Energy Consumption, Economic Growth and Prices: A Reassessment Using Panel VECM For Developed and Developing Countries
Energy Consumption, Economic Growth and Prices: A Reassessment Using Panel VECM For Developed and Developing Countries
Abstract
This paper reinvestigates the energy consumption–GDP growth nexus in a panel error correction model using data on 20 net energy
importers and exporters from 1971 to 2002. Among the energy exporters, there was bidirectional causality between economic growth and
energy consumption in the developed countries in both the short and long run, while in the developing countries energy consumption
stimulates growth only in the short run. The former result is also found for energy importers and the latter result exists only for the
developed countries within this category. In addition, compared to the developing countries, the developed countries’ elasticity response
in terms of economic growth from an increase in energy consumption is larger although its income elasticity is lower and less than
unitary. Lastly, the implications for energy policy calling for a more holistic approach are discussed.
r 2006 Elsevier Ltd. All rights reserved.
0301-4215/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enpol.2006.08.019
ARTICLE IN PRESS
2482 R. Mahadevan, J. Asafu-Adjaye / Energy Policy 35 (2007) 2481–2490
Table 1
Overview of selected studies
Kraft and Kraft (1978) Bivariate Sims causality test 1947–1974 USA Income-energy
Yu and Choi (1985) Bivariate Granger test 1954–1976 South Korea Income-energy
Philippines Energy-income
Erol and Yu, 1987 Bivariate Granger test USA Energy!income
Yu and Jin (1992) Bivariate Engle & Granger test 1974–1989 USA Energy!income
Stern (1993) Multivariate VAR 1947–1990 USA Energy-income
Masih and Masih (1996) Trivariate VECM 1955–1990 Malaysia, Singapore, & Philippines Energy!income
India Energy-income
Indonesia Income-energy
Pakistan Energy2income
Glasure and Lee (1997) Bivariate VECM 1961–1990 South Korea & Singapore Energy2income
Masih and Masih (1998) Trivariate VECM 1955–1991 Sri Lanka & Thailand Energy-income
Asafu-Adjaye (2000) Trivariate VECM 1973–1995 India & Indonesia, Energy-income
Thailand & Philippines Energy2income
Hondroyiannis et al. (2002) Trivariate VECM 1960–1996 Greece Energy2income
Soytas and Sari (2003) Bivariate VECM 1950–1992 Argentina Energy2income
South Korea Income-energy
Turkey Energy-income
Indonesia & Poland Energy!income
Canada, USA, & UK Energy!income
Fatai et al. (2004) Bivariate Toda and Yamamoto (1995) 1960–1999 Indonesia & India Energy-income
Thailand & Philippines Energy2income
Oh and Lee (2004) Trivariate VECM 1970–1999 South Korea Energy2income
Wolde-Rufael (2004) Bivariate Toda and Yamamoto (1995) 1952–1999 Shanghai Energy-income
Lee (2005) Trivariate Panel VECM 1975–2001 18 developing countries Energy-income
Al-Iriani (2006) Bivariate Panel VECM 1971–2002 Gulf Cooperation Countries Income-energy
Notes: - means variable x Granger causes variable y; 2 means bidirectional causality; !means no causality in any direction. VAR means vector
autoregression and VECM means vector error-correction model.
of low power associated with the traditional unit root and 1971–2002.3 The information-intensive difficulty of this
cointegration tests. Pooling increases the sample size exercise is compounded by the fact that the use of energy
considerably, allowing for higher degrees of freedom and sources (such as use of coal, oil, etc.) vary in these
hence more accurate and reliable statistical tests. It also economies and different prices exist for residents and
reduces collinearity between regressors. Another advantage industries. Furthermore, industries that are energy-inten-
of using panel cointegration is that it allows for hetero- sive may well be subsidised by the government and
geneity among the countries. therefore face different prices. Hence the consumer price
To the best of our knowledge, only Lee (2005) and Al- index is used instead as energy prices are expected to be
Iriani (2006) have used the panel causality tests but our sufficiently reflected in this index.
study differs from theirs in more ways than one. While the The second contribution of the study is in the check for
latter study uses a bivariate model (and only reports long robustness of the empirical outcome by a comparison of
run results) for six countries in the Gulf Cooperation the panel causality results (both short and long run) with
Council, the former uses a trivariate model with capital those from the separate estimation of a vector error
stock2 for 18 developing countries. The trivariate model correction model (VECM) for each country. In addition,
allows an additional channel of causality to be investigated. the impact on elasticity with respect to changes in GDP
Thus similar to the Lee study, we consider a trivariate growth and energy consumption are also discussed using
model but one that proxies energy prices. This is because both pooled and individual estimations.
