TESTING FOR AN ENVIRONMENTAL KUZNETS... 3Revista de Análisis Económico, Vol. 18, Nº 1, pp. 3-26 (Junio 2003)
INMACULADA MARTINEZ-ZARZOSO*
AURELIA BENGOCHEA-MORANCHO
Universitat Jaume I, Spain
TESTING FOR AN ENVIRONMENTAL KUZNETS CURVE
IN LATIN-AMERICAN COUNTRIES
* Department d’Economia. E-mail: martinei@eco.uji.es. The author acknowledges the support and
collaboration of Proyecto Bancaja-Castellon P-1B92002-11, Proyecto BEC 2002-02083 and SEC
2002-03651.
Abstract
This paper presents empirical estimates of Environmental Kuznets Curves
for a panel of Latin-American countries over the period 1975-1998. It
uses a new econometric technique that allows for more flexible assump-
tions in a panel data framework with a large time dimension. Unlike
most previous studies we test for slope heterogeneity of the income
coefficient in the search of a common empirical relation between car-
bon dioxide emissions and income. Our results point to the existence of
some heterogeneity among countries, but with specific patterns for those
sharing certain characteristics.
I. Introduction
According to some researchers, there is an inverse U-shaped pattern between
pollution and economic growth (see the survey presented by Stern, 1996). This
regularity implies that pollution increases with income until a “turning point” in
which pollution begins to decrease while income is still rising. Because of its
similarity to the relationship between income inequality and the level of income
(Kuznets, 1955), the inverted U-shaped curve that relates pollution with income is
called the Environmental Kuznets Curve (EKC).
4 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
Since the first EKC study, presented in 1991 by Grossman and Krueger, the
research in this field has grown very fast but empirical evidence does not support
the EKC hypothesis in a general way. Results are strongly dependent on the
pollutant indicators, the functional form, the econometric procedure and the ex-
planatory variables included in the regressions, the time period considered and the
countries included in the sample.
Over the last two decades there have also been extensive theoretical research
on the EKC hypothesis. Technological progress, structural changes, enforcement
of environmental regulations are some of the explanatory factors researchers have
pointed out to compensate the negative impact that a larger scale of economic
activity produces on the environment. Bulte and van Soest (2001) show an alter-
native explanation to EKC relative to environmental degradation in developing
countries based on imperfect markets for factors and commodities1.
Research on the EKC hypothesis is far from being an academic entertainment
since the existence or absence of such a curve has important policy implications.
If the EKC were a generalized phenomenon, environmental degradation will au-
tomatically fall in the long run as income becomes sufficiently high. However, if
the proposition does not hold, public intervention would be necessary to curb
pollution and make sustainable development a reality2.
The Earth Summit held in Rio de Janeiro in 1992 and the 1997 Kyoto Sum-
mit have called the international attention upon environment, particularly upon
the heating of the planet as a consequence of the greenhouse effect.
Since the main gas producing the greenhouse effect is carbon dioxide, we
focus on this pollutant to investigate the existence of an EKC in Latin American
countries. Previous research dealing with this topic is quite scarce (see Table 1).
Studies concerning developing countries have mainly focused on sulphur emis-
sions (de Bruyn et al., 1997; Panayotou, 1997; Kaufman et al., 1998) and also on
deforestation (Koop and Tole, 1999; Bhattarai and Hammig, 2001; Halkos and
Tsionas, 20013). However, there is an important debate about the role that should
be played by developing countries in curbing CO2 emissions. The Kyoto Protocol
contains a specific commitment taken by industrialised and transition economies
(the so-called Annex B countries)4 to reduce their emissions over the period 2008-
2012 down to the level attained in 1990, but no commitment exists for develop-
ing countries.
In Latin America and the Caribbean total CO2 emissions generated by the
energy sector have been steadily raising since 1970 as shown by the CEPAL
(2002). When emissions per unit of GDP are considered, a raising trend is ob-
served, but when emission intensity and per capita income are considered, the
path is not clear (pages 290 and 291, op. cit.). Anderson and Cavendish (2001)
present a dynamic simulation model to develop scenarios for SO2, and CO2 abate-
ment in developing countries. Their results show that, without a pollution abate-
ment policy, in a developing country with an initial per capita income of $ 2,500
and growth rate of 4%, emissions will rise to over five times today levels in the
present century.
TESTING FOR AN ENVIRONMENTAL KUZNETS... 5
The main aim of this paper is to investigate the relationship between eco-
nomic growth and CO2 emissions in Latin American countries in order to com-
pare the observed patterns with those followed by Annex B countries.
We use a fresh methodology, the pooled mean group estimator, based on
Pesaran et al. (1999) that allows for slope heterogeneity in the short run imposing
restrictions only in the long run and testing for their validity. To our knowledge,
this methodology has only been applied to EKCs for sulphur emissions in Perman
and Stern (1999) and to income growth equations in Bassanini and Scarpeta (2001).
There are 19 countries5 in the sample (only Guatemala is excluded due to missing
data), including some Caribbean countries. We consider the evolution of income
and CO2 emissions from 1975 to 1998.
The paper is structured as follows. Section II summarises previous research
in CO2 EKCs. Section III presents the econometric approach, Section IV shows
the empirical findings and, finally, Section V concludes.
II. Exploring the Evidence
Since the early 90’s a number of authors have estimated EKCs for various
indicators of environmental degradation. In this section we briefly review, in a
chronological order, the studies which are more relevant for our analysis. Table 1
summarises some results of previous EKC studies on CO2 emissions conducted
along the last decade.
Shafik and Bandyopadhyay (1992) estimated EKCs for ten different environ-
mental indicators from 1961 until 1986. They found that sulphur oxides conform
to the EKC hypothesis with a turning point at $ 5,000 per capita income, but they
did not find evidence for carbon emissions per capita which increase unambigu-
ously with rising income.
Holtz-Eakin and Selden (1995) formulated a quadratic EKC function for car-
bon dioxide emissions, with an estimated turning point at $ 35,428 per capita
income, and a log-quadratic function, which showed a very high turning point
($ 8 million). Although they concluded that there is a diminishing marginal pro-
pensity to emit CO2 as economies develop. They predicted that emissions growth
will continue because output and population will grow more rapidly in lower-
income nations that have the highest marginal propensity to emit. Agras (1995)
found a turning point for SO2 at $ 6.654 per capita income but for CO2 he ques-
tions the existence of an EKC. However, Sengupta (1996) found evidence for this
gas as well as Schmalensee et al. (1998).
Moomaw and Unruh (1997) analysed per capita CO2 emissions in a set of
countries over the period 1950 to 1992. They found a great heterogeneity among
them. OECD member states showed a discontinuous transition in which the CO2/
GDP relation changed from a strong positive covariance to a negative or weakly
correlated relation. Another subset of countries, dominated by centrally planned
economies and some developing countries, showed a positive correlation. Finally,
6 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
TA
BL
E 
1
CO
2 
EK
C 
ST
U
D
IE
S 
IN
 A
 C
H
RO
N
O
LO
G
IC
A
L 
O
RD
ER
No
te
s:
a:
 I
nt
er
na
tio
na
l E
ne
rg
y 
A
ge
nc
y:
 G
re
en
ho
us
e 
G
as
 E
m
iss
io
ns
: T
he
 E
ne
rg
y 
D
im
en
sio
n 
(P
ari
s, 
OE
CD
, 1
99
1),
 b
: O
ak
 R
idg
e N
ati
on
al 
La
bo
rat
or
y,
 c
: C
ar
bo
n 
D
io
xi
de
In
fo
rm
at
io
n 
A
na
ly
sis
 C
en
tre
, e
: W
or
ld
 R
es
ou
rc
es
 I
ns
tit
ut
e.
