Microsoft Word - 00_tresc.docx © 2013 Nicolaus Copernicus University. All rights reserved. http://www.dem.umk.pl/dem D Y N A M I C E C O N O M E T R I C M O D E L S DOI: http://dx.doi.org/10.12775/DEM.2013.003 Vol. 13 (2013) 51−68 Submitted October 15, 2013 ISSN Accepted December 30, 2013 1234-3862 Monika Papież, Sławomir Śmiech* Economic Growth and Energy Consumption in Post-Communist Countries: a Bootstrap Panel Granger Causality Analysis∗∗ A b s t r a c t. The aim of this paper is to identify Granger causality between energy consump- tion and economic growth in post-communist countries in the period 1993 to 2011. Bootstrap panel Granger causality test was used as a research tool in order to accommodate for country- specific heterogeneity and to avoid the problem of cross-sectional dependence. The analysis allowed for the verification of the hypothesis regarding the links between eco- nomic growth and energy consumption in nine countries. The hypotheses were confirmed: the growth hypothesis in three countries and the feedback hypothesis in one country. K e y w o r d s: energy consumption, economic growth, bootstrap panel Granger causality test, energy efficiency. J E L Classification: C33, Q43. Introduction Improving energy security is the priority of the EU policy. Growing de- pendence of the EU on energy import (its volume is predicted to increase from 50% at present to 80% in 2035) makes it imperative to limit energy * Correspondence to: Monika Papież, Cracow University of Economics, Department of Statistics, Rakowicka 27, 31-510 Krakow, Poland e-mail: papiezm@uek.krakow.pl; Sławomir Śmiech, Cracow University of Economics, Department of Statistics, Rakowicka 27, 31-510 Krakow, Poland e-mail: smiechs@uek.krakow.pl. ∗∗ We are grateful for the financial support of the Polish National Science Centre (the pro- ject DEC-2011/03/B/HS4/01134). Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 52 consumption (measured by CO2 emission) and replace non-renewable en- ergy sources with renewable ones. On the other hand, energy consumption is inextricably linked with economic growth (the Environmental Kuznets Curve). Dynamically developing countries consume more and more energy, and curbing energy consumption can lead to stagnation or the drop in eco- nomic growth rate. The analysis of links between economic growth and energy consumption was addressed by numerous research studies beginning with a pioneering study by Kraft and Kraft (1978). Four hypotheses regarding causal relations between energy consumption and economic growth can be found in the lit- erature dealing with this topic: the growth hypothesis, the conservation hy- pothesis, the feedback hypothesis and the neutrality hypothesis. The growth hypothesis assumes that there are countries in which the growth of energy consumption is an important element of their economic development. In this case, the energy conservation policy advocating the introduction of limits in energy consumption will have a negative impact on economic growth. The growth hypothesis is based on unidirectional Granger causality running from energy consumption to economic growth. The conservation hypothesis claims that the changes in energy consumption stem from the changes in economic activity, and energy conservation policy does not negatively affect economic growth. This hypothesis is confirmed if unidirectional Granger causality running from GDP growth to energy consumption can be observed. The feedback hypothesis assumes that there are countries with bi-directional Granger causality between energy consumption and economic growth. The neutrality hypothesis states that there are countries in which GDP does not depend on energy consumption and vice versa. As Karanfil (2009) demonstrates in his survey of empirical literature devoted to this issue, the relations between economic growth and energy consumption are not unambiguous. The differences can be attributed to dif- ferent econometric approaches, differently specified time frames and differ- ent sets of variables used in each of those studies. The occurrence of the relationship between economic growth and energy consumption is related to the changes in energy efficiency. Energy efficiency aims at reducing the amount of energy required to provide products and ser- vices in a given country. Most studies underline a positive influence of en- ergy efficiency on economy and the environment. For example, Sarkar and Singh (2010) show that energy efficiency programmes can: conserve natural resources, reduce the environmental pollution and carbon footprint of the energy sector, reduce a country’s dependence on fossil fuels, thus enhancing its energy security, ease infrastructure bottlenecks and impacts of temporary Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 53 power shortfalls, as well as improve industrial and commercial competitive- ness through reduced operating costs. However, rebound effects (see Turner, 2009; Turner and Hanley, 2011 for a recent review) can appear. When en- ergy becomes more productive, and its price falls, the increase of energy use through the substitution effect can be observed. This paper examines the relationship between energy consumption and economic growth in nine Eastern European countries and the Baltic states. It focuses on the production side model of