The nexus between financial development and energy consumption in the EU: A dynamic panel data analysis

Energy Economics 39 (2013) 81–88

Contents lists available at SciVerse ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneco

The nexus between financial development and energy consumption in the EU: A dynamic panel data analysis Serap Çoban, Mert Topcu ⁎ Nevsehir University, Department of Economics, Nevsehir, Turkey

a r t i c l e

i n f o

Article history: Received 30 January 2013 Received in revised form 29 March 2013 Accepted 5 April 2013 Available online 12 April 2013 JEL classification: C33 O52 Q43

a b s t r a c t The relationship between financial development and energy consumption has newly started to be discussed in energy economics literature. This paper investigates this issue in the EU over the period 1990–2011 by using system-GMM model. No significant relationship is found in the EU27. The empirical results, however, provide strong evidence of the impact of the financial development on energy consumption in the old members. Greater financial development leads to an increase in energy consumption, regardless of whether financial development stems from banking sector or stock market. By contrast, we find for the new members that the impact of financial development on energy consumption depends on how financial development is measured. Using bankindex the impact of financial development displays an inverted U-shaped pattern while no significant relationship is detected once it is measured using stockindex. © 2013 Elsevier B.V. All rights reserved.

Keywords: Financial development Energy consumption System-GMM EU

1. Introduction In the European Union (2020) energy strategy, energy-efficiency is listed among five priorities. The European Commission (2010) Energy report propounds that the EU is aiming for a 20% cut in Europe's annual primary energy consumption by 2020. For this purpose, it is very crucial to determine the dynamics of energy consumption in the EU. This learning also helps to understand how energy demand in the EU countries is going to change in the future. The relationship between energy consumption and economic growth is certainly well-documented both theoretically and empirically (see, for example, Apergis and Payne, 2010; Apergis and Tang, 2013; Narayan and Smyth, 2008; Narayan et al., 2010; Sari and Soytas, 2007, etc.). There also exist several studies exploring this issue in the case of the EU (see, for example, Ciaretta and Zarraga, 2010; Menegaki, 2011; Pirlogea and Cicea, 2012). On the other hand, a consensus has emerged about the vital role of financial development on economic growth in recent years (see, for example, Al-Yousif, 2002; Fung, 2009; Kar et al., 2011; Masten et al., 2008, etc.) and it is also possible to find some

⁎ Corresponding author at: Nevsehir University, Faculty of Economics and Administrative Sciences, 50300 Nevsehir, Turkey. Tel.: +90 3842512007x1552. E-mail address: [email protected] (M. Topcu). 0140-9883/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.eneco.2013.04.001

empirical evidences from the EU (see, for example, Antonios, 2010; Wu et al., 2010; Zagorchev et al., 2011). Over the last three decades, the papers focusing the nexus between energy consumption and economic growth or financial development and economic growth have commonly found a significant relationship. It is therefore rational to anticipate a significant running between financial development and energy consumption as well. Theoretically, Sadorsky (2011) explains how financial development affects energy consumption in three ways (see, for detailed discussion, Sadorsky (2011)). These ways, presented in Table 1, could be named as effect channels. Nonetheless, Sadorsky (2010) points out that the theoretical relationship between the variables in question is unclear, and it could only be resolved through empirical analysis. In energy economics, the relationship between financial development and energy consumption has been attractive in recent years. Chtioui (2012) finds a uni-directional causal running from energy consumption to financial development both in the short and in the long run in Tunisia. Shahbaz and Lean (2012) find a long run relationship among energy consumption, economic growth, financial development, industrialization and urbanization in Tunisia. However, a long run bi-directional causality is also found between energy consumption and financial development, which do not support Chtioui (2012). Sadorsky (2010) examines the impact of financial development on energy consumption in twenty two emerging economies by using various

82

S. Çoban, M. Topcu / Energy Economics 39 (2013) 81–88

Table 1 Effect channels of financial development over energy consumption. Effect channels

Reflection of the effects

Direct effect

When there is improved financial development, consumers can borrow easier and cheaper in order to buy durable consumer goods which consume energy a lot. Improved financial development helps businesses in accessing to financial capital easily and less costly. Additionally, stock market development can also affect businesses by providing them additional funding source. These ones can let businesses expand their present business potential and increase energy demand. Increased stock market activity usually affects the confidence of consumers and businesses by creating wealth effect. Increased economic confidence may expand the economy and promote energy demand.

Business effect

Wealth effect

Source: Sadorsky (2011): 1000.

financial development indicators over the period 1990–2006. It is found that financial development affects energy consumption positively in this panel sample. In his further study, Sadorsky (2011) investigates the same issue on nine frontier economies of Central and Eastern Europe (CEE) during the period 1996–2006. This time he divides financial development indicators into two parts and shows that the impact of financial development has a significant positive impact when financial development is measured using bank related variables. Measurement using stock market variables, however, reveals that only stock market turnover has a significant positive impact on energy consumption. Islam et al. (2013) examine the relationship among energy consumption, aggregate production, financial development and population in Malaysia and they find that economic growth and financial development are effective on energy use both in the short and in the long run. Tang and Tan (2012), another paper in Malaysia, find that energy consumption and financial development are correlated in the long run and energy is a prominent resource for financial sector development. Xu (2012) investigates the nexus between financial development and energy consumption during the period 1999–2009 with a panel data set in 29 provinces of China, using system-GMM. The results indicate a positive significant relationship between financial development and energy consumption when financial development is measured using the ratio of loans in financial institution to GDP and the ratio of FDI to GDP. On the other hand, the results of Jalil and Feridun (2011) suggest that financial development leads to a decrease in environmental pollution in China. Kakar et al. (2011) imply that while financial development affects energy consumption in the long run; there is no significant relation in the short run in Pakistan. Moreover, the results fail to prove any causal running between financial development and energy consumption. Ozturk and Acaravcı (2013) yield evidence of a long run relationship between per capita energy consumption and financial development in Turkey. In addition, they reveal that while per capita energy consumption causes financial development in the short run; financial development does cause changes in per capita energy consumption in the long run. Mehrara and Musai (2012) conclude that financial development and energy consumption is cointegrated in the case of Iran. Al-mulali and Sab (2012a) find that energy consumption enables selected 19 countries to achieve high economic and financial development. In another paper, Al-mulali and Sab (2012b) also conclude that energy consumption plays an important role in increasing financial development in the Sub Saharan African countries. Mielnik and Goldemberg (2002) point out that foreign direct investment as an indicator of financial development has a significant negative impact on energy intensity on a sample of twenty developing economies. The purpose of this study is to examine the impact of financial development on energy consumption in the EU. In this regard we employ generalized method of moments (GMM) during the period 1990–2011. We further divide sample size into two groups as the old members,

