Understanding the Causality between Carbon Dioxide Emission, Fossil Energy Consumption and Economic Growth in Developed Countries: an Empirical Study

Issues on climate change have been recognized as serious challenges for regional sustainable development both at a global and local level. Given the background that most of the artificial carbon emissions are resulted from the energy consumption sector and the energy is also the key element resource for economic development, this paper investigated the relationship between CO 2 emission, fossil energy consumption, and economic growth in the period 1970–2008 of nine European countries, based on the approach of Granger Causality Test, followed by the risk analysis on impacts of CO 2 reduction to local economic growth classified by the indicator of causality degree. The results show that there are various feedback causal relationships between carbon emission, energy consumption and economic growth, with both unidirectional and dual-directional Granger causality. The impact of reducing CO 2 emission to economic growth varies between countries as well.


Introduction
Global climate change has been a hot topic since the end of the 20th century (see, for instance, [1][2][3]), and it is currently also a major concern and challenge confronting countries in the world [4], for example, the report of IPCC4 pointed that it was very likely (more than 90% of probability) that global warming was related to the increasing of greenhouse gases (GHG) over the past 50 years [5].Concerning climate change mitigation and adaption, the Kyoto Protocol is a very important step for reducing GHG emissions, and, moreover, broader participation and deeper cuts in GHG emissions are essential for any post-Kyoto agreement [6].However, the common recognition on reducing GHG emissions from the past several Conferences of Parties (COP) to the United Nations Framework Convention on Climate Change, such as the COP 15 in 2009 of Copenhagen, COP 17 in 2011 of Duban and COP19 in 2013 of Warsaw are not sufficient enough.The key issue for reaching a consensus on GHG emission reduction is the concern on its impact to local economic development [7], either in developed regions or in developing and emerging countries.For example, van Renssen explored the worries voiced within industry and argued that Europe must come up with effective climate, environment and energy policies that do not jeopardize economic competitiveness [8], Wang et al. [9] stated that China's economic growth and export trade had significantly promoted its carbon emissions, and the rapid economic growth is the main determinant that causes an increase in carbon emissions [10].Major emitters seem unwilling to accept binding emissions reductions targets as their economies have stagnated [11].Given the common background that most of the artificial carbon emission results from the energy consumption sector and that energy is also the key element resource for economic development [12], the causal relationship between carbon emission, energy consumption and economic growth should be investigated.This could offer better understanding on the interaction impacts to local sustainable development [13][14][15].
As for developing and emerging countries, such as China, India and Brazil, the question as to how to delink the energy consumption and carbon emission from the economic growth has been the key issue faced by policy makers and scientific communities.However, currently, in the face of international pressure to curb carbon releases as well as the tight domestic fossil-energy supply [16], relative indicators are usually applied for measuring carbon reduction target in order to avoid the possible negative impacts on local economic development [17].For example, a case study for Brazil shows that the business-as-usual CO 2 emissions in 2040 from energy use and industrial processes would be almost three times as high as those in 2010, while the current policy aims to lower emissions intensity of the overall economy by 36-39 percent by 2020 [18].Another analysis on carbon reduction in China indicates that the region with a lower reduction benchmark and higher decoupling elasticity value has more difficulty in achieving the emission reduction targets without powerful relevant policy [19], and a provincial level case study illustrates that with the rapid economic development, CO 2 emission in Jiangsu province rose from 1, 88 × 10 8 t in 1995 to 5, 20 × 10 4 t in 2009, with an average annual growth rate of 7.54% [20].Even the Chinese central government has stated that in 2020 the carbon intensity will be reduced by 40%-45% compared to the 2005 level [21].An early study that aimed to examine the long-run relationship between output and energy consumption during the period 1971-1999 in Malaysia indicated that energy use was positively related to output in the long-run, and offered strong support for causalities running from economic growth to energy consumption growth both in the short-run and long-run [22].Research in the case of South Africa on the effects of economic growth and coal consumption on CO 2 emissions over the period 1965-2008 also shows that a rise in economic growth increases energy emissions, and coal consumption makes a significant contribution to deteriorating the environment [23], which was also verified by Menyah and Wolde-Rufael's research [24].Under such circumstances, investigation and insights into the relationship between carbon emission, energy consumption and economic growth in developed countries could offer further understanding for developing countries.
As the typical representatives of developed and industrialized countries in this world, Europe plays a leading role in climate change mitigation and adaptation.Thus, investigating the causal relationship between the mentioned elements is significant and could have policy implications for developing countries.Actually, few related studies have been done: Ang investigated the dynamic causal relationships between pollutant emissions, energy consumption, and output for France by using co-integration and vector error-correction modeling techniques for the period 1960-2000, and the causality results support the argument that economic growth exerts a causal influence on growth of energy use and growth of pollution in the long run [25].Marrero took the advantage of a panel data set of 24 European countries in the period of 1990-2006 and applied a Dynamic Panel Data framework to examine the relationship of GHG emissions, growth and the energy mix in Europe, and found evidence for the existence of conditional convergence in terms of GHG emissions among the EU27 countries [26].The change in energy intensity of the main economic activities in period of 1991-2005 of the EU15 countries was also investigated by Marrrero and Ramos-Real in 2013, and the results revealed four different typologies for EU15 countries and showed the importance of identifying those economic activities which are key to reducing energy consumption [27].
Herein, based on the statistical data in the open database of World Bank, nine European countries, namely, Denmark, France, Greece, Italy, Netherlands, Portugal, Spain, Sweden and United Kingdom are selected to investigate the causal relationship between GDP growth, energy consumption and CO 2 emissions in the period of 1970-2008, and followed by the evaluation on the possible impacts of CO 2 emissions reduction to economic development by ranking the classes based on the causality degree results.