Co
un
tri
es
11
8-
15
3
10
8
13
7
16
 D
ev
el
op
ed
 +
D
ev
el
op
in
g
7 
W
or
ld
R
eg
io
ns
16
 D
ev
el
op
ed
D
ev
el
op
ed
 +
D
ev
el
op
in
g
14
1
34 11
0
17
 D
ev
el
op
ed
11
8-
15
3
Sp
ai
n
16
0
EU
24
-O
EC
D
22
-O
EC
D
A
ut
or
s
Sh
af
ik
 a
nd
 B
an
dy
op
ad
hy
ay
(19
92
)
H
ol
tz
-E
ak
in
 a
nd
 S
el
de
n 
(19
95
)
Tu
ck
er
 (1
99
5)
Se
ng
up
ta
 (1
99
6)
Co
le
, R
ay
ne
r 
an
d 
Ba
te
s 
(19
97
)
M
oo
m
aw
 a
nd
 U
nr
uh
 (1
99
7)
R
ob
er
ts 
an
d 
G
rim
es
 (1
99
7)
Sc
hm
al
en
se
e,
 S
to
ke
r 
an
d 
Ju
ds
on
(19
98
)
A
gr
as
 a
nd
 C
ha
pm
an
 (1
99
9)
G
al
eo
tti
 a
nd
 L
an
za
 (1
99
9)
Pa
na
yo
to
u,
 P
et
er
so
n 
an
d 
Sa
ch
s
(20
00
)
H
ee
rin
k 
et
 a
l. 
(20
01
)
R
oc
a 
et
 a
l. 
(20
01
)
B
ai
oc
ch
i a
nd
 d
i F
al
co
 (2
00
1)
B
en
go
ch
ea
 e
t 
al
.
 
(20
01
)
D
ijk
gra
af 
an
d V
oll
eb
erg
h (
20
01
)
M
ar
tín
ez
-Z
ar
zo
so
 e
t 
al
.
 
(20
02
)
Tu
rn
in
g 
Po
in
ts
$7
M
ill
ion
$3
5,4
28
(le
ve
l)
$8
 M
ill
. (
log
s)
D
ec
re
as
in
g 
ov
er
tim
e
$8
,74
0
$2
5,1
00
(le
ve
ls)
$6
2,7
00
 (l
og
s)
$1
2,8
13
$8
,00
0-
$1
0,0
00
W
ith
in
 s
am
pl
e
$1
3,6
30
$1
5,0
73
-$
21
,75
7
$2
9,7
32
-$
40
,90
6
(19
50
-19
90
)
$6
8,8
71
Y
2  
n
o
n
 s
ig
n.
Y
2  
n
o
n
 s
ig
n.
$2
4,4
27
-$
73
,17
0
$2
0,6
47
$4
,91
4-
$1
8,3
64
PP
P
Y
es
Y
es
($1
,98
6)
Y
es
($1
,98
5)
N
o
($1
,98
5)
Y
es
Y
es
($1
,98
5)
Y
es
($1
,99
7)
Y
es
($1
,99
0)
Y
es
Y
es N
o
($1
,98
6)
Y
es
Y
es
($1
,99
3)
N
o
($1
,99
0)
Y
es
($1
,99
3)
D
at
a 
so
ur
ce
 f
or
 C
O
2
M
ar
la
nd
 (1
98
9)
O
RN
Lb
W
RI
 (1
99
4)
O
RN
Lb
M
ar
la
nd
  
et
 a
l. 
(19
94
)
W
or
ld
 B
an
k 
(19
92
)
O
RN
Lb
O
RN
Lb
IE
A
a  
an
d 
O
RN
Lb
IE
A
a
CD
IA
C 
c 
(19
97
)
M
ar
la
nd
 (1
98
9)
IE
A
a
W
or
ld
 R
es
ou
rc
es
In
st
itu
te
O
EC
D
 E
nv
iro
nm
en
ta
l
D
at
a
O
EC
D
 2
00
0
IE
A
 (1
99
1)a
W
or
ld
 D
ev
el
op
m
en
t
In
di
ca
to
rs
 2
00
1
Ti
m
e 
pe
rio
d
19
61
-8
6
19
51
-8
6
19
71
-9
1
19
60
-9
2
19
50
-9
2
19
62
-9
1
19
50
-1
99
0
19
71
-8
9
19
71
-9
6
18
70
-1
99
4
19
85
19
73
-9
6
19
80
-9
5
19
60
-9
7
19
75
-9
8
Es
tim
at
io
n 
te
ch
ni
qu
e
Fi
xe
d 
Ef
fe
ct
s, 
Ra
nd
om
Ef
fe
ct
s
Tw
o 
w
ay
s 
Fi
xe
d 
Ef
fe
ct
s
Y
ea
rly
 C
ro
ss
-s
ec
tio
na
l
an
al
ys
is.
 F
irs
t D
iff
er
.
Fi
xe
d 
Ef
fe
ct
s
G
en
er
al
iz
ed
 L
ea
st
Sq
ua
re
s
Fi
xe
d 
Ef
fe
ct
s
Cr
os
s-
se
ct
io
n 
an
al
ys
is
Tw
o 
w
ay
s 
Fi
xe
d 
Ef
fe
ct
s
A
ut
or
re
gr
es
siv
e-
D
ist
rib
ut
ed
La
g 
w
ith
 F
ix
ed
 E
ffe
ct
s
Le
as
t 
Sq
ua
re
s 
D
um
m
y
V
ar
ia
bl
e
Fe
as
ib
le
 G
en
er
al
iz
ed
Le
as
t 
Sq
ua
re
s
G
en
er
al
ise
d 
M
et
ho
d 
of
M
om
en
ts
Ti
m
e 
se
rie
s,
co
in
te
gr
at
io
n
N
on
pa
ra
m
et
ric
 m
et
ho
d
Fi
xe
d 
Ef
fe
ct
s, 
Ra
nd
om
Ef
fe
ct
s, 
 I
ns
tru
m
en
ta
l
V
ar
ia
bl
es
, F
irs
t D
iff
er
.
Fi
xe
d 
Ef
fe
ct
s, 
Se
em
ly
U
nr
el
at
ed
 R
eg
re
ss
io
n
Po
ol
ed
 M
ea
n 
G
ro
up
Fu
nc
tio
na
l f
or
m
Li
ne
ar
, Q
ua
dra
tic
 an
d
Cu
bi
c 
(lo
gs
)
Qu
ad
rat
ic 
(le
ve
ls 
an
d
lo
gs
)
Qu
ad
rat
ic
Qu
ad
rat
ic
Li
ne
ar
, Q
ua
dra
tic
(le
ve
ls 
an
d 
log
s)
St
ru
ct
ur
al
 T
ra
ns
iti
on
M
od
el
, C
ub
ic
 f
or
m
Qu
ad
rat
ic
Sp
lin
e 
Fu
nc
tio
n
Qu
ad
rat
ic 
(lo
gs
)
N
on
 li
ne
ar
 G
am
m
a
an
d 
W
ei
bu
ll
Qu
ad
rat
ic
Qu
ad
rat
ic 
(lo
gs
)
Li
ne
ar
, S
qu
ar
ed
 a
nd
Cu
bi
c 
(lo
gs
)
Lo
ca
l p
ol
yn
om
ia
l
Li
ne
ar
, q
ua
dr
at
ic
, c
ub
ic
Li
ne
ar
, q
ua
dr
at
ic
, c
ub
ic
.
Sl
op
e 
he
te
ro
ge
ne
ity
Li
ne
ar
, q
ua
dr
at
ic
, c
ub
ic
.