the energy consumption-growth nexus, with labour and capital included in it (Stern, 1993). The analysis cov- ers the period from 1993 to 2011. Such a choice was dictated by the need to cover the relations in the analysed countries, all of which witnessed rapid political and economic changes in the 1990s. The aim of the analysis was to investigate the relations between energy consumption and economic growth in selected Eastern European countries and the Baltic states on the basis of the relations between overall energy efficiency gains (industry, transport, households) in 2000–20101 and coun- tries' economic growth in the same period. The following research hypothe- ses were formulated: a) Countries which increased energy efficiency the most will confirm the growth hypothesis, feedback hypothesis or conservation hypothesis (bi- directional unidirectional causality between energy consumption and economic growth). b) Countries in which energy efficiency was not considerably increased will confirm the neutrality hypothesis. The hypotheses result from the following reasoning. The former group of countries had to bear the costs of the increase in energy efficiency on the one hand, and, on the other hand, modernisation allows for reducing the amount of energy used and, consequently, its costs, which should result in the ap- pearance of causal relations between energy production and economic growth. In the latter group of countries the effects mentioned are non- existent, which rules out any relations between them. We applied a bootstrap panel causality approach proposed by Kónya (2006), which allows for simultaneous examination of cross-sectional de- pendence. The paper consists of the following sections. Section 1 presents the most important findings from studies dealing with energy-economy nexus in Cen- tral and Eastern European countries. In Section 2 the relations between over- 1 The choice of the period for the analysis was dictated by the availability of the data pro- vided by http://www.odyssee-indicators.org. Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 54 all energy efficiency gains and economy growth are described. Section 3 introduces methodology used in the study, and Section 4 presents the data and results obtained. The paper ends with the conclusion and the interpreta- tion of the results. The paper contributes to the existing literature because the analysis fo- cuses on countries from Eastern Europe and the Baltic states. Most of them have not been analysed from this angle before. They have similar GDP and a similar level of energy consumption. As member states of the EU, they are obliged to follow a common energy policy. Thanks to these similarities, the data used were characterised by cross-sectional dependence. The application of the methodology suggested by Kónya (2006) made correct inference on causalities in this situation possible. Two additional variables, labour and capital, were used to compare the models, which makes this study more general in scope than other studies. Our findings may provide valuable information for developing more effec- tive energy policies with respect to both energy consumption and environ- mental protection. 1. Review of Literature Existing literature offers a wide range of perspectives and insights into the issue of energy consumption-growth nexus, which sometimes report contradicting results. It can be divided into country-specific case studies and multi-country studies (Karanfil, 2009). In both types various econometric methods, the choice of the period analysed, and the choice of control vari- ables can be found. Taking into consideration the methodological perspective, four genera- tions of contributions were identified (Belke et al., 2011; Costantini and Martini, 2010). The first-generation studies were based on VAR methodol- ogy (Kraft and Kraft, 1978) and assumed that the time series were stationary. The second-generation studies accounted for non-stationarity and applied Engle-Granger two-step procedure to test pairs of variables for cointegrating relationships. The third-generation studies used multivariate estimators (Johansen, 1991). This approach allowed for more than two variables in cointegration relationship and for analysing causality both in the short- and long-run simultaneously. The fourth-generation studies were based on panel methods testing for unit roots, cointegration and Granger causality. Using panel cointegration has several advantages. It allows for higher degrees of freedom, reduces multicollinearity between regressors, and improves the power of the cointegration test, especially in case of annual data. The main Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 55 disadvantage of this approach is the need to assume cross-sectional inde- pendence, which is difficult to satisfy in a panel data. What is more, different countries are treated as an entity. As a result, it is impossible to identify the difference in the dynamic relationship between energy consumption and economy (slope homogeneity). As a result, it is impossible to identify the difference in the dynamic relationship between energy consumption and economy (slope homogeneity). What is more, in most studies based on panel models (except panel VAR, see surveys by Canova and Ciccarelli (2013), which has not been used in such