EU15, that joined before 1996 and the new members that joined in 2004 and 2007 enlargements.1 Although there are some studies on this issue as is mentioned above; to the best of our knowledge there is no other published study in the case of the EU. This paper therefore aims to fulfill this gap and contributes to empirical literature. Moreover, we construct two indexes representing banking and stock market variables as an indicator of financial development employing Principle Component Analysis (PCA). The rest of the paper proceeds as follows: Section 2 describes empirical model and data, Section 3 presents PCA and GMM methodology, and Section 4 reveals empirical findings and discussion. Finally, we conclude in the last section. 2. Empirical model and data In this paper, we investigate the dynamic linkages between energy consumption and financial development for the EU countries during the period 1990–2011 using system-GMM model with a strongly balanced data. Following Sadorsky (2010), the empirical model is identified as a reduced form dynamic panel model of energy demand. Energy demand (energy) is described as a function of income (y), price (p) and financial development (fd). energyit ¼ α i energyit−1 þ βi1 yit þ βi2 pit þ βi3 f dit þ vi þ ϕt þ εit

ð1Þ

where i indicates the country (i = 1, …, 27) and t indicates the time period (t = 1990, …2011). We gathered annual data on energy consumption, real GDP per capita, energy prices and financial development indicators for the period 1990–2011. The data of energy consumption is measured as energy use in kg of oil equivalent per capita and real GDP per capita is measured as constant 2005 US dollars. Real oil prices are represented by energy prices.2 Since foreign direct investment is regarded as an emerging financial market determinant, it is added into the model. It is measured as net inflows as a share of GDP and denoted by fdi. All the data are sourced from the World Bank World Development Indicators (WDI) database. We divide financial development indicators into two main parts: baking variables and stock market variables as mentioned in the previous studies. Considering Beck and Demirguc-Kunt (2009), eleven banking variables and four stock market variables3 are chosen as financial development indicators. The bank related variables used in the paper are deposit money bank assets to GDP (dbagdp), financial system deposits to GDP (fdgdp), liquid liabilities as a share of GDP (llgdp), private credit to GDP (pcrdbgdp), bank overhead costs (overhead), net interest margin (netintmargin), concentration ratio (concentration), return on assets (roa), return on equity (roe), cost-income ratio, (costinc) and z-score. On the other hand, the stock market variables used in the paper are stock market turnover ratio (stturnover), stock market capitalization to GDP (stmktcap), stock market value traded to GDP (stvaltraded) and number of listed companies per 10,000 people (listco_pc). The data of these variables are gathered from the World Bank Financial Structure Database (2012). Modeling various related financial development indicators in a same equation could lead to multicollinearity. Additionally, observing aggregate effect of these indicators may be more efficient rather than 1 The 15 “old” member states of the EU are: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden and the United Kingdom. The 12 “new” member states of the EU are: Bulgaria, Cyprus, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Romania, Slovakia and Slovenia. 2 Energy price data are obtained by dividing Brent crude oil prices to each country's consumer price index (cpi, 2005 = 100). 3 Although a good number of stock market variables can be used as financial development indicator, only four of them are included in the study due to missing data matter for the most of the countries.

S. Çoban, M. Topcu / Energy Economics 39 (2013) 81–88

modeling each indicator separately. Thus, we bring banking and stock market variables together by employing “Principle Component Analysis (PCA)”. Hereby, two specific indexes are constructed for each group. In established models, all variables are in logarithmic form except fdi variable since it has negative values. Banking and stock market variables are also converted to logarithmic form after index values are obtained. Descriptive statistics of the data are presented in Table 2. Once we look at the mean values of the variables, lnenergy values are quite closer to each other in the new and the old members. While energy consumption is nearly same in the old and the new members, fdi in the new members equals to one fourth of the old members. This indicates that fdi remains lower in the new members relative to the old members. According to bankindex indicating the development of banking system, the new members seem to converge to the old members. In addition; stockindex in new members is one third of the old members, which demonstrates relatively underdeveloped system of the new members. 3. Methodology: PCA and system-GMM Understanding the impact of financial development on energy consumption and in this respect assessing the development of the financial sector in the EU require good measures of financial development (Levine, 2004). In this paper, to capture accurate and maximum effects of banking sector and stock market over financial development, eleven and four indicators are included respectively to construct indexes similar to some papers like Huang (2010), Saci and Holden (2008) and The World Economic Forum (2011). The method of principal component is used to extract a banking sector financial development index (bankindex) and stock market financial development index (stockindex). PCA is a modern tool of multivariate data analysis. It is a method to extract significant information from complex datasets. The main objective of PCA is to decrease the dimensionality in data. It is a technique that attempts to retain all the variation available in data even dealing with large set of variables. It transforms the data into new variables i.e. the principal components and they are not correlated. The maximum variation of the original variables is contained in the first few principal components (Jolliffe, 2002). PCA is normally applied as a method of variable reduction or for the detection of structure of relationship among

Table 2 Descriptive statistics. Obs.

Min.

Mean

Max.

Std. dev.

