Does electricity consumption panel Granger cause GDP? A new global evidence

Applied Energy 87 (2010) 3294–3298

Contents lists available at ScienceDirect

Applied Energy journal homepage: www.elsevier.com/locate/apenergy

Does electricity consumption panel Granger cause GDP? A new global evidence Paresh Kumar Narayan a,*, Seema Narayan b, Stephan Popp c a

School of Accounting, Economics and Finance, Faculty of Business and Economics, Deakin University, Melbourne, Australia School of Economics, Finance, and Marketing, RMIT University, Melbourne, Australia c Department of Economics, University of Duisburg-Essen, Germany b

a r t i c l e

i n f o

Article history: Received 21 December 2009 Received in revised form 20 March 2010 Accepted 24 March 2010 Available online 6 May 2010 Keywords: Real GDP Electricity consumption Panel Granger causality

a b s t r a c t The goal of this paper is to undertake a panel data investigation of long-run Granger causality between electricity consumption and real GDP for seven panels, which together consist of 93 countries. We use a new panel causality test and find that in the long-run both electricity consumption and real GDP have a bidirectional Granger causality relationship except for the Middle East where causality runs only from GDP to electricity consumption. Finally, for the G6 panel the estimates reveal a negative sign effect, implying that increasing electricity consumption in the six most industrialised nations will reduce GDP. Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction The literature on the relationship between electricity consumption and economic growth has mainly considered the Granger causality relationship (see, inter alia, [11]). A search of this literature reveals that there are about 18 studies on this relationship. Of these 18 studies, 12 find that electricity consumption Granger causes gross domestic product (GDP). The relevance of this finding is that electricity conservation policies (for these 12 countries) will retard economic growth. The goal of this paper is to examine the sign and direction of long-run panel Granger causality between electricity consumption and real GDP for seven different panels, namely Western Europe, Asia, Latin America, Africa, Middle East, and a G6 panel made up of the six largest industrialised countries. In sum, our panels consist of 93 countries and the time period considered is 1980–2006. There are four main contributions of this paper to the energy literature. First, a review of the literature seems to suggest that the bulk of the studies are concentrated on Asian countries (see [13,14,16]). A recent study by Narayan and Prasad [11], however, attempted to change this trend by considering the electricityGDP causality relationship for a group of 30 OECD countries. Our study is comprehensive in that it analyses this relationship for 93 countries. This represents a relatively detailed investigation of the electricity-GDP causality relationship.

* Corresponding author. E-mail address: [email protected] (P.K. Narayan). 0306-2619/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.apenergy.2010.03.021

Our second contribution is that we take a panel data modelling approach in this paper. In this literature, long-run panel Granger causality has not been considered to date. The motivation for undertaking panel Granger causality is strong and is rooted in the idea that a regional analysis of electricity-GDP causality will provide greater insights on the impact of electricity on GDP from a geographical point of view. For a regional analysis of the type considered in this paper, a panel data framework is needed and we investigate this to the fullest here. Our third contribution results from a key criticism of the extant literature. To-date, in modelling causality relationships, researchers have depended on economic theory to provide them guidance on the possible direction of the effect. However, due to the absence of appropriate modelling techniques, the theoretically motivated directional effect has not been tested in a Granger causality framework. Our novelty is that we test the sign effects in a panel Granger framework by adopting the procedure recently developed by Canning and Pedroni [4]. Fourth, existing studies (see, for instance, [1,9]) have mainly been able to test short-run Granger causality through either a vector autoregressive model or a vector error correction model. We do not do this. Our emphasis is on the long-run panel causality. This is a result of our use of a new method which only deals with long-run causality. The use of the Canning and Pedron [4] methodology, thus, turns out to be a strength in the energy literature given the lack of long-run causality modelling and non-existent panel causality modelling. Briefly foreshadowing the main results, we find that real GDP and electricity consumption for the seven panels are non-station-

3295

P.K. Narayan et al. / Applied Energy 87 (2010) 3294–3298

ary and panel cointegrated based on Pesaran [19] and Pedroni [17,18] tests, respectively. We find that in the long-run, there is bidirectional causality for all panels, except for the Middle East, where only GDP Granger causes electricity consumption. The sign effects reveal a positive relationship for the significant panels, except for the G6 which means that increasing electricity consumption in the G6 will reduce GDP. We organise the balance of the paper as follows. In the next section, we discuss the empirical model and econometric estimation technique. In Section 3, we discuss the results, while in the final section we provide some concluding remarks.