Sl
op
e 
he
te
ro
ge
ne
ity
EK
C
N
o
Y
es
In
 1
1 
ye
ar
s
Y
es
Y
es
N
 s
ha
pe
d
Y
es
, a
fte
r 
th
e 
70
s
Y
es N
o
Y
es
Y
es
 f
or
 D
ev
el
op
ed
Y
es N
o
N
o
Fo
r 
so
m
e
co
u
n
tr
ie
s
Y
es
 5
 r
ic
h
co
u
n
tr
ie
s
N
 s
ha
pe
d
A
dd
iti
on
al
 V
ar
ia
bl
es
Y
es
 (M
ark
et 
pre
mi
um
,
do
lla
r 
in
de
x)
N
o
N
o
N
o
Y
es
 (T
rad
e, 
po
p.d
.,
te
ch
)
N
o
N
o
N
o
Y
es
 (P
ric
e, 
tra
de
 v
ar.
)
N
o
Y
es
 (T
rad
e, 
K,
 p
op
. d
.)
Y
es
 (I
ne
qu
ali
ty)
Y
es
 (E
ne
rgy
 p
ric
es)
N
o
N
o
N
o
N
o
TESTING FOR AN ENVIRONMENTAL KUZNETS... 7
a third group described chaotic behaviour between CO2 emissions and income.
Thus, the authors conclude that neither the quadratic nor the cubic functional
formulation of the EKC hypothesis can provide a reliable indication of future
behaviour. Moreover, even in cases where transition is observed (an EKC exists),
decreasing of CO2 emissions does not appear to correlate with specific income
levels but with specific points in time in response to exogenous shocks influenc-
ing these economies as the petroleum crisis in 70s. In the same line, Roberts and
Grimes (1997), in their work about carbon dioxide emissions intensity, argued
that the inverted U curve reached statistical significance only after 1970. How-
ever, they consider that this fact is not the result of countries passing through
stages of development, but of efficiency improvements in a small number of
wealthy nations combined with worse performance in poor and middle-income
countries.
De Bruyn et al. (1998) considered three pollutants (CO2, NOx, and SO2) in
four countries (the Netherlands, the UK, the USA and the former West Germany).
The authors found that emissions were positively correlated with income but it is
possible to abate them because of technological progress and structural change.
Their conclusion is that emissions behaviour corresponding to EKC hypothesis
could be found, but estimations from panel data does not seem to be able to
capture all the dynamic processes involved, thus one can not accept as a general
rule that economic growth improves environmental quality.
Agras and Chapman (1999) test two EKCs using both per capita energy con-
sumption and per capita CO2 emissions to represent environmental degradation in
34 countries during the period 1971-1989. They observe that energy per capita
consumption decreases as gasoline prices increase, so they do not find significant
evidence for the existence of an EKC within the range of current incomes for
energy in the presence of price and trade variables. Despite the fact that they
report a $ 13,630 turning point for CO2 emissions, they conclude that, since en-
ergy use at all income levels is price elastic,  rising levels of GDP in the long run
can increase energy use and, therefore, Governments need to undertake policies
now to start to reduce levels of pollution and CO2 emissions.
Galeotti and Lanza (1999) considered 110 countries, 30 belonging to Annex
B and 80 coming from non Annex B. Despite their analysis conform the EKC
hypothesis, they forecast that future global emission between 2000 and 2020 will
rise as a result of the faster growth rate exhibited by developing countries.
Panayotou et al. (2000) also found a CO2 EKC for 17 developed countries.
Heerink et al. (2001) analyse the relationship between environmental degra-
dation, income and income inequality in a set of countries. They use several
environmental indicators and the EKC hypothesis holds for some of them. As far
as CO2  emissions are concerned, the relationship with income is positive and non
linear and income inequality is found to have a significant negative impact on the
level of emissions.
In a recent paper, Roca et al. (2001) analyse trends of annual emission flux
of six atmospheric pollutants in Spain. Only SO2 emissions can support the EKC
8 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
hypothesis. Looking at the evolution of per capita CO2 emissions, their work
confirms the Moomaw and Unruh’s theory. They find three stages corresponding
to the adjustment to the new situation of sharp increases in energy prices: strong
emissions growth until the end of the seventies, a subsequent relative emissions
stabilization during the 1980s and, finally, an increase in emission in the last
decade. Baiocchi and di Falco (2001) neither find an EKC for carbon dioxide.
Bengochea et al. (2001) find evidence only for a few countries in the Euro-
pean Union while for a 22-Annex B sample. On the other hand, Martínez-Zarzoso
et al. (2002) conclude that a cubic specification conforms better with the data.
Dijkgraaf and Vollebergh (2001) use panel data methodology to analyse CO2
emissions in 24 OCDE countries during the period 1960-1997. The EKC hypoth-
esis is only confirmed for eleven countries. The authors reject model specifica-
tions that feature homogeneity assumptions across countries. They calculate turn-
ing points with panel and time-series estimates. The unweighted mean turning
point in the panel general model is $ 20,647. When time-series estimates are used,
turning points vary among countries from $ 12,505 to $ 31,407.
III. The Econometric Approach
We estimate a standard EKC equation on the basis of annual data using pooled
cross-country time series. We assume that the long-run EKC function is
ln ln ln ln
, , ... , , , ..., .
co yh yh yh t
i N t T
it i i it i it i it i it= + + ( ) + ( ) + +
= =
? ? ? ? ? µ0 1 2 2 3 3 1
1 2 1 2               
(1)
where coit = CO2 emissions per capita, yhit = GDP per capita in dollars at 1993
PPP, ?0i is a country-specific intercept, t is a time trend and µit is an error term.
The two additional terms are the natural log of GDP per capita squared and cubed.
The cubed term has only been included in a few of the models tested in the
literature for CO2 emissions (Shafik and Bandyopadhyay, 1992; Moomaw and
Unruh, 1997; Roca et al, 2001; Bengochea et al., 2001; Dijkgraaf and Vollebergh,
2001). Equation (1) will be tested with and without this term for comparative
purposes.
We will assume that all these variables are I(1) and cointegrated for indi-
vidual countries, making the error term an I(0) process for all i. These assump-
tions are based on the test results shown in Appendix 3. According to the unit
root test results (Table A.1), the null of unit root cannot be rejected for the vari-
ables in levels for all the countries, whereas the first differences of the variables
are stationary (the null of unit root is rejected). Therefore, the univariate test
statistics strongly support the view that income and emissions are I(1) processes.
TESTING FOR AN ENVIRONMENTAL KUZNETS... 9
Concerning cointegration (Table A.2), we find clear evidence in favour of a sta-
tistically significant relationship for most countries.
Taking the maximum lag equal to one, the ARDL(1,1,1,1) equation is given
by
ln ln ln ln ln
ln ln ln
, ,
, ,
co yh yh yh yh
yh yh t co
it i i it i i t i it i i t
i it i i t i i t it
= + + + ( ) + ( )
+ ( ) + ( ) + + +
? ?
? ?
? ? ? ? ?
? ? ? ? ?
10 11 1 20
2
21 1
2
30
3
31 1
3
10 1
(2)
The error correction equation is
?
? ? ?
ln ln ln ln
ln ln ln
,
, , ,
co co yh yh yh
yh yh yh
it i i t i i it i it i it
i i t i i t i i t it
= ? ? ? ( ) ? ( )[ ]
? ? ( ) ? ( ) +
?
? ? ?
? ? ? ? ?
? ? ? ?
1 0 1 2
2
3
3
11 1 21 1
2
31 1
3 (3)
were:
?
?
? ?
? ?
? ?
? ?
? ?
? ?
?
? ?
0 1
10 11
2
20 21
3
30 31
1 1 1 1
1
i
i
i
i
i i
i
i
i i
i
i
i i
i
i i
=
?
=
+
?
=
+
?
=
+
?
= ? ?
, , , ,
( )
The empirical analysis of the EKC model specified above involves a system
of N*T equations that can be examined in several ways. The approach chosen
depends on part on the size of N and T and the quality of data across these two
dimensions. The main econometric approaches used in the literature have been
based on cross-section regressions and different forms of pooled cross-section
time-series regressions.
