Energy Economics 37 (2013) 52–59
Contents lists available at SciVerse ScienceDirect
Energy Economics
journal homepage: www.elsevier.com/locate/eneco
Do urbanization and industrialization affect energy intensity in developing countries?
Perry Sadorsky ⁎
Schulich School of Business, York University, 4700 Keele Street, Toronto, Ontario, Canada M3J 1P3
a r t i c l e
i n f o
Article history:
Received 17 May 2012
Received in revised form 17 January 2013
Accepted 20 January 2013
Available online 31 January 2013
JEL classification:
Q43
R11
O14
a b s t r a c t
Against a backdrop of concerns about climate change, peak oil, and energy security issues, reducing energy intensity is often advocated as a way to at least partially mitigate these impacts. This study uses recently developed
heterogeneous panel regression techniques like mean group estimators and common correlated effects estimators to model the impact that income, urbanization and industrialization has on energy intensity for a panel of 76
developing countries. In the long-run, a 1% increase in income reduces energy intensity by −0.45% to −0.35%.
Long-run industrialization elasticities are in the range 0.07 to 0.12. The impact of urbanization on energy intensity is mixed. In specifications where the estimated coefficient on urbanization is statistically significant, it is
slightly larger than unity. The implications of these results for energy policy are discussed.
© 2013 Elsevier B.V. All rights reserved.
Keywords:
Energy intensity
Developing countries
Industrialization
Urbanization
1. Introduction
Against a backdrop of concerns about climate change, peak oil, and
energy security issues, reducing energy intensity is often advocated as
a way to at least partially mitigate these impacts. Energy intensity
tends to correlate highly with income and higher income countries
have lower energy intensity than poorer countries. In addition to income, other factors like urbanization and industrialization may affect
energy intensity. The impact that urbanization has on energy intensity
is difficult to predict because urbanization increases economic activity
through a higher concentration of consumption and production but urbanization also leads to economies of scale and provides the opportunity for increases in energy efficiency.1 Industrialization, the introduction
of new equipment and techniques to make existing and new products,
increases industrial activity which uses more energy than does traditional agriculture or manufacturing implying that industrialization has
a positive impact on energy intensity.
Energy intensity for high income countries (HIC) has been falling
over the past 30 years and for major country income aggregates energy
⁎ Tel.: +1 416 736 5067; fax: +1 416 736 5687.
E-mail address: [email protected].
1
In this paper, urbanization refers to The World Bank's definition of the percentage
of the population living in urban areas as defined by national statistical offices (http://
data.worldbank.org/indicator/SP.URB.TOTL.IN.ZS). However, urban areas can be defined differently by different national statistical offices and in general there is no universally accepted definition of urbanization (eg. Vlahov and Galea, 2002). Moreover,
a country's definition of an urban area can change across time. I thank a well informed
reviewer for pointing this out.
0140-9883/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.eneco.2013.01.009
intensity today is lower than what it was in 1980 (Fig. 1). In general,
high income countries (HIC) are the most efficient at using energy
while low and middle income countries (LMY) tend to be the less efficient. In 2010, for example, energy use (kg of oil equivalent) per
$1000 GDP (constant 2005 PPP) in high income countries was 40%
lower than in 1980. For the world as a whole, energy intensity in 2010
was 27% lower than in 1980, while for the low and middle income countries energy intensity in 2010 was 23% lower than in 1980.2
This paper makes several important contributions to the literature. First, the relationship between urbanization and energy has
been studied by a number of authors (eg. Jones, 1989, 1991; Parikh
and Shukla, 1995; Poumanyvong and Kaneko, 2010; York, 2007) but
most of this research focuses on energy use rather than energy intensity. Jones (1991) appears to be the first to specifically investigate the
relationship between energy intensity, urbanization and industrialization for developing economies but on the whole, there is very little
known about how urbanization and industrialization affect energy intensity. It is important to have a better understanding of how income,
urbanization and industrialization impact energy intensity because
increases in energy efficiency are one way to mitigate concerns regarding climate change, peak oil, and energy security issues.
Second, while panel data techniques are becoming more common,
most existing models linking urbanization, industrialization and energy
use a static model applied to a panel data set. A panel data set offers advantages over a cross-section data set by including a time dimension.
2
Notice how energy intensity in the low and middle income country group spiked
following the 1991 breakup of the Soviet Union into 15 independent republics.
P. Sadorsky / Energy Economics 37 (2013) 52–59
53
LMY
HIC
240
360
220
320
200
280
180
240
200
1980
160
1985
1990
1995
2000
2005
2010
2000
2005
2010
140
1980
1985
1990
1995
2000
2005
2010
WORLD
240
230
220
210
200
190
180
1980
1985
1990
1995
Fig. 1. Energy use (kg of oil equivalent) per $1000 GDP (constant 2005 PPP).
(Data sourced from http://www.worldbank.org/data/onlinedatabases/onlinedatabases.html).
2. The impact of urbanization, industrialization and income on
energy use
According to data sourced from the United Nations Population
Division (2007), while the most developed regions (MDR) of the
world have higher urbanization (% of population living in urban
areas) than the less developed regions (LDR) of the world, urbanization for both groups is expected to continue rising (Fig. 2) with urbanization in the LDRs rising the fastest. The year 2010 marks a
milestone because this was the year when the world urbanization
passed 50%. It is expected that in the year 2020, urbanization in the
less developed regions of the world will pass 50%. Notice that for
the 100 year time period shown in Fig. 2, urbanization in the less
90
80
70
60
50
40
30
20
10
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
2020
2025
2030
2035
2040
2045
2050
This increases the number of observations and allows for variation in
both the cross-section and time dimension. Static models cannot, however, capture dynamic relationships. This present paper uses a dynamic
framework to model the impact of income, urbanization and industrialization on energy intensity. Dynamic models are advantageous because
both long-run and short-run impacts (elasticities) are modeled.
