Energy consumption, financial development and economic growth nexus: an empirical evidence from panel data for EU countries Iuliana Mateia

Abstract: The paper explores the energy-income nexus, energy-finance nexus, and financegrowth nexus for three groups of countries: 25 European Union (EU) countries, 10 Emerging European (EE) countries and 15 Eurozone (EZ) countries. Using dynamic panel estimation techniques, including the Pooled Mean Group estimator (Pesaran, Shin and Smith, 1999) over the period 1990-2012, the findings validate a feed-back effect between energy consumption and economic growth for 25 EU countries and Eurozone countries (i.e., the bi-directional causality hypothesis) and a unidirectional causality from economic growth to energy consumption in the long-run and short-run for EE countries (i.e., the conservation hypothesis). These outcomes suggest that energy consumption is a key input in the production function and that energy saving policy and efficiency improvement will favorably influence the economic growth and climate goals. Furthermore, results show long run causality from energy consumption to financial development for EU countries, EZ countries, and EE countries, and respectively, a short-run causality only for the last two groups of countries. Regarding the finance-growth nexus, findings indicate positive long-run causality, and respectively, negative or positive short-run causality from economic growth to finance (i.e., the demand following channel) in the case of 25 EU countries, EZ countries, but, also for EE states.

Keywords: energy consumption, economic growth, financial development, causality relationships, dynamic panel data and EU countries.

JEL Code: C32, E21, C01, C33, O13, Q43

______________________________ a

Adjunct Professor, IESEG School of Management - Paris, France. Email: [email protected] and [email protected].

1

1. Introduction

The paper investigates the energy-income nexus, energy-finance nexus, and finance-growth nexus for three groups of countries: 25 European Union countries, 10 Emerging European countries and 15 Eurozone countries. Due to its importance in the production function and accordingly, for the policy makers, the relationship between energy consumption and economic growth, referred as the “energyincome nexus”, has received a great attention over the past decade in the expanding empirical literature. Four major hypotheses with respect to the energy-growth nexus and related policy implications of each of them were revealed by this literature: “the growth assumption”, “the conservation assumption”, “the neutrality assumption” and “the feedback assumption”. The first hypothesis implies a unidirectional causality from energy consumption to economic growth which means that a decline in energy consumption negatively influences economic growth. The second hypothesis supposes a unidirectional causality from economic growth to energy consumption implying that energy conservation policy has no impact on growth. The third hypothesis supposes that there is no causality between energy consumption and economic growth and therefore, that energy policy aiming to conserve energy will not retard economic growth. Finally, the last hypothesis involves a bi-directional causality between energy consumption and economic growth which is a relationship of complementarity: increasing economic growth stimulates energy consumption and, at the same time, growing energy consumption accelerates growth. Therefore, the direction of causality is essential for both the energy efficiency design and climate goals: a country which is energy dependent will have a cautious energy policy as any negative energy supply shock (e.g., an increase in the oil prices) involves decreasing economic growth while for a country that is not energy dependent, an energy conservation policy will have little effect on growth.

Because the relationship between finance and economic growth is controversial in the theoretical literature, many empirical studies analyzed in great detail this relationship for the financial intermediaries, the bond and stock financing. Regarding the finance-growth nexus, the literature distinguishes four major assumptions: the first hypothesis considers that finance is an essential resource of funding for economic growth (the supply-leading channel). The second hypothesis suggests that finance does not cause economic growth but it rather follows GDP growth (the demand-following channel). The third hypothesis observes a bi-directional 2

causality between finance and economic growth. This interdependence hypothesis is explained by Patrick (1966) by the stage of development of the country: at the initial stage, financial development leads to economic growth while as real growth takes place in the economy this linkage becomes less important and growth will enhance the demand for financial services. Therefore, the lack of financial institutions in underdeveloped countries simply indicates the lack of demand for such services. The last hypothesis shows that financegrowth nexus is rather a complementary relationship as it may arise multiple equilibrium phenomena (e.g., Berthélemy and Varoudakis, 1997). But, the recent empirical literature adds to this list new hypothesis such as: a negative causality from finance to economic growth and no causal links between them.