Authors who have focused their research on a small number of countries,
such as those in the OECD area, have often exploited the time dimension of the
data (Hilton and Levinson, 1998; Roca et al., 2001) or have used pooled cross-
country and time-series data (Suri and Chapman, 1998). There are also some
examples of EKC regressions based on cross-section data (Tucker, 1995; Roberts
and Grimes, 1997).
Most recently, researchers have used techniques based on panel data method-
ology as in Cole et al. (1997), Stern et al. (1998), de Bruyn et al. (1998) and
Dijkgraaf and Vollebergh (2001). The main advantage of these aforementioned
techniques for the analysis of EKC equations is that the country-specific effects
can be controlled for by using static fixed-effect (SFE) or dynamic fixed-effect
estimators (DFE). The SFE or DFE estimators generally impose homogeneity of
all slope coefficients, allowing only the intercepts to vary across countries. DFE
imposes (N–1)(2k + 2) restrictions on the unrestricted model in equation (3): k
10 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
long-run coefficients, k short-run coefficients and the convergence coefficient and
the common variance. So, the validity of DFE, in particular, depends critically on
the assumptions of common estimated parameters, that in turn requires both com-
mon income elasticity and common EKC patterns across countries. Since the
evolution of CO2 emissions and income differ across countries, these assumptions
are difficult to reconcile with observed emissions patterns across countries. Then,
as Pesaran and Smith (1995) pointed out, under slope heterogeneity, estimated
coefficients will be affected by an heterogeneity bias.
At the other extreme of the SFE and DFE estimators we find the mean-group
approach (MG) that consists of estimating separate regressions for each country
and calculating averages of the country-specific coefficients. There are N(2k + 3)
parameters to be estimated. Each equation has 2k coefficients on the exogenous
regressors, an intercept, a coefficient on the lagged dependent variable and a
variance. The small-sample downward bias in the coefficient of the lagged depen-
dent variable remains. Furthermore, although this estimator is still consistent, it is
likely to be inefficient in small country samples, where any country outlier could
badly influence the averages of the country coefficients.
The pooled mean group estimator (PMG) involves both pooling and averag-
ing. It is an intermediate estimator between the DFE and the MG that allows
short-run coefficients, the speed of adjustment and error variances to differ across
countries but imposes common long-run coefficients. The PMG estimator is spe-
cially suited for panels with large T and N. Pesaran et al. (1999) show that for T
and N greater than 20 the PMG estimator is clearly superior to estimators such as
the SFE, DFE and the MG. However, they also use the PMG to estimate energy
demand functions for smaller sample sizes: N = 10 and T = 17, and also in this
case the PMG estimator performs better than alternative estimators. The advan-
tages of this method are that it does not impose homogeneity of slopes in the
short-run and it allows for dynamics. Thus, we have choose this procedure to
estimate the relationship between carbon dioxide emissions and economic growth.
Next section summarises the results obtained.
IV. Empirical Results
The 19 countries under study do not exhibit a single behaviour. On the con-
trary, a great heterogeneity is observed in the scatter diagrams shown in Appen-
dix 2 relating emissions and income, both in logarithms.
Equation (1)-(3) in various forms has been estimated for the sample of Latin-
American and Caribbean countries over the period 1975-1998. Several specifica-
tions have been tested allowing for a linear, quadratic and cubic form for the
income-emissions relationship, each of them with and without a time trend.
First, a common ARDL (1,1,1) was run for all countries. The best specifica-
tion in terms of diagnostic test was the quadratic form without time trend6. Table 2
presents results for an EKC for three alternative pooled estimates MG, PMG and
DFE. Results for a linear and cubic specification are available upon request.
TESTING FOR AN ENVIRONMENTAL KUZNETS... 11
Results vary significantly with respect to the estimation method, from MG
(the least restrictive, but potentially not efficient) to PMG and to DFE that only
allows intercepts to vary across countries. Moving from MG to PMG (imposing
only long-run homogeneity to the income variable (lnyh) reduces the standard
errors and the speed of convergence and reduces the size of the estimated long
run parameters. The Hausman test indicates that this restriction (equality of slopes
for the income coefficients) is rejected at 1% significance level in both specifica-
tions (with and without a linear trend). We also tested for homogeneity in the
speed of convergence and short-term dynamics. Moving from PMG to DFE esti-
mators significantly reduces the speed of convergence due to a downward bias in
dynamic heterogeneous panel data. Furthermore, the sign and significance of the
long-run coefficients change in both specifications. We performed a sensitivity
analysis of the PMG results to changes in the lag structure of the dependent and
independent variables. Table 3 presents the estimated coefficients when the Schwarz
Criterion (SBC) has been used to select the ARDL specifications for each coun-
try. The estimated coefficients are very different with respect to the ARDL(1,1,1)
specification (Table 2). The estimated speed of convergence is higher for the
PMG because the SBC criterion chooses the static model for some countries (with
instantaneous adjustment). The maximum likelihood increases for MG estimates
and also for PMG when the SBC is applied. Pesaran et al. (1999) argued that,
TABLE 2
QUADRATIC SPECIFICATION.  SELECTION OF THE ESTIMATION METHOD
ONLY Ly RESTRICTED IN THE LONG-RUN
Dep. variable: lco Without time trend. One lag (1,1,1)
Mean Pooled Hausman Dynamic Fixed
Group Mean Group test Effects1
Convergence
coefficient – 0.38** – 0.28** – 0.23**
Long run coefficients
Ly 1.58** 1.48** 0.00  – 2.78**
Ly^2 0.005 0.003 10.24**
Short run coefficients
?ly 4.23** 1.08 – 2.88*
?ly^2 – 0.21** 0.007 0.24*
Nº of countries 19 19 19
Nº of obs. 380 380 380
Log likelihood 478.46 436.64 – 130.64
Note: 1 t-stat. Calculated using heteroskedasticity consistent standard errors.
12 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
because dynamic specification and homogeneity restriction interact in a complex
way, what may be the optimal order for country-specific estimates may not be
optimal when cross-country homogeneity restrictions are imposed.
Diagnostic tests are reported in Table 4 for the ARDL(1,1,1) and in Table 5
for the SBC criterion used to select lags. There is evidence of serial correlation
of the residuals in two countries (Tables 4,5); functional form misspecification in
six countries (Table 4) and in seven countries (Table 5); evidence of non-normal-
ity of residuals in only one case (Tables 4, 5); finally, there is evidence of hetero-
skedasticity in only one case (Table 4).
We run the regression, keeping only those countries whose results do no present
any specification problems. Our results indicate that PMG estimates are only
slightly different showing similar significance levels. We confirm that PMG seems
to be robust to outliers and to the choice of lag order as stated by Pesaran et al.
(1999).
When a linear form was estimated, we obtained a positive and significant
long-run coefficient for income per capita (1.54) and the time trend was also non
significant but negatively signed. However, the fit of the individual regressions
was, in general, very poor in terms of adjusted R2 and log likelihood. Moreover,
the corresponding diagnostic statistics reported problems for a higher number of
countries than when estimating a quadratic EKC. Results for a cubic EKC speci-
fication showed estimated coefficients on income per capita and squared income
TABLE 3
QUADRATIC SPECIFICATION. SELECTION OF THE ESTIMATION METHOD. ONLY Ly
RESTRICTED IN THE LONG-RUN
SBC criterion used to chose the lag order
Dep. variable: lco Without time trend. SCQ Crit
Mean Pooled Hausman Dynamic Fixed
Group Mean Group test Effects1
Convergence
coefficient – 0.41 – 0.25**
Long run coefficients
Ly – 5.937 – 10.513* 0.87 – 3.07**
Ly^2 0.508* 0.789* 0.26*
Short run coefficients
?ly 10.02 11.49 – 13.82
?ly^2 – 0.509 – 0.58 2.08
Nº of countries 19 19 19
Nº of obs. 380 380 380
Log likelihood 526.70 471.35 – 294.01
Note: 1 t-stat. Calculated using heteroskedasticity consistent standard errors.