EU27 Lnenergy lngdp Lnoilprice Fdi Lnbankindex lnstockindex

561 579 594 560 550 500

3.195 3.772 −1.876 −29.229 0.598 −1.171

3.521 4.312 −0.786 14.101 1.612 0.095

3.974 4.870 6.654 564.916 2.133 0.323

0.167 0.221 0.891 60.691 0.223 0.187

Old members Lnenergy lngdp Lnoilprice Fdi Lnbankindex lnstockindex

315 320 330 317 314 311

3.224 4.209 −1.849 −14.922 0.945 −2.061

3.595 4.460 −1.000 20.937 1.639 1.519

3.974 4.870 −0.055 564.916 2.143 2.118

0.159 0.118 0.491 79.832 0.214 0.451

New members Lnenergy lngdp Lnoilprice Fdi Lnbankindex lnstockindex

246 259 264 243 235 211

3.195 3.772 −1.876 −29.229 0.711 −3.291

3.426 4.128 −0.520 5.183 1.427 0.675

3.800 4.435 6.654 52.052 2.136 1.863

0.123 0.176 1.167 6.720 0.274 0.867

83

the included variables. The information available in a group of variables is summarized by a number of mutually independent principal components. Each principal is basically the weighted average of the underlying variables. The first principal component always has the maximum variance for any of the combination. If more than one principal component is generated they are uncorrelated. For each principal component the eigenvalue (variance) indicates the percentage of variation in the total data explained. We employ the most recent data sets to construct our two main financial development indexes from 1990 to 2011 with 22 annual observations for each EU27, old members and new members. The output from a PCA is a table of factor scores or weights for each variable (see Table 3). Using the weights obtained from PCA, independent variables referring bankindex and stockindex can then be constructed for each country. These independent variables can be regarded as the countries' ‘financial development in banking sector’ and ‘financial development in stock market’ scores, and countries can be ranked according to mean values of these scores (see Table 4). Dynamic panel data estimation is more appropriate in cases where some unobservable factors affect both the dependent variable and the explanatory variables and some explanatory variables are strongly related to past values of the dependent variable. This is likely to be the case in regressions of financial development on energy consumption. Dynamic panel data model in Eq. (1) is proposed by Blundell and Bond (1998) and the extended version of the GMM estimator also known as system-GMM (sys-GMM). It is derived from the estimation of a system of two simultaneous equations: one in levels (with lagged first differences as instruments) and the other in first differences (with lagged levels as instruments). In presence of heteroscedasticity and serial correlation the two-step sys-GMM uses a consistent estimate of the weighting matrix taking the residuals from the one-step estimate (Davidson and MacKinnon, 2004). Though asymptotically more efficient, the two-step GMM carries out estimations of the standard errors that tend to be critically downward biased. However it is possible to overcome this problem using the finite-sample correction to the two-step covariance matrix developed by Windmeijer (2005) which can make two-step robust GMM estimates more efficient than one-step robust ones especially for sys-GMM (Roodman, 2009b). Another weakness of GMM estimations is too many instrument problem. There are various methods that are used to reduce instrument variable number. The first one is to use only certain lags instead of all available lags for instruments (limited lags). The second one, called as collapsing, is to combine instruments by adding them into smaller sets. Another way is to use the two techniques together (Roodman, 2009b). We use the sys-GMM to estimate the models in the present paper. Lagged energy consumption, income, and price are each treated as endogenous in the estimation of Eq. (1), considering the relations between each other. There are several reasons for preferring a dynamic sys-GMM panel model. First, static panel estimation omits dynamics causing dynamic panel estimation bias (Baum, 2006; Bond, 2002). Omitted dynamics means that such models are misspecified, because they pass over the impacts of lagged dependent variable as a right-hand-side variable on dependent variable (Bond, 2002). Second, the endogeneity problem which occurs when the independent variable is correlated with the error term in a regression model can be solved easier in dynamic panel data models than in the static models. Third, in multivariable dynamic panel models the sys-GMM estimator is known to perform better than the differenced-GMM (DIF-GMM) proposed by Arellano and Bond (1991). The sys-GMM estimation is more appropriate when variables are “random walk” or close to be random walk variables (Bond, 2002; Roodman, 2009a, 2009b) because DIF-GMM estimator can suffer from a weak instruments problem in that case (Sarafidis et al., 2009). Fourth, sys-GMM is a more consistent estimator when series are persistent, in which the lagged levels of variables are weak instruments for subsequent changes, and there is a dramatic reduction in the

84

S. Çoban, M. Topcu / Energy Economics 39 (2013) 81–88

Table 3 Weights obtained from principal components method for country groups. Banking sector variables Llgdp Fdgdp Dbagdp Pcrdbgdp Overhead Netintmargin Concentration Roa Roe Costinc Zscore

Stock market variables EU27

Old members

New members

0.1001 0.0825 0.1128 0.1151 0.0819 0.0909 0.1163 0.0891 0.0621 0.1152 0.0339

0.0945 0.0954 0.1126 0.1121 0.0788 0.0534 0.0562 0.1115 0.1143 0.0747 0.0964

0.1178 0.1191 0.1204 0.1171 0.0670 0.1022 0.0584 0.1129 0.1040 0.0309 0.0503

Stmktcap Stvaltraded Stturnover listco_pc

EU27

Old members

New members

0.1767 0.2999 0.2400 0.2834

0.2774 0.3574 0.3519 0.0134

0.1332 0.3239 0.2520 0.2910

Note: All outputs of PCA are not reported. They are available upon request.

finite sample bias due to the exploitation of additional moment conditions (Blundell and Bond, 1998, 2000; Blundell et al., 2000; Roodman, 2009a). Apart from the reasons mentioned above, if one works with an unbalanced panel data then it is better to avoid DIF-GMM estimation, which has a weakness of magnifying gaps (Roodman, 2009b). Although our panel data is strongly balanced, we avoid using DIF-GMM estimation. However, GMM estimators assume that the disturbances of error terms are cross-sectionally independent. In the econometric literature several tests for cross section dependence have been proposed (see, for example, Breusch and Pagan, 1980; Frees, 1995; Friedman, 1937; Pesaran, 2004, 2006; Ng, 2006). One of these tests is CD test proposed by Pesaran (2004) which can be applied to a wide variety of models, including heterogeneous dynamic models with small/large N and T. In this paper, the CD test is used to detect whether there is any cross sectional dependence in the data. In the presence of cross-sectional dependence, it is widely applied including time dummies or crosssectionally demeaning data in order to alleviate any common timevarying shocks. Applying this transformation one is able to coped with cross-sectional dependence, unless their impact differs across individual cross-sections (heterogeneous cross-sectional dependence)

(Sarafidis et al., 2009). After transforming data or including time dummies and applying GMM, however, there can be still crosssectional dependence in error terms which is the case under heterogeneous error cross-sectional dependence. If only homogenous crosssectional dependence is present, the inclusion of time dummies or demeaning data would be sufficient to remove any bias in the estimation approach (Sarafidis and Robertson, 2009). Sarafidis et al. (2009) proposed a new testing procedure for error cross section dependence after estimation of a linear dynamic panel data model with covariates by GMM. This test is valid when N > T (see, for details, Sarafidis et al., 2009). The test has following simple (C-Statistic based) form: d

2

C CDGMM ¼ ðSar Unres −Sar Res Þ → X hd

ð2Þ

where hd is the number of degrees of freedom of the test statistic as difference between the set of instruments (number of moment conditions) in the unrestricted model (SarUnres) and the restricted model (SarRes), where the GMM model has either DIF-GMM or sys-GMM form augmented by time dummies. The null hypothesis of this test is Sargan's difference test where there is homogenous cross-section

Table 4 Mean values of banking and stock market development indexes and ranking of countries. No.