Third, previous studies have assumed that the impact of urbanization and industrialization on energy use is homogeneous across countries. This is a very strong assumption to make and one that is unlikely
to hold across a large grouping of countries. In this present paper
panel regression models are estimated using recently developed techniques like mean group estimators that allow for heterogeneity in the
estimation of the slope coefficients. If panel data exhibits cross-section
dependence, estimating models with homogeneous slope coefficients
(as in the case of pooled OLS, fixed effects, or GMM) may yield misleading results. In order to account for cross-section dependence, models
are estimated using the mean group (MG) estimator of Pesaran and
Smith (1995), Pesaran's (2006) Common Correlated Effects Mean
Group (CCEMG) estimator, and the Augmented Mean Group (AMG) estimator of Eberhardt and Teal (2010) and Bond and Eberhardt (2009).
The purpose of this paper is to investigate the relationship between energy intensity, income, urbanization and industrialization
for a panel of 76 developing economies. Empirical models are estimated using heterogeneous panel regression techniques. The following sections of the paper set out the contextual material, the empirical
model, data, empirical results, implications, and conclusions.
LDR
MDR
WORLD
Fig. 2. Percent urban in less developed regions (LDR), most developed regions (MDR)
and the World.
(Data sourced from http://esa.un.org/unup/.)
54
P. Sadorsky / Energy Economics 37 (2013) 52–59
developed regions of the world is expected to more than triple, from
18% in 1950 to 67% in 2050.
Urbanization is associated with large scale movements of the labor
force from the country side into urban areas the result of which is to
increase population density in the urban areas. This increase in population density puts stress on the local environment. Pacione (2009),
for example, estimates that cities account for 75% of the world's consumption of natural resources yet cities cover only 2% of the world's
surface.
There are several channels by which urbanization can affect energy
use (Jones, 1989, 1991; Madlener, 2011; Madlener and Sunak, 2011;
Parikh and Shukla, 1995). First, urbanization can affect energy use by
its impact on production. Urbanization is associated with a concentration of economic activity in cities and metropolitan areas which leads
to economies of scale in production. Production shifts from less energy
intensive agriculture to more energy intensive manufacturing. Urbanization brings about fuel switching as decentralized rural energy sources
like traditional wood burning are switched for centralized energy
sources. Increased production in urban areas can also lead to an increase
in the informal economy, which may be a further source of increased energy use (Schneider and Enste, 2000). Second, urbanization affects mobility and transport by increasing the amount of motorized traffic into
and out of urban areas and this increase in traffic increases the demand
for energy. Transporting raw materials into the urban area production
centers and then transporting the finished goods to other destinations
increases the demand for energy. Urbanization also separates the consumers of food from the producers of food and energy use increases as
food products are transported into urban areas. Provided there is adequate mass transit infrastructure, mass transit may offer some relief
from commuter chaos and reduce the demand for energy (transportation fuel) compared to how much energy would be consumed if every
commuter travelled in their own car. Third, increased urbanization increases the demand for infrastructure. Growing cities, for example, increase the demand for energy intensive products and materials as
infrastructure is built. On the other hand, recently developed green
building codes like LEEDS certification will help to reduce the energy intensity of buildings. Fourth, urbanization can impact energy demand
through its impact on private consumption patterns. Urbanization is accompanied by economic development and as urban dwellers become
wealthier their consumption patterns change to include more energy intensive products. Obvious examples include refrigerators, air conditioning and automobiles. The impact that urbanization has on energy
intensity is difficult to predict because urbanization increases economic
activity through a higher concentration of consumption and production
but urbanization also leads to economies of scale and provides the opportunity for increases in energy efficiency.
Industrialization is a term that usually refers to an increase in industrial activity and most authors assume that industrialization leads to higher
energy usage because higher value added manufacturing uses more energy than does traditional agriculture or basic manufacturing. For example,
industries like petroleum refining, primary metals, chemicals, and paper
and allied products tend to be more energy intensive than agriculture
or textile industries (eg. Jones, 1991; Samouilidis and Mitropoulos,
1984). In empirical models, industrialization is usually measured by industry value added as a % of GDP. According to Blanchard (1992) this indicator represents internal manufacturing specialization. The evolution of
this indicator across time measures the productive restructuring effort.
Some authors like Parikh and Shukla (1995) use this indicator as a measure of structural change and the expectation is that larger shares of industrial activity in the economy create a demand for more energy.
Blanchard (1992), however, points out that the evolution of this ratio
across time may be due to structural changes but may also be due to divergent price effects between the two variables used to construct the
ratio.
Malenbaum (1978) was the first to show how resource intensity
changed with increases in per capita income. In his study of the resource
intensity of 12 major minerals over the period 1950 to 1975 across 10
world regions, he found that resource intensity initially increased with increases in income, reached a plateau and then begin to decline with further increase in income. This idea was applied to energy intensity and
according to Bernardini and Galli (1993) there are three reasons for the
decline in energy intensity as income increases. First, there are changes
in the structure of final demand as economies pass through the stages
of pre-industrialization, industrialization and post-industrialization. In
the pre-industrialization stage, agriculture is the dominant industry and
economic growth is driven by basic needs that can be met with low energy intensity. In the industrialization stage, the infrastructure network is
built up to facilitate mass production and mass consumption. The initial
build up of the capital stock associated with industrialization can increase
energy intensity but eventually a saturation point is reached where the
consumption of materials is more oriented to replacement of durables
rather than the creation of durables. In the post-industrialized stage,
manufacturing declines in relationship to services and energy intensity
in service oriented economies is lower than in manufacturing based economies. Second, technological progress leads to increases in energy efficiency. Third, technological progress leads to the usage of substitute
materials which are less energy intensive.