This paper studies the relationship and the causality between energy consumption and economic growth, the energy consumption and financial development, and respectively, the financial development and economic growth over the 1990-2012 period for EU’s countries by using dynamic panel data techniques. The contribution to the existing empirical literature is threefold: firstly, at my best knowledge, no study has analyzed the energy-income nexus for EU’s countries although the most part of these economies are completely dependent on energy imports. Furthermore, some Emerging European countries are of growing importance for the greater integration of the European internal energy market and for the stabilization and adequacy of the energy supply. Some of these states are not only consumers of non-renewable energy but also key producing and transit countries for several energy-consuming nations of EU. Secondly, from a methodological point of view, we investigate the panel causality between energy consumption and economic growth by applying recent dynamic panel models which account for cross-section dependence between countries - the Pooled Mean Group estimator of Pesaran, Shin and Smith (1999). Compared to the traditional literature on dynamic panels, this method allows estimating a different slope parameter for each country when investigating the energy-income nexus but also, the energy-finance nexus, and respectively, finance-growth nexus. Thirdly, to take into account the “lump-together” problem (Ozturk, Aslan and Kalyoncu (2010)), I work on data divided into two main sub-panels (the EZ countries, and respectively, the EE countries) based on the difference in income levels before starting estimations. By restricting the analysis also to these two sub-panels, the study proposes not only robustness checks but, also extends this empirical literature in the case of countries with different development levels.

3

Outcomes show evidence in favor of feed-back effect in the case of 25 EU countries, and respectively, of EZ countries; but, also for a long-run and short-run causality from economic growth to energy consumption for the EE countries. Regarding the relationship between the financial development and the energy consumption, my results indicate a long run causality from energy consumption to financial development for the EU countries, EZ countries, and respectively, for Emerging Europe, and a short-run causality only in the case of our subsample (EE and EZ countries). Finally, findings concerning the finance-growth nexus suggest a long-run causality, and respectively, a short-run causality from economic growth to finance (i.e., the demand following channel) in the case of all considers samples. The rest of the paper is structured as follows. In the next section, we describe the empirical methodology and the employed data. In section 3, we present and discuss the empirical results. The last section concludes the paper.

2. Methodology and data Econometric specification We investigate the long-run and short-run effects of economic growth on energy consumption (and vice-versa), of financial development on economic growth (and vice-versa), and of energy consumption on financial development (and vice-versa). To this end, we employ a Pooled Mean Group (PMG) model, developed by Pesaran, Shin and Smith (1999). Previous studies have widely employed static panel models and generalized methods of moments (GMM, developed by Arellano and Bond, 1991) to study the energy-income nexus, financegrowth nexus and energy consumption-finance nexus. Although the GMM method allows estimating country-specific intercepts, the PMG estimator has the advantage that it imposes the long-run homogeneity but, distinguishes different short-run dynamics in each country. We present the models analyzing only the energy- consumption nexus; but, similar equations were constructed for the relationships between financial development and economic growth, and respectively, between energy consumption and financial development. We consider the following models: (1) (2) 4

where subscripts and intercept, and

are respectively, country and time period,

is the country-specific

are the error terms. The first equation gives the dependent variable, the

energy consumption (

as a function of GDP per capita (

equation shows the alternative case where the energy consumption (

whilst the second become the

independent variable. The autoregressive distributive lag dynamic specification ARDL (1,1,1), i.e. with one lag, associated to equations (1) and (2) can be expressed as follows: =

+ =

+

+

(3)

+

(4)

Using the PMG procedure, we estimate the related error correction equilibrium models: (5) where

=

,

=

and

=

. (6)

where

=

,

=

and

=

.

The PMG estimator allows assessing two types of causality: a short-run causality by testing the significance of the coefficients related to the lagged differences of economic and energy variables ( adjustment coefficient

and a long-run causality related to the speed of that has to be negative for that variables exhibit a long-run

equilibrium. A larger value of

implies a stronger response of the variable to the deviation

from long-run equilibrium whilst a low value means that any deviation from long-run equilibrium of the GDP growth or the energy consumption needs much longer time to force the variables back to the long-run equilibrium. If the speed of adjustment coefficient appears significant in both equations, bidirectional causality between energy consumption and economic growth takes place. Thus, we estimate the long-run and short-run effects of energy consumption on GDP per capita growth (and vice-versa) over the period 1990-2012. Before turning to estimations, we check for the cross-section dependence hypothesis, the stationarity of our variables (i.e., the order of integration of the series) under cross-sectional dependence by using several panel unit root tests. Results provide evidence that most of the variables are integrated of order one and are cointegrated. Therefore, these results enable to 5

test the PMG model to evaluate the energy-income nexus, the energy-finance nexus, and the finance-economic growth nexus in the case of EU countries.