Ranking of EU27

No

Bankindex 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

Lux. Cyprus UK Netherlands Spain Malta Portugal Germany Belgium Austria Denmark Ireland France Italy Finland Greece Slovak R. Sweden Estonia Czech R. Hungary Bulgaria Slovenia Lithuania Latvia Poland Romania

92.54 78.80 64.13 62.44 58.56 58.03 56.43 51.79 50.96 48.99 48.46 47.06 43.86 43.20 43.05 42.94 36.87 36.83 36.40 35.74 32.60 32.12 29.60 28.36 27.31 26.01 19.53

Ranking of old members

Stockindex UK Netherlands Spain Sweden Finland France Germany Italy Denmark Lux. Greece Ireland Portugal Belgium Hungary Czech R. Cyprus Austria Poland Estonia Slovak R. Malta Slovenia Lithuania Bulgaria Romania Latvia

77.20 68.62 65.93 61.18 56.25 44.05 43.73 36.48 34.29 26.97 25.48 25.33 23.33 22.56 20.81 18.95 17.70 17.14 14.28 12.21 8.21 7.88 7.25 5.77 5.35 4.86 4.30

No

Bankindex 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Lux. UK Netherlands Spain Portugal Germany Austria Belgium Ireland Denmark France Italy Finland Greece Sweden

93.66 59.70 56.26 55.45 52.31 47.09 45.68 44.88 44.04 43.24 39.35 39.07 36.82 35.39 32.41

Ranking of new members

Stockindex UK Netherlands Spain Sweden Finland France Germany Italy Denmark Lux. Greece Ireland Portugal Belgium Austria

106.21 94.14 90.36 84.35 77.69 61.41 61.23 51.33 48.16 41.57 36.02 36.01 33.14 32.53 24.65

Bankindex 1 2 3 4 5 6 7 8 9 10 11 12

Cyprus Malta Czech R. Slovak R. Estonia Hungary Bulgaria Slovenia Poland Latvia Lithuania Romania

80.80 58.13 30.37 29.16 28.50 27.40 27.40 23.79 20.91 20.59 19.68 14.32

Stockindex Hungary Czech R. Cyprus Poland Estonia Slovak R. Slovenia Malta Lithuania Bulgaria Romania Latvia

21.01 18.85 16.73 14.21 11.64 8.36 6.79 6.48 5.34 5.09 4.64 4.15

S. Çoban, M. Topcu / Energy Economics 39 (2013) 81–88

dependence in the model versus the alternative of heterogeneous cross-sectional dependence as: H 0 : Var ðϕi Þ ¼ ∑ϕ ¼ 0

H 1 : Var ðϕi Þ ¼ ∑ϕ≠0:

ð3Þ

The restricted set of moment conditions thereby only includes instruments lagged values of income and price in Eq. (1). In this paper, this test is applied only for the EU27 due to data structure where N is larger than T. For the subsamples Sargan's standard difference tests are calculated as the difference of two Sargan/Hansen J-statistics for an unrestricted and a restricted sys-GMM model. 4. Empirical findings and policy implications The empirical analysis is applied using sys-GMM estimator for the EU27 and two sub-groups including the old members and the new members. Since EU27 includes new members, potential crosssectional dependences are investigated to ensure validity of the statistical results. According to CD test of Pesaran (2004), all models contain cross-sectional dependence. 4 However, following Sarafidis et al. (2009), time dummies are included into the models for EU27 in order to eliminate cross-sectional dependence. CCD-GMM test, as described above, is then used to investigate whether error cross-sectional dependence still exists or not. The test results do not indicate any sign of misspecification after including period-fixed effects for standard significance levels. As seen in Table 5, it is not rejected the null of homogenous cross-sectional dependence in these tests results. Results of AR(1) and AR(2) tests support the validity of the model specification for the EU27, old members and the new members. Hansen J-test, the most commonly used diagnostic in GMM estimations, explores the validity of the full set of overidentifying restrictions (instruments). This test confirms the validity of instruments in the all models. Following the suggestion of Roodman (2009b) for sys-GMM estimations, difference-in-Hansen test for the full set of instruments for levels equation, as well as the subset based on the dependent variable are reported. According to test results, all model specifications are valid. The “steady state” assumption suggested by Roodman (2009a) requires a kind of steady state in the sense of deviations from longterm values that are not systematically correlated with the fixed effect. In other words, the estimated coefficient on the lagged dependent variable in the model should point out convergence by having a value less than (absolute) unity (Roodman, 2009b). Since the estimated coefficients on the lagged energy variables in all models lie between 0.747 and 0.987, the steady state assumption which is used to check for the validity of instruments in sys-GMM holds. Sys-GMM panel estimation results for the EU27 are shown in Table 5. Consistent with the most existing literature, each specification includes a lagged value of energy use, and current period income and prices. In each regression, therefore, model 1 includes basis variables which commonly determine energy use. Models 2–6, on the other side, adds financial development indicators to basis model in order to monitor the nexus between energy consumption and financial development. For the EU27 lagged energy demand is significant, positive and highly persistent for each model. It shows that energy consumption in a certain year is strongly affected by its previous value. While real GDP is significant in basis model and model 4, it is not statistically significant for the rest of the models. This evidence implies that real income is not a clear determinant of energy consumption in the EU27 despite of the expectation which propounds a significant positive relationship between income and energy demand. Oil price has a statistically significant negative impact on energy consumption 4 Pesaran's CD test results are not reported herein. They can be provided from the authors.