2. Empirical model In this section, we present our testable regression model. We have a bivariate representation given that the new panel Granger causality test that we are using, only considers bivariate cases for panel long-run causality. Our proposed long-run regression model takes the following form:

ln EC t ¼ a0 þ a1 ln GDPt þ et

ð1Þ

The panel version of Eq. (1) is:

ln EC it ¼ a0 þ a1 ln GDPit þ eit

ð2Þ

Here ln GDP is the natural log of the real GDP and ln EC is the natural log of total electricity net consumption (henceforth, electricity consumption). The time series component of the data is for the period 1980–2006, while the cross-section component of the data, represented by subscript i, is countries (i = 1, . . ., 93). The sample size is dictated by data availability. The electricity consumption data is downloaded from the International Energy Agency’s webpage, while the real GDP data (measured in millions of 1990 US$) is obtained from the Total Economy Database [15] published online at www.conference-board.org/economics/. Using these datasets, Eq. (2) is estimated for seven different regions, namely Western Europe, which consists of 20 countries; Asia and Latin America, which consist of 17 countries each; Africa, which consists of 25 countries; Middle East, which consists of 12 countries; a G6 panel made up of the six largest industrialised countries; and a global panel including all the 93 countries.

t is the mean of the variable yt. The unit root null hypothesis Here, y amounts to testing:

H0 : bi ¼ 0 for all i

ð4Þ

against the possibly heterogenous alternatives:

H1 : bi < 0; i ¼ 1; 2; . . . ; N1 ;

bi ¼ 0;

i ¼ N1 þ 1;

N1 þ 2; . . . ; N

The t-test is used to examine the null hypothesis. However, the distribution is non-normal, hence critical values for various cases are produced by Pesaran [19]. Pesaran [19] shows that the test can be extended to a more general case of a panel unit root through a cross-sectionally augmented version of the Im et al. (IPS, [7]) panel unit root test. This can be termed the CIPS test, which is what we use in our application. More technical details can be found in Pesaran [19]. The results from the Pesaran [19] test are reported in Table 1.1 We report panel unit root tests for seven regions, namely the Western European region, the Asian region, the Latin American region, the Middle Eastern region, the African region, the G6 group of countries, and a global panel consisting of all the 93 countries. Our model, for both variables GDP and electricity consumption, includes a linear trend and intercept. The results are reported for up to four lags. The critical values for testing the null hypothesis are extracted from Table 2c in Pesaran ([19]: 281). Beginning with results for the log of real GDP, we find that we cannot reject the unit root null hypothesis at any of the four lags for any of the panels, except for Africa. For Africa, we reject the null at the 5% level but only at one of the four lags. So, there is still strong evidence that Latin American’s GDP is nonstationary. Next, we consider the results for the log of electricity consumption reported in the second half of the table. We find that for five of the seven panels, the unit root null hypothesis cannot be rejected at conventional levels regardless of the lags. In the remaining two panels, namely Western Europe and the G6, the null is only rejected at lag one at the 5% level; at lags 2–4, the null hypothesis is not rejected. In sum, then, it is clear that electricity consumption, like real GDP, is panel nonstationary for all the seven panels. This paves the way for panel cointegration analysis, which we perform next. 3.2. Panel cointegration

3. Estimation technique and results This section has two objectives: first, to present a discussion of the econometric estimating techniques, and second to discuss the main findings. This section is accordingly divided into three sections: panel unit root, panel cointegration, and panel long-run causality. 3.1. Panel unit root There are now several panel unit root tests. However, these tests do not account for cross-sectional dependence. Failure to do so leads to large size distortions (see [3,10]). In our empirical analysis, we use a recent panel unit root test developed by Pesaran [19], which accounts for cross-sectional dependence. Given that this test is relatively new and less used in the applied economics literature, we provide some more details on this test here. Pesaran [19] builds a panel unit root test by augmenting the augmented Dickey–Fuller (ADF) regression with cross-sectional averages of lagged levels and first differences of the individual series. This test is known as the cross-sectionally augmented ADF (CADF) test. The CADF test has the following regression representation:

Dyit ¼ ai þ bi yi;t1 þ gi yt þ ji Dyt þ eit

ð3Þ

In this section, we aim to search for any possible long-run relationship between real GDP and electricity consumption for our seven panels. To test for panel cointegration, we use the Pedroni [17,18] procedure. Pedroni’s test is residual based, in the spirit to the Engle and Granger [6] residual based test. Pedroni recommends testing the null hypothesis of no cointegration by using seven test statistics, namely the panel v-statistics, the panel Phillips–Perron rho-statistic, the panel Phillips–Perron t-statistic, the panel augmented Dickey–Fuller (ADF) t-statistic, the group Phillips–Perron rho-statistic, the group Phillips–Perron t-statistic, and the group ADF t-statistic. The first four tests are known as the ’within dimension’ panel tests while the last three are known as the ’between dimension’ group tests. The ‘within dimension’ statistics are based on pooling the residuals of the regression along the within dimension of the panel, whereas the ’between dimension’ statistics are based on pooling the residuals of the regression along the between dimension of the panel ([18]: 603). Features common to both tests are that they accommodate (a) individual specific short-run dynamics, (b) individual 1 For an application of this test in the energy economics literature, see Chen and Chen [5].

3296

P.K. Narayan et al. / Applied Energy 87 (2010) 3294–3298

Table 1 Panel unit root test results (linear trend case) for various panels, 1980–2006. Panel

N

T

ln GDP

ln EC CIPS

Western Europe Asia Latin America Middle East Africa G6 Global * ** ***

20 17 17 12 25 6 93

27 27 27 27 27 27 27

CIPS

IPS

p=1

p=2

p=3

p=4

IPS

p=1

p=2

p=3

p=4

3.56*** 4.46 3.60*** 0.60 0.07 2.49*** 1.90**

1.521 1.711 2.461 2.844 2.767** 1.773 2.420

1.253 1.505 2.322 2.706 2.540 1.366 2.207

1.511 1.263 2.317 2.375 2.663* 1.377 2.252

1.472 1.104 1.960 1.870 2.436 1.896 1.905

0.61 1.75 4.48*** 1.99** 1.10 2.80 2.43***

2.792** 1.891 2.564 2.363 2.676* 3.191** 2.056

2.228 1.425 2.089 2.411 2.360 2.775 1.722

2.064 1.385 1.879 2.440 2.184 2.278 1.520

2.016 1.154 1.699 2.009 1.735 1.880 1.322

Significance at 10% level. Significance at 5% level. Significance at 1% level.

Table 2 Pedroni residual cointegration test results (deterministic intercept and trend case) for various panels, 1980–2006. Panel

N

T

Pedroni residual cointegration tests Panel v-stat

Western Europe Asia Latin America Middle East Africa G6 Global * ** ***

20 17 17 12 25 6 93

27 27 27 27 27 27 27

***

4.324 3.983*** 1.025 0.153 3.324*** 4.614*** 5.707***

Panel rho-stat ***

7.242 6.695*** 6.663*** 5.646*** 8.120*** 3.954*** 2.042**

Panel PP-stat 0.508 0.280 1.611 0.346 0.051 0.634 0.718

Panel ADF-stat

Group rho-stat

**

***

8.306 7.682*** 7.645*** 6.456*** 9.271*** 4.550*** 4.085***

2.04 1.729* 3.027*** 0.131 1.306 1.925* 3.684***

Group PP-stat

Group ADF-stat

0.055 1.697* 1.031 0.624 0.712 0.144 0.226

4.528*** 1.535 3.044*** 1.111 1.804* 2.221** 5.575***

Significance at 10% level. Significance at 5% level. Significance at 1% level.

specific fixed effects, and (c) deterministic trends as well as individual specific slope coefficients. Our results are reported in Table 2. We find that for Western Europe, Asia, G6, and the global panel, the null of no cointegration is rejected by five of the seven test statistics at conventional levels of significance. For Latin America and Africa, four of the seven tests reject the null of no cointegration at conventional levels, and for the Middle East only two tests reject the null hypothesis. Taken together, there is reasonable evidence from Pedroni’s cointegration test that real GDP and electricity consumption are panel cointegrated for all the seven panels considered here. Having found cointegration implies that there is causality in a Granger sense. However, the direction and sign of causality is unknown, and this is what we explore next.