Variables and data To study these key issues developed in the current economic literature, we consider both time and cross-country variation in the data. Our data sample covers 25 EU’s countries and the period from 1990 until 2012. Table 1 displays a brief summary of used variables and data sources. We employ yearly data on per capita energy, per capita GDP, and on domestic credit to private sector (in % GDP) provided by World Development Indicators database of the World Bank. We use data on 25 EU countries: Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovak Republic, Spain, Sweden, and United Kingdom. For robustness checks (but, not only), we also consider the sub-samples formed by some Emerging European states: Bulgaria, Romania, Slovakia, Slovenia, Poland, Hungary, Czech Republic, Estonia, Latvia and Lithuania, and respectively, and by 15 euro area’s advanced economies such as: Austria, Belgium, Cyprus, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Malta, Netherlands, Portugal, Slovak Republic and Spain. Some of these countries are not integrated in the full sample because of the availability of data on the entire period whilst other integrated at the end of the selected time period the euro area. Per capita energy use is measured in kg of oil equivalent and per capita GDP is measured in constant 2005 US dollar. We estimate our models using variables transformed in natural logarithms. Accordingly, each estimated coefficient should be interpreted as a constant elasticity of the dependent variable with respect to the independent variable. Table 2 displays the main descriptive statistics of variables integrated in estimations.

6

Table 1: Data sources of explained and explanatory variables Variable Energy consumption

Explanation, computation Kg of oil equivalent per capita

Data source World Development Indicators, yearly data

GDP per capita

Per capita GDP in constant 2005 US dollar

World Development Indicators, yearly data

Financial Development Domestic credit to private sector (GDP %)

World Development Indicators, yearly data

Table 2: Descriptive statistics of selected variables (log values) 1990 – 2012

UE_25 Energy Consumption Financial development GDP per capita Emerging Europe Energy Consumption Financial development GDP per capita Euro Zone - 15 Energy Consumption Financial development GDP per capita

Obs

Mean Std Dev

Min

Max

575 575 575

8.10 4.19 9.83

0.39 0.75 0.84

7.39 1.96 7.81

9.15 5.72 1.36

230 230 230

7.94 3.52 8.87

0.28 0.65 0.52

7.39 1.96 7.76

8.75 4.66 9.94

345 345 345

8.17 4.46 10.20

0.41 0.53 0.49

7.42 3.20 8.82

9.15 5.72 11.36

3. Estimation results and discussion 3.1. Exploring cross-sectional dependence Before investigating the stationarity of our variables and checking for their cointegration, it is necessary to account for the dependence hypothesis between countries in the panel. Interdependencies between EU countries may arise after common shocks with heterogeneous impacts across countries (e.g., the sovereign debt crisis), spatial spillover effects and other unobserved components due to the economic and financial integration experienced in the last decade by EU countries. To this end, I apply the Pesaran (2004) test based on pair-wise correlation coefficients and report the results in the table 3. The test results strongly reject the 7

null hypothesis of no cross-sectional dependence at the 1% level of significance for all variables in the case of EU panel, and respectively, EE and EZ sub-panels.

Table 3: Cross section dependence results of Pesaran (CD) PANEL A: VARIABLES IN LOG avg ρ avg ǀρǀ CD p-value UE - 25 Energy Consumption GDP per capita Financial Development

