The Relationship between Energy Consumption, Energy Efficiency and Economic Growth: The Case of China, Germany and Russia

(1)

The Relationship between Energy

Consumption, Energy Efficiency and

Economic Growth: The Case of China,

Germany and Russia

August 2008

Author Research Supervisor Methodology Instructor

M.G. Popescu Dr. C.J. Jepma Dr. H.W.A. Dietzenbacher

S1744607 Faculty of Economics Faculty of Economics Winschoterdiep 46, 9723 AC LR Landleven 5, 9747 AD Landleven 5, 9747 AD Groningen,The Netherlands Groningen,The Netherlands Groningen, The Netherlands

(2)

2

Abstract

(3)

3

1. Introduction

The four generally recognized factors of production in theory are land, labor, capital, and entrepreneurship (Encyclopedia of Business and Finance). Before the twentieth century only three of the four factors were recognized as making up the ‗classical triad‘ land, labor and capital with entrepreneurship being introduced only later. Recently, specialists have begun to argue that it would be high time for another addition to the list of production factors. The new factor under discussion is energy consumption and its effects on economic growth. Therefore this paper, in its first part, discusses the causality relationship between energy consumption and economic growth. It analyzes whether economic growth is caused by energy consumption and if in turn economic growth determines energy consumption on both a short and a long term perspective and how the causality between energy consumption and GDP can influence energy policy. The topic of energy and GDP causality has come to the attention of energy economists for a number of years now. The interest in the subject has been stimulated by two major factors. The first factor is the unprecedented oil price increase that started in the early 1970s, which substantially increased the energy bill in oil-importing countries and secondly the very real threat of global warming. Hence, the increasingly pressing need to cut energy consumption and reduce emissions to help stem climate change sprung up. Despite strong reasons both economical and ecological to reduce energy consumption some concerns regarding the adverse side-effects of adopting conservation policies on economic growth have been raised. Therefore due to their complex implications, strong measures andcommitment towards energy conservation have yet to gain momentum.

As inferred by Ghali & El-Sakka, 2004 and Wolde-Rufael, 2005 whether the energy consumption of a country stimulates or not its GDP has been an ongoing debate among specialists. On one side there are those like Lee & Chang 2007 who argue that energy is as vital an input for GDP as all other factors of production such as labor and capital. Therefore, energy is an essential requirement for economic and social development and should be considered a potentially ―limiting factor to economic growth‖ Ghali & El-Sakka, 2004. On the other hand, due to the very small proportion energy holds in GDP it is argued that its impact on economic growth should not be of grate significance. This approach is described as the ―neutral impact of energy on growth‖ Ghali & El-Sakka, 20041

Summing up, is it vital to established whether energy consumption is a stimulus to economic growth or not? The answer to the above question holds major implications for policy makers. On one hand if energy determines GDP then reducing the energy consumption could lead to

1

(5)

5

economic slowdown and even regression. Whereas on the other hand if economic growth is not caused by energy consumption then energy conservation policies may be initiated without fears of plummeting economic growth and employment Masih & Masih 1997. Therefore it isessential to establish whether there is indeed a causal link running from energy consumption to GDP. This is also supported by the current debate on global warming and the need to reduce Greenhouse Gas Emissions and conserve energy, given that any limitations put on energy consumption to help reduce emissions will effect economic growth and development if causality between energy and GDP exists.

The seminal paper discussing causality running in the opposite direction was written by Kraft and Kraft 1978. The authors employ bivariate causality procedures on data for the period 1947-1974 and find evidence of causality running from GNP to energy consumption in the US. Ever since, a number of studies on the matter have proliferated using different techniques, sample and time periods.2

The importance of this issue being given, it comes as no surprise that many studies have been attempting to quantify the energy-GDP relationship for different countries. Chontanawat, Hunt & Pierse 2006 summarize a comprehensive list cataloging the efforts and investigations made by economists and econometricians in the debate to establish the causality between energy and GDP. The authors highlight that the results are mixed and a clear consensus fails to emerge. However differences in results are expected due to the ―many institutional, structural, and policy differences‖ Masih & Masih 1997 . What comes as a surprise tough are the different results obtained for counties with very similar characteristics and sometimes even for the same country. This lack of consensus is explained as argued by Masih & Masih 1997 is due to the differences in methodology (variable specification, the techniques and lag structures employed.)

In order to estimate the effect energy conservation policies might have on economic growth more accurately, the direction and strength of causality between the two variables needs to be tested. If unidirectional causality runs from energy consumption to economic growth then the implementation of conservation policies threatens to reduce economic growth. On the contrary, if unidirectional causality runs in the opposite direction from economic growth to energy consumption,then conservation policies may be introduced with little or no unfavorable impacts on economic growth. A lack of causality in either direction implies that energy consumption has no effect on economic growth. Mahadevan & Asafu-Adjaye 2006. The direction of causality is especially important because once revealed, reasonable energy conservation policies can be implemented without hampering economic growth. On the other hand, if the causality relationship between energy and GDP proves elusive designing energy conservation policies becomes more difficult in itself and implementing these policies might prove ill inspired on

2

(6)

6

behalf of the country‘s economy. One important thing to bear in mind is that innovative technologies might encourage energy savings and at the same time encourage economic growth. This paper therefore attempts to address the issues of energy-GDP causality, energy efficiency and conservation by studying the cases of three major players on the global energy market namely Russia, China and Germany. The former plays a major role in energy supply and maybe more importantly energy reserves, the second is responsible for the world‘s second largest and fastest to increase energy demand while the ladder contrast and makes a good case for comparing a major player in the global economy that is completely dependent on energy imports.

After having discussed the nature of the relationship between energy consumption and economic growth and the effects that reducing energy consumption might have on economic growth, this paper discusses how the same energy can be used more efficiently to support economic growth. In order to do so we analyze and discuss the levels of energy efficiency and some of the major factors that determine it. Energy efficiency as defined by The World Development Indicators 2007 is the amount of energy expressed in tons of oil equivalent is used to produce 1 unit of USD of GDP.

Dubbed in some specialist‘s circles as ‗the fifth fuel‘ it is argued that energy efficiency is able to cover the growing global demand for energy just as good as oil, gas, coal or uranium. Moreover, in present times when environmental consciousness is raising energy efficiency has ‗been climbing the rankings‘. While the classic burning of fossil fuels discharges greenhouse gases into the atmosphere thus contributing to global warming and nuclear plants produce hazardous waste that are life-threatening, the byproduct of energy efficiency comes only in the form of wealth due to lower fuel bills, smaller environmental problems and so forth.

(7)

7

similar point, they argue that if made properly energy efficiency investments would allow a significant cut in emission without impending the economical growth.

In conclusion, the investments made in energy efficiency would not only pay for themselves but would actually prove profitable. Although these implies a healthy sum to be invested, by MGI‘s calculations $170 billion yearly until 2020, this only represents 1.6% of the present annual investment in fixed capital. Furthermore with significant profits on the perspective financing these investments should prove manageable.

