رابطه سببی مشاهده شده رشد اقتصادی و مصرف انرژی از طریق یک پنجره راه انداز غلتان

One puzzling results in the literature on energy consumption-economic growth causality is the variability of results particularly across sample periods, sample sizes, and model specification. In order overcome these issues this paper analyzes the causal links between energy consumption and economic growth for G-7 countries using bootstrap Granger non-causality tests with fixed size rolling subsamples. The data used includes annual total energy consumption and real Gross Domestic Product (GDP) series from 1960 to 2006 for G-7 countries, excluding Germany, for which the sample period starts from 1971. Using the full sample bootstrap Granger causality test, we find that there is predictive power from energy consumption to economic growth only for Canada. However, parameter instability tests show that none of the estimated models have constant parameters and hence the full sample results are not reliable. Analogous to the full sample results, the results obtained from the bootstrap rolling window estimation indicate no consistent causal links between energy consumption and economic growth. We, however, find that causal links are present between the series in various subsamples. Furthermore, these subsample periods correspond to significant economic events, indicating that the findings are not statistical artefacts, but correspond to real economic changes. Our results encompass previous findings and offer an explanation to varying findings.

The linkage between energy consumption and economic growth has been a topic studied in depth in the field of energy economics because of the importance it has in the present-day economies, ranging from small developing economies to the developed ones. Most of these studies used the Granger causality test to investigate the direction of the link between energy consumption and real GDP. The main motivation behind testing whether energy consumption Granger causes real GDP or not is rooted in energy conservation policies. It has been well-established that a positive long-run link between energy consumption and real GDP exists. Then, if the direction of the causality runs from energy consumption to real GDP, energy conservation policies will have a negative effect on real GDP growth. A number of studies followed the pioneering paper appeared in 1978 by Kraft and Kraft. Ozturk (2010) provides an extensive survey of this literature on causal nexus between energy consumption and economic growth. Using the US annual data over the period 1947–1974 and applying Granger causality test, they found a unidirectional causality from economic growth to energy consumption. In the following year, Akarca and Long (1979), using the US monthly data from 1973 to 1979, found a unidirectional causality from energy consumption to employment, which is used as a proxy for economic growth. Besides, just a year later, Akarca and Long (1980), using monthly data from 1973 to 1979, which was shorter compared to the study of Kraft and Kraft (1978), found bidirectional causality between energy consumption and GDP, implying that temporal instability was not taken into consideration in the study of Kraft and Kraft (1978). Since the early 1980s, a number of studies, using different time periods and different techniques, on the linkages between energy consumption and economic growth appeared in the literature. Using different techniques and the time periods these studies obtained surprisingly different results, with one contradicting another. Despite the fact that the linkage between the energy consumption and economic growth is well-studied, the findings are still controversial. To illustrate, some studies found unidirectional Granger causality from energy consumption to economic growth for the Philippines (Yu and Choi, 1985), India (Masih and Masih, 1996), Singapore (Glasure and Lee, 1997), and for France, West Germany, Japan, and Turkey (Soytas and Sari, 2003). On the other hand, there are studies, where Granger causality is from economic Growth to energy consumption, such as those by Yu & Choi, 1985 and Soytas & Sari, 2003 for South Korea, and Cheng and Lai (1997) for Taiwan. There are also studies, where causality is in both directions and supports the neutrality hypothesis, such as for the US by Stern, 1993, Stern, 2000 and Cheng, 1995, for Thailand and the Philippines by Asafu-Adjaye (2000), and for Taiwan by Yang (2000). Some studies employed a multivariate cointegration test to examine the relationship between economic growth and energy consumption, such as Stern (2000), and Ghali and El-Sakka (2004). Yu and Jin (1992), using cointegration tests, analyzed the long-run linkage with the level of energy consumption and employment for the US and found no evidence of a cointegrating relationship. On the other hand, using Engle-Granger cointegration test for a group of six Asian economies Masih and Masih (1996) found no cointegration between energy consumption and economic growth for Malaysia, Singapore, and the Philippines but significant cointegration for India, Pakistan, and Indonesia. Using panel data of energy consumption and GDP for 82 countries from 1972 to 2002 and GMM (Generalized Method of Moment), Huang et al. (2008) found no evidence in favour of the hypothesis that energy consumption leads to economic growth. Chiou-Wei et al. (2008), using both linear and non-linear Granger causality tests, found evidence supporting the neutrality hypothesis for the US, Thailand and South Korea, while they found a unidirectional causality from economic growth to energy consumption for the Philippines and Singapore. Bowden and Payne (2009) examined the casual relationship between energy consumption and real GDP using aggregate and sectoral primary energy consumption measures within a multivariate framework. They found evidence that industrial energy consumption Granger causes real GDP. The results in the literature on the energy consumption-growth causality are at large variability, particularly with respect to the sample period covered. The subsample variability is particularly evident in papers using data that sets that includes 1980s. There are two important issues relating to the data used in these studies. First is the small sample size and second one is the structural changes or regime shifts. A further variability in the results is due to the handling of the trending properties of the data. The results using cointegrated models are mostly different than results of those ignoring the integration-cointegration properties of the data. This paper takes these three major issues into account by using bootstrap tests and rolling window estimation. In order to be robust against small sample and integration-cointegration properties of the data we use bootstrap causality tests. In order to show the subsample variability of the Granger causality tests and bring an explanation to variability we use subsample rolling bootstrap tests. The main objective of this paper is to investigate causal nexus between energy consumption and economic growth using a bootstrap causality test, applied to both full sample and rolling window subsamples, for G-7 countries. By causality, we mean causality in the Granger's sense (Granger, 1969). As we discussed above the previous literature on causality between energy consumption and GDP all tested for Granger causality in the full sample using several variants of the Granger causality test. In this study, we adopt a different approach and investigate the causality in a time-varying fashion using bootstrap Granger causality test. All type of Granger causality tests assumes (non)existence of one “causal” relationship for the whole sample. However, the causal relationship might be time-varying where a variable may be Granger causing the other in some periods and not in others, or there may be bidirectional causality in some periods. The results of the previous literature discussed above on the relationship between energy consumption and GDP seems sensitive with respect to changes in the sample period. When the causal relationship between two variables is not stable and the non-causality is not rejected, then it is not clear what has been rejected. Due to policy changes and the significant changes in the level energy consumption in response to changes in energy prices the causal relationship between energy consumption and economic growth may shift in time. Volatile periods during recessions may also be radically different than other periods. Ozturk (2010) points out to the conflicting results in the literature and need to consider alternative methods for analyzing the causal links between energy consumption and economic growth. Our results complement this conclusion. To investigate whether the causal relationships between energy consumption and economic growth is stable over the sample period, we consider bootstrap Granger non-causality tests computed from rolling subsamples of a fixed size (Thoma, 1994, Swanson, 1998 and Psaradakis et al., 2005). Using bootstrap rolling window approach, we analyse the time-varying causal links between economic growth and energy consumption, which is the main empirical contribution of this paper. The data used in this study are annual time series from 1960 to 2006 of total energy consumption and real GDP for G-7 countries, except Germany, for which the sample period covers 1971-2006. Total Energy Consumption is measured in kilotons of oil equivalency. We use parameter constancy tests to establish the stability of the estimated vector autoregressive (VAR) models. The evidence obtained from the stability tests show that energy consumption and GDP do not maintain a stable dynamic relationship for none of the G-7 countries. Therefore, the causality inference based on full sample VAR estimates is not reliable. When the full sample bootstrap Granger causality test is used, results of this paper indicate that there is only predictive power from energy consumption to economic growth for Canada. Excluding this finding, the findings from full sample estimation results indicate that there is no predictive content between the series for G-7 countries. Similar to the full sample results, results of the rolling window bootstrap tests show no evidence of consistent causal links between energy consumption and economic growth. The paper is organised as follows: Section 2 explains the methodology employed, Section 3 provides empirical results and the last section presents the conclusion.