0.16 0.92 0.53

0.39 0.92 0.64

13.00a 76.04a 43.78a

0.00 0.00 0.00

Emerging Europe - 10 Energy Consumption GDP per capita Financial development

0.46 0.94 0.40

0.67 0.94 0.60

14.70a 30.19a 12.89a

0.00 0.00 0.00

Euro Zone - 15 Energy Consumption GDP per capita Financial Development

0.36 0.95 0.57

0.48 0.95 0.67

17.65a 46.74a 27.91a

0.00 0.00 0.00

Notes: a significant at the 1% level; b at the 5% level; c at the 10% level

3.2. Panel unit root investigation We check for the stationarity of the selected variables (i.e., the order of integration of these series) by applying different panel unit root tests (PURT). Given that the previous subsection shows evidence in favor of cross-section correlation, we use the second-generation tests of Pesaran (2007) and Hadri (2000) that take into account the dependence between countries. The Pesaran (2007) test assumes that countries can respond differently to unobserved common factors whilst the Hadri (2000) LM tests the null hypothesis that data are stationary versus the alternative that at least one panel contains a unit root. Table 4 summarizes the results of these two tests. Overall, both tests indicate that all variables are integrated of order 1 - I(1). The results of Pesaran (2007) and Hadri (2000) tests strongly reject the null hypothesis that all panel’s series are stationary in level (excluding Pesaran (2007) results corresponding to the EU financial development variable).

8

Table 4: The second generation PURT results Tests

Pesaran (2007) CIPS

UE-25 Energy consumption GDP per cap Financial development

2.52 (0.99) 2.79 (0.99) -3.84 (0.00)

Emerging Europe Energy consumption GDP per capita Financial development Euro Zone - 15 Energy consumption GDP per capita Financial development

Hadri (2000) LM 22.30 (0.00) 62.61 (0.00) 48.35 (0.00)

1.11 (0.87)

8.35 (0.00)

2.31 (0.99) -1.19 (0.12)

39.44 (0.00) 28.74 (0.00)

-1.09

(0.14)

21.85 (0.00)

1.59 (0.94) -0.058 (0.48)

48.66 (0.00) 39.78 (0.00)

No of periods 23 23 Notes: i) only the first specification includes trends and constant for Emerging Europe; the other results take into account only the models with constant and trend; but, the results still hold with alternative specifications; ii) the values in brackets are the associated probabilities

3.3. Panel cointegration investigation The previous section has shown that the hypothesis of strong interdependencies between cross-sectional units was accepted by the Pesaran (2004) test. Consequently, we can apply second-generation cointegration tests assuming the cross-section dependence in cointegrating vectors. We apply the panel cointegration test by Westerlund (2007) imposing as null hypothesis the absence of cointegration. This test also assumes the existence of an error correction for individual panel members (with the group-mean statistics - Gt and Ga) and/or for the panel as a whole (with the panel statistics - Pt and Pa) without any common-factor restriction. As explained by Westerlund (2007), these tests are general enough to allow for a large degree of heterogeneity, both in the long-run cointegrating relationship and in the shortrun dynamic, and for dependence within, as well as across, the cross-sectional units. Table 5, 6 and 7 report the Westerlund (2007) results for the UE-25 panel, Emerging Europe sub-panel and Eurozone sub-panel corresponding to the energy consumption-income nexus. The null hypothesis of no cointegration between the energy consumption and GDP is strongly rejected by all test statistics (except Ga - may be because the sample size is smaller). These results should be taken as evidence of cointegration for the panel as a whole and/or at least for one of the countries in these panels (as shown by the Gt statistic).

9

Table 5: The Westerlund (2007) cointegration test results: UE-25 Statistics with constant only Gt Ga Pt Pa

Value -2.805 -7.806 -13.377 -12.294

Z-value -5.693 -0.558 -5.927 -8.650

P-value 0.000 0.289 0.000 0.000

Statistics with constant and trend Gt Ga Pt Pa

Value -3.645 -11.005 -16.406 -17.021

Z-value -8.027 0.671 -6.816 -6.753

P-value 0.000 0.749 0.000 0.000

Note: i) the p-values are for the test based on the normal distribution; ii) we find similar results under different specifications for one lag with/without one lead; iii) the Ga statistic may reject the null hypothesis of no cointegration in small panel data (Westerlund, 2007).

Table 6: The Westerlund (2007) cointegration test results: Emerging Europe Statistics with constant only Gt Ga Pt Pa

Value -2.506 -8.339 -9.265 -10.646

Z-value -2.564 -0.695 -4.715 -4.570

P-value 0.005 0.243 0.000 0.000

Statistics with constant and trend Gt Ga Pt Pa

Value -2.866 -12.715 -10.308 -13.747

Z-value -1.943 -0.327 -4.148 -2.481

P-value 0.026 0.372 0.000 0.007

Note: i) the p-values are for the test based on the normal distribution; ii) we find similar results under different specifications for one lag with/without one lead; iii) the Gt and Ga statistics may reject the null hypothesis of no cointegration in small panel data (Westerlund, 2007).

