Anda di halaman 1dari 3

Economics Letters 117 (2012) 848850

Contents lists available at SciVerse ScienceDirect

Economics Letters
journal homepage: www.elsevier.com/locate/ecolet

Urbanization and CO2 emissions: A semi-parametric panel data analysis


Hui-Ming Zhu a, , Wan-Hai You a , Zhao-fa Zeng b
a

College of Business Administration, Hunan University, Changsha 410082, PR China

College of Finance and Statistics, Hunan University, Changsha 410082, PR China

article

info

Article history:
Received 19 June 2012
Received in revised form
15 August 2012
Accepted 3 September 2012
Available online 7 September 2012

abstract
This paper investigates the relationship between urbanization and CO2 emissions in a sample of 20
emerging countries over the period 19922008 using the semi-parametric panel data model with fixed
effects, proposed by Baltagi and Li (2002). We find little evidence in support of an inverted-U curve,
and thus the Kuznets hypothesis is not confirmed by our analysis. Our findings shed new light on the
urbanization-CO2 emissions nexus.
2012 Elsevier B.V. All rights reserved.

JEL classification:
C14
O5
Keywords:
CO2 emissions
Urbanization
Emerging countries
Semi-parametric regression
Panel data

1. Introduction
The relationship between urbanization and CO2 emissions
has been extensively investigated in recent years. The empirical
results, however, are mixed. For example, Cole and Neumayer
(2004) and Liddle and Lung (2010) demonstrate a positive
correlation between urbanization and CO2 emissions, while Fan
et al. (2006) find a negative correlation between urbanization and
CO2 emissions in developing countries. Poumanyvong and Kaneko
(2010) argue that the assumption that the relationship between
urbanization and CO2 emissions is homogenous for all countries
may be unreasonable. They examine the effects of urbanization on
CO2 emissions for low-, middle-, and high-income group, and find
that while a positive relationship exists for all income groups, it is
most prominent in the middle-income group.
The vast majority of existing literature has assumed that
there exists a linear relationship between urbanization and CO2
emissions. Ehrhardt-Martinez et al. (2002) argue that urbanization
is a good proxy for modernization, and thus the relationship
between urbanization and CO2 emissions may vary across different
stages of development. Martinez-Zarzoso and Maruotti (2011)

Corresponding author. Tel.: +86 731 88823670; fax: +86 731 88823670.
E-mail addresses: zhuhuiming1999@yahoo.com.cn, zhuhuiming@hnu.edu.cn
(H.-M. Zhu).
0165-1765/$ see front matter 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.econlet.2012.09.001

find an inverted U-shaped relationship between urbanization and


CO2 emissions. To explore a clear relational structure between
urbanization and CO2 emissions, we re-investigate this topic by
employing the semi-parametric panel fixed effects regression
model developed by Baltagi and Li (2002). Within this framework,
a no priori parametric functional form is assumed for modeling
the relationship between urbanization and CO2 emissions, which
allows us to obtain the true functional form of the urbanizationCO2 emissions nexus. Our empirical results show that the
relationship between urbanization and CO2 emissions is not a
simple inverted U-shape, and thus our findings are inconsistent
with the results obtained using the conventional polynomial
functional form specification.
The rest of the paper is organized as follows. Section 2 describes
the model framework and methods. Section 3 discusses the results,
and Section 4 offers concluding remarks.
2. Theoretical framework and econometric methods
The IPAT model (I = PAT) proposed by Ehrlich and Holdren
(1971) has been widely used to assess the impact of population
size, affluence, and other factors on the environment. Although
the IPAT model is a very useful framework, it does have some
limitations (Liddle and Lung, 2010). To address those limitations,
Dietz and Rosa (1997) formulate a stochastic version of the IPAT
model, named STIRPAT (Stochastic Impacts by Regression on

H.-M. Zhu et al. / Economics Letters 117 (2012) 848850

Population, Affluence, and Technology). The model specification


for a single year is as follows
Ii = aPib Aci Tid i ,

(1)

where I, P, A, and T denote environmental impact, population,


affluence, and technology; a, b, c, and d are the estimated
parameters; and denotes the disturbance term.
The STIRPAT model can be used to assess the effects on
environmental impact not only of the core components, population
size and affluence, but also of other factors, such as modernization
(York et al., 2003). To examine the impact of urbanization on
the environment, we have estimated an extended version of the
STIRPAT model. Thus, the following augmented model is estimated
in the environmental Kuznets curve (EKC) hypothesis framework
ln(CO2it ) = i + 1 ln(Pit ) + 2 ln(Ait ) + 3 ln(EIit )