price responses have been argued to have a crucial role in The third contribution lies in the sample that considers a
affecting income and energy consumption directly (Dun- mix of countries comprising both net energy producers and
kerley, 1982; Hoa, 1993). Although data on energy prices consumers, as well as developing and developed countries.
would be ideal to use, given the multicountry nature in a Most previous studies have either focussed on single
panel estimation framework, it is not possible to obtain a countries or groups of countries of a similar level of
comparable series on energy prices for all 20 countries over economic development. Here, we examine countries at two
different stages of development within the group of energy
2
Although capital formation is a relevant variable, it reflects an
3
investment decision for energy production which may not directly affect This is clearly too tedious a task even for the World Development
household energy consumption. The latter is determined more by prices. Indicators to compile!
ARTICLE IN PRESS
R. Mahadevan, J. Asafu-Adjaye / Energy Policy 35 (2007) 2481–2490 2483
exporter and importers. This has important policy implica- Thus we adopt the IPS test which allows for a
tions for energy use considering the differences in produc- heterogeneous coefficient of yit#1. This is a more reason-
tion, consumption and institutional structures. able proposition because heterogeneity could arise from
The analysis of the sample mix adds to the existing different economic conditions and levels of development in
debate by considering the following questions. First, what each country. IPS propose averaging P the augmented
pi
are the similarities and/or differences in the causality Dickey–Fuller (ADF) tests, that is, !it ¼ j¼1 jij !it#j þ uit
behaviour between the variables for net energy exporters while allowing for different orders of serial correlation.
and importers? Second, within this category, does the Substituting this expression into Eq. (1), we get
causality relationship depend on the level of economic pi
X
development of the economy? Third, what implications yit ¼ ri yit#1 þ jij !it#j þ X it di þ !it , (2)
does this have on elasticities of income and energy j¼1
consumption as well as for energy policy? The paper
proceeds as follows. Section 2 provides an overview of the where ri is the number of lags in the ADF regression. The
panel VECM used in the study. Section 3 discusses the null hypothesis is that each series in the panel contains a
data. Section 4 presents and discusses the empirical results unit root, i.e. H0: ri ¼ 1 for all i. The alternative hypothesis
while Section 5 concludes. is that at least one of the individual series in the panel is
stationary, i.e. H1: rio1 for at least one i. IPS define a t-bar
statistic as the average of the individual ADF statistic
2. Estimation method
1X N
t̄ ¼ tr , (3)
Following established procedures, we conduct the test of N i¼1 i
the causal relationship between economic growth (GDP)
and energy consumption in three stages. First, we test for where tri is the individual t-statistic for testing H0: ri ¼ 1
the order of integration in the GDP, energy consumption for all i in Eq. (2). The t-bar statistic has been shown to be
and price series. Next, we employ panel cointegration tests normally distributed under H0 and the critical values for
to examine the long-run relationships among the variables. given values of N and T are provided in Im et al. (2003).
Finally, we use dynamic panel causality tests to evaluate
the short run cointegration and the direction of causality
2.2. Panel cointegration
among the variables. These results are then compared to
the conventional VECM separately estimated for each
If it is established from the unit root tests that the
country.
variables are integrated of order one, then the next step is
to apply cointegration analysis to determine whether a
2.1. Panel unit root tests long-run relationship exists among them. This is done by
applying the Johansen and Juselius (1992) maximum
Recent developments in the literature suggest that panel- likelihood approach to identify the number of cointegrat-
based unit root tests have higher power than unit root tests ing relationships between the three variables of interest.
based on individual time series. Newly developed panel The empirical model for this test is based on the following
unit root tests include Breitung (2000), Hadri (2000), Levin equation:
et al. (LLC) (2002), and Im et al. (IPS) (2003). Let us
gdpit ¼ ai þ dt þ benit þ gpit þ !it , (4)
consider the following autoregressive model
where gdp, en, and p are the natural logarithms of GDP,
yit ¼ ri yit#1 þ di X it þ !it , (1)
energy consumption, and prices, respectively; and a and d
where i ¼ 1; 2; . . . ; N represent countries observed over are country and time fixed effects, respectively. With a
periods t ¼ 1; 2; . . . ; T, Xit are exogenous variables in the dynamic panel containing a large cross-section dimension,
model including any fixed effects or individual trend, ri are Johansen’s procedure is likely to be infeasible and therefore
the autoregressive coefficients, and eit is a stationary panel cointegration methods are more appropriate.