Table 7: The Westerlund (2007) cointegration test results: Eurozone Statistics with constant only Gt Ga Pt Pa

Value -3.618 -8.813 -8.395 -9.173

Z-value -7.952 -1.152 -2.649 -4.066

P-value 0.000 0.125 0.004 0.000

Statistics with constant and trend Gt Ga Pt Pa

Value -5.503 -14.024 -11.910 -11.665

Z-value -15.177 -1.238 -4.341 -1.758

P-value 0.000 0.108 0.000 0.039

Note: i) the p-values are for the test based on the normal distribution; ii) we find similar results under different specifications for one lag with/without one lead; iii) the Gt and Ga statistics may reject the null hypothesis of no cointegration in small panel data (Westerlund, 2007).

Similar results are obtained for energy consumption-finance nexus (tables 8, 9 and 10), and finance–growth nexus (tables 11, 12 and 13). Results concerning the Eurozone linkages, and respectively, the EU linkages between finance and energy consumption are generally 10

statistically significant in the case of models with constant only (table 8), and the results regarding the finance-growth nexus, are mostly significant only for the model with constant and trend in the case of EU countries and EZ countries (see table 11 and table 13). Table 8: The Westerlund (2007) cointegration test results: UE-25 (EC – FIN) Statistics with constant only Gt Ga Pt Pa

Value -2.259 -8.028 -23.774 -20.658

Z-value -2.624 -0.762 -16.190 -17.765

P-value 0.004 0.223 0.000 0.000

Statistics with constant and trend Gt Ga Pt Pa

Value -2.841 -8.401 -11.433 -10.423

Z-value -3.020 2.629 -1.013 -1.230

P-value 0.001 0.996 0.156 0.109

Note: the Ga statistic may reject the null hypothesis of no cointegration in small panel data Table 9: The Westerlund (2007) cointegration test results: Emerging Europe (EC – FIN) Statistics with constant only Gt Ga Pt Pa

Value -3.409 -9.873 -18.823 -24.974

Z-value -5.749 -1.559 -13.978 -14.210

P-value 0.000 0.060 0.000 0.000

Statistics with constant and trend Gt Ga Pt Pa

Value -3.631 -11.858 -9.532 -11.509

Z-value -5.021 0.019 -3.321 -1.353

P-value 0.000 0.508 0.000 0.088

Note: the p-values are for the test based on the normal distribution.

Table 10: The Westerlund (2007) cointegration test results: Eurozone (EC – FIN) Statistics with constant only Gt Ga Pt Pa

Value -1.669 -5.091 -7.299 -8.343

Z-value 0.541 1.508 -1.568 -3.365

P-value 0.706 0.934 0.059 0.000

Statistics with constant and trend Gt Ga Pt Pa

Value -3.044 -6.037 -6.799 -6.647

Z-value -3.320 3.413 1.612 1.496

P-value 0.001 1.000 0.947 0.933

Note: the p-values are for the test based on the normal distribution.

11

Table 11: The Westerlund (2007) cointegration test results: UE-25 (GDP – FIN) Statistics with constant only Gt Ga Pt Pa

Value -1.893 -2.793 -9.068 -4.954

Z-value -0.562 4.068 -1.674 -0.652

P-value 0.287 1.000 0.047 0.257

Statistics with constant and trend Gt Ga Pt Pa

Value -2.841 -14.179 -19.840 -24.098

Z-value -3.018 -1.714 -10.804 -12.677

P-value 0.001 0.043 0.000 0.000

Note: i) the p-values are for the test based on the normal distribution; ii) we find similar results under different specifications for one lag with/without one lead; iii) the Ga statistic may reject the null hypothesis of no cointegration in small panel data (Westerlund, 2007).