85

from model 1 to model 4. Models 5 and 6 proves that the impact of oil price on energy consumption is negative but insignificant. This indicates that once oil prices as a measure of energy prices increase demand for energy decreases. Despite of the significant and positive impact, foreign direct investments affect energy demand extremely little and one would argue that it is not a dynamic of energy demand in the EU27. Although the relationship between financial development and energy demand is estimated under linear and non-linear equations, 5 neither banking nor stock market indexes have a significant impact on energy use. In general, sys-GMM panel estimation results do not show much indication that financial development variables impact energy consumption in the EU27. Considering related literature, two reasons could be addressed why energy demand is inexplicable with financial development. The first one is about model. While previous papers add banking and stock market variables into the model, we bring each indicator together employing PCA and construct an index for each group. The second one is about union's structure. Since EU consists of various members who have different economic structures, evaluating EU as a whole lead not to find any significant relationship between financial development and energy consumption. Given the fact that old members have advanced economic potential compared to new members, dividing EU into two parts could help to obtain more significant results. Tables 6 and 7 report sys-GMM panel estimation results for the old members and the new members, respectively. Specific to old members, basis and additional models indicate that lagged value of energy demand and income has statistically significant and positive impact on energy demand. Oil price, on the other hand, has a statistically significant and negative impact on energy consumption for each model in line with the theory. The coefficient of foreign direct investment variable is significant and positive although its effect is very limited. The strongest evidence supporting the hypothesis that financial development increases energy consumption comes from models 3 and 4. They present that banking and stock market indexes are statistically significant and have positive impact on energy consumption. Recent papers in energy economics have found that growth stimulates energy consumption in developed economies (see, for example, Mahadevan and Asafu-Adjaye, 2007; Tugcu et al., 2012; Yildirim and Aslan, 2012). As economic growth is positively correlated with baking sector and stock market developments, finding a positive relation between banking/stock market index and energy consumption in the case of the old members is well-balanced. This is also consistent with the previous literature covering emerging and European frontier economies (Sadorsky, 2010, 2011). In Table 7, model 1 reports that lagged value, income and energy price have significant and expected impacts on energy consumption. According to model 2, foreign direct investment has no significant impact anymore in the new members. This evidence clearly points out 5 Several empirical specifications related to financial development, especially the papers which investigate the relationship between financial development and inequality, include squared terms for the financial development variables (see, for example: Batuo et al., 2010; Jauch and Watzka, 2012; Liang, 2006). However, sectoral composition of an economy can provide clues about the potential for economy-wide growth and economic growth increases the demand for energy use. Transition process, which causes a gradual shift from manufacturing to services, has changed the sectoral output composition in transition economies. As our data include the new EU members, one would expect financial development to have a nonlinear impact on the new members' energy consumption via this sectoral shift. In addition, the literature on financial development-economic growth nexus has suggested that financial development may have a nonlinear impact on economic growth (see, for example, Asal, 2012; Hung, 2009; Masten et al., 2008). In the countries with very low levels of financial development, additional improvement in financial markets might have an uncertain effect on growth (Rioja and Valev, 2004: 429). Similarly, one would argue that the reason for the nonlinearity of finance–growth relationship might be that financial development helps catch up to the productivity frontier but limited or no growth effect for the countries that are close or at the frontier (Akhion et al., 2005). We therefore estimate nonlinear regression for the samples where the new members are included.

86

S. Çoban, M. Topcu / Energy Economics 39 (2013) 81–88

Table 5 Two step sys-GMM panel estimation regression results for the EU27.

Lnenergy (−1) Lngdppcap Lnoilprice Fdi Lnbankindex Lnbankindexsq Lnstockindex Lnstockindexsq AR(1) (p-value) AR(2) (p-value) CCD-GMM (p-value) Hansen J-test (p-value) Difference Hansen tests (p-value) All sys-GMM instruments Those based on lagged energy only Instruments Observations

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

0.949⁎⁎⁎ (33.12) 0.030⁎ (1.91) −0.019⁎⁎ (−2.24)

0.951⁎⁎⁎ (32.09) 0.020 (1.30) −0.021⁎⁎⁎ (−3.26) 0.00004(1.49)

0.987⁎⁎⁎ (40.82) −0.024 (−0.64) −0.022⁎ (−2.04)

0.972⁎⁎⁎ (51.88) 0.032⁎ (1.78) −0.013⁎ (−1.78)

0.972⁎⁎⁎ (21.12) −0.019 (−0.54) −0.002 (−0.12)

0.951⁎⁎⁎ (51.85) 0.048 (1.19) −0.003 (−0.19)

0.044 (1.03)

0.453(1.13) −0.115 (−0.99) −0.025 (−1.68)

(0.014) (0.464) (0.204) (0.363)

(0.017) (0.532) (0.304) (0.391)

(0.009) (0.565) (0.184) (0.283)

(0.016) (0.578) (0.217) (0.526)

(0.005) (0.988) (0.302) (0.942)

−0.029 (−0.84) 0.024 (0.51) (0.010) (0.514) (0.175) (0.821)

(0.363) (0.633) 27 435

(0.391) (0.669) 27 407

(0.283) (0.505) 27 409

(0.526) (1.000) 28 423

(0.942) (0.963) 29 437

(0.726) (0.953) 29 398

Note: Robust t-statistics are reported in parentheses of estimated coefficients. GMM type variables are lnenergy, lngdppcap and lnoilprice and their lags' range is set to from one to four in all models. Following the suggestions of Roodman (2009b), the standard type instrumental variables are first, second, and third lags of fdi, lnbankindex, lnstockindex, lnbankindexsq, lnstockindexsq and time dummies. Time dummies which are not reported in the table are involved in the regressions for eliminating cross-sectional dependence following Sarafidis et al. (2009). AR(1) and AR(2) are tests for autocorrelation in differences. CCD-GMM is Sargan's difference test proposed by Sarafidis et al. (2009). Hansen J-test is a test for over identification restrictions. p-Values for these tests are shown in parentheses. ⁎ p b 0.10. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.

that foreign direct investments cannot be regarded as a financial development indicator in contrast to EU15 and EU27 due to relative importance of portfolio investments in emerging countries. Models 3 and 4 respectively show that banking and stock market indexes do not have significant impact on energy consumption. Non-linear estimations, reported in models 5 and 6, prove that the relationship

Table 6 Sys-GMM panel estimation regression results for the old members. Model 1

Model 2

Model 3

Model 4

lngdppcap

0.954⁎⁎⁎ (47.22) 0.042⁎⁎

0.932⁎⁎⁎ (43.33) 0.052⁎⁎⁎

0.747⁎⁎⁎ (6.55) 0.168⁎⁎

0.831⁎⁎⁎ (12.89) 0.124⁎⁎

lnoilprice

(2.66) −0.009⁎⁎⁎

(3.11) −0.0096⁎⁎⁎ (−3.20) 0.00003⁎⁎ (2.57)