The prestigious publication ‗The Economist‘ takes it upon itself and addresses in a recent article the issue of energy efficiency. A question is raised that despite the numerous opportunities and benefits promised by energy efficiency how come investors are not taking advantage of them? According to Art Rosenfeld of the California Energy Commission to some extent they are. At a global level the world energy intensity – the amount of energy used to generate one dollar of GDP- is falling by around 1,5% each year. The example of America is given where before the oil shock of 1973 this variable was falling by a mere 0.4% each year had this been the continuing trend today the USA would be spending 12% of its GDP on energy instead of sent 7% its spending in present.

However MIG points out that there are still hundreds of billions of dollars worth of potential opportunities in energy efficiency. Specialists argue that the main answer to the questions what is holding investors back is prices. To many consumers energy often is too cheap to be worth saving, especially in countries that indulge energy subsidies. For example Russian industrialists can afford to be wasteful when using natural gas mainly because gas sells in Russia at a fracture of its international price whereas in Qatar drivers have little incentives to save up on petrol when they pay just over a dollar per barrel for it. Therefore it comes as little surprise that energy is used more efficiently in countries where it has bigger prices hence it is not by chance that Denmark has high energy prices and high energy efficiency levels. George David of United Technologies states that high fuel and power prices are the only things needed to achieve energy efficiency.

Summing up, this paper will analyze the relationship between energy consumption and economic growth. Having established the nature of this relationship and the role it plays in the adoption of energy conservation policies, this paper goes a step forward and analyzes not just the effects of reducing energy consumption by energy conservation policy but how efficient countries use the same amount of energy to produce economic growth and finally discusses some of the major factors that determine the levels of energy efficiency.

(8)

8

endowment and completely dependent on energy imports. The choice of countries was made according to their energy resource endowment and tries to cover the whole spectrum by choosing Russia, on one hand on of the richest in energy resources country in the world, on the other hand Germany situated close to the opposite end of the spectrum and somewhere in the middle is China, relatively well endowed from an energetic point of view but whit an energy hungry economy.

2. Literature Review

In the past decades following the two energy crises in 1974 and 1981, the energy consumption—economic growth relationship has become the focus of numerous empirical studies using either income or employment as a proxy for the latter. Such studies are Erol & Yu 1987, Masih & Masih 1996, Asafu-Adjaye 2000, Morimoto & Hope 2004, Lee 2006, Lee & Chang, in press-a and Lee & Chang, in press-b). The seminal article on this topic was published in the late seventies by Kraft & Kraft (1978) who used data for the 1947–1974 period and found evidence of unidirectional causality running from GNP to energy consumption in the United States.

Investigating the causality relationship running in the opposite direction from energy consumption to economic growth Akarca and Long 1979 use month to month data for the period of 1973–1978 in the US and found that energy consumption does Granger-cause economic growth reflected here by rate of employment with a causality coefficient of −0.1356. Similar results obtained by Stern 1993, Stern 2000, Oh & Lee 2004, Ghali & El-Sakka 2004, Wolde-Rufael 2005 and Beaudreau, 2005 support the existence of a causality relationship running from energy consumption to economic growth and indicate that energy is indeed an essential factor of production.

However, these studies have been challenged by the works of Akarca & Long 1980, Erol & Yu 1987, Yu & Choi 1985, and Yu & Hwang 1984 who found no trace of causality between energy consumption and economical growth (proxy by income). Evidence supporting the ‗neutrality hypothesis‘ was presented by Erol & Erol, Yu & Jin 1992, Ghali& El-Sakka 2004 and Yu et al. 1988 who found neutrality in the nature of the causality relationship between energy consumption with respect to economic growth (proxy by employment)

(9)

9

First, when considering the perspective Lee & Chang 2007 argue that the majority of previous studies have investigated the energy - GDP relationship from one of two perspectives. The first approach to this relies on a demand perspectives or energy demand function whereas the second approach is based on a production perspective that deals with an aggregate production function. For the first approach Oh & Lee 2004 highlight the necessity to use of a tri-variat model with energy, GDP and energy price, proxied by the consumer price index (CPI) as variables. On the other hand, the production perspective implies a multivariate model consisting of energy, GDP, capital stock and labor. Making exception of the works of Masih & Masih 1998, Asafu-Adjaye 2000 and Fatai et al. 2004 who approach the issue from the demand side perspective, most of the other studies approach the causality between energy consumption and economic growth from a production side perspective. The majority of the works are focused either a small data sample or just a singular country in example we have the work of Stern 1993 and Stern 2000 and Oh & Lee 2004

Second, recent studies aimed at the energy and GDP causality relationship have empathized the distinctions found between developed and developing countries. Chontanawat, Hunt and Pierce 2006 after analyzing data from over 100 countries conclude that when compared to the developing (non-OECD) countries the causality relationship from energy to GDP is stronger in the developed (OECD) countries. Mahadevan & Adjaye 2006 findings corroborate however they argue that while the elasticity of economic growth in response to an increase in energy consumption is greater in developed countries then in the developing ones, its income elasticity is much lower and also sub unitary and call for a more holistic approach to policy building. Third, there are studies that distinguish between energy exporting and energy importing countries. So far empirical testing for the energy–GDP causality relationship in oil-importing countries has yielded mixed results in so doing adding to the confusion about the effects of energy conservation policies. The mixed findings are most likely a result of different methodologies and the use of different data. Less attention has been paid to the to the energy-GDP relationship in the oil-exporting countries. Al-Iriani 2006 uses recently developed panel cointegration and causality techniques to reveal the direction of energy–GDP causality in the six countries of the Gulf Cooperation Council (GCC). Mahadevan & Adjaye 2005 use a panel error correction model to analyze data collected from 20 net energy importers and exporter between 1971 and 2002. Their findings indicate that in the case of energy exporters there is bidirectional causality between economic growth and energy consumption in the developed countries in both the short term as well as long term, while for the developing countries causality is found only in the short run. While former result is also found to be true for energy importers, the latter result is valid solely for developed countries within this category.

(10)

10

consumption and economic growth. For example Al-Iriani 2006 uses newly developed techniques of panel unit root test, cointegration and causality. The authors argue that by using these new methods, the low power associated to traditional unit root and cointegration tests is avoided. Other studies also support the higher power of panel-based test over those based on individual series. Perron 1991 suggest that the cointegration test‘s power is influenced significantly by the data span and that the use of panel cointegration allows for heterogeneity between panel members. Pedroni, 2000, Pedroni 2001 and Pedroni 2004 highlight the undesirable effects of wrongly imposing homogeneity across panel members. We can find his approach being applied with success by Harb (2003) when testing for cointegration among variables of the money demand function in the six Gulf Cooperation Council (GCC) countries.

Lee & Chang 2007 also use the newly developed panel unit root, panel cointegration and panel error correction model techniques to investigate the causality relationship between energy consumption and economic growth. The authors argue against the studies that use OLS and do not test for stationarity. This argument is also supported in the works of Granger & Newbold 1974 who highlight the fact that ignoring non-stationarity issues may lead to deceptive conclusions on the relationships between variables. Furthermore when dealing with cointegration between series Asafu-Adjaye 2000 suggest the need for an alternative and superior method to the vector autoregressive VAR method. An alternative technique is employed by Lee & Chang 2007 who use the VECM error correction model that is able to distinguishing between short-run and long-run relationships amid variables in so doing they are able to identify the sources of causation Oh and Lee 2004b.