TESTING FOR AN ENVIRONMENTAL KUZNETS... 13
TABLE 4
GROUP ESTIMATES AND DIAGNOSTIC STATISTICS FOR 19 COUNTRIES.
QUADRATIC FORM WITHOUT TREND. FIXED LAG STRUCTURE (1,1,1)
D.Var. lco Phia lyb ly^2c SIGMAd Ch-SCe CH-FFf CH-NOg CH-HEh RBARSQi LLj
Argentin – 0.190 1.487 – 0.086 0.037 2.14 1.13 0.58 0.03 0.29 42.76
(0.151) (0.622) (0.037)
Brazil 0.018 1.487 – 0.003 0.047 2.14 1.10 3.29 0.47 0.99 34.76
(0.043) (0.622) (0.084)
Chile – 0.523 1.487 – 0.186 0.561 3.47 4.99 0.11 0.04 0.48 – 12.37
(0.205) (0.622) (0.055)
Colombia 0.043 1.487 – 0.050 0.038 3.75 1.49 0.75 0.00 – 0.17 42.61
(0.110) (0.622) (0.107)
CostaRic – 0.301 1.487 – 0.031 0.186 1.29 0.48 0.02 0.02 – 0.19 8.64
(0.233) (0.622) (0.133)
Dominica – 0.088 1.487 – 0.077 0.094 3.85 0.60 0.01 0.57 0.54 21.58
(0.080) (0.622) (0.065)
Dom-Rep – 0.060 1.487 – 0.067 0.092 3.24 10.09 8.85 0.22 0.98 22.04
(0.047) (0.622) (0.057)
Ecuador – 0.324 1.487 0.026 0.389 0.38 21.25 2.10 17.85 0.45 – 5.41
(0.139) (0.622) (0.074)
Guyana – 0.105 1.487 0.106 0.112 0.00 18.89 0.23 0.27 0.97 18.23
(0.069) (0.622) (0.071)
Haiti – 0.054 1.487 0.069 0.074 0.00 0.74 0.58 0.66 0.98 26.19
(0.087) (0.622) (0.135)
Honduras 0.024 1.487 0.475 0.094 4.41 15.35 2.78 0.17 0.97 21.58
(0.044) (0.622) (0.792)
Mexico – 0.082 1.487 – 0.058 0.067 3.17 2.32 0.02 0.30 0.44 27.95
(0.123) (0.622) (0.127)
Nicaragu – 0.491 1.487 – 0.068 0.051 0.00 0.52 3.60 1.45 0.20 36.39
(0.177) (0.622) (0.041)
Paraguay – 0.040 1.487 0.097 0.091 0.10 1.57 0.66 0.50 0.99 22.22
(0.028) (0.622) (0.084)
Panama – 0.332 1.487 – 0.022 0.076 3.29 5.42 0.47 0.63 1.00 25.62
(0.078) (0.622) (0.038)
Peru – 0.703 1.487 0.023 0.246 3.67 4.63 0.09 0.26 0.85 3.33
(0.267) (0.622) (0.039)
Salvador – 0.609 1.487 – 0.067 0.026 6.49 1.82 0.55 0.58 0.46 50.14
(0.162) (0.622) (0.045)
Uruguay – 0.031 1.487 – 0.030 0.137 0.60 15.92 0.17 0.28 0.98 14.42
(0.083) (0.622) (0.114)
Venezuel – 0.295 1.487 0.010 0.049 0.44 0.94 0.53 0.69 0.45 33.96
(0.104) (0.622) (0.038)
Notes:
Figures in brackets are the standard errors.
a. Convergence coefficient
b. Estimated coefficient for income per capita
c. Estimated coefficient for income per capita squared
d. Standard error of the regression
e. Godfrey´s test of residual serial correlation
f. Ramsey´s Reset test of functional form
g. Jarque-Bera test of normality of regression residuals
h. Lagrange multiplier test of homoscedasticity
i. Adjusted R2
j. Log-likelihood
14 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
TABLE 5
GROUP ESTIMATES AND DIAGNOSTIC STATISTICS FOR 19 COUNTRIES.
QUADRATIC FORM WITHOUT TREND. LAGS (GROUP-SPECIFIC ESTIMATES
OF THE LONG-RUN COEFFICIENTS BASED ON ARDL SPECIFICATIONS SELECTED
USING THE SCHWARZ CRITERION)
D.Var. lco Phia lyb ly^2c SIGMAd Ch-SCe CH-FFf CH-NOg CH-HEh RBARSQi LLj
Argentin – 0.345 – 10.513 0.595 0.031 0.67 0.03 0.74 0.03 0.51 44.39
(0.125) (0.728) (0.041)
Brazil – 0.337 – 10.513 0.632 0.025 0.02 0.00 0.46 0.72 1.00 45.81
(0.052) (0.728) (0.040)
Chile – 1.000 – 10.513 0.725 0.477 0.54 7.46 1.39 4.09 0.63 –11.27
(NA) (0.728) (0.057)
Colombia – 0.012 – 10.513 0.835 0.031 9.36 0.56 0.49 2.15 0.25 45.49
(0.017) (0.728) (0.096)
CostaRic – 0.346 – 10.513 0.819 0.174 1.27 0.76 0.13 0.19 – 0.04 8.53
(0.214) (0.728) (0.079)
Dominica – 0.231 – 10.513 0.649 0.086 0.52 0.49 1.18 0.45 0.64 23.14
(0.096) (0.728) (0.047)
Dom-Repu – 0.202 – 10.513 0.663 0.080 0.52 9.58 0.68 0.51 0.99 25.19
(0.066) (0.728) (0.042)
Ecuador – 0.235 – 10.513 1.065 0.245 3.00 25.05 0.27 0.22 0.79 6.02
(0.090) (0.728) (0.159)
Guyana – 0.054 – 10.513 0.920 0.116 0.02 21.59 0.18 0.19 0.97 17.61
(0.056) (0.728) (0.171)
Haiti – 0.140 – 10.513 0.685 0.040 3.15 0.53 0.80 0.46 1.00 37.55
(0.042) (0.728) (0.043)
Honduras 0.024 – 10.513 1.363 0.094 3.95 15.45 3.31 0.17 0.97 21.50
(0.068) (0.728) (1.163)
Mexico – 0.299 – 10.513 0.861 0.060 1.52 1.00 9.99 0.10 0.58 29.57
(0.108) (0.728) (0.061)
Nicaragu – 0.231 – 10.513 0.712 0.053 0.30 6.75 3.14 0.03 0.11 33.97
(0.104) (0.728) (0.049)
Paraguay – 0.040 – 10.513 0.800 0.085 0.51 3.02 1.48 0.38 0.99 23.19
(0.017) (0.728) (0.069)
Panama – 0.130 – 10.513 0.701 0.094 4.27 3.71 0.05 0.60 0.99 22.24
(0.051) (0.728) (0.045)
Peru – 1.000 – 10.513 0.748 0.210 3.23 3.32 0.30 0.33 0.89 5.48
(NA) (0.728) (0.044)
Salvador 0.143 – 10.513 0.789 0.028 1.21 0.18 1.18 3.26 0.17 47.61
(0.050) (0.728) (0.053)
Uruguay – 0.068 – 10.513 0.701 0.122 0.01 14.99 0.37 0.35 0.98 16.04
(0.066) (0.728) (0.060)
Venezuel – 0.231 – 10.513 0.725 0.056 0.59 3.47 0.44 0.69 0.32 29.31
(0.198) (0.728) (0.049)
Notes:
Figures in brackets are the standard errors.
a. Convergence coefficient
b. Estimated coefficient for income per capita
c. Estimated coefficient for income per capita squared
d. Standard error of the regression
e. Godfrey´s test of residual serial correlation
f. Ramsey´s Reset test of functional form
g. Jarque-Bera test of normality of regression residuals
h. Lagrange multiplier test of homoscedasticity
i. Adjusted R2
j. Log-likelihood
TESTING FOR AN ENVIRONMENTAL KUZNETS... 15
per capita are both significant with a negative and positive sign respectively.