analyses so far) different countries are treated as an entity. In spite of a substantial number of studies concerning relations between energy consumption and economic growth, not all Eastern European coun- tries were analysed, for example, the Baltic states were not included in any of them, and among the countries from Central and Eastern Europe only the following ones were studied: Poland (Gurgul and Lach, 2011a, 2011b, 2012), Romania (Apergis and Danuletiu, 2012), Albania, Bulgaria, Hungary, and Romania (Ozturk and Acaravci, 2010). Gurgul and Lach (2011b) found that energy consumption Granger caused GDP in Poland during the last dec- ade. They (Gurgul and Lach, 2011a) also investigated causal relations be- tween coal consumption and economic growth. In another paper Gurgul and Lach (2012) investigated causal interdependences between electricity con- sumption and GDP in Poland. Apergis and Danuletiu (2012) showed that energy consumption Granger caused GDP in Romania in the period 2000- 2011. However, Ozturk and Acaravci (2010) did not find any relationship between energy consumption and real GDP in Romania and Bulgaria, while found bidirectional strong Granger causality between these variables in Hungary in the period 1980–2006. Using a two-way fixed effects model, Menegaki and Ozturk (2013) confirmed bidirectional causality between growth and political stability, capital and political stability, and capital and fossil energy consumption for 26 European countries in a multivariate panel framework over the period 1975–2009. 2. Energy Efficiency in Eastern European Countries and the Baltic States Energy efficiency is considered to be one of the most cost effective ways of meeting the demands of sustainable development and lower fossil fuel dependence. So, the efficient use of energy is an important topic in public policy debates. Unfortunately, in literature dealing with energy issues there is no consensus on the appropriate method for defining and measuring en- Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 56 ergy efficiency. The authors of this paper adopted the definition of energy efficiency given by the Odyssee project2. The “ODEX” energy efficiency indicator provides an overall perspective of energy efficiency trends by sec- tor and combines the trends of indicators by end-use or sub-sector. It repre- sents a better proxy to evaluate energy efficiency trends at an aggregate level (overall economy, industry, households, transport, services). The ODEX indicators by sector (industry, transport, households) are calculated from unit consumption trends by sub –sector (or end-use or mode of transport) by ag- gregation of unit consumption indices by sub-sector in one index for the sector on the basis of the current weight of each sub-sector in the sector’s energy consumption. The ODEX can be defined as the ratio between the actual energy consumption of the sector in year t and the sum of fictive en- ergy consumption of each underlying sub-sector/end-use that would have been observed in year t had the unit consumption of the sub- sector been that of a reference year. The energy efficiency gains are calculated from ODEX and reflect efficiency gains since 2000. Figure 1 presents relations between overall energy efficiency gains (in- dustry, transport, households) in the period 2000–2010 and real growth of GDP per capita in the same period for Eastern Europe and the Baltic states. This diagram identifies two groups of countries with similar levels of energy efficiency and a similar level of economic development. The first group comprises countries with high overall energy efficiency gains (above 18 percent) and high (and medium) growth of GDP per capita in the period 2000–2010. The second group includes countries with low overall energy efficiency gains (below 14 percent) and low (and medium) growth of GDP per capita in the period 2000–2010. In the analysed Eastern European countries and the Baltic states in the period 2000–2010 the mean energy efficiency index for the whole economy (ODEX) decreased by 15.3 percent. Countries with the highest improvement in energy efficiency in the period analysed include: Poland (25.7 percent), Bulgaria (23 percent), Latvia (21 percent), Lithuania (19.8 percent), and Romania (19.4 percent). Countries with the lowest improvement in energy efficiency in the same period include: Slovakia (3.7 percent), the Czech Re- public (5.2 percent), Estonia (7.3 percent), and Hungary (13 percent). It should be noticed that countries which did not improve their energy effi- ciency substantially also had a lower increase growth in GDP per capita 2 ODYSSEE is a project between ADEME, the EIE programme of the European Commis- sion/DGTREN and energy efficiency agencies, or their representative, in the 27 countries in Europe plus Norway and Croatia (http://www.odyssee-indicators. org). Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 57 (except for Slovakia3) than countries with a considerable increase in energy efficiency. Figure 1. Relations between overall energy efficiency gains (industry, transport, households) in the period 2000-2010 and growth of GDP per capita in the period 2000–2010 3. Methodology The choice of a suitable method allowing for the analysis of causality for panel data requires the assessment of cross-sectional dependence. If cross- sectional dependence exists, the seemingly unrelated regressions (SUR) are more efficient then the ordinary least-squares (OLS) (Zellner, 1962). Kónya (2006) proposed a method which takes into account the characteristics of cross-sectional dependence. Therefore, before considering causality, we investigated the characteristics of panel data. The tools used for bootstrap panel causality test are presented below. 