(1.90) −0.026⁎⁎ (−2.76)

(2.57) −0.015⁎⁎ (−2.64)

lnenergy (−1)

(−3.13) Fdi lnbankindex

0.079⁎⁎⁎ (3.82)

(0.002) (0.659) (0.272)

(0.002) (0.693) (0.230)

(0.001) (0.423) (0.431)

0.016⁎⁎ (2.11) (0.002) (0.738) (0.315)

(0.441)

(0.477)

(0.634)

(0.801)

(0.690)

(0.674)

(0.309)

(0.685)

15 263

15 262

13 261

11 264

lnstockindex AR(1) (p-value) AR(2) (p-value) Hansen J-test (p-value) Difference Hansen tests (p-value) All sys-GMM instruments Those based on lagged energy only Instruments Observations

Note: Robust t-statistics are reported in parentheses of estimated coefficients. GMM type variables are lnenergy, lngdppcap and lnoilprice and their lags' range is set to from one to four in all models. Following the suggestions of Roodman (2009b), the standard type instrumental variables are first, second, and third lags of fdi, lnbankindex, lnstockindex, lnbankindexsq, and lnstockindexsq. Hansen J-test is a test for over identification restrictions. p-Values for these tests are shown in parentheses. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.

between energy consumption and bankindex displays an inverted U-shaped pattern. This finding may be about transition from manufacturing industry to services in the CEE countries. It makes sense that greater banking activity leads to more manufacturing and energy consumption. And a gradual shift from manufacturing sector to services leads to a drop in energy consumption. However, financial development does not affect the demand for energy when it is measured using stockindex. For the nonlinear relationship between bankindex and energy consumption, we calculate the turning point for bankindex 6 so as to find out whether there is any country in the sample that has passed the turning point. Comparison results 7 show that Cyrus, Hungary and Malta have passed the turning point. This indicates that banking expansion and access to financial capital in these new members accelerates energy consumption much more than those of the rest. In line with Table 4, this finding points out the relative importance of banking activities in the financial system of these countries. Comparing the findings of the old members with the new members reveals an obvious distinction. Greater financial development accelerates energy consumption in the EU15, regardless of whether financial development is measured using bankindex or stockindex. On the other side, stockindex is no longer an effective indicator on energy consumption for the new members. This shows that while an improvement in financial sector results in increased energy demand in industrialized EU countries, the impact of financial development on energy consumption is sensitive to how it is measured in emerging countries. Since stock market activities are not as effective as banking activities in emerging countries, this evidence supports the argument in Sadorsky (2011) that banking sector has to be developed first in order to have a well-functioning stock market. In general, as concluded by Caporale et al. (2009), the evidences from the new EU members suggest that the stock markets are still underdeveloped and their contribution to economic growth and energy demand is limited because of a lack of financial depth. However, a more efficient banking sector is found to have accelerated growth and energy consumption. 6

Turning point for bankindex is calculated as 0.336 / (2 × 0.124) = 1.354. Country level financial development results which are not reported herein can be provided from the authors. 7

S. Çoban, M. Topcu / Energy Economics 39 (2013) 81–88

87

Table 7 Sys-GMM panel estimation regression results for the new members.

lnenergy (−1) Lngdppcap Lnoilprice Fdi Lnbankindex Lnbankindexsq Lnstockindex Lnstockindexsq AR(1) (p-value) AR(2) (p-value) Hansen J-test (p-value) Difference Hansen tests (p-value) All sys-GMM instruments Those based on lagged energy only Instruments Observations

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

0.945⁎⁎⁎ (47.69) 0.044⁎⁎ (2.72) −0.013⁎⁎⁎ (−10.27)

0.968⁎⁎⁎ (71.49) 0.025⁎⁎ (2.34) −0.005⁎⁎ (−2.43) −0.0002(−0.69)

0.941⁎⁎⁎ (18.14) 0.055 (1.15) −0.002 (−0.42)

0.982⁎⁎⁎ (9.21) 0.016 (0.18) 0.005 (0.53)

0.863⁎⁎⁎ (31.66) 0.059⁎⁎ (2.34) −0.001 (−0.33)

0.961⁎⁎⁎ (51.85) 0.036 (0.66) 0.008 (−1.09)

0.336⁎⁎ (2.54) −0.124⁎⁎ (−2.72)

−0.020 (−1.03) −0.004 (−0.27) (0.041) (0.423) (0.762)

(0.044) (0.348) (0.403)

(0.044) (0.383) (0.317)

(0.049) (0.314) (0.541)

(0.040) (0.439) (0.820)

−0.005 (−0.71) −0.003 (−1.45) (0.052) (0.326) (0.536)

(0.769) (0.467) 9 230

(0.226) (0.452) 12 197

(0.488) (0.126) 11 202

(0.433) (0.265) 11 191

(0.759) (0.893) 11 201

(0.382) (0.331) 11 191

Note: Robust t-statistics are reported in parentheses of estimated coefficients. GMM type variables are lnenergy, lngdppcap and lnoilprice and their lags' range is set to from one to four in all models. Following the suggestions of Roodman (2009b), the standard type instrumental variables are first, second, and third lags of fdi, lnbankindex, lnstockindex, lnbankindexsq, and lnstockindexsq. Hansen J-test is a test for over identification restrictions. p-Values for these tests are shown in parentheses. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.

Energy conservation policies may deteriorate economic growth in the EU15. On the other side, limited financial development in the new EU members leads them not to access to technology which provides energy efficiency due to high costs. This also results in that energy intensity in these countries is above the average of the EU. Growth can be promoted on the condition that a well-functioning baking system improves stock market effectiveness, and thus; demand for energy in these countries remain increasing in the future. In making energy saving policies, a great importance should be given to the position of these countries in terms of energy consumption. Additionally, taking required measures could also contribute to decrease EU's global gas emission.

empirical evidences from the EU such as Wu et al. (2010), Zagorchev et al. (2011), Pirlogea and Cicea (2012). Furthermore, such a nonlinear relationship may explain why existing energy consumption-growth literature in emerging countries appears to exhibit dissimilar results. Finally, the results of this study contribute to the results of Sadorsky (2011) by suggesting that considering financial development indicators into two parts may yield more effective results if the cross-section unit of the panel shows homogenous structure. In this respect, researchers who will investigate an issue related to energy economics in the EU in the future should take this heterogeneity into account.