Table 12: The Westerlund (2007) cointegration results: Emerging Europe (GDP – FIN) Statistics with constant only Gt Ga Pt Pa

Value -2.506 -8.339 -9.265 -10.646

Z-value -2.564 -0.695 -4.715 -4.570

P-value 0.000 0.243 0.000 0.000

Statistics with constant and trend Gt Ga Pt Pa

Value -3.169 -8.245 -6.630 -5.770

Z-value -4.895 -0.609 -1.941 -0.975

P-value 0.000 0.271 0.026 0.165

Note: i) the p-values are for the test based on the normal distribution; ii) we find similar results under different specifications for one lag with/without one lead; iii) the Gt and Ga statistics may reject the null hypothesis of no cointegration in small panel data (Westerlund, 2007).

Table 13: The Westerlund (2007) cointegration results: Eurozone (GDP – FIN) Statistics with constant only Gt Ga Pt Pa

Value -2.206 -4.591 -6.413 -6.238

Z-value -1.797 1.865 -0.693 -1.589

P-value 0.036 0.969 0.244 0.056

Statistics with constant and trend Gt Ga Pt Pa

Value -3.584 -13.615 -11.825 -11.658

Z-value -5.922 -1.000 -4.243 -1.754

P-value 0.000 0.159 0.000 0.040

Note: i) the p-values are for the test based on the normal distribution; ii) we find similar results under different specifications for one lag with/without one lead; iii) the Gt and Ga statistics may reject the null hypothesis of no cointegration in small panel data (Westerlund, 2007).

12

3.4 Panel causality investigation 3.4.1. Panel DOLS estimates The cointegration framework initiated by Westerlund (2007) in the second step of estimation permits the investigation of the long-run relationship between energy consumption and economic growth. But, Kao and Chiang (2000) proposes a more powerful test for finite samples to determine the cointegration vector, the Dynamic Ordinary Least Squares (DOLS), with homogeneous long-run covariance structure across cross-sectional units (i.e., the countries) but, with heterogeneous coefficients in the short run. This method allows us to test whether or not strong relationship between energy consumption and GDP per capita growth are consistently present for all countries of the panel. The DOLS estimator is given by the following equation: (7) where The

is the log energy consumption per capita and are cointegrated with the slopes parameters

is the log GDP per capita. . Furthermore, the

is the

coefficient of lead and lag first differenced explanatory variables. Similar equations can be constructed to capture the energy-finance nexus, and the finance-economic growth nexus in the case of EU countries and Emerging Europe states. Table 14 provides the results of DOLS estimator on the whole period. They show whether economic growth strongly stimulates energy consumption or not (and vice versa) in the case of 25 European Union countries, 10 Emerging European countries and Eurozone.

Table 14: The DOLS test results for EU -25, for Emerging Europe and for Eurozone Dependent variables Independent variable – EC Dependent variables Independent variable – GDP

GDP EU-25

GDP Emerging Europe

GDP Eurozone

1.502*** (0.181)

1.212*** (0.279)

0.884*** (0.193)

EC EU-25

EC Emerging Europe

EC Eurozone

0.342*** (0.043)

0.338*** (0.045)

0.639*** (0.065)

Notes: *** significant at the 1% level; standard errors are in parenthesis; DOLS results are given for the two selected sub-panels

13

The results from table 14 show that a 1% increase in per capita income raises in the long-run per capita energy consumption by 0.342% (in EU-25), by 0.338% (in Emerging Europe) and by 0.639% (in Eurozone-15). In addition, there is positive causality running from GDP to EC. Furthermore, a 1% increase in per capita energy consumption rises in the long-run per capita GDP by 1.50% (in EU-25), by 1.21% (in Emerging Europe), and by 0.884% (in euro area). Table 15 shows us the DOLS results in the case of energy - finance nexus, and the table 16 in the case of finance - growth nexus.

Table 15: The DOLS test results for EU -25, Emerging Europe and Eurozone Dependent variables

EC EU-25

EC Emerging Europe

EC Eurozone

Independent variable - FIN

0.245*** (0.019)

1.212*** (0.279)

0.127*** (0.027)

FIN EU-25

FIN Emerging Europe

FIN Eurozone

1.072 * (0.609)

0.338*** (0.045)

0.135 (0.384)

Dependent variables Independent variable – EC

Notes: *** significant at the 1% level; standard errors are in parenthesis; DOLS results are given for the two selected sub-panels

