توضیح (غیر) علیتی بین مصرف انرژی و رشد اقتصادی در US-A تجزیه و تحلیل بخشی چند متغیره
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|12304||2012||11 صفحه PDF||سفارش دهید||11826 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Energy Economics, Volume 34, Issue 2, March 2012, Pages 489–499
The rapidly growing literature on the relationship between energy consumption and economic growth has not univocally identified the causal relationship yet. We argue that bivariate models, which analyze the causality only at the macro level, are eventually misleading, especially in cases where both variables do not cover the same scope of economic activity. After discussing appropriate pairs of variables, we investigate Granger causality between energy and growth in the U.S. for the period from 1970 to 2007 for three sectors, industry, commercial sector, transport, as well as on the macro level. Using the recently developed ARDL bounds testing approach by Pesaran and Shin (1999) and Pesaran et al. (2001), we find evidence for unidirectional long-run Granger causality in the commercial sector from growth to energy, as well as evidence for bi-directional long-run Granger causality in the transport sector. The dependence of causality on the level of aggregation is interpreted as evidence for ‘Simpsons' Paradox’. The choice of control variables is based on findings from the Environmental Kuznets Curve literature: we find that controlling for the increasing energy productivity of production as well as trade significantly improves the fit of the bivariate model.
What is the causality between energy consumption and economic growth? It is the key question of the empirical energy growth literature initiated by Kraft and Kraft (1978), which has been left unanswered univocally—after more than three decades of empirical research. It has been discussed that conflicting results may arise due to different time periods of the studies, countries' characteristics, variables used, and different econometric methodologies see Ozturk (2010) and Payne (2010) for an overview. In this paper we will argue that another, even more important, reason for the weak evidence is the level of aggregation. Recent studies (e.g. Bowden and Payne, 2009 and Zachariadis, 2007) investigate the causality between energy consumption and economic growth both on the macro as well as on the sector level. While the relationship between energy and growth seems to be neutral on the macro level, both studies find evidence for Granger causality for a lower level of aggregation in some cases. In statistical analyses it is not uncommon that evidence can be found for a lower level aggregation, although the results for the total population suggests the opposite. This phenomenon has been named ‘Simpson's Paradox’ after E. Simpson (1951)1. However, if the results for Granger causality tests are found to be dependent on the level of aggregation and not on the variables, it is necessary to analyze the causal relationship at the correct level of aggregation. Otherwise, the results are spurious and policy advice should be given with caution. The paradox becomes even more severe if the pairs of variables for Granger causality analyses are not matching. For this reason, we will extend Zachariadis' notion of appropriate pairs for causality analyses. The fact that sectors differ with respect to their relationship between energy and growth, is well known in the Environmental Kuznets Curve (EKC) literature: changes of the industry composition have a changing impact on the energy demands of the economy over time. In the early phases of modern economic growth, when a country industrializes, structural change is believed to increase these demands. Later on when the country enters the post-industrial phase, or the service economy, the energy demands are believed to decline (e.g. Kahn, 1979, Panayotou, 1993, Panayotou et al., 2000, Schäfer, 2005 and Smil, 2000)2. The resulting divergence between energy and economic growth is also a challenge for Granger causality analyses. In order to account for the decreasing energy intensity in a Granger causality framework, we suggest to include major findings from the EKC literature: one major finding is the role of the increasing energy productivity of production, which leads to the divergence between energy and growth. Another main finding is the role of trade, especially for goods producing industries, where energy intensive production is being offshored according to the Pollution Haven Hypothesis (PHH). For our analysis we use the recently developed autoregressive distributed lag (ARDL) bounds testing approach as proposed by Pesaran and Shin (1999) and Pesaran et al. (2001). We analyze the evidence for long-run as well as short-run Granger causality between final energy consumption and GDP for the U.S. from 1970 to 2007 at the macro level as well as for the industry sector, the commercial sector, and the transport sector. After identifying appropriate pairs of variables for the Granger causality test, we test bivariate as well as multivariate specifications of the model in order to avoid omitted variable bias. The choice of additional control variables is based on major findings of the EKC literature as well as its limitations discussed in the literature. In line with the majority of studies for the U.S., we find neutrality between energy and growth at the macro level. In addition, we find evidence for unidirectional long-run Granger causality in the commercial sector from growth to energy. We also find evidence for bi-directional Granger causality in the transport sector. Adding or removing additional control variables is found to create or break long-run causality. This finding is important especially in the transport sector, where controlling for the increasing energy productivity of production neutralizes the long-run relationship when growth is the dependent variable. For the industry sector we find that controlling for trade is important for identifying short-run Granger causality when growth is the dependent variable. We conclude that some of the divergence across sectors can be explained by the fundamental differences between goods and service producing industries. In various specifications energy productivity is found to Granger cause growth as well as energy. The latter is interpreted as evidence for ‘Jevon's Paradox’3. We find only weak evidence for the impact of energy prices on energy consumption for the transport sector. Given the evidence of long-run Granger causality on the sector level, compared to a neutral relationship on the macro level, we conclude that the Granger causality between energy and growth should only be analyzed on the sector level. Otherwise, results for the total economy are spurious. The paper is organized as follows: first, we discuss reasons for the inconclusive evidence for Granger causality in the existing empirical literature. We further elaborate Zachariadis' identification of appropriate pairs for causality analyses and use those pairs we consider appropriate for our analysis. We also discuss our extensions of the basic bivariate models mostly used in the empirical literature. Section 3 describes the econometric methodology. We investigate the causal relationship between energy consumption and economic growth in the U.S. for the period 1970–2007 and three economic sectors as well as for the macro level. Cointegration tests are based on the ARDL bounds testing procedure as proposed by Pesaran and Shin (1999) and Pesaran et al. (2001). Afterward, we analyze the existence of long-run and short-run Granger causality. In Section 4 we discuss our findings and the final section concludes.