5. Conclusion

Acknowledgements

This study is probably the first to investigate the dynamic relationship between financial development and energy consumption in the case of the EU. System-GMM model for a panel of twenty seven member states is employed based on annual data ranging from 1990 to 2011. There is no much indication that financial development affects energy consumption in the EU. Nevertheless, when the sample is divided into two parts as the old and the new members, we achieve to obtain expressive results. The empirical results provide strong evidence of the impact of the financial development on energy consumption in the case of the old members. Regardless of whether financial development stems from banking sector or stock market, greater financial development leads to an increase in energy consumption. This finding is consistent with the bulk of the financial development-energy literature (see, for example, Islam et al., 2013; Kakar et al., 2011; Ozturk and Acaravcı, 2013; Sadorsky, 2010, 2011; Xu, 2012). Contrary to the old members, we find strong support for the new members that the impact of financial development on energy consumption depends on how financial development is measured. When it is measured using stock market variables, no significant relationship is found. On the other hand, however, when it is measured using banking variables the impact of financial development displays an inverted U-shaped pattern. As similar with Hassan et al. (2011), the inverted U-shaped relationship in emerging countries shows that a well-functioning banking system is a necessary but not a sufficient condition to reach steady economic growth which then triggers energy consumption. Hence, the difference between the newer and the older members strongly implies the role of economic development, which is supported by some

We would like to gratefully acknowledge the kind guidance and help of Perry Sadorsky during the early stages of this study. We also wish to thank anonymous reviewers for their constructive suggestions. Any remaining errors are the authors' own responsibility.

References Akhion, H., Howitt, P., Mayer-Foulkes, D., 2005. The effect of financial development on convergence: theory and evidence. Q. J. Econ. 120, 173–222. Al-mulali, U., Sab, C.N.B.C., 2012a. The impact of energy consumption and CO2 emissions on the economic and financial development in 19 selected countries. Renew. Sustain. Energy Rev. 16, 4365–4369. Al-mulali, U., Sab, C.N.B.C., 2012b. The impact of energy consumption and CO2 emissions on the economic growth and financial development in the Sub Saharan African countries. Energy 39, 180–186. Al-yousif, Y.K., 2002. Financial development and economic growth: another look at the evidence from developing countries. Rev. Financ. Econ. 11 (2), 131–150. Antonios, A., 2010. Financial development and economics growth a comperative study between 15 European Union member states. Int. Res. J. Financ. Econ. 35, 143–149. Apergis, N., Payne, J.E., 2010. Energy consumption and growth in South America: evidence from a panel error correction model. Energy Econ. 32 (6), 1421–1426. Apergis, N., Tang, C.F., 2013. Is the energy-led growth hypothesis valid? New evidence from a sample of 85 countries. Energy Econ. 38, 24–31. Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58, 277–297. Asal, M., 2012. Non-linear growth impacts of financial development in Euro Area. Int. J. Econ. Financ. 4 (12), 23–38. Batuo, E., Guidi, F., Mlambo, K., 2010. Financial development and income inequality: evidence from African countries. MPRA Paper, No: 25658. (http://mpra.ub.unimuenchen.de/25658/). Baum, F.C., 2006. An Introduction to Modern Econometrics Using Stata. Stata Press, Texas.

88

S. Çoban, M. Topcu / Energy Economics 39 (2013) 81–88

Beck, T., Demirguc-Kunt, A., 2009. Financial institutions and markets across countries and over time: data and analysis. World Bank Policy Research Working Paper 4943, May 2009. Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. J. Econ. 87, 115–143. Blundell, R., Bond, S., 2000. GMM estimation with persistent panel data: an application to production functions. Economet. Rev. 87, 115–143. Blundell, R.W., Bond, S.R., Windmeijer, F., 2000. Estimation in dynamic panel data models: improving on the performance of the standard GMM estimator. In: Baltagi, B. (Ed.), Nonstationary Panels, Panel Cointegration, and Dynamic Panels. Advances in Econometrics, 15. JAI Press, Elsevier Science. Bond, S., 2002. Dynamic panel models: a guide to micro data methods and practice. Institute for Fiscal Studies. Department of Economics, UCL, CEMMAP (Centre for Microdata Methods and Practice) Working Paper No. CWPO9/02. Breusch, T., Pagan, A., 1980. The lagrange multiplier test and its applications to model spesification in econometrics. Rev. Econ. Stud. 47 (1), 239–253. Caporale, G.M., Rault, C., Sova, R., Sova, A., 2009. Financial development and economic growth: evidence from ten new EU members. German Institute for Economic Research, Working Paper No: 940. Chtioui, S., 2012. Does economic growth and financial development spur energy consumption in Tunisia? J. Econ. Int. Financ. 4 (7), 150–158. Ciaretta, A., Zarraga, A., 2010. Economic growth-electricity consumption causality in 12 European Countries: a dynamic panel data approach. Energy Policy 38 (7), 3790–3796. Davidson, R., MacKinnon, J.G., 2004. Econometric Theory and Methods. Oxford University Press. European Commission, 2010. Energy report. http://ec.europa.eu/energy/observatory/ annual_reports/annual_reports_en.htm. European Union 2020. Energy strategy. http://ec.europa.eu/energy/strategies/2010/ 2020_en.htm. Frees, E.W., 1995. Assessing cross-sectional correlation in panel data. J. Econ. 69, 393–414. Friedman, M., 1937. The use of ranks to avoid the assumption of normality implicit in the analysis of variance. J. Am. Stat. Assoc. 32, 675–701. Fung, M.K., 2009. Financial development and economic growth: convergence or divergence? J. Int. Money Finance 28 (1), 56–67. Hassan, M.K., Sanchez, B., Yu, J.S., 2011. Financial development and economic growth: new evidence from panel data. Q. Rev. Econ. Financ. 51, 88–104. Huang, Y., 2010. Determinants of Financial Development. Palgrave Macmillan, USA. Hung, F.S., 2009. Explaining the nonlinear effects of financial development on economic growth. J. Econ. 97 (1), 41–65. Islam, F., Shahbaz, M., Ahmed, A.U., Alam, M.M., 2013. Financial development and energy consumption nexus in Malaysia: a multivariate time series analysis. Econ. Model. 30, 435–441. Jalil, A., Feridun, M., 2011. The impact of growth, energy and financial development on the environment in China: a cointegration analysis. Energy Econ. 33, 284–291. Jauch, S., Watzka, S., 2012. Financial development and income inequality: a panel data approach. CESifo Working Paper, No: 3687, October 2012. Jolliffe, I.T., 2002. Principal Component Analysis, 2nd ed. Springer-Verlag, New York. Kakar, Z.K., Khilji, B.A., Khan, B.J., 2011. Financial development and energy consumption: empirical evidence from Pakistan. Int. J. Trade Econ. Financ. 2 (6), 469–471. Kar, M., Nazlioglu, S., Agır, H., 2011. Financial development and economic growth nexus in the MENA countries: bootstrap panel granger causality analysis. Econ. Model. 28 (1–2), 685–693. Levine, R., 2004. Finance and growth: theory and evidence. NBER Working Paper No. W10766. Liang, Z., 2006. Financial development and income distribution: a system GMM panel analysis with application to urban China. J. Econ. Dev. 31 (2), 1–21. Mahadevan, R., Asafu-Adjaye, J., 2007. Energy consumption, economic growth and prices: a reassessment using panel VECM for developed and developing countries. Energy Policy 35 (4), 2481–2490. Masten, A.B., Coricelli, F., Masten, I., 2008. Non-linear growth effects of financial development: does financial integration matter? J. Int. Money Finance 27 (2), 295–313. Mehrara, M., Musai, M., 2012. Energy consumption, financial development and economic growth: an ARDL approach for the case of Iran. Int. J. Bus. Behav. Sci. 2 (6), 92–99.