process. If rio1, yi is said to be weakly trend-stationary. In this study, we use Pedroni’s (1999, 2000) method as it
On the other hand, if ri ¼ 1, then yi contains a unit root. allows for heterogeneity across individual members of the
The LLC, Breitung, and Hadri tests assume that the eit are panel. He considers seven different test statistics, four (the
IID (0, se2) and ri ¼ r for all i. This implies that the panel statistics in Table 6) of which are based on pooling
coefficient of yit#1 is homogeneous across all cross-section the residuals of the regression along the within-dimension
units of the panel and that individual processes are cross- of the panel, and the other three (the group statistics in
sectionally independent. Pesaran and Smith (1995) stress Table 6) are based on pooling the residuals of the
the importance of parameter heterogeneity in dynamic regression along the between-dimension of the panel. In
panel data models and analyse the potentially severe biases both cases, the basic approach is first to estimate the
that could arise from including it in an inappropriate hypothesised cointegrating relationship separately for each
manner. panel member and then to pool the resulting residuals for
ARTICLE IN PRESS
2484 R. Mahadevan, J. Asafu-Adjaye / Energy Policy 35 (2007) 2481–2490
conducting the panel tests. See Pedroni (1999) for the Table 2
details on these tests and the relevant critical values. List of selected countries
Table 4
Results of the IPS unit root tests for the full sample
Note: Numbers in parentheses are lag levels determined by the Schwarz Bayesian Criterion.
***Indicates significance at the 1% level.
Table 5
Results of Johansen’s cointegration tests
Country H0 Trace statistics Critical values Country H0 Trace statistics Critical values
capita energy consumption of 7.1 kg of oil equivalent the null hypotheses of a unit root cannot be rejected at
followed by Saudi Arabia and Singapore. South Korea their levels. However, upon taking first differences, the null
again had the highest annual GDP growth in the period of unit roots is rejected at the 1% significance level.
under study followed by Singapore. In particular, the Therefore, it is concluded that all the series are nonsta-
rapidly growing East Asian economies of South Korea, tionary and integrated of order one.
Singapore, Malaysia and Thailand, had the highest per Having established that energy consumption, GDP and
capita energy consumption and GDP growth, as well as prices are I(1), we next proceed to test whether a long-run
fairly low inflation rates. On the other hand, the less relationship exists between them. Here, we report both the
developing countries such as Nigeria, Ghana, Argentina results of Pedroni’s heterogeneous panel test as well as
and Senegal tended to have low energy consumption, low Johansen’s tests for the individual countries for compar-
GDP growth and high inflation. Overall, net energy ison. The Johansen test results in Table 5 indicate that in 15
importers have relatively higher per capita energy con- out of the 20 countries, the null hypothesis of no
sumption and GDP growth rates and lower inflation than cointegration can be rejected at the 5% significance level.
net energy exporters. Within the group of energy exporters, Nigeria and
Venezuela do not exhibit a long-run relationship and it is
4. Empirical results possible that the political environment, as well as the high
level of corruption in these economies, has blurred the
The results of the IPS panel unit root tests for the full
sample are presented in Table 4.4 For all three variables, (footnote continued)
space. For the same reason, the unit root tests for each individual country
4
Similar results are obtained for the two separate samples of energy using ADF and the Perron (1989) tests are also not reported as they
importers and exporters but these have not been reported to conserve indicated that the variables were I(1).
ARTICLE IN PRESS
2486 R. Mahadevan, J. Asafu-Adjaye / Energy Policy 35 (2007) 2481–2490
Table 7
Wald F-test statistics from panel VECM estimation
the short run. In the long run however, significant changes South Africa. This reflects the effects of gold mining
in investment and consumption patterns in energy use energy-intensive industry’s expansion as a result of
pertaining to price changes can take place. Although not economic growth. On the other hand, the causality results
reported, evidence shows that the difference among the from pooled and unpooled data on the developed countries
energy importers is that the developed countries have a are quite similar regardless of whether they are energy
lower (elasticity) response in terms of a fall in energy exporters or importers.