However, the estimated coefficient on cubed income per capita was non-signifi-
cant and negative signed. Nevertheless, the cubic specification seems to perform
better for a number of countries indicating a reduction in the emissions when
income is rising but an increase of them when income overpasses a certain level.
Appendix 2 shows scatter diagrams for GDP per capita against CO2 emis-
sions per capita for the 19 countries considered in our study. We have grouped
the graphs in four categories according to the shape of the curves. We observe
that 9 countries present a N-shaped curve, 2 countries show a curve with decreas-
ing trend, other 2 countries show a U-shaped curve and 6 countries present an
almost lineal up-ward slopping curve. Evidence confirms the need to develop
studies that consider the existence of heterogeneity in country-panels and also the
appropriateness of single country studies.
V. Conclusions
This paper presents empirical estimates of Environmental Kuznets Curves for
a panel of 19 Latin-American and Caribbean countries over the period 1975-1998.
A new econometric technique is applied, that allows for more flexible assump-
tions in a panel data framework with a large time dimension. Unlike most previ-
ous studies we test for slope heterogeneity of the income coefficient. A number
of functional forms have been tested. A quadratic specification seems to be the
more appropriate, although not all the coefficients are significant at conventional
levels. Therefore, there is not a clear pattern related to the carbon dioxide emis-
sions path in contrast with Martínez-Zarzoso et al. (2002) where a N-shaped EKC
was shown for the most part of 22 Kyoto protocol Annex B countries. This find-
ing is in accordance with Perman and Stern (1999)’s study on SO2 emissions.
These authors consider a large sample of countries and conclude that in many
cases the cointegrating relation between sulphur emissions and income are not
consistent with the EKC hypothesis.
The 19 countries we have studied do not exhibit a single behaviour since a
great heterogeneity is observed in the scatter diagrams showing the shape of the
relationship between emissions and income. Nevertheless, there is a common fact:
emissions have been growing continuously since 1975 to the present time.
Although the levels of emissions in Latin-American and Caribbean countries
are still lower than OECD’s levels (1.68 and 8.62 annual tons per capita, respec-
tively), similar to Dijkgraaf and Vollebergh (2001), our results might imply that
there exists a serious risk that the environmental problem of climate change will
not become internalised automatically if countries grow richer. Governments have
the responsibility of enacting the commitments reached in the Kyoto protocol.
We are concerned that other explanatory variables like population density,
openness to international trade, structural change or variables indicating political
reforms could also help to improve the fit of the EKC estimations. We leave
these questions open for further research.
16 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
Notes
1 These authors focus on natural capital as an input in production. Assuming that households affect
the environment through their production and consumption decisions, they apply a dynamic model
to analyse the environmental impact of such decisions. The authors conclude that the EKC hypoth-
esis is supported or reversed depending on the indicator used to represent environmental pressure.
Second, in a partial equilibrium context, the EKC only holds when the household faces an imper-
fect set of markets.
2 Even if the EKC holds, for different reasons, such avoiding safety and health risks, public inter-
vention is still desirable to reduce faster environmental degradation.
3 These authors also study carbon dioxide emissions.
4 Austria, Belgium, Canada, Denmark, Finland, France, Germany, Iceland, Ireland, Italy, Japan,
Luxembourg, Netherlands, Norway, New Zealand, Portugal, Sweden, Spain, Switzerland, United
Kingdom and United States.
5 Argentina, Brazil, Chile, Colombia, Costa Rica, Dominica, Dominican-Republic, Ecuador, Guyana,
Haiti, Honduras, Mexico, Nicaragua, Paraguay, Panama, Peru, El Salvador, Uruguay and
Venezuela.
6 The time trend was not statistically significant.
References
AGRAS, J. (1995). “Environment and Development: An Economic Analysis of Pollution, Growth and
Trade”. Master’s Thesis. Department of Agricultural, Resource and Managerial Economics, Cornell
University.
AGRAS, J. and D. CHAPMAN (1999). “A Dynamic Approach to the Environmental Kuznets Curve
Hypothesis”. Ecological Economics, 28 (2), pp. 267-277.
ANDERSON, D. and W. CAVENDISH (2001). “Dynamic Simulation and Environmental Policy Analy-
sis: Beyond Comparative Statics and the Environmental Kuznets Curve”. Oxford Economic Pa-
pers, 53 (4), pp. 721-746.
BAIOCCHI, G. and S. DI FALCO (2001). “Investigating the Shape of the EKC: A Nonparametric
Approach”. Nota di Lavoro 66.01. Fondazione Eni Enrico Mattei.
BASSANINI, A. and S. CARPETA (2001). “Does Human Capital Matter for Growth in OECD Coun-
tries? Evidence from Pooled Mean-Group Estimates”. OECD Economics Department, Working
Paper Nº 282.
BENGOCHEA-MORANCHO, A.; F. HIGON-TAMARIT and I. MARTINEZ-ZARZOSO (2001). “Eco-
nomic Growth and CO2 Emissions in the European Union”. Environmental and Resource Eco-
nomics, 19 (2), pp. 165-172.
BHATTARAI, M. and M. HAMMIG (2001). “Institutions and the Environmental Kuznets Curve for
Deforestation: A Cross-Country Analysis for Latin America, Africa and Asia”. World Develop-
ment, 29 (6), pp. 995-1010.
BULTE, E.H. and D.P. VAN SOEST (2001). “Environmental Degradation in Developing Countries:
Households and the (Reverse) Environmental Kuznets Curve”. Journal of Development Econom-
ics, 65 (1), pp. 225-235.
CEPAL (2002). Globalización y Desarrollo. Documento LC/G.2157(SES.29/3).
COLE, M.A.; A.J. RAYNER and J.M. BATES (1997). “The Environmental Kuznets Curve: An Em-
pirical Analysis”. Environment and Development Economics, 2 (4), pp. 401-416.
DE BRUYN, S.M. (1997). “Explaining the Environmental Kuznets Curve: Structural Change and
International Agreements in Reducing Sulphur Emissions”. Environment and Development Eco-
nomics, 2 (4), pp. 485-503.
DE BRUYN, S.M.; J.C. VAN DEN BERGH and J.B. OPSCHOOR (1998). “Economic Growth and
Emissions: Reconsidering the Empirical Basis of Environmental Kuznets Curves”. Ecological
Economics, 25 (2), pp. 161-175.
TESTING FOR AN ENVIRONMENTAL KUZNETS... 17
DICKEY, D. A. and W. A. FULLER (1981). “Likelihood Ratio Statistics for Autoregressive Time
Series with a Unit Root”. Econometrica, 49 (4), pp. 1057-1072.
DIJKGRAAF, E. and H.R.J. VOLLEBERGH (2001). “A Note on Testing for Environmental Kuznets
Curves with Panel Data”. Nota di Lavoro 63.01. Fondazione Eni Enrico Mattei.
GALEOTTI, M. and A. LANZA (1999). Richer and Cleaner? A Study on Carbon Dioxide Emissions
in Developing Countries. Nota di Lavoro 87.99. Fondazione Eni Enrico Mattei.
GROSSMAN, G.M. and A.B. KRUEGER (1991). “Environmental Impacts of a North American Free
Trade Agreement”. NBER Working Paper Series WP 3914.
HALKOS, G.E. and E.G. TSIONAS (2001). “Environmental Kuznets Curves: Bayesian Evidence from
Switching Regime Models”. Energy Economics, 23 (2), pp. 191-210.
HEERINK, N.; A. MULATU, and E. BULTE (2001). “Income Inequality and the Environment: Ag-
gregation Bias in Environmental Kuznets Curves”. Ecological Economics, 38 (3), pp. 359-367.
