subject/country selections, data time spans, empirical econometric model settings or other explanatory variable selections. Therefore, the main purpose of this study is to use the meta-analysis approach together with the multinomial logit model (hereafter MNL) to estimate the effects of potential factors on these controversial conclusions in regard to the issue of the relationship between energy consumption and economic growth. To this end, the 39 related studies with 174 samples that test the relationship between GDP and energy consumption will be examined while the multinomial logit model will be applied to investigate the factors impacting the relationships between GDP and energy consumption. The empirical results will show how such relationships are affected by these factors. In the sections that follow, the approach to data collection adopted in our research will be introduced in Section 2, and the detailed methodology of the MNL will be developed and illustrated in Section 3. The empirical results will be presented and discussed in Section 4 and the concluding remarks will appear in the Section 5.

In order to solve the controversial issue of the relationship between GDP and EC, a Meta-analysis with a Multinomial Logit Model and 174 sampling sizes from the previous literature related to this issue are applied and collected. Four types of this relationship including T1, T2, T3, and T4 are defined while the absolute and relative probability ratio equations related to these four types using NML are estimated. Such estimation outcomes can show how these four types of relationship are ranked and affected by GDP, EC, the time period, economic development, estimation methods, and greenhouse gas emission reduction plans including a carbon tax and an Annex I country list. The first contribution of this paper is to indicate how the factors affect the four types of relationship involving GDP and EC. The empirical results show that Y2000, GDP, EC, Method, Develop, OPEC, Annex I, and Tax significantly affect the individual's behavior when choosing among the options for the relationships between GDP and EC. The data involving the year 2000 raise the probabilities of choosing T3 (EC affects GDP) and T4 (GDP and EC exhibit mutual causality). The higher GDP enhances the country's preference for the T2 (GDP affects EC) and T4 (GDP and EC exhibit mutual causality) types, but diminishes the preference for the T1 (GDP and EC exhibit no causality) and T3 (EC affects GDP) types. The greater value of EC leads to the probabilities of T3 (EC affects GDP) and T4 (GDP and EC exhibit mutual causality) being increased, and the probabilities of T1 (GDP and EC exhibit no causality) and T2 (GDP affects EC) being decreased. If the Granger-causality test approach is adopted, the probabilities of T2 (GDP affects EC), T3 (EC affects GDP), and T4 (GDP and EC exhibit mutual causality) are increased, but the probability of T1 (GDP and EC exhibit no causality) is decreased. The developed country prefers to choose the T1 (GDP and EC exhibit no causality), T2 (GDP affects EC), and T4 (GDP and EC exhibit mutual causality) types. To the OPEC member states, only the probability of T3 (EC affects GDP) increases, and the other three types decrease. However, when a country joins the greenhouse gas reduction plan such as those countries in the Annex I list, it raises the probability of picking the T3 (EC affects GDP) type, and lowers the probability of the other three types. As for the carbon tax, imposing the carbon tax leads to a rise in the probability of T3 (EC affects GDP), and a drop in the probabilities of T1 (GDP and EC exhibit no causality), T2 (GDP affects EC), and T4 (GDP and EC exhibit mutual causality). The second contribution of this paper is that it indicates how the probability of each type of relationship involving GDP and EC is affected by these factors. For the T1 (GDP and EC exhibit no interaction), developed countries have the positive impact. The probability that developed countries will choose T1 reflects an increase of 27.1% compared to non-developed ones. However, there is no factor with significant positive effects on the T1 option. The developed countries have the most positive influences on the T2 (GDP affects EC) option, and the OPEC member states and the countries with the carbon tax have the most negative influences on the T2 option. The probability of T2 for the developed countries is an increase of 33.8% compared to the non-developed countries. The probability of T2 for the OPEC member states reflects a decrease of 29.2% compared to the non-OPEC member states. GDP has the most negative impacts on T3 (EC affects GDP), and the countries in ANNEX I have the most positive impact on T3. The probability that countries in ANNEX I will choose T3 reflects an addition of 55.0% compared to countries not in ANNEX I. When GDP grows by 1%, the probability of choosing T3 is reduced by 57.6%. As to T4 (GDP and EC interact), the estimation method and the carbon tax have the most positive and negative impacts, respectively. The probability of selecting T4 (GDP and EC exhibit mutual causality) with the Granger-causality test reflects a rise of 52.3% compared to the other causality tests. The probability of picking T4 for the country with the carbon tax reflects a decline of 29.7% compared to without the carbon tax. With the reliable empirical observations collected from the developed countries that have the most positive influences on the T2 option (GDP affects EC). We could conclude that the reliance on energy consumption is increased as GDP develops for developed countries. The developed countries will not choose to increase energy consumption to support economic development but they cannot avoid increasing energy consumption as their GDP growth increase. In addition, we could find that the OPEC countries have the positive impact on T3 that makes sense that economy growth in energy exporting countries will be affected by energy industry development or energy consumption. As the energy consumption increases, the GDP will be sharply affected. Countries in the list of ANNEX I will choose T3 than other types since the efficiency of energy usage could promote their economy growth. Similar situation occurs for country with carbon tax. Countries with the imposing carbon tax may prefer T3 but for T4, which indicates that the carbon tax is an effective environmental policy for policy maker to consider as a conservative resource use policy. T2 (GDP affects EC) is preferred by developed countries but not listed in ANNEX I or OPEC may consider adopting energy conservation policy. Such energy conservation policy could reduce the increasing speed of energy consumption due to increasing economy growth. On other hand, countries with lower GDP and listed in ANNEX I or OPEC may promote their energy consumption in order to a higher economy growth. Countries with alternative economic and environmental conditions may adopt different energy and environmental policies. Taking policy plan on the reduction of greenhouse gas emission as an example, countries adopt such a plan that may push or depress their economy growth and will depend on their economic and environmental conditions. Countries with higher GDP but not listed in ANNEX I or OPEC could be benefited from this greenhouse gases emission plan but not for lower GDP countries. Eggoh et al. (2011) have such similar finding. Therefore, the conflict on energy development between developed and developing countries is obvious and how to avoid damage on economy when they obey the greenhouse reduction plan should be widely discussed in the future. Therefore, the differences among the adopted variables, the characteristics of the country, and the econometric methods all have influences on the estimation of the relationships between GDP and EC. As Masih and Masih (1997, pp. 419) indicate: “One of the major reasons for these apparently conflicting statistical findings, apart from many institutional, structural, and policy differences, is due to methodological differences—the definitional specifications of the variables used and, most importantly, the type of causality techniques, tests, and lag structures employed.” Here, we provide evidence to explain why the controversial results exist and give evidence regarding the point made by Masih and Masih (1997). Hence, further research on the issue of the relationship between GDP and EC should focus on the criteria for some of the variables that we provided. Because there are so many inconsistent and uncertain empirical results regarding the causality nexus between energy consumption and economic growth, we utilize a new econometric model to provide an entirely different policy guide for the policy maker, especially in relation to environmental care.