نتیجه گیری انگلیسی
We analyzed the Granger causality between energy and growth in the U.S. for the period from 1970 to 2007 on the macro level as well as for the industry sector, the commercial sector, and the transport sector. For our analysis we used the recently developed ARDL bounds testing approach as proposed by Pesaran and Shin (1999) and Pesaran et al. (2001). Our paper contributes to the literature in several ways: (1) based on Zachariadis (2007), we analyze the existing energy growth literature for the U.S. with respect to the choice of appropriate variable pairs for causality analyses and discuss why the evidence is ambiguous. We conclude that only sectoral value added together with sectoral final energy consumption covers the same scope of economic activity and that the sector level should be preferred to the macro level. (2) We also emphasize the fundamental differences between goods and service producing industries and its implications for the relationship between energy and growth in each sector. We argue that, due to the inseparability of the production chain of service producing industries, there exists a closer relationship between energy and growth than in goods producing industries. As the energy intensive production of intermediate goods can be offshored to developing countries, the relationship between industry value added (accounted for in national statistics) and energy consumption, whereas the indirect consumption of energy is not accounted for, is weaker. (3) We combine the well-established methodology from the energy growth literature with major findings from the EKC literature as well as its limitations discussed in the literature. We show that augmenting the basic bivariate model with variables controlling for trade and energy productivity significantly improves the fit of several model specifications. (4) We find that Granger causality between energy consumption and economic growth is not always forced by the same (control) variables. This is the case when we do not find cointegration or the BIC is not minimized for the same model where energy and growth are the dependent variables. In contrast to most bivariate analyses at the macro level, we conclude that the causal relationship between energy consumption and economic growth is much closer than is normally assumed. Our results confirm that long-run Granger causality between energy consumption and economic growth can rather be found on the sectoral level. We find evidence for bi-directional long-run Granger causality in the transport sector. However, once the increasing energy productivity of the capital stock is controlled for, the relationship breaks. In the commercial sector we find evidence for long-run Granger causality from growth to energy, if energy productivity is controlled for. The fundamental difference between goods and service producing industries also shows the differential impact of trade on the relationship between energy and growth. Once trade is controlled for, we find evidence for short-run Granger causality running from energy and trade to growth in the industry sector. Concerning the implications, which can be drawn from the results, we strongly recommend the choice of an appropriate level of aggregation for Granger causality analyses in the energy growth literature. If evidence for Granger causality cannot be found at the macro level, the implication that no causality exists at all is myopic (Simpson's paradox). Even though no evidence for long-run Granger causality can be found at the macro level, policies which aim at the reduction of energy consumption could, in fact, affect individual sectors, both in the long run as well as in the short run. International policies which aim at stricter environmental regulations for developing countries would also indirectly affect the home country if the indirect consumption of energy is not internalized. Finally, the long-run relationship between energy and growth is not carved in stone. We show that efforts to increase the energy productivity of the capital stock allow to break the long-run relationship between energy and growth in the transport sector. As long as the ‘rebound effect’ of increasing energy productivity does not outweigh the conservation of energy, a de-coupling between energy consumption and economic growth is possible. However, for this purpose we have to be aware of the ‘real’ relationship between energy consumption and growth, which tends to be undervalued in inappropriate model specifications.