Galli (1998) studied the possibility of an inverted U shaped pattern for energy intensity in a sample of 10 Asian economies over the
period 1973–1990. Panel models are estimated using fixed effects
and random coefficients estimators. There is some evidence of an
inverted U shaped relationship between energy intensity and income
for the fixed effects specification but no statistically significant evidence of this relationship in the random coefficients specification.
Jones (1989) studies the impact of urbanization on energy use for
a cross section of 59 developing countries for the year 1980. Regression results are presented for the dependent variable measured either
as energy use per capita or energy per dollar of GDP. Energy use is
also classified as either modern (commercially traded fossil fuels) or
total (which includes traditional fuels like wood and biomass along
with modern fuels). Explanatory variables include income, industrial
structure, urbanization, and population density. Income elasticities
of modern and total (modern plus traditional) energy consumption
are estimated in the range of 0.64 to 1.10. Industrialization elasticities
of energy consumption are estimated to range between 0.83 and 1.08.
Holding other variables constant, urbanization elasticities of energy
consumption are estimated between 0.30 and 0.48. Jones (1991),
using a similar data set for energy intensity as used in Jones (1989),
finds, in the long-run, an income elasticity of 0.77, an urban elasticity
of 0.35 and an industrialization elasticity of 1.35. In their study of the
Greek economy, Samouilidis and Mitropoulos (1984) find long-run
elasticities of industrialization of energy intensity in the range 0.90
to 1.96 and short-run elasticities in the range 0.17 to 0.46. Parikh
and Shukla (1995) use a pooled data set of developed and developing
countries over the years 1965–87 to investigate the impact of urbanization on energy consumption. For total energy consumption models, they find the income elasticity varies between 0.25 and 0.47,
while the urbanization elasticity varies between 0.28 and 0.47. In addition to including explanatory variables for income and urbanization, they include variables for population density and the share of
agriculture in GDP. Liddle (2004) finds that urbanization and population density have a negative impact on per capita road transportation
energy use. This implies that populous, highly urban cities have less
demand for personal transport. Cole (2006) investigates the relationship between trade and per capita energy use or energy intensity in a
panel of 32 developed countries for the period 1975–1995. Income
elasticities range from − 1.1 to − 0.1 depending on the specification
of the regression model. York (2007) uses panel data techniques to
investigate the determinants of energy consumption in the European
Union over the period 1960 – 2025. Income elasticities vary between
0.52 and 0.69. Urbanization elasticities vary between 0.29 and 0.56.
Population elasticities, ranging from 2.56 to 2.75, are much larger
P. Sadorsky / Energy Economics 37 (2013) 52–59
than income or urbanization elasticities indicating that slowing population growth in the European Union is expected to play a large
role in reducing energy consumption. Mishra et al. (2009) study the
impact that urbanization has on per capita energy use in a sample
of Pacific Island economies. They find that urbanization has a negative
impact on energy use in New Caledonia, but a positive impact in Fiji,
French Polynesia, Samoa and Tonga. Poumanyvong and Kaneko
(2010) use panel data techniques to estimate the impact of income,
urbanization, industrialization, and population on energy use in a sample of 99 countries covering the period 1975 – 2005. They find that the
impact of urbanization on energy use varies by income class. Urbanization decreases energy use in the low-income group, while it increases
energy use in the middle- and high-income groups. The impact of the
share of industrial activity in the economy on energy consumption is
positive, but statistically significant for only the low- and middleincome groups. Krey et al. (2012) use integrated assessment models
to analyze the impact of urbanization on residential energy use in
China and India. They find that residential energy use is not very sensitive to urbanization directly but the relationship between urbanization
and energy use depends upon how labor productivity affects economic
growth. O'Neill et al. (2012) use a computable general equilibrium
model to investigate the impact of urbanization of energy use in China
and India. They find that the direct impact of urbanization on energy
use is not that strong and much of the impact of urbanization on energy
use comes through the impact that an increased labour supply has on
economic growth. Stern (2012) uses a stochastic production frontier
approach to model energy efficiency trends in 85 countries over a
37 year period. The empirical results show that countries with higher
total factor productivity, undervalued currencies and smaller fossil
fuel reserves have higher energy efficiency. These models do not, however, include specific variables for urbanization.
3. Empirical model
Following Jones (1991), the relationship between the logarithm of
energy intensity (e), logarithm of income (y), logarithm of urbanization (u) and logarithm of industrialization (d) is specified as:
eit ¼ β1i yit þ β2i uit þ β3i dit þ ν i þ εit :
ð1Þ
In Eq. (1), countries are denoted by the subscript i (i = 1,…,N) and
the subscript t (t = 1,…,T) denotes the time period. Country specific
effects are included through νi and εit represents the random error
term. All variables are expressed in natural logarithms and as a result
the estimated coefficients can be interpreted as elasticities. A distinction can be made between models with homogeneous slope coefficients (β1i = β1, β2i = β2, β3i = β3, β4i = β4,) and models with
heterogeneous slope coefficients (β1i, β2i, β3i, β4i). If the assumption
of homogeneous slope coefficients is made then these models can
be estimated using standard panel regression techniques like pooled
OLS (POLS) and various fixed effects (FE) or GMM specifications.