3 A specific situation of Slovenia is a result of numerous phenomena which are described in detail in the study: Energy Efficiency Policies and Measures in Slovakia in 2012, ODYSSEE- MURE 2010, http://www.odyssee-indicators.org/publications/PDF/slovakia_nr.pdf [30.12.2013]. Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 58 3.1. Tests of Cross-Sectional Dependence An important issue to be considered in a panel data analysis is testing for cross-sectional dependence across countries, because a shock that affects one country may spillover on other countries. Let us consider the standard panel data model: ititiiit uy ++= x'βα , (1) where Ni ,,2,1 …= represents the cross section dimension, Tt ,,2,1 …= refers to the time series dimension, itx is a (K × 1) vector of observed re- gressors (individual-specific as well as common regressors). The individual intercepts iα and the slope coefficients iβ are defined on a compact set and allowed to vary across i. For each i, ( )2,0~ iuit IIDu σ for all t, although they may exhibit cross- sectional dependence. The null hypothesis of no-cross-sectional dependence – ( ) 0:0 =jtituuCovH for all t and ji ≠ – is tested against the alternative hy- pothesis of cross-sectional dependence – ( ) 0:1 ≠jtituuCovH , for at least one pair of ji ≠ . In literature several tests for error cross-sectional dependence have been proposed. Breusch and Pagan (1980) proposed a Lagrange multiplier (LM) statistic for testing the null hypothesis of no-cross-sectional dependence, which is defined as: ∑ ∑ − = += = 1 1 1 2ˆ N i N ij ijTLM ρ , (2) where ijρ̂ is the sample estimate of the pair wise Pearson correlation coeffi- cient of the residuals from the Ordinary Least-Squares (OLS) estimation of Eq. (1) for each i. LM is asymptotically distributed as chi-squared with 2/)1( −NN degrees of freedom under the null hypothesis, as ∞→T , with N fixed. It is important to note that the LM test is valid for relatively small N and sufficiently large T. So, where ∞→T and ∞→N , Pesaran (2004) proposed the following LM statistic for the cross-sectional dependence test (the so-called CD test): ( ) ( )∑ ∑ − = += − − = 1 1 1 2 1ˆ 1 1 N i N ij ijlm TNN CD ρ . (3) Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 59 Under the null hypothesis, the lmCD test converges to the standard normal distribution. However, this test is likely to exhibit substantial size distortions for N large and T small, a situation that can frequently arise in empirical applica- tions. To overcome this problem, Pesaran (2004) proposed the following simple alternative test, which is based on the pair-wise correlation coeffi- cients rather than their squares used in the LM test: ( ) ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ − = ∑ ∑ − = += 1 1 1 ˆ 1 2 N i N ij ijNN T CD ρ , (4) and showed that under the null hypothesis of no cross-sectional dependence ( )1,0NCD d→ for ∞→N and T sufficiently large. Pesaran (2004) con- cluded that it is also clear that, since the mean of CD is exactly equal to zero for all fixed 1+> KT and N, the test is likely to have good small sample properties (for both N and T small). In Pesaran et al. (2008) the authors concluded that the CD test has an important drawback, namely it will lack power in certain situations where the population average pairwise correlations are zero, although the underly- ing individual population pairwise correlations are non-zero. That is why Pesaran et al. (2008) proposed a bias-adjusted test, which is a modified ver- sion of the LM test, by using the exact mean and variance of the LM statis- tic. The bias-adjusted LM test is as follows: ( ) ( ) ∑ ∑ − = += −− − = 1 1 1 2ˆ 1 2 N i N ij Tij Tijij adj v kT NN LM μρ , (5) where Tijμ and 2 Tijv are respectively the exact mean and variance of ( ) 2ˆijkT ρ− provided in Pesaran et al. (2008 p.108). Pesaran et al. (2008) showed that under the null hypothesis of no cross-sectional dependence with ∞→T first followed by ∞→N , the statistics adjLM follow an asymp- totic standard normal distribution. 3.2. Bootstrap Panel Granger Causality Test Taking into account cross-sectional dependence and heterogeneity across country groups requires a method of analysis which would be able to capture both these features. The bootstrap panel causality approach proposed by Kónya (2006) seemed to be a suitable method. This approach uses seemingly Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 60 unrelated regression (SUR) and, therefore, is able to deal with cross- sectional dependence across the members of the panel. The test for the direction of causality is based on Wald tests with coun- try-specific bootstrap critical values. That is why it does not impose a joint hypothesis across all members of the panel. Using Kónya (2006) approach allows for the identification of specific countries in which a Granger causal relationship exists. What is more, Kónya (2006) claimed that 'this approach