Methodologies also vary in the structure of their model for example Al-Iriani 2006, Mahadevan & Asafu-Adjaye 2006 and Chotanawat, Hunt & Pierse 2006 all use a bi-variant model with variable such as energy and GDP (energy-GDP). Energy price is used as additional in Asafu-Adjaye 2000‘s tri-variat model (Emergy-GDP-Price) Multivariate models checking also for labor and capital are employed by Lee & Chang 2007 (E-GDP-Labour-Capital)

(11)

11

On the other hand the literature on the energy efficiency topic is less extensive than that on the prior subject and most of the cases analyzed were related to the energy-GDP causality issue. To the best of our knowledge a similar model to the one proposed by this paper has not yet been built. However the issues approached in the model are separately discussed in the work of several authors.

First, energy abundance was initially discussed in the paper of Jeffrey Sachs and Andrew Warner ‗Natural resource abundance and economic growth‘ Sachs and Warner 1995 where they argue that energy abundance measured as the ration of natural resources (mainly energy) to GDP had a negative effect on the economical growth rates of the countries studied between the subsequent period between 1971 and 1989. Papyrakis and Gerlagh 2004 address the direct and indirect effects of resource abundance (especially energy resources) on economic growth, investigating the causes of mediocre performance of resource rich countries.

Secondly, energy imports are discussed by Oya S. Erdogdu. In their study the authors employ Granger causality tests to estimate the relationship between total energy consumption, imported energy and economic growth (gross domestic product, industrial production index) in Turkey. In their results they conclude that the total and the imported quantity of energy resources do effect economic growth, however the provenience of energy resources, whether they are of foreign or domestic origin has little influence on industrial production.

Finally the influence bared by the size of the industry sector on energy efficiency is discussed by Rosen and Houser (2007) who point out in their paper that energy consumption in China between 1978 and 2000 exceeded four times its predicted values amounting in 2006 to over 15% of the global energy demand). The authors argue in their work that the growth of the industry sector is the main reason for the increase in energy consumption this having a negative effect on energy efficiency. Furthermore Wang Yanjia (2006) argues that the most of the main industrial products in China have a considerably higher energy intensity compared to the developed countries3. The author‘s measurements refer to energy used per unit of physical output and given the results he concludes that there is plenty of room to improve the now low energy efficiency levels.

In addition after researching the literature on energy price subsidies we find that it fails to offer a comprehensive definition of what a subsidy constitutes of. The most clear-cut definition describes a subsidy as being a direct payment to either energy consumers and/or producers. However, beside a payment of this nature there are many other forms in which the state can support the local energy sector. For example $50 billion is spent by public policies to support the local coal industry. Similar methods are also used in Germany where the promotion of price regulations and purchase obligations has been gaining momentum Moor, 2001. The author argues that the wide pallet of methods used to subsidize the energy sector spreads from

3

(12)

12

market instruments like direct payments to command and control measures regarding quotas and consumption. On the other hand, in a study conducted by the OECD energy subsidies are described as any measures or policies intended to keep the energy prices below the market level OECD, 1998b. A more comprehensive classification is found in Von Moltke et al. (2004) here the system used classifies subsidies as following: (a) direct financial transfers, granting preferential loans whit lower interest rates to producers or consumers; (b) preferential tax treatments, constituting in or exemptions on various taxes or tariffs like royalt y or sales taxes, tax credits, or an accelerated depreciation policy; (c) trade restrictions, similar to quotas; (d) the energy-related services provided by the government at a lower price than the usual such as a direct investment in energy infrastructure; (e) the regulation of the energy sector— market access restrictions, price controls and demand guarantees Von Moltke et al. (2004).

3. Methodology

3.1 The Causality Relationship between Energy Consumption and Economical Growth

Most of the previous studies addressing the causality relationship between economic growth and energy consumption were based on the Granger-causality method (Granger, 1969) (Granger, 1969). The studies approach vary significantly, while some analyze the datasets for only one or a few countries others consider over a hundred different datasets. Furthermore, although Granger-causality represents the core of the analysis, as mentioned before in the literature there is a wide spectrum of different methodologies employed. Chotanawat, Hunt & Pierse 2006 argue that the different approaches to the mater of energy to GDP causality is explained mostly by the development of new econometric techniques. Furthermore the authors provide a classification of the studies according to the methodologies they employ:

• First are the studies that use the classic methodologies developed by Granger (1969) and Sims (1972), most of which are focused on developed countries with a specific interest in the United Stated covering the period between 1940 and 1980.

• The second group are the studies that use cointegration and the error correction techniques (Granger, 1988); these studies have a wider focus on both developed and some developing countries covering the period between 1950 and 2000.

(13)

13

referred to as the Hsiao (1981) technique and it is employed in studies on the USA, Latin America and several Asian countries covering the period between 1940s to the early 2000.

The Granger-causality method predicts causality rather than approaching it in a structural sense. It starts with the premise that ‗the future cannot cause the past‘; if event A occurs after event B, then A cannot cause B (Granger, 1969).Furthermore, event B causes event A if the current values of A can be better predicted by using past values of B then by not doing so. Therefore, in order to test the causality between energy and GDP the following bivariate equation is use:

ΔYit = α1 + ∑βi k ΔYit-k + ∑γi k ΔEit-k + εit (1)

ΔEit = α‘1 + ∑β‘ik ΔEit-k + ∑γ‘ik ΔYit-k + εit (2)

where Et = ln(ECt); Yt = ln(GDPt); ECt is the total energy consumption expressed in Kt of oil equivalent; GDPt the real GDP expressed in USD 2000; i indicates the country, t indicates the year, k represents the optimal lag number and ΔEit-k and ΔYit-k are log(Et/Et-k) and log(Yt/Yt-k).

Here the presence of Granger-causality is determined by the significance of the ΔEt-k in equation

(1) and ΔYt-k in equation (2).

This paper follows established procedures thus the causality relationship test between energy consumption and economic growth is carried out in three stages. First the order of integration in the energy consumption and GDP series is tested for. To do so we are going to check for the stationarity of the variables using the procedures employed by Maddala and Wu (1999) who propose a straightforward, nonparametric unit root test and advocate the use the augmented Dicky-Fuller (ADF) and Philip-Perron (PP) tests. In addition to this due to the fact that in this study we analyze data stretching from 1970 to 2005 we also employ panel unit root tests in order to correct for the relatively short time span of the data to establish the stationarity status of the series in an unquestionable fashion.

(14)

14

(Y) by using the Johansen4 technique. If co-integration is found it will suggests that over the long run the two variables move in the same direction.