Table 1
Summary statistics.
Obs
Mean
Std. D.
Min
55
Models with heterogeneous slope coefficients can be estimated
using mean group (MG) estimators (eg. Pesaran and Smith, 1995;
Pesaran, 1997) or variants on mean group estimators. Estimating
panel models with heterogeneous slope coefficients is currently an
active area of econometrics (eg. Coakley et al., 2006; Eberhardt and
Teal, 2011; Eberhardt et al., in press).
The relationship between energy intensity (e), income (y), urbanization (u) and industrialization (d) can be specified as a dynamic
panel data model:
eit ¼ α i eit−1 þ β1i yit þ β2i yit−1 þ β3i uit þ β4i uit−1 þ β5i dit
þ β6i dit−1 þ νi þ εit :
ð2Þ
Eq. (2) is an example of an autoregressive distributed lag (ARDL)
model of order one for each variable. 3 This model is a dynamic version of the static model originally proposed by Jones (1991). Dynamic
models are advantageous over static models because dynamic models
easily facilitate the calculation of both short-run and long-run elasticities. ARDL models can also be estimated assuming homogenous slope
coefficients or heterogeneous slope coefficients.
Models are estimated using the mean group (MG) estimator of
Pesaran and Smith (1995), Pesaran's (2006) Common Correlated Effects
Mean Group (CCEMG) estimator, and the Augmented Mean Group
(AMG) estimator (Bond and Eberhardt, 2009; Eberhardt and Teal,
2010). The mean group (MG) approach incorporates heterogeneity
across countries by allowing all slope coefficients and error variances
to vary across panels (or countries as is the case in this paper)
(Pesaran and Smith, 1995). The MG approach applies OLS to each
panel/country to obtain panel specific slope coefficients and then averages the panel specific coefficients. For large T and N the MG estimator is
consistent. For inference on the long-run parameters, Pesaran (1997)
and Pesaran and Shin (1999) show that including the appropriate number of lags in the order of the ARDL model can simultaneously correct
for the problem of residual serial correlation and endogenous regressors. The MG estimator does not, however, incorporate any information
on common factors that may be present in the panel data set. Common
factors are time specific effects that are common across countries. Examples include fluctuations in global energy prices, technological
change, and global business cycle conditions. Pesaran's (2006) Common
Correlated Effects Mean Group (CCEMG) estimator includes crosssectional dependence and heterogeneous slope coefficients. The crosssectional dependence is modeled using cross-sectional averages of the
dependent and independent variables. These cross-sectional averages
account for the unobserved common factors. The unobservable common factors may be nonlinear or non-stationary. As in the case of the
MG estimator, the slope coefficients are averaged across panel members. The CCEMG estimators are very robust to structural breaks, lack
of cointegration and certain serial correlation (Kapetanios et al.,
2011). The Augmented Mean Group (AMG) estimator is an alternative
to the Pesaran (2006) CCEMG estimator (Bond and Eberhardt, 2009;
Eberhardt and Teal, 2010). In the CCEMG approach the set of
unobservable common factors is treated as a nuisance. In the AMG approach the set of unobservable common factors are treated as a common dynamic processes that, depending upon the context, may have
useful interpretations.
Max
4. Data
Logarithms
Energy/GDP
GDP/POP
Urban
Industry
2039
2177
2356
2071
5.425
8.134
3.794
3.415
0.634
0.873
0.487
0.361
4.195
5.513
1.808
1.971
7.453
9.777
4.543
4.385
Growth rates
Energy/GDP
GDP/POP
Urban
Industry
1963
2101
2280
1995
−1.057
1.524
1.041
−0.130
7.269
6.542
1.196
10.323
−54.403
−60.377
−1.919
−95.702
54.323
64.432
11.657
168.339
The data set is an unbalanced panel of 76 developing countries
followed over the years 1980–2010. The dimensions of the panel
data set are chosen to include as many countries as possible each
with a reasonable time length of observations. The countries are
3
In practice, one can allow each right hand side variable in Eq. (2) to have different
lag lengths. In this present paper, the optimal combination of lags is chosen using the
SIC starting from a maximum of one lag on each variable.
56
P. Sadorsky / Energy Economics 37 (2013) 52–59
Table 2
Correlations.
Table 3
Tests for cross-section dependence and unit roots.
Energy
Energy
Income
Urban
Industry
1
−0.591
−0.404
−0.185
Income
1
0.793
0.503
Urban
1
0.459
Industry
Variable
CD-test
P-value
Corr
Abs (corr)
CIPS
P-value
1
Energy
Income
Urban
Industry
46.250
117.630
114.910
6.230
0.000
0.000
0.000
0.000
0.190
0.473
0.435
0.023
0.513
0.636
0.930
0.413
1.149
2.085
3.757
0.518
0.875
0.981
1.000
0.719
Nobs = 1985.
selected from the World Bank's classification of developing countries:
low income (LIC), lower middle income (LMI), upper middle income
(UMI). The list of countries and their country group based on income
classification is reported in Table A1 of the Appendix A.
In the empirical analysis, energy is the natural log of energy intensity (energy use in kg of oil equivalent per $1000 GDP (constant 2005
PPP)), income is the natural log of real per capita GDP (GDP per
capita, PPP (constant 2005 international dollars), urban is the natural
log of urbanization (measured by the fraction of the population living
in urban areas) and industry is the natural log of industrialization (industry, value added as a % of GDP). The data are obtained from the
World Bank (2011) World Development Indicators online data base.