HILTON, F.G. and A. LEVINSON (1998). “Factoring the Environmental Kuznets Curve: Evidence
from Automotive Lead Emissions”. Journal of Environmental Economics and Management, 35
(2), pp. 126-141.
HOLTZ-EAKIN, D. and T.M. SELDEN (1995). “Stoking the Fires? CO2 Emissions and Economic
Growth”. Journal of Public Economics, 57 (1), pp. 85-101.
IM, K. S.; M. H. PESARAN and Y. SHIN (1995). “Testing for Unit Roots in Heterogeneous Panels”.
DAE Working Paper, Serie Nº 9526. University of Cambridge.
KAUFMAN, R.K.; B. DAVIDSDOTTIR, S. GARNHAM and P. PAUL (1998). “The Determinants of
Atmospheric SO2 Concentrations: Reconsidering the Environmental Kuznets Curve”. Ecological
Economics, 25 (2), pp. 209-220.
KOOP, G. and L. TOLE (1999). “Is There an Environmental Kuznets Curve for Deforestation?”.
Journal of Development Economics, 58 (1), pp. 231-244.
KUZNETS, S. (1955). “Economic Growth and Income Inequality?” American Economic Review, 45 (1),
pp. 1-28.
MARTINEZ-ZARZOSO, I.; A. BENGOCHEA-MORANCHO and F. HIGON-TAMARIT (2002).
“Testing for Environmental Kuznets Curves: Empirical Evidence from Pooled Mean Estimates”.
Paper under revision in Environmental and Resource Economics.
MOOMAW, W.R. and G.C. UNRUH (1997). “Are Environmental Kuznets Curves Misleading us?
The Case of CO2 Emissions”. Environment and Development Economics, 2 (4), pp. 451-463.
PANAYOTOU, T. (1997). “Demystifying the Environmental Kuznets Curve: Turning a Black Box
into a Policy Tool”. Environment and Development Economics, 2 (4), pp. 465-484.
PEDRONI, P. (1999). “Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple
Regressors”, Oxford Bulleting of Economics and Statistics, 61 (0), pp. 653-670.
PERMAN, R. and D.I. STERN (1999). “The Environmental Kuznets Curve: Implications of Non-
Stationarity. The Australian National University. Center for Resource and Environmental Studies.
Working Papers in Ecological Economics Nº 9901.
PESARAN, M.H.; Y. SHIN and R. SMITH (1999). “Pooled Mean Group Estimation of Dynamic
Heterogeneous Panels”. Journal of the American Statistical Association, 94 (446), pp. 621-634.
PESARAN, M.H. and R. SMITH (1995). “Estimating Long-Run Relationships from Dynamic Hetero-
geneous Panels. Journal of Econometrics, 68 (1), pp. 621-634.
QUAH, D. (1994). “Exploiting Cross-Section Variation for Unit Root Inference in Dynamic Data”.
Economic Letters, 44 (1-2), pp. 9-19.
ROBERTS, J.T. and P.E. GRIMES (1997). “Carbon Intensity and Economic Development 1962-91:
A Brief Exploration of the Environmental Kuznets Curve”. World Development, 25 (2), pp. 191-
198.
ROCA, J.; E. PADILLA; M. FARRE and V. GALLETTO (2001). “Economic Growth and Atmo-
spheric Pollution in Spain: Discussing the Environmental Kuznets Curve Hypothesis. Ecological
Economics, 39 (1), pp. 85-99.
SCHMALENSEE, R.; T.M. STOKER and R.A. JUDSON (1998). “World Carbon Dioxide Emissions:
1950-2050”. Review of Economics and Statistics, 80 (1), pp. 15-27.
18 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
SENGUPTA, R. (1996). “CO2 Emission-Income Relationship: Policy Approach for Climate Control”.
Pacific and Asian Journal of Energy, 7 (2), pp. 207-229.
SHAFIK, N. and S. BANDYOPADHYAY (1992). “Economic Growth and Environmental Quality:
Time Series and Cross-Country Evidence”. The World Bank, Working Paper Series WP-904.
STERN D.I.; T. AULD; M. S. COMMON and K.K. SANYAL (1998). “Is There an Environmental
Kuznets Curve for Sulphur?”. The Australian National University. Center for Resource and
Environmental Studies, Working Papers in Ecological Economics, Nº 9804.
STERN D.I. and M.S. COMMON (2001). “Is There an Environmental Kuznets Curve for Sulphur?”.
Journal of Environmental Economics and Management, 41 (2), pp. 162-178.
SURI, V. and D. CHAPMAN (1998). “Economic Growth, Trade and Energy: Implications for the
Environmental Kuznets Curve”. Ecological Economics, 25 (2), pp. 195-208.
TUCKER, M. (1995). “Carbon Dioxide Emissions and Global GDP”. Ecological Economics, 15 (3),
pp. 215-223.
UNFCCC (1998). Report of the Conference of the Parties on its Third Session, Held at Kyoto. Doc.
FCCC/CP/1997/7/Add.1. New York: United Nations.
APPENDIX 1
DATA SOURCES
2000 World Development Indicators CD-ROM, The World Bank.
Series: CO2 emissions, industrial (metric tons per capita)
Carbon dioxide emissions from industrial processes are those stemming from the
burning of fossil fuels and the manufacture of cement. They include contributions
to the carbon dioxide flux from solid fuels, liquid fuels, gas fuels, and gas flaring.
Carbon dioxide (CO2) emissions, largely a by-product of energy production and
use account for the largest share of greenhouse gases, which are associated with
global warming.
GDP per capita based on purchasing power parity (PPP)
(current international $)
GDP PPP is gross domestic product converted to international dollars using pur-
chasing power parity rates. The data are based on a 1993 reference year. An
international dollar has the same purchasing power over GDP as the U.S. dollar
in the United States.
TESTING FOR AN ENVIRONMENTAL KUZNETS... 19
APPENDIX 2
SCATTER DIAGRAMS: GDP AND CO2 EMISSIONS
PER CAPITA IN 19 COUNTRIES
A: Argentina D: Dominica HO: Honduras PE: Peru
B: Brazil DR: Dominican-Republic MEX: Mexico SA: El Salvador
CH: Chile EC: Ecuador NIC: Nicaragua U: Uruguay
COL: Colombia GUY: Guyana P: Panama VE: Venezuela
CR: Costa Rica HAI: Haiti PA: Paraguay
a) Countries with N-shaped EKCs:
20 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
TESTING FOR AN ENVIRONMENTAL KUZNETS... 21
b) Countries with EKCs showing a decreasing trend:
c) Countries with EKCs showing a “U” shaped curve:
22 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
d) Countries with EKCs showing increasing trend:
TESTING FOR AN ENVIRONMENTAL KUZNETS... 23
APPENDIX 3
TESTING FOR UNIT ROOTS AND COINTEGRATION
Unit roots
The testing procedure employed here attempt to minimise the problem and
distortions caused by the presence of too many or too few deterministic variables.
It involves starting with the most general specification of the augmented Dickey-
Fuller test (Dickey and Fuller, 1981) and testing downwards. Essentially, moving
from Equation 1 to Equation 3:
? ?y y t B yt t i t
i
p
t= + + + +? ? +
=
?? ? ? ?0 1 2 1 1
1
(1)
? ?y y B yt t i t
i
p
t= + + +? ? +
=
?? ? ?0 1 1 1
1
(2)
? ?y y B yt t i t
i
p
t= + +? ? +
=
?? ?1 1 1
1
(3)
In these regressions many lags of the dependent variable are included to ensure
that the residuals are serially uncorrelated. Table A.1 summarises our unit root
test results. Null of unit root cannot be rejected for the variables in levels for all
the countries, whereas the first differences of the variables are stationary (the null
of unit root is rejected). Therefore, the univariate test statistics strongly support
the view that both variables (income and emissions) are I(1) processes. We do not
report tests statistics for ln(yh)2 since these are virtually the same as those for
ln(yh).