Methodologies and Results
Selected variables used in this study include: (1) GDP per capita (GDPP) which represents the local economic developmental level of studied countries, (2) CO 2 emissions per capita (PCO 2 ) which represents the contribution of economic activities to climate change, and (3) Energy Consumption Intensity (ECI) (fossil energy consumption per capita) which represents the fossil energy resources consumption affected by the type of the fossil energy mix and population.

Unit Root Test
Co-integration analysis introduces the idea that even if underlying time series are non-stationary, linear combinations of these series might be stationary.Therefore it is essential to verify that all variables are integrated of order one in levels before employing panel co-integration techniques.The main idea behind performing a unit root test while using ADF statistics is to ensure that the error term should be lagged independent [14].In this method, the unit root test is carried out by means of the following formulation [15].
Together with the fixed term regressions with trend, ADF statistics and McKinnon's critical values are derived.Where ∆Y t = Y t − Y t-i ,  is a drift term, and T is the time trend with the null hypothesis H 0 : ρ = 0 and H 1 its alternative hypothesis H 1 : ρ ≠ 0, n is the number of lags necessary to obtain white noise, and ε is the error term.Failure in rejecting H 0 implies that the time series is non-stationary.The unit root tests of all the three variables sets mentioned above for the nine countries are tested under differential levels of both first/second differences.From Table 1, we could conclude that most of the unit root hypothesizes have not been rejected when the variables are taken in various levels, but when the first difference was used, most of the hypothesizes of unit root non-stationary are rejected at the level 1% or level 5% of the significance, and few of the hypothesizes of unit root non-stationary are rejected under the second differences at level 1%.

Co-Integration Test
The co-integration test was employed in order to determine whether there exists a co-integration between the studied series.The aim of co-integration analysis is to model and estimate the long term correlation-ship among non-stationary time series.Herein, the Johansen co-integration procedure was performed for determining whether the co-integration exists.The first step in Johansen procedure is to determine the lag order.In this study, the significant and right lag period is finally selected after the numbers test from lag 1 to lag 10 and followed by the trace statistic and the maximum Eigen-value statistic (see the supplementary materials).The result shows that causality exists among the selected variables series; however, the co-integration test cannot tell the direction of the causal relationship.Thus, the following multivariate Granger-causality test based on the VECM was carried out to investigate both short-run and long-run causality.

Granger Causality
If the series X and Y are individually I( 1) and co-integrated as well, then the Granger causality tests could use I( 1) data because of the super consistency properties of estimation, whereas u t and v t are zero-mean, serially uncorrelated, random disturbances.The optimum lag lengths m, n, q and r are determined on the basis of Schwarz Bayesian and/or log-likelihood ratio test criterion: In the above equation, Y Granger causes X if H 0 : a 21 = a 22 = … a 2n = 0 is rejected, and H1: at least on a 2i ≠ 0, I = 1,2,3,…, n. and, correspondingly, in the following equation, X Granger causes Y if H 0 : b 21 = b 22 = … b 2n = 0 is rejected, and H 1 : at least on b 2i ≠ 0, i = 1,2,3,…, r: Results of the Granger-causality test for the 9 countries show that there are various feedback causal relationships between carbon emission, energy consumption and economic growth, with both unidirectional and dual-directional causalities (Table 2, and supplementary materials).In conclusion, (1) Granger Causality between energy consumption and CO 2 emissions: the null hypothesis of -ECI does not Granger Cause PCO 2 ‖ and -PCO 2 does not Granger Cause ECI‖ are rejected in countries of France, Spain and Sweden, but accepted in countries of United Kingdom, Italy, Portugal and Greece; while in Denmark, the null hypothesis of -ECI does not Granger Cause PCO 2 ‖ is rejected but that of -PCO 2 does not Granger Cause ECI‖ is accepted; (2) Granger Causality between economic growth and CO 2 emissions, the null hypothesis of -GDPP does not Granger Cause PCO 2 ‖ and -PCO 2 does not Granger Cause GDPP‖ are all accepted for most of the countries includes France, Italy, Spain, Sweden, Portugal, Netherlands and Greece; while for United Kingdom, the null hypothesis of -GDPP does not Granger Cause PCO 2 ‖ is rejected but that of -PCO 2 does not Granger Cause GDPP‖ is accepted; however, the situation in Denmark is opposite than that in the United Kingdom.
Comparing the differences based on the long term, conclusions are summarized as follows: (1) for France, Spain and Sweden, the causality relationship between ECI and PCO 2 is reciprocal relation, but has no causality relationship between GDPP and PCO 2 ; (2) for the United Kingdom, there exist non-causality relationship between ECI and PCO 2 , and GDPP is the Granger causality to PCO 2 while PCO 2 is not the Granger causality to GDPP; (3) for Italy, Portugal and Greece, there no causality exists between ECI and PCO 2, and between GDPP and PCO 2 .(4) for the Netherlands, ECI is not the Granger Cause to PCO 2 , but PCO 2 is the granger cause to ECI; in addition, there exists a non-causality relationship between GDPP and PCO 2 ; (5) for Denmark, ECI is the Granger Cause to PCO 2 , but PCO 2 is not the granger cause to ECI; also the same situation exists between GDPP and PCO 2 .

Table 1 .
Results of the unit root test in first difference.
a at level 5%; b tested in level 10%; c tested in 2nd difference.