Summary statistics for the variables are shown in Table 1. For the
variables measured in natural logarithms, income has the greatest
variability while industrialization has the least. For the complete
panel of countries, the average annual growth rate in energy intensity
is − 1.057% while the average annual growth rate in per capita income is 1.524%. Urbanization increased, on average, by 1.041% per
year while industrialization declined, on average, by 0.13% per year.
The correlation coefficients show that energy intensity correlates
the highest with income, while income correlates the highest with urbanization (Table 2). Industrialization correlates the highest with
income.
Unit root tests that assume cross-sectional independence can have
low power if estimated on data that have cross-sectional dependence.
In order to account for this possibility, Pesaran's (2004) cross-section dependence (CD) test was used to check for cross-sectional dependence.
The CD tests indicate that each series exhibits cross-sectional dependence
(Table 3). As a result, Pesaran's (2007) CIPS (Z(t-bar)) test for unit roots
was calculated. This is a unit root test that allows for cross-sectional dependence. These tests were estimated with a constant term and 2 lags.
The CIPS tests indicate that each series contains a unit root.
5. Empirical results
The empirical analysis is conducted by estimating a series of regression models under different assumptions. The first suite of results is for
static models with homogeneous slope coefficients. Empirical results
are presented for models estimated using pooled OLS (POLS), fixed
effects (FE), fixed effects instrumental variable (FE-IV), and fixed effects
first difference (FD) (Table 4). The residuals are tested for cross-sectional
dependence using Pesaran's (2004) CD test and stationarity using
Pesaran's (2007) CIPS. It is important to test for stationarity in the residuals because residual stationarity is an important part of a good fitting
econometric model.4
For the case of the static model with pooled estimates, the estimated
coefficient on the income variable covers a fairly tight range of between
−0.554 and −0.473 (Table 4). The estimated coefficient on the income
variable is negative and statistically significant at the 1% level in each
specification. The estimated coefficient on the industry variable is positive and statistically significant in each specification. The estimated
4
Following a suggestion by a reviewer, a series of regression models was estimated
to investigate the impact of each explanatory variable and its square on energy intensity. To account for heterogeneity, the regressions were estimated using the mean
group (MG) estimator (Pesaran and Smith, 1995). At the 10% level of significance there
is no evidence of quadratic terms.
coefficient on the urban variable is statistically significant in just one
of the specifications. Collectively, these results suggest that increases
in income reduce energy intensity while increases in industrialization
increase energy intensity. In three out of four of the specifications, urbanization has a statistically insignificant impact on energy intensity.
Applying the CD test to the regression residuals provides little evidence
of cross-section dependence (except in the case of POLS). More troubling is the CIPS test indicates that, except for the first difference specifications, all other regressions have non-stationary residuals and
non-stationary residuals indicate a poorly fitting model.
Empirical results for static models with heterogeneous estimates
are reported in Table 5. The estimated coefficient on the income variable is statistically significant at the 1% level and ranges between −
0.499 and − 0.434 (a closer range than what was observed in the
homogeneous static case). The estimated coefficient on the industrialization variable is positive and statistically significant in two of the
specifications. The estimated coefficient on the industrialization variable ranges from 0.011 to 0.096 (a closer range than what was observed in the homogeneous static case). The estimated coefficient
on the urbanization variable is positive and statistically significant
in two of the specifications. The CD test indicates some evidence of
cross-section dependence in the MG specification. The CIPS test indicates that the residuals from each specification are stationary, which
satisfies a requirement of a good fitting model.
The dynamic model presents a challenge for the unbalanced panel
data set of 76 countries because some countries have rather short
time series. The average observations per group is 24.9, the minimum
observations per group is 13 and the maximum observations per
group is 30. Taking this into account, the dynamic model in Eq. (2)
was chosen using the SIC starting from a maximum of one lag on
each variable. Heterogeneous parameter estimates for the dynamic
panel model are reported in Table 6. The estimated coefficient on
the lagged energy intensity variable is positive and statistically significant at the 1% level in each of the specifications. The estimated coefficient ranges from 0.218 to 0.533 indicating a low to moderate level
of persistence. The estimated coefficient on the income variable is
Table 4
Pooled estimates (static).
POLS
Income
Urban
Industry
Constant
RMSE
CD test
CIPS test
Observations
Countries
FE
a
−0.554
(−26.24)
0.162 a
(4.91)
0.261 a
(6.69)
8.27 a
(63.6)
0.504
0.000
0.892
1955
−0.542
(−6.98)
0.034
(0.19)
0.200 b
(2.43)
8.962 a
(10.92)
0.150
0.412
0.350
1955
76
FE-IV
a
FD
a
−0.473
(−24.07)
0.007
(0.15)
0.189 a
(8.78)
8.677 a
(38.61)
0.155
0.817
1903
76
−0.552 a
(−10.4)
0.295
(1.13)
0.040 b
(2.50)
−0.007
(−0.26)
0.060
0.095
0.000
1879
76
Estimation is from an unbalanced panel of 76 developing countries covering the period
1980–2010. POLS, pooled OLS; FE, fixed effects; FE-IV, fixed effects instrumental variables, FD, fixed effects first difference. For the FE-IV estimation, the instrument list includes year dummy variables, urban, industry and a one period lag of income. T
statistics reported in parentheses. The superscripts a, b and c denote significance at
the 1%, 5% and 10% levels respectively. P values reported for the CD and CIPS tests.
Year dummy variables are included in each specification.