The issue of nonstationarity can also be addressed directly within a panel
data framework. Recently, a number of researchers (Quah, 1994; Im, Pesaran and
Shin, 1996; Pedroni, 1998) have developed new tests which usually are more
powerful than those used for single time series. Panel data exploits more informa-
tion and therefore improves the power. However, since the results from these
tests usually reinforce the findings of the individual countries results and we al-
ready found that the variables are I(1), we find not need to perform them.
Cointegration
It is well known that the test for cointegration using the Engle-Granger meth-
odology suffer from low power compared to the test developed by Johansen. The
24 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
TABLE A.1
UNIT ROOT TEST RESULTS
Augmented Dickey-Fuller Statistic (5% critical Value)
Country
lco2 ?lco2 ly ?ly
Argentina – 2.73 (–3.66) –4.71** (–3.67) –3.21 (–3.64) –3.84** (–3.65)
Brazil 0.69 (–1.95) –2.71** (–1.95) –2.36 (–3.63) – 4.13** (–3.65)
Chile – 0.84 (–3.01) –3.24** (–3.02) –3.07 (–3.64) –3.30** (–3.01)
Colombia – 1.60 (–3.01) – 4.14** (–3.02) –2.66 (–3.63) – 4.38** (–3.64)
Costa Rica – 1.65 (–3.01) –2.78** (–1.96) –1.72 (–3.00) –2.91* (–2.65)a
Dominica 0.87 (–3.02) –3.65** (–3.04) 1.36 (–1.95) –2.65** (–1.96)
Dominican
Republic – 0.72 (–3.01) –4.99** (–3.02) –2.73 (–3.66) –3.83** (–3.66)
Ecuador – 2.45 (–3.01) –4.51** (–3.02) –2.64 (–3.01) –2.84* (–2.64)aPP
Guyana – 3.43 (–3.65) –7.18** (–3.67) –2.86 (–3.63) –3.29* (–3.26)a
Haiti – 1.41 (–3.04) –3.52** (–3.02) –1.75 (–3.63) –3.56* (–3.25)aPP
Honduras – 1.23 (–3.01) –3.93** (–3.02) –3.25 (–3.63) –3.85** (–3.02)
Mexico – 2.41 (–3.01) –2.38** (–1.96) –3.19 (–3.65) –3.50* (– 3.27)
Nicaragua – 2.06 (–3.01) –5.25** (–3.02) –2.11 (–3.64) –4.94** (–3.66)
Panama – 0.56 (–3.01) –2.39** (–1.96) –2.94 (–3.01) –3.44* (–3.27)a
Paraguay – 0.24 (–3.01) –3.12** (–3.02) –3.22 (–3.63) – 4.29** (–3.64)
Peru – 2.01 (–3.01) –3.50** (–3.02) –3.67 (–4.47)*** –2.76** (–1.95)
El Salvador – 0.93 (–3.01) –1.81* (–1.62) –3.03 (–3.63) –2.37** (–1.95)
Uruguay – 1.90 (–3.01) –3.85** (–3.02) –2.36 (–3.62)PP – 4.11** (–3.64)
Venezuela – 2.60 (–3.69) –3.27** (–1.96) –2.42 (–3.63) –2.41** (–1.95)
Notes: Five percent critical values are in parentheses. a indicates ten per cent critical values. Null of
unit root is rejected if the test statistic is smaller than the corresponding critical value. **, *
indicates rejection at the 5% and 10% level respectively. PP indicates that the Philips–Perron
test has been used. Notice that critical values depend on the number of observations and the
lag structure of error terms. The number of lags was selected by using a sequential search
procedure: one–step reductions of the lag length are made until they can no longer be rejected
in testing for the significance of the final included lag using a t test.
TESTING FOR AN ENVIRONMENTAL KUZNETS... 25
set of results presented are therefore obtained using the Johansen methodology:
See Johansen (1988) and (1991).
The empirical specifications examined are:
ln ln (ln )co yh yht t t t= + + +? ? ? ?0 1 2 2
ln ln (ln )co yh yht t t t t= + + + +? ? ? ? ?0 1 2 2
Table A.2 shows the results. It contains the test statistics to establish the
number of cointegrating vectors as well as the most stationary vector of coeffi-
cients from estimation. We find clear evidence in favour of a statistically signifi-
cant relationship for most countries. Only in one case, that of Brazil, can the null
of no cointegrating vectors be accepted.
Like the unit root test, single individual cointegration tests suffer from low
power. This low power may lead to reject cointegration far more often that should
be done.
From the panel unit root testing literature discussed above, some authors have
also developed panel tests for cointegration. Pedroni (1998) presents a survey of
the literature. They are residual-based tests of the null of no cointegration. Their
main advantage is the improved power with respect to the single individual
cointegration tests. Since we already find cointegrating vectors in all but one of
the countries using the traditional methodology, we do not consider necessary to
perform these tests.
26 REVISTA DE ANALISIS ECONOMICO, VOL. 18, Nº 1
TABLE A.2
JOHANSEN COINTEGRATION TEST (5% CRITICAL VALUE)
Country Ly ly2 No. CE(s) Eigenvalue Likel-R.5%
Notes: No. CE(s) denotes the number of cointegrating vectors. Standard errors are in brackets.
Argentina 0.04 (0.09) – 0.02 (0.01) 0 0.72 38.45**
1 0.28 12.48
2 0.26 5.93*
Brazil – – 0 0.56 30.85
1 0.34 14.09
2 0.25 5.88
Chile 0.0000 (0.05) – 0.07 (0.02) 0 0.75 43.27**
1 0.52 15.30
2 0.03 0.61
Colombia – 3.08 (0.52) 0.19 (0.03) 0 0.84 60.49**
1 0.56 23.11
2 0.29 6.78
Costa Rica 41.14 (4.71) – 2.51 (0.28) 0 0.83 47.19**
1 0.38 11.17
2 0.07 1.48
Dominica – 6.32 (1.05) 0.39 (0.06) 0 0.98 120.37**
1 0.72 30.12*
2 0.19 4.38
Dominican
Republic 5.04 (1.16) – 0.35 (0.07) 0 0.77 49.01**
1 0.50 19.22
2 0.23 5.33
Ecuador – 72.67 (39.67) 4.62 (2.50) 0 0.61 38.76*
1 0.46 19.87
2 0.30 7.30
Guyana 29.23 (10.76) – 1.81 (0.70) 0 0.63 35.39*
1 0.52 15.37
2 0.01 0.32
Haiti 832.4 (1728) – 56.77 (117) 0 0.98 86.46**
1 0.41 12.10
2 0.09 1.91
Honduras 34.12 (7.98) – 2.28 (0.52) 0 0.79 45.60**
1 0.50 15.43*
2 0.11 2.23
Mexico 21.26 (5.24) – 1.22 (0.30) 0 0.95 69.24**
1 0.40 10.15
2 0.01 0.24
Nicaragua 40.11 (81.05) – 2.83 (5.39) 0 0.85 48.60**
1 0.47 12.25
2 0.01 0.29
Panama 40.44 (14.47) – 2.52 (0.88) 0 0.78 37.55**
1 0.34 8.49
2 0.02 0.49
Paraguay 153.9 (105) – 9.22 (6.29) 0 0.93 68.58**
1 0.57 17.18*
2 0.04 0.96
Peru 12.51 (7.59) – 0.77 (0.48) 0 0.69 40.14*
1 0.48 17.88
2 0.24 5.35
El Salvador – 322 (–794) 20.54 (50.74) 0 0.77 43.76**
1 0.38 15.27
2 0.27 6.09
Uruguay 66.02 (21.80) – 3.79 (1.24) 0 0.86 68.69**
1 0.67 30.01**
2 0.37 8.93
Venezuela 2.99 (4.45) – 0.19 (0.26) 0 0.62 39.51*
1 0.55 20.85*
2 0.26 5.79