D ln GDPit ¼ b1i þ

k X

b11ij D ln GDPi;tj þ

j¼1

k X

b12ij D ln EC i;tj

j¼1

þ k2i eit1 þ l2it

ð6Þ

The one period lagged residual terms, captured by k1i and k2i, measure the speed of adjustment to equilibrium. For the long-run relationship to exist, the Granger representation theorem implies that at least one of the adjustment coefficients must be non-zero (see [6]). The null hypothesis is that there is no panel Granger causality. To test the null, Canning and Pedroni [4] develop two tests. The group mean (GM) test is one of them; its panel estimate is computed as follows:

k2 ¼ N1

N X

^k2i

ð7Þ

i¼1

3.3. Panel long-run granger causality

and its panel test statistic is computed as: In this section, we have two aims: (a) to identify the direction of the causality relationship between real GDP and electricity consumption for the seven panels that we have formed, and (b) to identify whether the sign of the causality relationship is positive or negative. Achieving these objectives are possible due to a recently developed panel causality test proposed Canning and Pedroni [4]. The Canning and Pedroni [4] test regression follows a dynamic error correction model, which is embedded in the pioneering work of Engle and Granger [6]. The key difference is that the Canning and Pedroni [4] specification is structured in a panel data framework. We will introduce the test method by specifying it in terms of our current research question: Why electricity consumption does not Granger causes GDP, and vice versa?, as follows:

D ln EC it ¼ b1i þ

k X

b11ij D ln EC i;tj þ

j¼1

þ k1i eit1 þ l1it

k X

b12ij D ln GDPi;tj

j¼1

ð5Þ

tk2 ¼ N1

N X

t k2i

ð8Þ

i¼1

where N is the number of countries in the panel, and tk2 is the individual country test for the null hypothesis that electricity consumption does not Granger causes GDP; that is, k2i = 0. The test statistic has a standard normal distribution. The second test is the Lambda–Pearson (LP) panel test. The LP test is computed as follows:

Pk2 ¼ 2

N X

ln pk2i

ð9Þ

i¼1

where ln pk2i is the log of the p-value, which results from the t-test statistic used to test the null hypothesis. It follows that the LP test essentially combines p-values associated with each of the individuals that make up the panel. The test has a chi–square distribution with 2N degrees of freedom.

3297

P.K. Narayan et al. / Applied Energy 87 (2010) 3294–3298

Finally, the coefficient on k2/k1 will have the same sign as the long-run effect of electricity consumption on GDP. So, as explained by Canning and Pedroni [4], the coefficient is interpreted as a test of the impact of the long-run and a test of the sign of the long-run impact. The long-run panel Granger causality results are reported in Table 3. Results are reported for each of the seven panels and their associated GM and LP test statistics. We first consider the panel long-run causality running from electricity consumption to real GDP reported in the first half of Table 3. Based on the GM test statistic, we only reject the null of no Granger causality for the G6 panel. For the rest of the panels, the GM test does not reject the null. However, in contrast, the LP test statistic rejects the null for all panels except the Middle East: the null is rejected at the 1% for Europe, Asia, Africa, G6, and the global panel, and at the 5% level for the Latin American panel. In choosing between the two tests, the nature of the panel is of importance. For example, as noted by Canning and Pedroni [4], if the panel consists of a heterogenous set of countries, then the GM test statistic, by its construction, can be dictated by those countries having high (in absolute values) t-statistics. In this case, because the LP test is based on p-values, it is a more reliable test. Even though our panels are based on regional location of countries, there is still significant inequalities in terms of per capita incomes and other socio economic conditions. Perhaps the only panels which are relatively homogenous are the G6 (made up of the six most industrialised countries) and the Middle East. For the G6, as we uncovered, both tests reject the null hypothesis and for the Middle East none of the tests reject the null hypothesis. Thus, taking evidence from the LP test, we find significant evidence of long-run panel Granger causality running from electricity consumption to GDP for all panels except for the Middle East. The result for Middle East, while to some extent contrary to a recent study by Narayan and Smyth [12] on Middle East, is consistent with the behaviour of the electricity market in the Middle East re-