We proceed to augmenting equations (1) and (2) as follows:

ΔYit = α1 + ∑λik ECTit-1 + ∑βik ΔYit-k + ∑γik ΔEit-k + εit (3)

ΔEit = α‘1 + ∑λ‘ik ECTit-1 + ∑β‘ik ΔYit-k + ∑γ‘ik ΔEit-k + εit (4)

ECT - error correction term from a cointegration equation of the form yt = βet + Et

In both Eq.3 and Eq.4 the presence of Granger-causality can have two different sources. First there is a short-term causality effect determined by the significance of the ΔEt-k in Eq.3 and ΔYt-k

in Eq.4. This effect is considered transitory and its presence can be established by testing the hypothesis H0: γi=0 in both equations. Secondly long term causality between energy

consumption and GDP is determined by the significance of the speed of adjustment λi the error

correction term‘s coefficient. The long-term relationship in the cointegrated process is established by the significance of λi therefore moves along this path can be considered

permanent. For long-run causality we test the hypothesis H0: λi =0 in both equations. Finally we

test for the variables short-run adjustments in order to re-establish long-term equilibrium. To do so we use a joint Wald F test on the error correction term and γi coefficient in (Eq.3) and βi

coefficient in (Eq.4)

3.2 The Energy Efficiency Model

After establishing the nature of the causality relationships between energy consumption and economic growth in the three countries, the paper then moves to examine and compare the trends in energy efficiency for each country over the studied period. To do so the evolution the evolution of energy used to produce 1 USD of GDP in each country is analyzed.

EEit =ECit / GDPit

4

(15)

15

Notes: EE represents the variable for energy efficiency, EC stands for total energy consumed expressed in Kt of oil equivalent and GDP is gross domestic product expressed in constant 2000 US$

The paper goes a step further and discussed some of the most important factors that influence and determine the level of energy efficiency in each country of the countries. By using cross-sectional time series from the World Development Indicators 2007 four major indices of the country‘s economy are being investigated.

The first issue addresses the abundance of energy in each country. To achieve this, the energy production capabilities of these countries expressed in volume of energy produced is analyzed. Energy production refers to forms of primary energy--petroleum (crude oil, natural gas liquids, and oil from nonconventional sources), natural gas, solid fuels (coal, lignite, and other derived fuels), renewable combustible, waste and primary electricity all converted into forms measured in Kt of oil equivalent. This is then compared to the actual size of the economy, GDP expressed in constant 2000 USD. A first hypothesis would be that scarcity determines high energy efficiency while abundance tolerates a more wasteful approach in both energy production and usage. Namely high energy abundance will have a negative effect on the energy efficiency level. To be noted is that after empirically testing it was established that the best results are obtained when using the formula below to describe energy abundance. Thus a low value of the EA variable suggests actually increased energy abundance whereas high values of the EA variable indicate low energy abundance.

EAit = GDPit / EPit

Notes: EA stands for energy abundance, GDP is gross domestic product expressed in constant 2000 US$ and EP is energy production measured is Kt of oil equivalent

(16)

16

IND_GDPit = INDit / GDPit

Notes: IND_GDP is the percentage held by industry sector in a country‘s GDP, GDP is gross domestic product expressed in constant 2000 US$ and IND is Industry value added expressed in constant 2000 US$

Furthermore, this paper discusses the issue of external energy dependency of each country. A country that is able to cover all or most of its energy demand with energy produced at a national level finds itself in a strong position where it has significant influence on the cost and hence price of energy in that country.

It is to be expected that an increase in dependency on external, non national sources or even a decrees in energy independency will encourage a boost in energy efficiency. Thus the more energy a country has to import the more carful will it have to be with its usage in so doing achieving a higher level of energy efficiency. To support this theory I want to analyze the evolution of energy imports in the three countries by using data from WDI 2007. One thing to have in mind here is that some countries produce considerable amounts of energy that are destined for exports but this differentiation lies outside of this paper‘s objectives thus the formula below is used to express energy imports:

EIit= ECit- EPit

Notes: EI is energy imports expressed in Kt of oil equivalent, EC stands for total energy consumed expressed in Kt of oil equivalent and EP is energy production measured is Kt of oil equivalent

Combining the above mentioned variables the following model is built:

EEit= αi + βiEAit + γi EIit + λi IND_GDPit + εit

We proceed to subjecting this model to several statistical tests. First, because panel data implies both a time dimension and a cross-sectional element, testing for autocorrelation and heteroskedasticity is imperative, to do so we employ the Durbin-Watson test and the White Heteroskedasticity test. Furthermore we test the model for multicollinearity by building a correlation matrix. Afterwards we use the Jarque-Bera statistics to evaluate the model‘s normality. Finally we use the Chow Breakpoint test to check for structural breaks.

(17)

17

build a variable and include it in our model. Despite this shortcoming the literature written on this topic does allow us make a theoretical point and to argue that in China energy prices are low mostly because of the low cost of coal extraction and production rather than price subsidies. On the other hand, in Russia gas prices are low due to heavy subsidising and despite international pressure on Russia to liberalize its energy prices it can be argued that if gas prices go up the economy will probably redirect itself towards the use of cheaper but more environmental hazardous coal resources.

4. Data and Results

This study analyzes data covering the period between 1970 and 2005. The length of the period studied is influenced by the availability of data on the variables used. Three different countries with contrasting energetic profiles are selected. The first one is Russia whit a rich endowment of energy resources and a major energy exporters, at the other end of the spectrum lies Germany with a poor resource endowment and completely dependent on energy imports and finally we analyze the case of China that despite its fairly rich endowment still struggles to cover its own energy demand. The data gathered comes from several sources however the biggest and most comprehensive such source is the 2007 World Development Indicators database. The difference in real GDP series (constant 2000 US$) is used as a proxy for economic growth while energy consumption is embodied by energy use expressed in Kt of oil equivalent. The rest of the series used were already described in the methodology part. It must be stated that the data used for Russia before 1988 refers actually to the USSR and was collected from ‗Energy Statistics and Energy Balance for the Soviet Union‘ OECD IEA, Paris 1991 a dummy variable is used to test for this.

4.1 Granger Causality Model Stationarity

(18)

18

In the table below we present both the traditional unit root tests as well as the newly developed panel unit root tests. Amongst the results we include statistical value and probability both in level and first difference.

The results from the unit root tests for both the full sample and specifically for each country are presented in the table below. For both the variables in either the general case or in each country‘s case, the null hypothesis of the unit root test is not rejected at their levels. However, when employing the first difference the null hypothesis can be rejected in most of the cases even at a 1% significance level. In conclusion, it can be stated that all the series are non stationary and integrated of order one I(1). It must also be said that in the table below there appear to be some isolated exceptions in the results, which however are easily corrected by employing second difference.

Table 1: Unit root test and Panel unit root test

Country Var Test Level 1st difference

Unit root test Panel Unit root test Unit root test Panel Unit root test

Statistic Prob. Statistic Prob. Statistic Prob. Statistic Prob.