does not require pretesting for unit roots and cointegration', which is impor- tant 'since the unit-root and cointegration tests in general suffer from low power'. On the other hand, ignoring potential (common) stochastic trends results in a situation in which the results of the suggested procedure can be used only for the evaluation of short-term causality (one-period-ahead fore- cast). Kónya's (2006) panel causality approach models the data as a system of two sets of the following equations4: 1 1 1 1 1, 1,1 1,1, 1, 1,1, 1, 1,1, 1, 1 1 1 1,1, 1, 1,1, 1 , mly mlx mlz t l t l l t l l t l l l l ml l t l t l y y x z v ν α β δ γ ϑ ε − − − = = = − = = + + + + + + ∑ ∑ ∑ ∑ …. (6) 1 1 1 1 , 1, 1, , , 1, , , 1, , , 1 1 1 1, , , 1, , 1 , mly mlx mlz N t N N l N t l N l N t l N l N t l l l l mlv N l N t l N t l y y x z v α β δ γ ϑ ε − − − = = = − = = + + + + + + ∑ ∑ ∑ ∑ and 2 2 2 2 1, 2,1 2,1, 1, 2,1, 1, 2,1, 1, 1 1 1 2,1, 1, 2,1, 1 , mly mlx mlz t l t l l t l l t l l l l mlv l t l t l x y x z v α β δ γ ϑ ε − − − = = = − = = + + + + + + ∑ ∑ ∑ ∑ …. (7) 4 It is possible to include a deterministic component into the system of equations. Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 61 2 2 2 2 , 2, 2, , , 2, , , 2, , , 1 1 1 2, , , 2, , 1 , mly mlx mlz N t N N l N t l N l N t l N l N t l l l l mlv N l N t l N t l x y x z v α β δ γ ϑ ε − − − = = = − = = + + + + + + ∑ ∑ ∑ ∑ where tiy , denotes economic growth (in country i and t period), tix , refers to energy consumption, tiz , is the capital formation, tiv , is the labour participa- tion rate5, N denotes the number of countries in the panel ( Ni ,,2,1 …= ), t is time period ( Tt ,,2,1 …= ), and l is the number of lags in equations. titi ,,2,,1 ,εε are supposed to be correlated contemporaneously across equations (due to common random shocks). The system of equations allows for testing unidirectional and bi- directional Granger causality for each country separately. There is unidirec- tional Granger causality running from economic growth to energy consump- tion (the equivalent of the conservation hypothesis) if in (7) not all i,2β 's are zero, but in (6) all i,1δ 's are zero. There is unidirectional Granger causality running from energy consumption to economic growth in country i (the equivalent of the growth hypothesis) if not all i,1δ 's are zero, but all i,2β 's are zero in (7). There is bi-directional Granger causality between en- ergy consumption and economic growth if neither all i,1δ 's nor all i,2β 's are zero. Finally, there is no Granger causality between energy consumption and economic growth if all i,1δ 's and all i,2β 's are zero. The country-specific bootstrap6 critical values are obtained as follows7: 1. A system of equations (6) is estimated under the null hypothesis of non- causality running from energy consumption to economic growth (that is imposing the 0,,1 =liδ restriction for all i and l) and the residuals are ob- tained: 5 Z and ν are treated as an auxiliary variable, and they will not be directly involved in the Granger causality analysis. 6 On bootstrapping in general see e.g. Horowitz (2003). On bootstrapping in SUR models see Atkinson et. al (1992), and Rilstone and Veall (1996). 7 We present a procedure for testing Granger causality running from X to Y. Similar steps are required for testing causality running from Y to X. Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 62 1 1 1 0 , , , 1, 1, , 1, 1, , 1, 1, , 1, 1 1 1 ˆ ˆˆ ˆ , mly mlz ml H i t i t i i l t l i l t l i l t l l l l e y y z v ν α β γ ϑ− − − = = = = − − − −∑ ∑ ∑ (8) for 1, ,i N= … and Tt ,,1 …= . From these residuals N×T matrix [ ]tiHe ,,0 is developed. 2. These residuals are re-sampled by randomly selecting a full column form the matrix [ ]tiHe ,,0 , and denote the selected bootstrap residuals as [ ]* ,,0 tiHe where *1, 2,3,..., .t T= 3. The bootstrap sample of Y is generated under the assumption of no cau- sality running from energy consumption to economic growth, i.e.: 1 1 1 0 * * * , 1, 1, , , 1, , 1, 1, , 1, , , 1 1 1 ˆ ˆˆ ˆ . mly mlz ml i t i i l i t l i l t l i l t l H i t l l l y y z v e ν α β γ ϑ− − − = = = = + + + +∑ ∑ ∑ (9) 4. *,tiy is substituted for tiy , and a system of equations is re-estimated (without any restrictions). The Wald test for each country is implied by the no-causality null hypothesis. 5. The empirical distributions of the Wald test statistics are developed by repeating steps 2 – 4. The bootstrap critical values are specified by se- lecting appropriate percentiles of these sampling distributions. Eventually, Wald test statistics obtained from original series are com- pared with the bootstrap critical values. Specifying the number of lags in all equations is a crucial step in Kónya's approach. Following Kónya (2006), we decided to allow for different lags in each system but did not allow for different lags across countries. Assuming that the number of lags ranges from 1 to 4, we estimated all equations and used the Akaike Information Criterion (AIC) to determine the optimal solu- tion. The Akaike Information Criterion8 (AIC) was evaluated as: T qN AICk 22 ||ln += W , (10) where W stands for estimated residual covariance matrix, N is the number of equations, q is the number of coefficients per equation, T is the sample size. 8 Kónya (2006) presented also Schwartz Information Criterion. Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 63 4. Data and Empirical Results The analysis of causal relationship between energy consumption and economic growth based on the annual panel data was carried out over the period 1993–2011 for nine European countries: Bulgaria, the Czech Repub- lic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, and Slovakia. Two variables from the World Bank Development Indicators were chosen for the analysis: real Gross Domestic Product per capita (GDP) in constant 2000 U.S. dollars and energy consumption (EC), represented by energy use in kg of oil equivalent per capita. Taking into consideration rapid economic changes experienced by the countries analysed, a set of variables was extended to include real gross fixed capital formation per capita (K) in constant 2000 US dollars as a proxy of capital and labour participation rate (L)9. All variables were in natural logarithms. Till 1989 Eastern European countries and the Baltic states were under the communist rule with centrally planned economies. In 1989 communism fell in Bulgaria, Czechoslovakia, Hungary, Poland, and Romania. After the dis- solution of the Soviet Union in 1991, Estonia, Latvia, and Lithuania reap- peared on the map, and in 1993 Czechoslovakia was divided into two coun- tries: the Czech Republic and Slovakia. That is why year 1993 was chosen as an initial period of the analysis of causality between economic growth and energy consumption. Table 1. Cross-sectional dependence tests Variable Cross-sectional dependence test LM CDLM CD LMadj GDP 150,62*** 13,51*** 6,23*** 21,654*** EC 324,84*** 30,50*** 17,84*** 36,375*** Note: ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively. The first step in analysing panel data Granger causality is testing for cross-sectional dependence. Table 1 shows the results obtained for four dif- ferent cross-sectional dependence test statistics: LM (Breusch and Pagan, 1980), CDLM (Pesaran, 2004), CD (Pesaran, 2004), and LMadj (Pesaran et al., 2008). The results indicate that for all countries with significance level p = 0.05 we reject the null hypothesis of no cross-sectional dependence among the four variable examined. These findings show that a shock which 9 The use of real gross fixed capital as a proxy of capital follows the works by Sari and Soytas (2007) in assuming that under the perpetual inventory method with a constant depre- ciation rate, the variance in capital is closely related to the change in investment. Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 64 occurred in one post-communist country will be transmitted to other coun- tries. The existence of cross-sectional dependence in these countries means that it is justified to use the bootstrap panel Granger causality testing method. For each system of equations the number of lags was chosen ac- cording to the AIC criterion10. Additionally, specifications incorporating deterministic trend were taken into account. The results from the bootstrap11 panel Granger causality analysis are reported in Table 2 and Table 3. Table 2. The bootstrap panel Granger causality analysis Countries H0: Energy consumption does not Granger cause GDP (H1: EC → GDP) Wald statis- tics Bootstrap critical value 10% 5% 1% Bulgaria 12.915** 7.638 11.186 21.785 Czech Republic 0.686 7.715 11.176 19.899 Estonia 0.574 10.351 16.084 35.753 Hungary 0.317 7.751 11.784 20.686 Latvia 9.265* 8.842 13.331 24.626 Lithuania 3.378 8.856 14.397 29.298 Poland 18.917** 6.597 10.029 20.283 Romania 8.372* 8.245 11.548 25.975 Slovakia 2.235 7.584 12.652 23.370 Note: ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively. Bootstrap critical values are obtained from 10,000 replications. Table 2 and Table 3 present the results obtained for nine transition coun- tries in Eastern Europe and the Baltic states. The results confirm the growth hypothesis for Bulgaria, Poland (both at the significance level 5%) and Ro- mania (at the significance level 10%). This means that economies in those three countries can be called ‘energy dependent’, and that energy consump- tion plays an important role in their economic growth, both directly and indi- rectly in the production process as a complementary factor to labour and capital. Consequently, we may conclude that energy is a limiting factor to economic growth and, hence, shocks to energy supply will have an impact 10 We used the AIC criterion to compare the specifications with and without a linear trend. Finally, we constructed SUR with one lag and without a linear trend. 11 Following the original paper of Kónya (2006) and several others, e.g. Nazlioglu et. al (2011), we used 10000 replications in the procedure. Andrews and Buchinsky (2000) provide an exact method of evaluating the adequacy of the chosen number of replications. Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 65 on economic growth. Additionally, it can be claimed that excessive energy protection and a reduction in energy consumption may lead to stagnation. Table 3. The bootstrap panel Granger causality analysis Countries H0: GDP does not Granger cause energy consumption (H1: GDP → EC) Wald statis- tics Bootstrap critical value 10% 5% 1% Bulgaria 0.425 13.234 20.697 41.039 Czech Republic 0.330 12.162 18.667 37.114 Estonia 0.696 8.989 13.623 29.087 Hungary 1.244 12.947 20.043 42.007 Latvia 23.841** 13.824 22.365 61.726 Lithuania 4.246 9.716 14.063 26.881 Poland 12.580 17.470 25.420 51.422 Romania 2.910 11.334 17.488 34.534 Slovakia 0.632 10.430 15.556 31.457 Note: ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively. Bootstrap critical values are obtained from 10,000 replications. The feedback hypothesis was confirmed only for Latvia. This means that energy consumption and economic growth are jointly determined and af- fected at the same time. The results support the neutrality hypothesis for 5 countries: the Czech Republic, Hungary, Estonia, Lithuania, and Slovakia. The neutrality hy- pothesis states that energy consumption and economic growth are not sensi- tive to one another. Therefore, any policy with respect to the consumption of energy, conservative or expansive, is expected to have a negligible effect on economic growth. Conclusions and Discussion We investigated the relations between energy consumption and eco- nomic growth. Labour and real gross fixed capital formations were added to the analysis in order to avoid the problem of impact of omitted-variables bias. The methodology used in the study, Kónya's procedure (2006), firstly, allowed for the assessment of causality in countries with cross-sectional dependence, and, secondly, avoided the problem of incorrect specification connected with unit root and cointegration. Empirical results confirm the linkages between energy consumption and economic growth in four of nine countries. The growth hypothesis was con- firmed in 3 countries: Bulgaria, Poland, and Romania. Energy consumption seems to be the bottleneck in their economic growth, and, hence, shocks to Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 66 energy supply will have an impact on this growth. Latvia is the only country for which the feedback hypothesis was confirmed. In such countries energy consumption and economic growth are complementary to each other. The results obtained in the study indicate that causality between energy consumption and economic growth can be equated with energy efficiency. The group of countries with the best energy efficiency indicators consisted of almost the same countries as the group in which the growth hypothesis, the conservation hypothesis or the feedback hypothesis were confirmed (the only exception was Latvia). Thus, it can be assumed that the hypotheses formulated in the introduction were confirmed. Unfortunately, the empirical strategy used in this study did not allow for showing the reactions of eco- nomic growth to the changes in energy consumption and vice versa. So, it is not possible to conclude whether modernisation of economies, which can be equated to energy efficiency, exerts a positive or negative impact on econ- omy (the rebound effect). In conclusion, a special situation of Poland and Bulgaria, countries con- firming the growth hypothesis, should be mentioned. They rely on coal as the most important source of energy. In 2011 in Poland it accounted for 56.2 percent and in Bulgaria for 42.3 percent of the country’s total primary en- ergy consumption. The pressure on those countries is especially heavy be- cause of the amount of their emission of carbon dioxide, which are the high- est for coal. The necessity of limiting the use of this energy source without access to alternative energy sources can be a serious threat for their econo- mies. References Apergis, N., Danuletiu, D. (2012), Energy Consumption and Growth in Romania: Evidence from a Panel Error Correction Model, International Journal of Energy Economics and Policy, 2 (4), 348-356, http://www.econjournals.com/index.php/ijeep/article/view/316 Andrews, D. W. K., Buchinsky, M. (2001), Evaluation of a Three-step Method for Choosing the Number of Bootstrap Repetitions, Journal of Econometrics, 103, 345–386, DOI: http://dx.doi.org/10.1016/S0304-4076(01)00047-1. Atkinson, S. E., Wilson P.W. (1992), The Bias of Bootstrapped versus Conventional Standard Errors in the General Linear and SUR Models, Econometric Theory, 8, 258-275, DOI: http://dx.doi.org/10.1017/S0266466600012792. Belke, A., Dobnik, F., Dreger, C. (2011), Energy Consumption and Economic Growth: New Insights into the Cointegration Relationship, Energy Economics, 33 (5), 782–789, DOI: http://dx.doi.org/10.1016/j.eneco.2011.02.005. Breusch, T., Pagan, A. (1980), The Lagrange Multiplier Test and its Application to Model Specifications in Econometrics, Reviews of Economics Studies, 47, 239–253. Canova, F., Ciccarelli, M. (2013), Panel Vector Autoregressive Models: A Survey, Working Paper Series 1507, European Central Bank, http://ssrn.com/abstract=2201610. Economic Growth and Energy Consumption in Post-Communist Countries... DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 67 Costantini, V., Martini, C. (2010), The Causality between Energy Consumption and Eco- nomic Growth: a Multi-sectoral Analysis Using Non-stationary Cointegrated Panel Data, Energy Economics, 32 (3), 591–603, DOI: http://dx.doi.org/10.1016/j.eneco.2009.09.013. Gurgul, H., Lach, Ł. (2011a), The Role of Coal Consumption in the Economic Growth of the Polish Economy in Transition, Energy Policy, 39, 2088–2099, DOI: http://dx.doi.org/10.1016/j.enpol.2011.01.052. Gurgul, H., Lach, Ł. (2011b), The Interdependence Between Energy Consumption and Eco- nomic Growth in the Polish Economy in the Last Decade. Managerial Economics, 