+ 4 ln(URBit ) + 5 ln(URBit )2 + it ,

(2)

where CO2 is the amount of CO2 emissions (in tons per capita) of
country i at year t; P denotes the population size; A is per capita
GDP; EI is energy use; URB is the urbanization level and i is the
country fixed effect. We also include in our empirical models a
time-period specific effect, which is a proxy for the variables that
are common across countries but vary over time. This effect could
be interpreted as the effect of technical progress over time (Stern,
2002).
Yatchew (1998) believes that most economic theories do not
identify a specific functional form in a model for the relationship
between the dependent variable and the independent variables. To
avoid possible model misspecification of the parametric domain,
a more flexible way of modeling the relationship between
urbanization and CO2 emissions is to estimate a semi-parametric
model. In addition, we can obtain a more accurate inference of
model parameters in the panel data framework (Hsiao, 2007).
Thus, we model the relationship between urbanization and CO2
emissions using the semi-parametric panel data model with fixed
effects, developed by Baltagi and Li (2002). The model is given by:
ln(CO2it ) = i + 1 ln(Pit ) + 2 ln(Ait )

+ 3 ln(EIit ) + f (ln(URBit )) + it ,

(3)

where the functional form f (.) is unspecified. The unobserved


heterogeneity effects (i ) can be eliminated by taking a first
difference. To estimate the first difference model consistently,1
Baltagi and Li (2002) propose approximating [f (ln(URBit )) f (ln
(URBit 1 ))] by the following series differences
pk (ln(URBit ), ln(URBit 1 ))

= [pk (ln(URBit )) pk (ln(URBit 1 ))],

(4)

where p (ln(URB)) are the first k terms of a sequence of function2


(p1 (ln(URB)), p2 (ln(URB)), . . .). In our empirical analysis, we use
a B-spline regression model with k = 4 (Desbordes and Verardi,
2012; Newson, 2001).
k

3. Data and empirical results


The panel data set consists of cross-country observations
covering the period 19922008 for 20 emerging countries.3 This
sample period is selected based on data availability. All of these

1 For a more detailed discussion, see Baltagi and Li (2002).


2 Desbordes and Verardi (2012) point out that a typical example of pk series is a
spline.
3 Emerging countries are identified as those 20 countries (except Taiwan) listed
in the MSCI: http://www.msci.com/products/indices/country_and_regional/em/.

849

Table 1
Summary statistics (19922008, observations = 340).
Variables

Mean

lnCO2
lnP
lnA
lnEI
lnURB
Countries
N = 20

1.1695
0.8398
0.2548
2.7037
17.9706
1.2691
16.1219
21.0044
8.8198
0.6874
7.1209
10.1471
5.2237
0.4483
4.1947
6.34298
4.0575
0.3245
3.2558
4.4823
Brazil, Chile, China, Colombia, Czech Republic,
Egypt, Hungary, India, Indonesia, Korea, Malaysia,
Mexico, Morocco, Peru, Philippines, Poland,
Russia, South Africa, Thailand, Turkey

Standard Deviation

Min

Max

Table 2
Estimation results for CO2 emission models.a
Variables

FE

Constant
lnP
lnA
lnEI
lnURB
lnURB2
Country dummies
Year dummies
R2

34.43067b (4.5871)
0.78642b (0.1941)
1.1216b (0.0682)
0.9706b (0.0661)
3.5523c (1.3675)
0.47198c (0.1934)
Yes
Yes
0.8915

Semi-parametric
0.9488b (0.2465)
1.1995b (0.0906)
0.9347b (0.0769)

Yes
Yes
0.6162

a
Robust standard errors are reported in parentheses. The time fixed effect is
included but not reported.
b
denotes the rejection of the null hypothesis at the 1% significance level.
c
denotes the rejection of the null hypothesis at the 5% significance level.

data are collected from World Bank World Development Indicators


(WDI). The gross domestic product is expressed in constant PPP
(2005US$), the urbanization level is measured by the percentage
of total population living in urban areas, and energy use (kg of oil
equivalent) is measured in per $1,000 GDP (constant 2005 PPP). All
of the variables are converted into natural logarithms. A summary
of the data, as well as a complete list of the sample countries, is
shown in Table 1.
The empirical results are reported in Table 2.4 The Hausman test
result (p = 0.0000) suggests the use of the fixed effects model.5
Column 1 reports the results of the fixed effects estimator in the
EKC hypothesis framework (Eq. (2)). The results indicate that all
the variables included are statistically significant at the 1% or 5%
significance level and that they show the expected signs. The results confirm the EKC hypothesis. Column 2 reports the coefficient
estimation of the control variables in the semi-parametric panel
data model. The coefficient estimation of all control variables is
close to the parametric model, and it is statistically significant at
the 1% significance level.
Fig. 1 illustrates the semi-parametric estimation of f (), and
shaded areas correspond to 95% confidence intervals. The points in
the graph are partial residuals for CO2 emissions, centered on the
mean, and the CO2 emissions level has been adjusted for the effects
of the other independent variables.6 The result, in accordance with
previous studies, shows that urbanization has a nonlinear effect
on CO2 emissions. However, we find little evidence to support the
EKC hypothesis. The inverted-U hypothesis is confirmed only when
lnURB reaches 3.5; when lnURB is below this level, CO2 emissions

4 The quadratic term of income is not included in our model because we find that
the relationship between income and CO2 emissions is monotone. More specifically,
we model the relationship between income and CO2 emissions in a semi-parametric
framework. We observe that the curve is increasing for all values of income (See
Fig. 2).
5 For an excellent textbook treatment of the panel data model, see Baltagi (2008).
6 In fact, the curve f () can be estimated by the following equation: u
it
ln(CO2it ) a i 1 ln(Pit ) 2 ln(Ait ) 3 ln(EIit ) = f (ln(URBit )) + it .