Focusing on the energy consumption-economic growth relationship, this study investigated a sample of G-7 countries using the bootstrap rolling window approach. The data used in the study are annual data of total energy consumption and real Gross Domestic Product for G-7 countries. Different from the previous studies about the dynamic relation between energy consumption and economic growth series, in this study, we use both full sample bootstrap Granger causality tests and bootstrap rolling window tests with fixed sample size, which allows inference without considering whether the series are integrated-cointegrated. Besides, in order to determine the magnitude and the direction of the dynamic relationship between the series considered, we estimate bootstrap estimates of sum of the rolling coefficients. While the results of full sample bootstrap Granger causality tests show that energy consumption series lead economic growth for Canada, they show no causality links between economic growth and energy consumption for the remaining G-7 countries. In a similar manner, the results from the bootstrap rolling window estimation show no consistent causality between the series under consideration. For most of the sample points we do not find any causality between energy consumption and economic growth. There are, however, subperiods during which strong unidirectional or bidirectional causality is observed. Importantly, the periods of causal links are associated with various significant economic changes, such as the as the oil price hikes. This result indicates that the results from bootstrap rolling tests are not statistical artefacts and correspond to real economic changes. To sum up, the results of this paper show that there is no stable causal relation between energy consumption and economic growth. In this framework, although there are causal relations confined to certain periods, the causal relation does not apply for other periods. The result that the neither energy consumption nor economic growth leads each other consistently has implications for both economic policy and empirical modelling. Although neutrality hypothesis may hold for some periods and empirical tests may not find any causal link between energy consumption and economic growth, this should not be taken to hold consistently. Volatile periods due to energy price fluctuations, particularly that of oil, do behave radically different. The neutral link between energy consumption and economic growth are easily broken, particularly during recessions and oil price hikes or collapses. Therefore, energy conservation policies that do not have initial significant effects may later surprisingly have large negative impact on economic growth. Moreover, there may not be a significant pressure on energy demand during periods of low growth, but this may easily reversed, and insufficient supply may seriously limit economic growth. Our findings also points towards the uncertainty related to the impact of cuts in CO2 emissions brought by lower energy consumption. The impact of CO2 emission cuts will vary temporally and across countries. The dynamic links between energy consumption and economic growth is non-linear, asymmetric, and time-varying. The CO2 emissions cuts that are initially not detrimental to economic growth may be quite unfavourable in later periods. Linear models are not appropriate to analyze the lead-lag relationship between energy consumption and economic growth and researchers should consider models that can explicitly incorporate regime changes, such as the nonlinear models. The findings that the lead-lag relationship between energy consumption and economic growth are not permanent, volatile periods behave radically different, and that the relationship is likely to be non-linear, asymmetric, and time-varying suggests that time series analyses of energy consumption-economic growth that ignore these features may be misspecified. Moreover, the predictions of linear models that assume a permanent relationship may be highly misleading.