P. Sadorsky / Energy Economics 37 (2013) 52–59
Table 5
Heterogeneous estimates (static).
Income
Urban
Industry
Constant
RMSE
CD test
CIPS test
Observations
Countries
Table 7
Energy intensity elasticities.
MG
CCEMG
AMG
Elasticities
MG
CCEMG
AMG
−0.482a
(−9.55)
1.896 b
(1.98)
0.096 b
(2.39)
1.273
(0.32)
0.063
0.011
0.000
1955
76
−0.434 a
(−6.12)
1.915
(1.62)
0.011
(0.27)
1.719
(0.30)
0.041
0.829
0.000
1955
76
−0.499 a
(−9.18)
1.821 a
(2.72)
0.072 b
(2.00)
1.955
(0.74)
0.052
0.198
0.000
1955
76
Short-run
Income
Urban
Industry
−0.53
0.44
0.06
−0.54
−0.02
0.06
−0.57
1.19
0.05
Long-run
Income
Urban
Industry
−0.35
0.95
0.12
−0.45
−0.02
0.07
−0.35
2.11
0.09
Estimation is from an unbalanced panel of 76 developing countries covering the period
1980–2010. MG, mean group; CCEMG, cross correlated effects mean group; AMG, augmented mean group. T statistics reported in parentheses. The superscripts a, b and c denote significance at the 1%, 5% and 10% levels respectively. P values reported for the CD
and CIPS tests. Estimated coefficients are un-weighted averages across countries.
negative, statistically significant, and ranges in value between −0.573
and −0.528. These values are slightly smaller than those estimated
under the static heterogeneous specifications. The estimated coefficient
on the lagged income variable is positive and statistically significant in
all three specifications. The estimated coefficient on the industry variable is positive, statistically significant, and ranges in value between
0.052 and 0.056. The estimated coefficient on the industry variable is
very similar across the three dynamic specifications. These values are
of the same sign and magnitude as those estimated under the static heterogeneous specifications. Only in the case of the AMG is the estimated
coefficient on the urban variable statistically significant. This indicates
that the estimated coefficient for the urban variable is more sensitive
to the estimation technique than the estimated coefficients on the
other variables. The CD test indicates little evidence of cross-section dependence in the CCEMG or AMG specifications. The CIPS test indicates
that the residuals from each specification are stationary, which satisfies
a requirement of a good fitting model.
The difficulties in finding a strong relationship between energy intensity and urbanization are consistent with the findings of Krey et al.
(2012) and O'Neill et al. (2012). Krey et al. (2012) use integrated assessment models to analyze the impact of urbanization on residential
energy use in China and India while O'Neill et al. (2012) use a
Table 6
Heterogeneous estimates (dynamic).
Energy (−1)
Income
Income (−1)
Urban
Industry
Constant
RMSE
CD test
CIPS test
Observations
Countries
57
MG
CCEMG
0.533a
(16.50)
−0.528 a
(−9.76)
0.366 a
(6.58)
0.442
(0.76)
0.056 b
(2.16)
1.737
(0.73)
0.0434
0.003
0.000
1890
76
0.218
(5.03)
−0.538
(−8.42)
0.185
(3.62)
−0.016
(−0.08)
0.056
(1.75)
5.766
(6.36)
0.031
0.076
0.000
1890
76
AMG
a
a
a
c
a
0.437 a
(10.90)
−0.573 a
(−10.59)
0.378 a
(6.58)
1.19 b
(2.06)
0.052 b
(1.99)
−0.439
(−0.18)
0.039
0.332
0.000
1890
76
Estimation is from an unbalanced panel of 76 developing countries covering the period
1980–2010. MG, mean group; CCEMG, cross correlated effects mean group; AMG, augmented mean group. T statistics reported in parentheses. The superscripts a, b and c denote significance at the 1%, 5% and 10% levels respectively. P values reported for the CD
and CIPS tests. Estimated coefficients are un-weighted averages across countries.
computable general equilibrium model to investigate the impact of
urbanization of energy use in China and India. Even though the approaches are different, both studies find that energy use is not very
sensitive to urbanization. The results in this present paper are consistent with Poumanyvong and Kaneko (2010) who use pooled static
models and find that unlike income and industrialization, the estimated coefficient on urbanization is sensitive to the estimation
technique.
6. Implications
The empirical results reported in Table 6 can be used to calculate
short-run and long-run energy intensity elasticities (Table 7). When
the dependent variable is measured as per capita energy use, the income
variable captures both scale and technique effects (Cole, 2006). When
the dependent variable is measured as energy intensity, the income variable measures only a technique effect and not a scale effect. The
short-run income elasticity ranges between −0.57 and −0.53 while
the long-run income elasticity ranges between −0.45 and −0.35.
These results suggest that the technique effect is similar in both the
short-run and long-run. The short-run industry elasticity ranges between 0.05 and 0.06 while the long-run elasticity ranges between 0.07
and 0.12. The long-run industry elasticity is slightly larger than the
short-run industry elasticity. The short-run industrialization elasticities
are smaller than the 0.17 to 0.46 range found by Samouilidis and
Mitropoulos (1984). The long-run industrialization elasticities are
smaller than the 0.90 to 1.96 range found by Samouilidis and
Mitropoulos (1984) or the value of 1.35 found by Jones (1991). The
long-run income elasticity ranges between −0.45 to −0.35 and the
long-run industry elasticity ranges between 0.07 and 0.12, suggesting
that a 1% increase in income combined with a 1% increase in industrialization will lead to lower energy intensity when the impact of urbanization on energy intensity is statistically insignificant from zero.