9, 25–48, http://www.managerial.zarz.agh.edu.pl/EM_on_line/Managerial%20Econo- mics%20%209%20%282011%29.pdf. Gurgul, H., Lach, Ł. (2012), The Electricity Consumption versus Economic Growth of the Polish Economy, Energy Economics, 34(2), 500–510, DOI: http://dx.doi.org/10.1016/j.eneco.2011.10.017. Horowitz, J. L. (2003), The Bootstrap in Econometrics, Statistical Science, 18, 211–218, DOI: http://dx.doi.org/10.1214/ss/1063994976. Johansen, S. (1991), Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models, Econometrica, 59 (6), 1551–1580, DOI: http://dx.doi.org/10.2307/2938278. Karanfil, F. (2009), How Many Times Again Will We Examine the Energy–income Nexus Using a Limited Range of Traditional Econometric Tools?, Energy Policy, 36, 1191–1194, DOI: http://dx.doi.org/10.1016/j.enpol.2008.11.029. Kónya, L. (2006), Exports and Growth: Granger Causality Analysis on OECD Countries with a Panel Data Approach, Economic Modelling, 23, 978–992, DOI: http://dx.doi.org/10.1016/j.econmod.2006.04.008. Kraft, J., Kraft, A. (1978), On the Relationship between Energy and GNP, Journal of Energy Development, 3, 401–403. Menegaki, A. N., Ozturk, I. (2013), Growth and Energy Nexus in Europe Revisited: Evidence from Fixed Effects Political Economy Model, Energy Policy, 61, 881–887, DOI: http://dx.doi.org/10.1016/j.enpol.2013.06.076. Nazlioglu, S., Lebe, F., Kayhan, S. (2011), Nuclear Energy Consumption and Economic Growth in OECD Countries: Cross-sectionally Dependent Heterogeneous Panel Cau- sality Analysis, Energy Policy, 39, 6615–6621, DOI: http://dx.doi.org/10.1016/j.enpol.2011.08.007. ODYSSEE- MURE 2010, (2012), Energy Efficiency Policies and Measures in Slovakia in 2012, Monitoring of EU and National Energy Efficiency Targets, Slovak Innovation and Energy Agency, Bratislava, http://www.odyssee-indicators.org/publications/PDF/slovakia_nr.pdf [30.12.2013] Ozturk, I., Acaravci, A. (2010), The Causal Relationship between Energy Consumption and GDP in Albania, Bulgaria, Hungary and Romania: Evidence from ARDL Bound Test- ing Approach, Applied Energy, 87, 1938–43, DOI: http://dx.doi.org/10.1016/j.apenergy.2009.10.010. Pesaran, M. H. (2004), General Diagnostic Tests for Cross Section Dependence in Panels. Cambridge Working Papers in Economics No. 0435. Faculty of Economics, University of Cambridge, DOI: http://www.dspace.cam.ac.uk/handle/1810/446. Pesaran, M. H., Ullah, A., Yamagata, T. (2008), A Bias-adjusted LM Test of Error Cross- section Independence, The Econometrics Journal, 11, 105–127, DOI: http://dx.doi.org/10.1111/j.1368-423X.2007.00227.x. Monika Papież, Sławomir Śmiech DYNAMIC ECONOMETRIC MODELS 13 (2013) 51–68 68 Rilstone, P., Veall, M. R. (1996), Using Bootstrapped confidence intervals for improved inferences with seemingly unrelated regression equations, Econometric Theory, 12, 569–580, DOI: http://dx.doi.org/10.1017/S026646660000685X. Sarkar, A., Singh, J. (2010), Financing Energy Efficiency in Developing Countries—Lessons Learned and Remaining Challenges, Energy Policy, 38, 5560–5571, DOI: http://dx.doi.org/10.1016/j.enpol.2010.05.001. Soytas, U., Sari, R. (2007), The Relationship Between Energy and Production: Evidence from Turkish Manufacturing Industry, Energy Economics, 29, 1151–1165, DOI: http://dx.doi.org/10.1016/j.eneco.2006.05.019. Stern, D. I. (1993), Energy and Economic Growth in the USA, Energy Economics, 15, 137–150, DOI: http://dx.doi.org/10.1016/0140-9883(93)90033-N. Turner, K. (2009), Negative Rebound and Disinvestment Effects in Response to an Improve- ment in Energy Efficiency in the UK Economy, Energy Economics, 31, 648-666, DOI: http://dx.doi.org/10.1016/j.eneco.2009.01.008. Turner, K., Hanley, N. (2011), Energy Efficiency, Rebound Effects and the Environmental Kuznets Curve, Energy Economics, 33, 709–720, DOI: http://dx.doi.org/10.1016/j.eneco.2010.12.002. Zellner, A. (1962), An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias, Journal of the American Statistical Association, 57, 348–368, DOI: http://dx.doi.org/10.1080/01621459.1962.10480664. Wzrost gospodarczy i zużycie energii w krajach postkomunistycznych - bootstrapowa panelowa analiza przyczynowości Z a r y s t r e ś c i. Celem artykułu jest identyfikacja zależności przyczynowych (w sensie Grangera) pomiędzy zużyciem energii i wzrostem gospodarczym w krajach Europy Środko- wo Wschodniej oraz w krajach Bałtyckich w okresie 1993–2011. Jako narzędzie badawcze wykorzystano procedurę bootstrapowej panelowej analizy przyczynowości zaproponowaną przez Kónya (2006). Procedura ta pozwala na wnioskowanie w przypadku występowania zależności przestrzennych w badanej próbie i nie wymaga wstępnej analizy stacjonarności oraz umożliwia opis relacji dla poszczególnych analizowanych obiektów. Przeprowadzone badanie wskazuje na prawdziwość hipotezy wzrostu w przypadku trzech krajów oraz hipotezy sprzężenia zwrotnego w przypadku jednego kraju. Otrzymane wyniki pozwalają podejrzewać, że zależność przyczynowa pomiędzy wzrostem gospodarczym i konsumpcją energii jest związana z zmianami efektywności energetycznej. S ł o w a k l u c z o w e: zużycie energii, wzrost gospodarczy, bootstrapowa panelowa analiza przyczynowości, efektywność energetyczna. Acknowledgements We thank the anonymous referee for providing constructive comments and suggestions and help in significantly improving the contents of this paper.