General Y ADF -1.108896 0.7101 7.50164 0.2769 -5.598134 0.0000 16.6604 0.0106 PP -1.217946 0.6646 4.09073 0.6644 -5.832925 0.0000 23.4909 0.0006 E ADF 1.354594 0.9987 9.13362 0.1662 -4.469152 0.0005 19.8498 0.0029 PP 1.269739 0.9984 4.12854 0.6593 -4.231869 0.0011 19.2505 0.0038 China Y ADF 0.797637 0.9925 4.24324 0.1198 -3.010646 0.0442 4.40683 0.1104 PP 1.701967 0.9995 2.62476 0.2692 -3.859032 0.0057 8.82559 0.0121 E ADF 0.616638 0.9880 3.91332 0.1413 -2.455545 0.1355 0.45423 0.0968 PP 0.454990 0.9824 1.18664 0.5525 -2.455545 0.1355 1.97926 0.0717 German y Y ADF -0.928577 0.7663 1.46322 0.4811 -3.973696 0.0044 9.12112 0.0105 PP -1.358365 0.5907 0.24696 0.8838 -3.931439 0.0049 7.60247 0.0223 E ADF -3.650920 0.0099 5.21191 0.0738 -4.104660 0.0034 11.4160 0.0033 PP -3.248462 0.0257 2.94088 0.2298 -5.004253 0.0003 15.4286 0.0004 Russia Y ADF -2.421631 0.1436 1.79518 0.4076 -3.759500 0.0074 3.13243 0.0488 PP -2.252377 0.1925 1.21901 0.5436 -3.759500 0.0074 7.06281 0.0293 E ADF 0.317207 0.9681 0.00839 0.9958 -2.562052 0.0370 7.97956 0.0185 PP 1.064634 0.9957 0.00102 0.9995 -2.380257 0.0593 1.84266 0.0980 Cointegration

Having established that GDP and energy consumption series are integrated of order one we move to test the existence of a long-term relationship between them. Johansen procedure is employed to establish the cointegration relationship between energy consumption and GDP in each of the three countries. A summary of these test results is presented in the below table. The table reports in its second column (H0) the number of cointegration relations under the null hypothesis, it the

(19)

19

level. The results suggest a rejection of the null hypothesis that states that there is no cointegration at a 5% significance level for all of the three countries analyzed. Therefore a conclusion that a cointegration relationships exists between our variables and that over long term periods energy consumption and GDP move in the same direction in all three countries.

Table 2: Johansen‘s test results for cointegration between energy consumption and GDP Country H0 Trace 0.05 Crit.val. Prob. l-Max stat. 0.05 Crit.val. Prob.

China None 15.89661 15.49471 0.0435 14.92661 14.26460 0.0392 At most 1 0.970006 3.841466 0.3247 0.970006 3.841466 0.3247 Germany None 28.50406 15.49471 0.0003 27.56080 14.26460 0.0002 At most 1 0.943258 3.841466 0.3314 0.943258 3.841466 0.3314 Russia None 44.15993 15.49471 0.0000 41.63101 14.26460 0.0000 At most 1 2.528926 3.841466 0.1118 2.528926 3.841466 0.1118 Causality relationship

Once established the cointegration relationship between the function‘s variables, following the example of Lee & Chang 2007 we move to perform a panel-based error correction model to analyze the short-term and long-term causality relationship between energy consumption and economic growth as well as the direction of these relations. In order to accomplish this, the two-step procedure developed by Engle and Granger (1987) is being employed. First we estimate the long-term model for equations (1) and (2) so that we obtain the estimated residuals (error correction term – ECT). The second step in the procedure is to estimate the Granger causality with a dynamic error correction model in equations (3) and (4). Because of the correlation that might arise between the dependant variables and the error term in the panel data model an instrumental variable estimator must be used. After careful examination and numerous empirical attempt s it is found necessary that the classical assumption that the error term has the lag 3 (k=3), these findings being in accordance also with those of Lee & Chang 2007.

(20)

20

In table 3 the short term results are determined by analyzing the lagged ΔE and ΔY dependant variables in equations 3 and 4 however they are irrelevant, a Wald F test like the one employed in table 4 is required to establish the short term causality relationship. On the other hand the results presented in the long term section refer to the t-statistics values of the lagged ΔE and ΔY dependant variables equations 1 and 2. Although the long term results would have been considered significant earlier, newly developed error correction model techniques use Wald F test to better estimate the causality relationship, therefore we use the results presented in table 4 to discuss the causality relationship both in the short term as well as in long term.

Table 3: Panel causality test results t-statistics

Country Variables Short-term Long-term

Δ Y Δ E ECT Δ Y Δ E General Δ Y - -0.31 (-17.41)*** 0.54 (19.30)** - -0.09 (-0.64) Δ E -0.19 (-7.31)*** - 0.19 (7.23)** 0.02 (0.53) - China Δ Y - -0.07 (-1.05) 0.49 (7.00)** - -0.06 (-0.60) Δ E -0.28 (-2.73)** - 0.37 (3.50)** 0.07 (1.92) - Germany Δ Y - 2.77 (4.93)** -5.49 (-5.29)** - 0.22 (-0.73) Δ E -0.47 (-5.39)*** - 0.48 (5.36)** 0.01 (0.48) - Russia Δ Y - -0.24 (-7.31)*** 0.56 (9.06)** - 0.19 (2.72)** Δ E -0.12 (2.53) - 0.13 (-2.08)** 0.18 (4.62)*** -

Note: t-Values are in parentheses and ** indicates statistical significance at the 5% level,

(21)

21

Table 4: Panel causality test results f-statistics

Country Variables Short-term Long-term

Δ Y Δ E ECT Joint (ECT and

Δ Y) Joint (ECT and Δ E) General Δ Y - -0.31 (303.1694)*** 0.54 (372.5588)*** - (187.6480)*** Δ E -0.19 (53.47012)*** - 0.19 (52.37382)** (26.73541)*** _- China Δ Y - -0.07 (1.106881) 0.49 (49.09864)*** - (27.71433)*** Δ E -0.28 (7.500430)** - 0.37 (12.25797)*** (7.349544)*** _- Germany Δ Y - -0.11 (7.535660)** 0.48 (67.00327)*** - (33.64184)*** Δ E -0.47 (29.12418)*** - 0.48 (28.81279)*** (14.60807)*** _- Russia Δ Y - -0.24 (13.95998)*** 0.56 (82.16806)*** - (41.09209)*** Δ E -0.12 (4.351890) - 0.13 (6.450474)** (5.663727)** _-

Note: f-Values are in parentheses and ** indicates statistical significance at the 5% level,

additionally ***indicates statistical significance at the 1% level.

Looking first at the short-term results for China (Table 4) we find that despite E‘s insignificant in the GDP equation (Eq.3), Y is significant in the energy equation (Eq. 4) suggesting in the short run the presence of unidirectional granger causality from GDP to energy. Looking at the coefficients of ECT in both equations we find them to be positive and significant meaning that given a deviation of one of the variables from the long run equilibrium relationship the other variable would react to restore that equilibrium. In the last two columns, the results from the Walt F test suggest that in the long run there is a bidirectional granger causality relationship between energy consumption and GDP.

In the case of Germany (Table 4) it can be deducted from the income equation (Eq. 3) in both the short-term and long-term energy Granger-causes income. Furthermore the energy equation (Eq.4) results indicate that also income Granger-causes energy consumption in both the short and long run thus it can be concluded that a bidirectional causality relationship exists between energy consumption and economic growth in Germany.