The urbanization elasticities are presented for completeness but
caution needs to be used in interpreting the results since the estimated coefficient on the urban variable is statistically significant only in
the case of the AMG specification. Using the results from the AMG
specification, the short-run urbanization elasticity is 1.19 while the
long-run urbanization elasticity is 2.11. The results from Tables 4–6
show that when the estimated coefficient on the urbanization variable was statistically significant it was positive and slightly larger
than unity. A long-run urbanization elasticity of 2.11 implies that a
1% increase in urbanization increases energy intensity by 2.11% in
the long-run and this has serious implications for energy intensity
in developing countries since the long-run urbanization elasticity is
one order of magnitude larger than the long-run elasticities of either
income or industrialization.
7. Conclusions
While there has been some work done studying the impact of
urbanization and industrialization on energy use much less is known
about how urbanization and industrialization affect energy intensity in
58
P. Sadorsky / Energy Economics 37 (2013) 52–59
developing countries. It is expected that urbanization and industrialization will continue rising in developing countries and understanding
how urbanization and industrialization affect energy intensity is an important topic to study because reducing energy intensity is one way to
partially mitigate the impacts of climate change, peak oil and energy security issues.
This paper reports results from estimating a variety of static and dynamic panel data models of energy intensity. A dynamic model is useful
because both short-run and long-run impacts (elasticities) of income,
urbanization, and industrialization on energy intensity can be captured
in one model. One of the novel features of this paper is the use of recently
developed econometric techniques that facilitate heterogeneous parameter estimates. Results, presented for a large panel of developing
countries, show that heterogeneous parameter models yield more
favorable diagnostic results than do pooled parameter models.
The estimated coefficient on the income variable is negative and
statistically significant for each specification (homogenous/static,
heterogeneous/static, heterogeneous/dynamic). This result is important in establishing that increases in income reduce energy intensity.
Using results from the dynamic model with heterogeneous parameters, the short-run income elasticity of energy intensity varies
between − 0.57 and − 0.53 while the long-run income elasticity
varies between − 0.45 and − 0.35. When the dependent variable is
measured as energy intensity, the income variable measures only a
technique effect and not a scale effect. These results suggest that the
technique effect is similar in both the short-run and long-run.
The estimated coefficient on the industrialization variable is positive and statistically significant in most specifications. Using results
from the dynamic model with heterogeneous parameters, the
short-run industrialization elasticity of energy intensity varies between 0.05 and 0.06 while the long-run industrialization elasticity
varies between 0.07 and 0.12. These results are important in
establishing that higher industrialization increases energy intensity
in both the short-run and the long-run.
The impact of urbanization on energy intensity is mixed. The estimated coefficient on the urbanization variable is statistically significant in less than half of the specifications estimated but only
significant in one out of three dynamic models. When the estimated
coefficient on the urbanization variable is significant, the value is positive and greater than unity. The strongest evidence for urbanization
affecting energy intensity comes from heterogeneous static models,
but these models lack dynamics. Urbanization increases economic activity through a higher concentration of consumption and production
but urbanization also leads to economies of scale and provides the opportunity for increases in energy efficiency. A positive and statistically
significant coefficient on urbanization implies that the net effect of
these two impacts is to increase energy intensity.
Reducing energy intensity is often advocated as a way to at least
partially mitigate concerns about climate change, peak oil, and energy
security issues. The results of this paper show that increasing income
reduces energy intensity in developing countries. From a policy perspective this means that economic policies that increase income in
developing countries will reduce energy intensity. Empirical results
are presented showing that increases in industrialization will increase energy intensity. Thus, industrial policy aimed at speeding
up industrialization will increase energy intensity. Urbanization is
expected to increase in developing countries. The combined effect
of increasing income, industrialization, and urbanization will lead
to a fall in energy intensity as long as income growth is sufficiently large enough to offset the impact of urbanization and
industrialization.
Acknowledgments
My thanks to an anonymous reviewer for very helpful comments.
Appendix A
List of countries by income classification.
LIC
LMI
UMI
Benin
Cambodia
Congo, Dem. Rep.
Eritrea
Ethiopia
Kenya
Kyrgyz Republic
Mozambique
Nepal
Tajikistan
Tanzania
Togo
Angola
Armenia
Bolivia
Cameroon
Congo, Rep.
Cote d'Ivoire
Egypt, Arab Rep.
El Salvador
Georgia
Honduras
India
Indonesia
Moldova
Mongolia
Morocco
Nicaragua
Pakistan
Paraguay
Philippines
Senegal
Sri Lanka
Sudan
Syrian Arab Republic
Turkmenistan
Ukraine
Uzbekistan
Vietnam
Yemen, Rep.
Zambia
Albania
Algeria
Argentina
Azerbaijan
Belarus
Bosnia and Herzegovina
Botswana
Brazil
Bulgaria
Chile
China
Colombia
Costa Rica
Dominican Republic
Gabon
Iran, Islamic Rep.
Jamaica
Jordan
Kazakhstan
Latvia
Lebanon
Lithuania
Macedonia, FYR
Malaysia
Mexico
Namibia
Panama
Peru
Russian Federation
South Africa
Thailand
Tunisia
Turkey
Uruguay
Venezuela, RB
Countries grouped by World Bank classification (GNI per capita). LIC (low income,
$1005 or less), LMI (lower middle income, $1006 to $3975), UMI (upper middle income, $3976 to $12,275). http://data.worldbank.org/about/country-classifications/
country-and-lending-groups#Low_income.