gion. Next, we attempt to explain this difference in the results for Middle East in some detail. First, we wish to make a comparison of our results with a recent study by Narayan and Smyth [12]. Narayan and Smyth [12] examine the causal relationship between electricity consumption, exports and GDP for a panel of Middle Eastern countries. Their panel includes six countries and the time period is 1974–2002. While they find short-run Granger causality running from electricity to GDP, their panel long-run estimate reveals that the long-run elasticity on electricity consumption is only 0.04 (see their Table 4). Most interestingly, two of the countries in their panel have an insignificant impact of electricity on GDP and in two cases electricity consumption has a statistically significant but negative effect on GDP. Hence, it follows that the overall positive growth is mainly a result of Israel which has a relatively large and statistically significant positive coefficient on electricity. So, when the Narayan and Smyth [12] results are examined in detail, it seems that our results are not so much different and if anything the Narayan and Smyth results may be overstating the impact of electricity on GDP. Apart from this, a major difference between the Narayan and Smyth [12] and our study is that we consider twice as many countries (12 in total) for our Middle Eastern panel and while our panel in terms of time is relatively short, it is nonetheless most up-todate. These differences may also be contributing to the marginal difference in results of the two studies. Second, there are some irregularities in the functioning of the Middle Eastern electricity market which maybe an obstacle for the full impact of electricity consumption to be felt on GDP. The electricity market is heavily subsidised in the Middle East [2], which to a large extent erodes the potential positive effects of electricity on GDP. The other issue facing the Middle Eastern electricity market is weak electricity bills collection mechanism; as a result, only a few power stations are able to provide a reasonable measure of self-financing to power extension [8].

Table 3 Long-run panel Granger causality test.

GM LP GM LP GM LP GM LP GM LP GM LP GM LP

Estimate k2:ECit ? GDPit

Test Europe

p-value

Estimate k 1:GDPit ? ECit

Test

p-value

Median k2/k1

0.15

0.89 81.26***

0.81 0.00

0.35

0.91 68.57***

0.18 0.00

0.06(0.20)

k2:ECit ? GDPit 0.05

Asia 0.58 51.54**

0.72 0.03

1.23 100.2***

0.11 0.00

k2:ECit ? GDPit 0.03

Latin America 0.09 53.44**

0.46 0.02

1.47 81.53***

0.07 0.00

k2:ECit ? GDPit 0.05

Middle East 0.27 23.47

0.39 0.49

1.66** 64.68***

0.05 0.00

k2:ECit ? GDPit 0.10

Africa 0.66 115.6***

0.75 0.00

2.25*** 225.4***

0.01 0.00

k2:ECit ? GDPit 0.13

G6  1.42* 30.52***

0.08 0.00

1.73** 31.78***

0.04 0.00

k2:ECit ? GDPit 0.05

Global 0.56 470***

0.71 0.00

1.27* 490***

0.10 0.00

Note: Both variables are in natural logarithmic form. * Statistical significance at 10% level, respectively. ** Statistical significance at 5% level, respectively. *** Statistical significance at 1% level, respectively.

k 1:GDPit ? ECit 0.29 k1:GDPit ? ECit 0.35 k1:GDPit ? ECit 0.34 k1:GDPit ? ECit 0.59 k 1:GDPit ? ECit 0.52 k1:GDPit ? ECit 0.34

k2/k 1 0.10(0.19) k 2/k1 0.02(0.11) k 2/k1 0.08(0.23) k 2/k1 0.18(0.09) k2/k 1 0.34(0.16) k 2/k1 0.10(0.08)

3298

P.K. Narayan et al. / Applied Energy 87 (2010) 3294–3298

Table 4 Electricity power transmission and distribution losses (% of output) for selected Middle Eastern countries. Country

2000

2006

Bahrain Iran Iraq Israel Jordon Kuwait Oman Qatar Saudi Arabia Syria UAE Yemen

8.845 15.844 8.589 3.3447 10.969 10.824 17.199 6.930 7.236 30.697 3.382 27.776

4.999 20.018 6.479 2.654 12.777 10.939 16.474 7.922 6.757 23.678 7.309 22.728

Source: World Bank [20].