Menegaki, A.N., 2011. Growth and renewable energy in Europe: a random effect model with evidence for neutrality hypothesis. Energy Econ. 33 (2), 257–263. Mielnik, O., Goldemberg, J., 2002. Foreign direct investment and decoupling between energy and gross domestic product in developing countries. Energy Policy 30 (2), 87–89. Narayan, P.K., Smyth, R., 2008. Energy consumption and real GDP in G7 countries: new evidence from panel cointegration with structural breaks. Energy Econ. 30 (5), 2331–2341. Narayan, P.K., Narayan, S., Popp, S., 2010. A note on the long-run elasticities from the energy consumption-GDP relationship. Appl. Energy 87 (3), 1054–1057. Ng, S., 2006. Testing cross-section correlation in panel data using spacings. J. Bus. Econ. Stat. 24, 12–23. Ozturk, I., Acaravcı, A., 2013. The long run and causal analysis of energy, growth, openness and financial development on carbon emissions in Turkey. Energy Econ. 36, 262–267. Pesaran, M.H., 2004. General diagnostic tests for cross section dependence in panels. Unpublished manuscript, Faculty of Economics and Politics, University of Cambridge. Pesaran, M.H., 2006. Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica 74, 967–1012. Pirlogea, C., Cicea, C., 2012. Econometric perspective of the energy consumption and economic growth relation in European Union. Renew. Sustain. Energy Rev. 16 (8), 5718–5726. Rioja, F., Valev, N., 2004. Does one size fit all?: a reexamination of the finance and growth relationship. J. Dev. Econ. 74, 429–447. Roodman, D.M., 2009a. How to do xtabond2: an introduction to difference and system GMM in Stata. Stata J. 9 (1), 86–136. Roodman, D.M., 2009b. A note on the theme of too many instruments. Oxf. Bull. Econ. Stat. 71 (1), 135–158. Saci, K., Holden, K., 2008. Evidence on growth and financial development using principal components. Appl. Financ. Econ. 18 (19), 1549–1560. Sadorsky, P., 2010. The impact of financial development on energy consumption in emerging economies. Energy Policy 38 (5), 2528–2535. Sadorsky, P., 2011. Financial development and energy consumption in Central and Eastern European frontier economies. Energy Policy 39 (2), 999–1006. Sarafidis, V., Robertson, D., 2009. On the impact of error cross-sectional dependence in short dynamic panel estimation. Econ. J. 12, 62–81. Sarafidis, V., Yamagata, T., Robertson, D., 2009. A test of cross section dependence for a linear dynamic panel model with regressors. J. Econ. 148, 149–161. Sari, R., Soytas, U., 2007. The growth of income and energy consumption in six developing countries. Energy Policy 35 (2), 889–898. Shahbaz, M., Lean, H.H., 2012. Does financial development increase energy consumption? The role of industrialization and urbanization in Tunisia. Energy Policy 40, 473–479. Tang, C.F., Tan, B.W., 2012. The linkages among energy consumption, economic growth, relative price, foreign direct investment, and financial development in Malaysia. Qual. Quant. http://dx.doi.org/10.1007/s11135-012-9802-4. The World Economic Forum, 2011. The Financial Development Report 2011. World Economic Forum USA Inc., New York, USA. Tugcu, C.T., Ozturk, I., Aslan, A., 2012. Renewable and non-renewable energy consumption and economic growth relationship revisited: evidence from G7 countries. Energy Econ. 34 (6), 1942–1950. Windmeijer, F., 2005. A finite sample correction for the variance of linear efficient twostep GMM estimators. J. Econ. 126 (1), 25–51. World Bank Financial Structure Database, 2012. World Bank World Development Indicators (WDI) Database. http://econ.worldbank.org/ (http://data.worldbank.org/ data-catalog/world-development-indicators). Wu, J.L., Hou, H., Cheng, S.Y., 2010. The dynamic impacts of financial instutuions on economics growth: evidence from the European Union. J. Macroecon. 32, 879–891. Xu, S., 2012. The impact of financial development on energy consumption in China: based on SYS-GMM estimation. Adv. Mater. Res. 524–527, 2977–2981. Yildirim, E., Aslan, A., 2012. Energy consumption and economic growth nexus for 17 highly developed OECD countries: further evidence based on bootstrap-corrected causality tests. Energy Policy 51, 985–993. Zagorchev, A., Vasconcellos, G., Bae, Y., 2011. Financial development, technology, growth and performance: evidence from the accession to the EU. J. Int. Financ. Mark. Inst. Money 21 (5), 743–759.