References
Bernardini, O., Galli, R., 1993. Dematerialization: long term trends in the intensity of
use of materials and energy. Futures 431–448 (May).
Blanchard, O., 1992. Energy consumption and modes of industrialization: Four developing countries. Energy Policy 20, 1174–1185.
Bond, S., Eberhardt, M., 2009. Cross-section dependence in nonstationary panel
models: a novel estimator. Paper Presented at the Nordic Econometrics Conference
in Lund Sweden.
Coakley, J., Fuertes, A.-M., Smith, R.P., 2006. Unobserved heterogeneity in panel time
series models. Comput. Stat. Data Anal. 50, 2361–2380.
Cole, M., 2006. Does trade liberalization increase national energy use? Econ. Lett. 92,
108–112.
Eberhardt, M., Teal, F., 2010. Productivity analysis in global manufacturing production.
Economics Series Working Papers, 515. University of Oxford, Department of
Economics.
Eberhardt, M., Teal, F., 2011. Econometrics for grumblers: a new look at the literature
on cross-country growth empirics. J. Econ. Surv. 25, 109–155.
Eberhardt, M., Helmers, C., Strauss, H., in press. Do spillovers matter when estimating
private returns to R&D? Rev. Econ. Stat.
Galli, R., 1998. The relationship between energy intensity and income levels: forecasting long term energy demand in Asian emerging countries. Energy J. 19, 85–105.
Jones, D.W., 1989. Urbanization and energy use in economic development. Energy J. 10,
29–44.
Jones, D.W., 1991. How urbanization affects energy-use in developing countries. Energy Policy 19, 621–630.
Kapetanios, G., Pesaran, M.H., Yamagata, T., 2011. Panels with nonstationary multifactor error structures. J. Econ. 160, 326–348.
Krey, V., van O'Neill, B.C., Ruijven, B., Chaturvedi, V., Daioglou, V., Eom, J., Jiang, L.,
Nagai, Y., Pachauri, S., Ren, X., 2012. Urban and rural energy use and carbon dioxide
emissions in Asia. Energy Econ. 34 (S3), S272–S283.
P. Sadorsky / Energy Economics 37 (2013) 52–59
Liddle, B., 2004. Demographic dynamics and per capita environmental impact: using
panel regressions and household decompositions to examine population and
transport. Popul. Environ. 26, 23–39.
Madlener, R., 2011. The impact of urbanization on urban structures and energy demand in developing countries. Smart Energy Strategies Conference 2011 Zurich,
September 21–23, 2011.
Madlener, R., Sunak, Y., 2011. Impacts of urbanization on urban structures and energy
demand: what can we learn for urban energy planning and urbanization management? Sustain. Cities Soc. 1, 45–53.
Malenbaum, W., 1978. World Demand for Raw Materials in 1985 and 2000. McGrawHill, New York.
Mishra, V., Smyth, R., Sharma, S., 2009. The energy-GDP nexus: evidence from a panel
of pacific island countries. Resour. Energy Econ. 31, 210–220.
O'Neill, B.C., Ren, X., Jiang, L., Dalton, M., 2012. The effect of urbanization on energy use
in India and China in the iPETS model. Energy Econ. 34 (S3), S339–S345.
Pacione, M., 2009. Urban Geography — A Global Perspective. Routledge, New York.
Parikh, J., Shukla, V., 1995. Urbanization, energy use and greenhouse effects in economic
development — results from a crossnational study of developing countries. Glob.
Environ. Chang. 5, 87–103.
Pesaran, M.H., 1997. The role of economic theory in modelling the long run. Econ. J.
107, 178–191.
Pesaran, M.H., 2004. General diagnostic tests for cross section dependence in panels. Cambridge Working Papers in Economics, No. 0435. University of Cambridge (June).
Pesaran, M.H., 2006. Estimation and inference in large heterogeneous panels with a
multifactor error structure. Econometrica 74, 967–1012.
59
Pesaran, M.H., 2007. A simple panel unit root test in the presence of crosssection dependence. J. Appl. Econ. 22, 265–312.
Pesaran, M.H., Shin, Y., 1999. An autoregressive distributed lag modelling approach to
cointegration analysis. In: Strom, S. (Ed.), Econometrics and Economic Theory in
the 20th Century: the Ragnar-Frisch Centennial Symposium. Cambridge University
Press.
Pesaran, M.H., Smith, R.P., 1995. Estimating long-run relationships from dynamic heterogeneous panels. J. Econ. 68, 79–113.
Poumanyvong, P., Kaneko, S., 2010. Does urbanization lead to less energy use and
lower CO2 emissions? A cross-country analysis. Ecol. Econ. 70, 434–444.
Samouilidis, J.E., Mitropoulos, C.S., 1984. Energy and economic growth in industrialized
countries: the case of Greece. Energy Econ. 6, 191–201.
Schneider, F., Enste, D.H., 2000. Shadow economies: size, causes, and consequences. J.
Econ. Lit. 38, 77–114.
Stern, D.I., 2012. Modeling international trends in energy efficiency. Energy Econ. 34,
2200–2208.
United Nations Population Division, 2007. World urbanization prospects — the 2007
revision population database. Retrieved March 2012, from http://esa.un.org/
unup/.
Vlahov, D., Galea, S., 2002. Urbanization, urbanicity, and health. J. Urban Health 79,
S1–S12.
World Bank, 2011. World Development Indicators. Accessed at http://www.worldbank.
org/data/onlinedatabases/onlinedatabases.html.
York, R., 2007. Demographic trends and energy consumption in European Union Nations, 1960–2025. Soc. Sci. Res. 36, 855–872.