(22)

22

relationship over the long term (table 3) such a relationship being revealed only by the significant coefficients of the ECT, in the case of Russia a bidirectional causality relationship between energy consumption and economic growth is suggested by both the f-statistic and t-statistic results.

The results of this paper concur with those of Masih and Masih (1997), Hwang and Gum (1992) Glasure and Lee (1997) and Asafu-Adjaye 2000. The first three found bidirectional causality between energy and GDP in the cases of South Korea and Taiwan. Furthermore these studies refute the neutrality hypothesis between energy and income proposed by Erol and Yu in respect to the US. The findings of the forth paper also suggest bidirectional causality between energy and income in the cases of India, Indonesia, Thailand and The Philippines. In addition, Chotanawat, Hunt & Pierse 2006 in their study on 100 countries find a causality relationship between energy and GDP in the case of Germany however they do not specify the direction nor weather if its long term or short term.

In conclusion there seem to be, on a long term perspective a bidirectional causality between energy consumption and GDP in all the three countries studied, whereas in the short run a unidirectional causality relationship can be found for China and Russia and bidirectional causality in the case of Germany. These results indicate that policy makers should strongly consider economical effects when designing energy policy. Furthermore, given the tight relationship between energy consumption and economic growth, the implementation of restrictive policies alone may not be opportune and might prove to bare negative effects on economic growth. Therefore given the conditions that a growing economy cannot afford to consume the same amount of energy it used to let alone more, it is imperative to the economy‘s well being that it manages to use the same energy more efficient. This paper discusses in its next section the levels of energy efficiency in every country and the factors that determine it.

4.2 Energy Efficiency Model

The three hypotheses regarding energy efficiency and its determinants expressed in the methodology are restated and tested below:

H1: High energy abundance will have a negative effect on the energy efficiency level.

H2: The higher the proportion of industrial output in total GDP is the lower the energy efficiency level.

(23)

23

It must be stated that because energy efficiency estimates the energy required to produce 1 USD dollar worth of GDP a lower value of the energy efficiency variable indicate that the same quantity of energy is used to produce more GDP then before. Therefore reduced values of the EE variable indicate high levels of energy efficiency whereas increased values of the EE variable suggest modest levels of energy efficiency.

First we estimate the model in its original form. Dependent Variable: EE

Method: Least Squares Date: 06/18/08 Time: 18:39 Sample (adjusted): 2 109

Included observations: 82 after adjustments

Variable Coefficient Std. Error t-Statistic Prob. EA -3.11E-07 1.69E-08 -18.40055 0.0000 EI 1.99E-06 3.73E-07 5.342192 0.0000 IND_GDP -12.42847 1.000214 -12.42582 0.0000

C 6.968132 0.387423 17.98585 0.0000

R-squared 0.865301 Mean dependent var 1.432297 Adjusted R-squared 0.860120 S.D. dependent var 1.130941 S.E. of regression 0.422977 Akaike info criterion 1.164555 Sum squared resid 13.95498 Schwarz criterion 1.281956 Log likelihood -43.74675 F-statistic 167.0229 Durbin-Watson stat 0.401474 Prob(F-statistic) 0.000000

(24)

24

be noted here is that because of on unavailable information, mainly some data for the industry sector‘s output in Russia between 1970 and 1988 the number of observations have slightly decreased.

The good fit of the model is reflected by the high squared value of 0.865301 and Adjusted R-squared of 0.860120, saying basically that approximately 86% of the variation in a country‘s energy efficiency levels can be explained by the three variables included.

Autocorrelation and heteroskedasticity

Two important issues must be tested when performing regression analysis namely Autocorrelation and Heteroskedasticity (Carter Hill 2001). While the former is encountered when performing time series analysis, the latter is usually associated to cross-sections analysis. Because panel data implies both a time dimension and a cross-sectional element, testing for autocorrelation and heteroskedasticity is imperative.

Autocorrelation:

As argued by Carter Hill 2001 the error terms of panel data cannot be correlated, the sample randomness implies that the error terms of the observations are uncorrelated. However when observing time data series it might emerge that the observations follow an ordering through time in so doing generating a possibility that successive errors might present correlations . This violates the simple linear regressions model‘s hypothesis that covariance is null (Carter Hill 2001).

(25)

25

model becomes even more fitted whit a high R-square value of 0.938044 meaning that 93.8% of the variation in a country‘s energy efficiency levels can be explained by the three variables included in the model. On the other hand applying logarithmising and differentiating techniques has also somewhat of an obstructive effect. When combined with the relatively short time span of the data available applying these methods does correct for autocorrelation but in the same time in significantly reduces the number of observations available. To further correct for autocorrelation and to increase the Durbin-Watson test values the inclusion of an additional variable reflecting the energy price levels and subsidies in each country might prove opportune. However data of this nature proved elusive to collect thus such a variable can only be constructed in future studies.

Dependent Variable: LG_EE(-3) Method: Panel Least Squares Date: 06/20/08 Time: 19:00 Sample (adjusted): 1980 2006 Cross-sections included: 1

Total panel (balanced) observations: 27

Variable Coefficient Std. Error t-Statistic Prob. LG_EA(-9) 0.097626 0.256256 0.380969 0.7067 IND_GDP(-1) 16.98606 1.348265 12.59846 0.0000 LG_EI(-9) 1.075916 0.419289 2.566050 0.0173 C -21.04783 3.595789 -5.853466 0.0000 R-squared 0.938044 Mean dependent var -1.363978 Adjusted R-squared 0.929963 S.D. dependent var 0.496040 S.E. of regression 0.131274 Akaike info criterion -1.087104 Sum squared resid 0.396357 Schwarz criterion -0.895128 Log likelihood 18.67590 F-statistic 116.0780 Durbin-Watson stat 1.641339 Prob(F-statistic) 0.000000

Heteroskedasticity:

(26)

26

covariance matrix estimator derived by White (1980) is used to testing for Heteroskedasticity. This is supposed to offer accurate estimates of the coefficient covariance in the presence of Heteroskedasticity of unknown form.

White Heteroskedasticity Test:

F-statistic 1.304933 Probability 0.265479 Obs*R-squared 7.751180 Probability 0.256905

When performing a White Heteroskedasticity test one can see that the value of Prob (F-statistic) is 0.265479 meaning that we accepts the null hypothesis stating that there is no Heteroskedasticity

Multicollinearity:

Because of similarities between the independent variables used, testing the model for Multicollinearity is a must. To do this a Correlation Matrix is built as shown below

Correlation Matrix

EA EI IND_GDP

EA 1.000000 0.724082 -0.461410 EI 0.724082 1.000000 -0.148072 IND_GDP -0.461410 -0.148072 1.000000

(27)

27 Dependent Variable: EE

Method: Least Squares Date: 06/20/08 Time: 00:48 Sample (adjusted): 2 109

Included observations: 82 after adjustments

Variable Coefficient Std. Error t-Statistic Prob. EA -2.43E-07 1.30E-08 -18.67390 0.0000 IND_GDP -10.80403 1.106560 -9.763615 0.0000

C 6.149438 0.413215 14.88193 0.0000

R-squared 0.816017 Mean dependent var 1.432297 Adjusted R-squared 0.811359 S.D. dependent var 1.130941

Normality:

To establish whether a random variable is or is not randomly distributed a normality test must be employed. The most common application of the normality test is usually performed on residuals of linear regression models. In case the residuals are found not to be normally distributed they it is indicated that they should not be used Z tests or other tests derived from the normal distribution like t-tests, f-tests and chi-square tests. When the residuals are not normally distributed this suggest that the dependent variable or at least one of the explanatory variables might either be of the wrong functional form or an essential variable might be missing. In this case to order to obtain normally distributed residuals correcting one or more of the above mentioned systematic errors is indicated.