850

H.-M. Zhu et al. / Economics Letters 117 (2012) 848850

panel data model with fixed effects proposed by Baltagi and Li


(2002). We find a nonlinear relationship between urbanization
and CO2 emissions. However, unlike previous studies, we find
little evidence to support an inverted-U relationship between
urbanization and CO2 emissions.
Acknowledgments

Fig. 1. Partial fits of the relationship between urbanization and CO2 emissions.
Note: The points in the graph are partial residuals for CO2 emissions in the
semi-parametric model, and the partial residuals are centered on the mean. The
green dash line is the curve generated by the parametric model (adjusted for the
effects of the other independent variables). The maroon curve represents the semiparametric estimation of f (). Shaded areas correspond to 95% confidence intervals.

Fig. 2. Partial fits of the relationship between CO2 emissions and income. (Nonparametric variable = Income, including control variables in Eq. (2).)

decrease with urbanization. For comparison purposes, we also


provide the curve generated by the parametric model (FE model
in Table 2) in Fig. 1 (the green dash line). The result of an inverted
U-shaped relationship between urbanization and CO2 emissions
is inconsistent with the result obtained by the semi-parametric
analysis. Thus, our findings shed new light on the urbanization
levelCO2 emissions nexus.
4. Conclusion
This paper examines the urbanization-CO2 emissions nexus
under the STIRPAT framework, using the semi-parametric

The authors wish to thank the editors and an anonymous


referee for very constructive comments. We are also indebted
to Dr. Vincenzo Verardi and Dr. Franois Libois for providing
the program code and for helpful comments. This research is
supported by the National Natural Science Foundation of China
under grants No. 71171075, No. 71221001, and No. 71031004.
References
Baltagi, B.H., Li, D., 2002. Series estimation of partially linear panel data models with
fixed effects. Annals of Economics and Finance 3, 103116.
Baltagi, B.H., 2008. Econometric Analysis of Panel Data, 4th Ed.. John Wiley & Sons,
Chichester.
Cole, M.A., Neumayer, E., 2004. Examing the impact of demographic factors on air
pollution. Pollution and Environment 26, 521.
Desbordes, R., Verardi, V., 2012. Refitting the Kuznets curve. Economics Letters 116,
258261.
Dietz, T., Rosa, E.A., 1997. Effects of population and affluence on CO2 emissions. In:
Proceedings of the National Academy of Sciences USA 94, pp. 175179.
Ehrhardt-Martinez, K., Crenshaw, E.M., Jenkins, J.C., 2002. Deforestation and the
environmental Kuznets curve: a cross-national investigation of intervening
mechanism. Social Science Quarterly 83, 226243.
Ehrlich, P.R., Holdren, J.P., 1971. Impact of population growth. Science 171,
12121217.
Fan, Y., Liu, L.-C., Wu, G., Wei, Y.-M., 2006. Analyzing impact factors of CO2
emissions using the STIRPAT model. Environmental Impact Assessment Review
26, 377395.
Hsiao, C., 2007. Panel data analysis-advantages and challenges. Mathematics and
Statistics 16, 122.
Liddle, B., Lung, S., 2010. Age-structure, urbanization, and climate change in
developing countries: Revisiting STIRPAT for disaggregated population and
consumption-related environmental impacts. Population Environment 31,
317343.
Martinez-Zarzoso, I., Maruotti,, 2011. The impact of urbanization on CO2 emissions:
Evidence from developing countries. Ecological Economics 70, 13441353.
Newson, R., 2001. B-splines and splines parameterized by their values at reference
points on the X-axis. Stata Technical Bulletin 10, 2027.
Poumanyvong, P., Kaneko, S., 2010. Does urbanization lead to less energy use
and lower CO2 emissions? A cross-country analysis. Ecological Economics 70,
434444.
Stern, D.I., 2002. Explaining changes in global sulphur emissions: an econometric
decomposition approach. Ecological Economics 42, 201220.
Yatchew, A., 1998. Nonparametric regression techniques in economics. Journal of
Economic Literature 36, 669721.
York, R., Rosa, E.A., Dieta, T., 2003. STIRPAT, IPAT and ImPACT: analytic tools for
unpacking the driving forces of environmental impacts. Ecological Economics
46, 351365.

Anda mungkin juga menyukai