consumption due to a price increase relative to the The coefficients from the VECM estimations also
developing countries. This could be due to a difference in provide information on elasticities. Here, we only focus
the composition of energy consumers (industrial and on the long-run cointegrating relationships as they show
residential) in the two types of economies. For instance, stable behaviour. For most countries, increase in energy
the OECD (1992) reports that in the developed countries, consumption has a significant positive effect on GDP and
residential and industrial consumers respectively account vice versa. However, as these elasticity values vary
for 32% and 41% of total electricity consumption while the depending on the choice of the variables in the model,
corresponding proportions are 24% and 51% in the they are best interpreted in terms of relative and not actual
developing countries. Hondroyiannis et al. (2002) show magnitudes. Let us first consider the responsiveness of
that the two groups of energy consumers have very GDP growth to a 1% increase in energy consumption. The
different price elasticities. Thus, the results relating to energy exporting developed countries show an elasticity
price effects need to be interpreted with caution. value greater than one for both the pooled and unpooled
estimations. Saudi Arabia, a developing economy on the
4.2. Comparison of pooled and individual results other hand, has a value less than one. A similar result is
obtained for all the energy importers. But within the latter
Here, the causality results from the panel VECM are group, the developed economies of Sweden, Japan and the
compared against those obtained from the separate USA experience a greater increase in output growth than
estimation of the VECM for each economy in Table 8. the developing economies for a 1% increase in energy use,
For some economies such as Singapore and Malaysia, possibly reflecting energy efficiency measures in place.
bidirectional causality between energy consumption and These results are robust to panel and separate estimation of
GDP are observed in the long run unlike the generalised the VECM.
results from the panel estimations. Perhaps the fact that With income elasticity, the increase in energy consump-
these economies are newly industrialising economies makes tion brought about by a 1% increase in GDP is close to
it inappropriate to categorise them as developing countries unity for energy exporting developing countries except for
as was done in the panel analysis. These two economies Malaysia. Given that energy prices are heavily subsidised
together with the USA are heavily involved in information in these developing economies, energy is not perceived as a
and communication industries that require electricity and limiting source and there is a tendency to overuse it.
as such economic growth can be expected to lead to However, the developed economies which export energy
increased energy consumption in the long run. A similar have an income elasticity that is less than unitary as
result is obtained using unpooled data for Ghana and efficiency in energy management (as discussed earlier) may
ARTICLE IN PRESS
2488 R. Mahadevan, J. Asafu-Adjaye / Energy Policy 35 (2007) 2481–2490
modest increases in petroleum prices over the past two Dunkerley, J., 1982. Estimation of energy demand: the developing
years, a sharp rollback of subsidies was announced in countries. Energy Journal 3, 79–99.
September 2005 and this more than doubled the retail price Engle, R.F., Granger, C.W.J., 1987. Cointegration and error-correc-
tion: representation, estimation and testing. Econometrica 55,
of gasoline and diesel in Indonesia. Thus it remains to be 251–276.
seen how these structural reforms in the energy sector Erol, U., Yu, E.S.H., 1987. On the causal relationship between energy and
affect energy use and its subsequent impact on GDP income for industrialised countries. Journal of Energy and Develop-
growth especially now that Indonesia has reportedly ment 9, 75–89.
become a net importer of oil in 2004. Malaysia is another Fatai, K., Oxley, L., Scrimgeour, F.G., 2004. Modelling the causal
relationship between energy consumption and GDP in New Zealand,
economy which has considered reforms (but this is still at
Australia, India, Indonesia, the Philippines and Thailand. Mathe-
an early stage) to its power sector to make it more matics and Computers in Simulation 64, 431–445.
competitive and lower costs, and is expected to be net Glasure, Y.U., Lee, A.R., 1997. Cointegration, error-correction, and the
importer of oil by 2008.8 Before 1994, three state-owned relationship between GDP and energy: the case of South Korea and
utilities dominated power generation and distribution in Singapore. Resource and Energy Economics 20, 17–25.
Malaysia but since the market has been opened to Hadri, K., 2000. Testing for stationarity in heterogeneous panel
application. Econometric Journal 3, 148–161.
independent power producers.
Hoa, T.V., 1993. Effects of oil on output growth and inflation in
In addition, in the area of energy supply infrastructure, developing countries: the case of Thailand. International Journal of
there are also major plans underway to increase energy Energy Research 17, 29–33.
capacity with programs such as the ASEAN National Holtz-Eakin, D., Newey, W., Rosen, H., 1988. Estimating vector
Power Grid and power grid within the Gulf countries. This autoregressions with panel data. Econometrica 56, 1371–1395.
Hondroyiannis, G., Lolos, S., Papapetrou, G., 2002. Energy consumption
may help strengthen the causality from economic growth
and economic growth: assessing the evidence from Greece. Energy
(due to autonomous expenditure in capacity generation) to Economics 24, 319–336.
energy consumption. This has implications for energy Im, K.S., Persaran, M.H., Shin, Y., 2003. Testing for unit roots in
utilisation and hence energy conservation policy. heterogeneous panels. Journal of Econometrics 115, 53–74.