In Table 4, we report some statistics relating to electric power transmission and distribution loss for Middle Eastern countries considered in our panel. The loss is measured as a percentage of output and are for years 2000 and 2006. We notice that while in seven of the 12 countries the loss as a percentage of GDP has declined over the two years, they are still high. Out of the five countries where loss has increased over the period, it has risen significantly in Iran, from around 16% to 20%. On average for the panel, the loss from electricity transmission and distribution in 2006 was around 12%. It follows that the Middle Eastern electricity market is in need of reforms to boost its performance. Our result that electricity consumption has an insignificant effect on GDP when measured against this background reflects the fact that the Middle Eastern electricity market is functioning inefficiently. The sign effect based on the ratio of the lambda coefficients which proffer the sign of the long-run causality reveal a positive sign for all panels for which long-run causality was found, except for the G6. For the G6, evidence suggests that electricity consumption will reduce GDP. In terms of the panel long-run Granger causality running from real GDP to electricity consumption, we reject the null of no causality (from both tests) for the Latin America panel, Middle East panel, Africa panel, the G6, and the global panel. For the Asian and the Europe panels, only the LP test rejects the null. For the heterogeneity reason alluded to above, on the whole these results mean that for these panels of countries GDP Granger causes electricity consumption in the long-run. 4. Concluding remarks The goal of this paper was to examine the Granger causal relationship between electricity consumption and real GDP. This objective was aided by the recent development of a panel Granger causality test that proffers both the long-run causality and the sign on the direction of causation. Subsequently, our work makes a fresh contribution to the literature. Equally importantly, we under-

take a large panel data study, encompassing a total of 93 countries, divided into seven panels (Western Europe, Asia, Africa, Middle East, Latin America, G6, and global). Our empirical analysis, thus, is relatively comprehensive. We find evidence of electricity consumption Granger causing real GDP positively for all panels except Africa and the G6. For the G6 the causality has a negative sign whereas for the Middle East the relationship is statistically insignificant, implying no evidence of long-run causality. Meanwhile, we find that GDP Granger causes electricity consumption in the long-run in all panels. It follows that, in terms of policy implications, energy conversation policies that reduce electricity consumption in the G6 and Middle East will not retard economic growth. Generally, the results also suggest that as the global economy grows it will put more demand on electricity supply which will have consequences for global warming and indeed the environment. References [1] Abosedra S, Dah A, Ghosh S. Electricity consumption and economic growth, the case of Lebanon. Appl Energy 2009;86:429–32. [2] Al-Iriani M. Energy–GDP relationship revisited: an example from GCC countries using panel causality. Energy Policy 2006;34:3342–50. [3] Banerjee A, Marcellino M, Osbat C. Testing for PPP: should we use panel methods? Empirical Econ 2005;30:70–91. [4] Canning D, Pedroni P. Infrastructure, long-run economic growth and causality tests for cointegrated panels. The Manch Sch 2008;76:504–27. [5] Chen S-S, Chen H-C. Oil prices and real exchange rates. Energy Econ 2007;29:390–404. [6] Engle RF, Granger CWJ. Cointegration and error correction representation, estimation and testing. Econometrica 1987;55:251–76. [7] Im K, Pesaran H, Shin Y. Testing for unit roots in heterogeneous panels. J Econom 2003;115:53–74. [8] Khatib H. MENA electrical power sector: challenges and opportunities. Middle East Econ Surv 2005;48:1–4. [9] Lean HH, Smyth R. On the dynamics of aggregate output, electricity consumption and exports in Malaysia: Evidence from multivariate Granger causality tests. Appl Energy 2010;87:1963–71. [10] Maddala GS, Wu S. A comparative study of unit roots with panel data and a new simple test. Oxford Bull Econ Stat 1999;61:631–51. [11] Narayan PK, Prasad A. Electricity consumption-real GDP causality nexus: evidence from a bootstrapped causality test for 30 OECD countries. Energy Policy 2008;36:910–8. [12] Narayan PK, Smyth R. Multivariate granger causality between electricity consumption, exports and GDP: evidence from a panel of Middle Eastern countries. Energy Policy 2009;37:229–36. [13] Narayan PK, Wong P. A panel data analysis of the determinants of oil consumption: the case of Australia. Appl Energy 2009;86:2771–5. [14] Narayan PK, Narayan S, Popp S. A note on the long-run elasticities from the energy consumption–GDP relationship. Appl Energy 2010;87:1054–7. [15] The Conference Board and Groningen Growth and Development Centre. Total Economy Database; January 2009. [16] Payne JE. A survey of the electricity consumption-growth literature. Appl Energy 2010;87:723–31. [17] Pedroni P. Critical values for cointegration tests in heterogenous panels with multiple regressors. Oxford Bull Econ Stat 1999;61:653–70. [18] Pedroni P. Panel cointegration: asymptotics and finite sample properties of pooled time series tests, with an application to the PPP hypothesis. Econom Theory 2004;20:597–625. [19] Pesaran MH. A simple panel unit root test in the presence of cross-section dependence. J Appl Econom 2007;22:265–312. [20] World Bank. World Bank, Washington, (DC): World Development Indicators; 2009.