(28)

28 0 2 4 6 8 10 12 14 16 -0.5 -0.0 0.5 1.0 Series: Residuals Sample 2 109 Observations 82 Mean -1.83e-16 Median -0.008490 Maximum 1.144972 Minimum -0.836249 Std. Dev. 0.415071 Skewness 0.143808 Kurtosis 2.867345 Jarque-Bera 0.342759 Probability 0.842502

The figure obtained does resemble a bell shape and the p-value of the Jarque-Bera statistic is 0.842502 therefore we cannot reject the null hypothesis, the null hypothesis is normally distributed error terms this implies that the model has normally distributed error terms.

Chow Breakpoint Test

One of the most important assumptions of model that uses time series is that the underlying process remains constant over all observations in the sample. Therefore it is recommended that a careful analysis is made of time series data that include periods of severe change and crisis. The Chow test proves to be a very useful tool particularly in this regard. This test is usually employed to test the structural changes in one, more ore even all of the parameters of a model in cases where the disturbance term is expected to be identical for both periods

In this particular case the Chow Breakpoint test is employed to evaluate whether there is a structural break in the series at the time when the USSR disbanded in 1989.

Chow Breakpoint Test: 89

(29)

29

The null hypothesis of a Chow test is ‗no structural break‘, the p-values in the output below however indicate that this hypothesis is rejected at a very low significance level meaning that there is a structural break in the series at the point of the USSR‘s collapse in 1989.

Country Cases:

Table 5: country results Country / Variable EA (-) EI (-) IND_GDP (+)

China -2.40 (-4.412541)** 6.92 (3.067203)** -4.45 (-1.807583)* Germany 1.29 (6.970352)** -7.29 (-4.491055)** 11.29 (11.49746)** Russia -2.47 (-0.213331) 2.26 (0.732245) -3.97 (-0.728018)

Note: t-Values are in parentheses and ** indicates statistical significance at the 5% level *

indicates statistical significance at the 10% level.

When discussing the determinant of energy efficiency for every country in particular one finding is that in two of the three country‘s energy abundance (EA) has a significant effect on the level of energy efficiency. In the case of China EA has a negative coefficient therefore supporting the hypothesis that energy abundance tolerates a wasteful attitude in both energy production and usage thus having a negative effect on the energy efficiency level of the country. However when, looking at the results for Germany we discover the EA variable to have a significant and positive coefficient and in so doing contradicting the same hypothesis. What concerns Russia the p-value of EA is 0.8354 thus rendering it insignificant at any reasonable level of significance. One explanation for this is that bad policy and energy subsidies in Russia significantly distort the relationship between the country‘s energy abundance and the efficiency with which it is used. Furthermore when analyzing the energy imports (EI) effects on energy efficiency we find that in the case of China imports have a positive and significant effect on the energy efficiency variable in so doing contradicting the hypothesis that the more energy a country imports at world prices the more careful it has to be whit its usage thus determining a higher energy efficiency level. Not surprisingly the same hypothesis is being confirmed by the findings in Germany where the energy imports variable has a significant and negative effect on energy efficiency variable. In the case of Russia EI it is found to be insignificant. One explanation for this might be that considering the fact that Russia has been and continues to be one of the world‘s biggest energy exporters the concept of energy imports applies to it only in a modest manner.

(30)

30

energy efficiency in that country. Again, the findings for Germany are according to expectations and suggest that the IND_GDP variable has a negative effect on the energy efficiency levels in Germany thus corroborating the above mentioned hypothesis. In the case of Russia, similar to the EA and EI variables the coefficients of IND_GDP variable are insignificant with a high p-value of 0.4833 that makes it insignificant at any reasonable level of significance.

The above results show that in the case of Germany, the last two of the three hypotheses are confirmed. Some interesting results are also encountered in the case of China. In this case according to our findings two of the three hypotheses are contradicted by the results. Here the questions of what these results actually mean must be set. Do they mean that in China a higher proportion of industrial output will actually determine a more efficient use of energy or that the more energy is imported at high international prices, the more wasteful will it be in using it? Probably not, a much more likely explanation would be that the extremely low prices of the main energy resource in China coal, distorts the whole energetic sector and its relationship whit the above mentioned variables. Finally in the case of Russia the three variables prove to be insignificant as stated above, one plausible explanation for this might be that bad policy and considerable amounts of energy subsidies significantly distort the relationship between the country‘s energy efficiency and the determinants available to our model.

Finally we argue that energy efficiency is also supposed to be tightly related to the volume of energy subsidies in that country, however because of reduced data availability an exact quantitative analysis of the effects energy price subsidies have on energy efficiency was not possible. There are several scholars that do address this issue among them is Rosen and Houser (2007). In their work the authors argue that despite the fact that the once highly subsidized energy prices in China have been over the past three decades in principal converging with world prices factors like an accurate estimation local idiosyncrasies in pricing, dual supply channels and arrears (with and without permission) and other such factors make estimating accurately how much a specific company actually pays for oil, gas coal or electricity intricate. The prices for raw energy commodities in China is considerably lower than in the OECD countries especially in the interior provinces of the mainland, in close proximity of the resources deposits. In the case of coal which according to CIEC in 2005 covered for 69.5% of the total energy consumed in China, the low prices are not as much a result of energy subsidies but rather of the low extraction costs in areas which are isolated from international markets. However with transportation bottlenecks easing up the prices of coke will continuously converge upwards towards international levels. On the other hand, natural gas prices have been heavily subsidized by Chinese authorities. This is to keep the gas prices competitive with other developing countries mainly in response to the recent coming under pressure from the Middle East in a attracting and maintaining gas intensive industries.

(31)

31

raises questions on Russia‘s energy price policies5

these price differentials were also widely discussed in the frame of Russia‘s assertion to the WTO. The OECD has tried to quantify hidden subsidies on natural gas in Russia. The aggregate subsidies to Russia‘s industry were estimated around $1.7-3.5 billion (OECD, 2004a)

Moor (2001) in his paper argues that promoting the consumption of natural gas in Russia has improved the quality of the environment at a time when environmental policy and regulation enforcement was weak. Thus a significant increase in natural gas prices may encourage an increase in coal consumption which, without a sound environmental policy system, would in turn contribute significantly to air pollution. One of the main assumptions behind the suggestions that Russia should increase its prices for natural gas is the existence of a uniform international market for natural gas. Moor (2001) disproves the existence of such a market and argues that in the absence of unity among the regional markets of natural gas and lack of a world price for natural gas there would be no reason for Russia to raise its gas prices without taking into consideration other factors.