Johansen, S., Juselius, K., 1992. Testing structural hypotheses in a
multivariate cointegration analysis of the purchasing power parity and
5. Conclusion the uncovered interest parity for the UK. Journal of Econometrics 53,
211–244.
Jumbe, C.B.L., 2004. Electricity consumption and GDP: empirical
At the outset it must be noted that the purpose of this evidence from Malawi. Energy Economics 26, 61–68.
paper is not to resolve the conflicting evidence on the Kraft, J., Kraft, A., 1978. On the relationship between energy and GNP.
relationship between energy consumption and economic Journal of Energy and Development 3, 401–403.
growth. Instead, the study provides some broad and useful Lee, C.C., 2005. Energy consumption and GDP in developing countries: A
generalisations about the relationship in both developed cointegrated panel analysis. Energy Economics 27, 415–427.
Levin, A., Lin, C.F., Chu, C., 2002. Unit root tests in panel data:
and developing economies within the group of net energy
asymptotic and finite-sample properties. Journal of Econometrics 108,
exporters and importers. The results on causality and the 1–24.
elasticity response of changes in energy consumption and MacKinnon, J.G., Haug, A.A., Michelis, L., 1999. Numerical distribution
economic growth on each other are compared and functions of likelihood ratio tests for cointegration. Journal of Applied
discussed using both pooled and unpooled estimation Econometrics 14, 563–577.
techniques in an attempt to reconcile country-specific Masih, A.M.M., Masih, R., 1996. Energy consumption, real income and
temporal causality: results from a multi-country study based on
analysis before the generalisations can be substantiated.
cointegration and error-correction modelling techniques. Energy
Lastly, the form in which energy policy takes effect is Economics 18, 165–183.
crucial. While it is difficult to be definitive about energy Masih, A.M.M., Masih, R., 1998. A multivariate cointegrated modelling
policy, it must be acknowledged that such a discussion approach in testing temporal causality between energy consumption,
needs a holistic setting to be more effective and not just real income and prices with an application to two Asian LDCs.
based on the empirical evidence on causality between Applied Economics 30, 1287–1298.
OECD/International Energy Agency (IEA), 1992. Global Energy: the
energy consumption and economic growth. Changing Outlook. OECD, Paris.
Oh, W., Lee, K., 2004. Energy consumption and economic growth in
Korea: testing the causality relation. Journal of Policy Modeling 26,
References 973–981.
Pedroni, P., 1999. Critical values for cointegration tests in heterogeneous
Al-Iriani, M.A., 2006. Energy–GDP relationship revisited: an example panels with multiple regressors. Oxford Bulletin of Economic and
from GCC countries using panel causality. Energy Policy 34 (17), Statistics 61, 653–678.
3342–3350. Pedroni, P., 2000. Fully modified OLS for heterogeneous cointegrated
Asafu-Adjaye, J., 2000. The relationship between energy consumption, panels. Advances in Econometrics 15, 93–130.
energy prices and economic growth: time series evidence from Asian Perron, P., 1989. The great crash, the oil price shock and the unit root
developing countries. Energy Economics, 615–625. hypothesis. Eonometrica 57, 1361–1401.
Breitung, J., 2000. The local power of some unit root tests for panel data. Pesaran, M.H., Smith, R.J., 1995. Long-run relationships from dynamic
Advances in Econometrics 15, 161–177. heterogeneous panels. Journal of Econometrics 68, 79–113.
Shiu, A., Lam, P.L., 2004. Electricity consumption and economic growth
8
See ASEAN Energy Bulletin, May 2003, vol.7, no.1: 8–10. in China. Energy Policy 32, 47–54.
ARTICLE IN PRESS
2490 R. Mahadevan, J. Asafu-Adjaye / Energy Policy 35 (2007) 2481–2490
Soytas, U., Sari, R., 2003. Energy consumption and GDP: causality Yoo, S.-H., 2006. The causal relationship between electricity consumption
relationship in G-7 countries and emerging markets. Energy Econom- and economic growth in the ASEAN countries. Energy Policy
ics 25, 33–37. forthcoming.
Toda, H.Y., Yamamoto, T., 1995. Statistical inference in vector Yu, E.S.H., Choi, J.Y., 1985. The causal relationship between energy and
autoregressions with possibly integrated processes. Journal of Econo- GNP: an international comparison. Journal of Energy and Develop-
metrics 66, 225–250. ment 10, 249–272.
Wolde-Rufael, Y., 2004. Disaggregated energy consumption and GDP, Yu, E.S.H., Jin, J.C., 1992. Cointegration tests of energy consumption,
the experience of Shangai, 1952–1999. Energy Economics 26, 69–75. income and employment. Resources and Energy 14, 259–266.