The results from table 15 show that a 1% increase in per capita energy raises in the long-run the financial resources by 1.07% (in EU-25), and by 0.338% (in Emerging Europe). In addition, there is positive causality running from EC to FIN. Furthermore, a 1% increase in financial resources rises in the long-run per capita consumption energy by 0.25% (in EU-25), by 1.21% (in Emerging Europe), and by 0.13% (in the euro area countries). Table 16: The DOLS test results for EU -25, Emerging Europe and Eurozone Dependent variables

GDP EU-25

GDP Emerging Europe

GDP Eurozone

Independent variable - FIN

0.901*** (0.032)

0.557*** (0.049)

0.503*** (0.054)

FIN EU-25

FIN Emerging Europe

FIN Eurozone

0.642*** (0.205)

0.728*** (0.255)

0.468* (0.268)

Dependent variables Independent variable - GDP

Notes: *** significant at the 1% level; standard errors are in parenthesis; DOLS results are given for the two selected sub-panels

Similarly, the results from table 16 indicate us that a 1% increase in per GDP raises in the long-run the financial resources by 0.642% (in EU-25), by 0.728% (in Emerging Europe), and by 0.50% (in Eurzone countries). There is also a positive causality running from GDP to FIN. Also, a 1% increase in the financial resources rises in the long-run per capita GDP by 0.90% (in EU-25), by 0.56% (in Emerging Europe), and by 0.47% (in euro area countries).

14

3.4.2. Panel PMG estimates This section presents the causality results between energy consumption and GDP growth based on the PMG estimators exposed in the equations (5) and (6). Tables 17 and 18 report the results based on the PMG estimator with the long-run and the short-run parameter estimates for 25 EU countries (column 2), and for the Emerging European states (column 3) and the euro area countries (column 4). Regarding the tables 17 and 18, we can observe that there is evidence for long-run adjustment in economic growth (only for Emerging Europe) and respectively, in energy consumption (only for the UE_25 and Eurozone) because the speed of adjustment coefficients (i.e., the error correction terms) are statistically significant at 1% level in almost all selected panels. However, the link from energy consumption to economic growth is stronger than from economic growth to energy consumption as the speed of adjustment terms are greater (except for the euro area countries). Results of table 17 show the long-run and short-run causality running from economic growth to energy consumption for all samples considered in the study as the error correction coefficient is negative and statistically significant. In the table 18, we find also evidence in favor of a long-run (only for Eurozone) and short-run causality coming from energy consumption to economic growth only for the 25 EU countries and euro area countries. The speed of adjustment coefficient is positive and statistically significant at 1% level for Emerging European countries meaning that there is no long-run causality in this case. Overall, the results validate a feed-back effect between energy consumption and GDP growth only in the case of the whole panel (EU-25) and euro area panel (Eurozone). This result is in line with that obtained by Ozturk, Aslan and Kalyoncu (2010) in middle income economies, Mahadevan and Asafu-Adjaye (2007) in the case of developed economies and Lee et al. (2008) in OECD countries. The conservation hypothesis implying a unidirectional causality running from economic growth to energy consumption is confirmed for Emerging Europe and Eurozone in both long-run and the short-run horizons. These results are in the same vein with those demonstrated by Al-Iriani (2006), Mehrara (2007) and Damette and Seghir (2013) for Arabian oil exporting countries, Huang et al. (2008) for middle and high income countries, Apergis and Payne (2009) for central American countries, Ozturk, Aslan and Kalyoncu (2010) for low income economies. In terms of policy implications, the feed-back effect suggests that economic policy should favor the development of the energy sector because the

15

energy sector seems to be a key input in the production function, and that energy efficiency will favorably influence economic growth. Table 17: PMG model: long-run and short-run estimates with the dep. var., the energy use Whole period Long-run term - GDP Growth Error correction term (ECT)

UE-25

Emerging Europe

Eurozone

-0.009 (0.025) -0.291*** (0.046)

0.087** (0.040) -0.263*** (0.050)

1.543*** (0.078) -0.117*** (0.030)

Short-run term – Δ GDP Growth Constant No. obs. No. groups

0.573*** (0.072) 0.425*** (0.077) 0.112** (0.050) 2.369*** (0.368) 1.868*** (0.365) -0.255*** (0.068) 550 220 330 25 10 15 Note: ECT – the speed of adjustment coefficient, * p