It is desired that more data will be made available on these issues and that a variable reflecting energy subsidies could be included in future studies.

5. Conclusions and discussion

Establishing the exact nature of the relationship between energy consumption and economic growth is of great interest to policy makers. This paper expands the literature on the energy to GDP relationship by investigating stationarity, cointegration and causality between energy consumption and economic output by analyzing in a bivariate model comprised of the economies of China, Germany and Russia between 1970 and 2005. These countries are characterized by their contrasting energetic profiles when it comes to resource endowment. Our paper analyzes the data set from an aggregate perspective as well as for each country in particular. Given the relatively short time span of the typical dataset we make use of new panel cointegration and panel-based error correction model techniques to acquire consistent and reliable results that reinvestigate the relationship between energy consumption and economic growth in the three countries.

When considering the energy consumption and GDP dynamics from the both a shortrun/long -run as well as from the direction of the causality perspective in all the three countries we refute the neutrality hypothesis advanced before. In the long run there seems to be a bidirectional causality relationship between energy consumption and GDP in all of the three countries. From

5

(32)

32

the short run perspective the nature of the causality relationship differs slightly for every country. While in the case of Germany in the short run as in the long run bidirectional causality is present between energy consumption and economic growth. In the cases of China and Russia we find that for the first there is unidirectional causality running from GDP to energy in the short run, where as for the latter it is concluded that in the short run energy consumption causes economic growth. Of the most interest to policymakers are the findings of long term granger causality running from energy consumption to GDP. Our current results provide solid support that changes in energy consumption have significant influence on economic output and that continuous energy use generates a continuous rise in economic growth. This means that economic growth is driven by energy consumption and despite the rising necessity for energy conservation, policymakers must take great care when designing conservation policies to try and avoid compromising economic output.

In addition, the findings of this paper can be used when debating over the disparity of previous conclusions on the causal relationship between energy consumption and economic growth. The nature of this relationship has been the subject of many studies. Due to a rich variety of methodologies previously employed, empirical results have so far been mixed when analyzing different countries. Furthermore using different variables and methodologies often leads to obtaining different results for the same country. However, because of the different approaches used, these conflicting results must not necessarily be looked upon as inconsistencies and further studies might be needed to analyze the issue.

(33)

33

mentioned variables. Similarly in the case of Russia a plausible explanation to the insignificance of the three hypotheses might be that bad policy and considerable amounts of energy subsidies significantly distort the relationship between the country‘s energy efficiency and the determinants available to our model.

(34)

34

References

Akarca, A.T. and Long, T.V., 1979. Energy and employment: a time series analysis of the causal relatonship. Resources Energy 2, pp. 151–162. Abstract | Abstract + References | PDF (841 K) | View Record in Scopus | Cited By in Scopus (11)

Akarca, A.T. and Long, T.V., 1980. On the relationship between energy and GNP: a re-examination. J. Energy Dev. 5, pp. 326–331.

Akaike, 1970 H. Akaike, Statistical predicator identification, Annals of the Institute of Statistical

Mathematics 21 (1970), pp. 203–207.

Al-Iriani 2006 Mahmoud A. Al-Iriani. Energy–GDP relationship revisited: An example from GCC countries using panel causality. 2005

Asafu-Adjaye 2000 John Asafu-Adjaye , . The relationship between energy consumption, energy prices and economic growth: time series evidence from Asian developing countries*1. 2000

Baltagi 1995 Baltagi, Badi H. ―Econometric Analysis of Panel Data.‖ John Wiley & Sons LTD, 1995.

Beaudreau, 2005 B.C. Beaudreau, Engineering and economic growth, Structural Change and

Economic Dynamics 16 (2005), pp. 211–220. SummaryPlus | Full Text + Links | PDF (104 K) | View Record in Scopus | Cited By in Scopus (3)

Brüderl, 2005 Brüderl, Josef. ―Panel Data Analysis.‖ University of Mannheim, March 2005. Carter Hill 2001 Carter Hill, R., William E. Griffiths and George G. Judge, Undergraduate

econometrics, 2001

Chotanawat, Hunt & Pierse 2006 Jaruwan Chontanawata, , , Lester C. Huntb and Richard Piersec. Does energy consumption cause economic growth?: Evidence from a systematic study of over 100 countries. 2006

Encyclopedia of Business and Finance http://www.enotes.com/business-finance-Encyclopedia/factors-production

Energy Statistics and Energy Balance for the Soviet Union‘ OECD IEA, Paris 1991

(35)

35

Erol and Yu, 1987 U. Erol and E.S.H. Yu, On the causal relationship between energy and income for industrialised countries, Journal of Energy and Development 9 (1987), pp. 75–89. Abstract | Abstract + References | PDF (925 K) | View Record in Scopus | Cited By in Scopus (13)

Fatai et al., 2004 K. Fatai, L. Oxley and F.G. Scrimgeour, Modelling the causal relationship between energy consumption and GDP in New Zealand, Australia, India, Indonesia, the Philippines and Thailand, Mathematics and Computers in Simulation 64 (2004), pp. 431–445. SummaryPlus | Full Text + Links | PDF (121 K) | View Record in Scopus | Cited By in Scopus (12)

Ghali and El-Sakka, 2004 K.H. Ghali and M.I.T. El-Sakka, Energy use and output growth in Canada: A multivariate cointegration analysis, Energy Economics 26 (2004), pp. 225–238. SummaryPlus | Full Text + Links | PDF (116 K) | View Record in Scopus | Cited By in Scopus (15)

Glasure and Lee, 1997 Y.U. Glasure and A.R. Lee, Cointegration, error-correction, and the relationship between GDP and energy: the case of South Korea and Singapore, Resource and

Energy Economics 20 (1997), pp. 17–25.

Granger, 1969 C.W.J. Granger, Investigating causal relations by econometric models and cross-spectral methods, Econometrica 36 (1969), pp. 424–438. Full Text via CrossRef

Granger, 1988 C.W.J. Granger, Some recent developments in a concept of causality, Journal of

Econometrics 39 (1988), pp. 199–211. Abstract | Abstract + References | PDF (872 K) | MathSciNet

Granger and Newbold, 1974 C. Granger and P. Newbold, Spurious regressions in econometrics,

Journal of Econometrics 2 (1974), pp. 111–120. Abstract | Abstract + References | PDF (618 K)

Harb, 2003 N. Harb, Money demand function: a heterogeneous panel application, Applied

Economics Letters 11 (2003) (9), pp. 551–555.

Hsiao, 1981 C. Hsiao, Autoregressive modelling and money-income causality detection, Journal

of Monetary Economics 7 (1981), pp. 85–106. Abstract | Abstract + References | PDF (1112 K)

Kraft and Kraft, 1978 J. Kraft and A. Kraft, On the relationship between energy and GNP,

Journal of Energy and Development 3 (1978), pp. 401–403.

The Relationship between Energy Consumption, Energy Efficiency and Economic Growth: The Case of China, Germany and Russia