Ecological Economics 45 (2003) 133
/148

www.elsevier.com/locate/ecolecon

ANALYSIS

Determinants of CO2 emissions in a small open economy

Birgit Friedl, Michael Getzner *
Department of Economics, University of Klagenfurt, Universitaetsstrasse 65-67, A-9020 Klagenfurt, Austria

Received 17 June 2002; received in revised form 18 December 2002; accepted 6 January 2003

Abstract

The aim of the paper is to explore the relationship between economic development and carbon dioxide (CO2)
emissions for a small open and industrialized country, Austria. We test whether an Environmental Kuznets Curve
relationship also holds for a single country rather than concentrating on panel or cross-section data for a set of
countries. A cubic (i.e. N-shaped) relationship between GDP and CO2 emissions is found to fit the data most
appropriately for the period 1960
/1999, and a structural break is identified in the mid-seventies due to the oil price
shock. Furthermore, two variables are additionally significant: import shares reflecting the well-known pollution haven
hypothesis, and the share of the tertiary (service) sector of total production (GDP) accounting for structural changes in
the economy. Emission projections derived from this single country specification support the widely held opinion that
significant policy changes are asked for when implementing the Kyoto Protocol in order to bring about a downturn in
future carbon emissions.
# 2003 Elsevier Science B.V. All rights reserved.

Keywords: CO2 emissions and economic growth; Environmental Kuznets curve; Stationarity; Cointegration

JEL classification: Q2; O1; C2

1. Introduction for a variety of pollutants, such as nitrogen oxide,

The Environmental Kuznets Curve (EKC here-
after) summarizes what has meanwhile become
well known as the inverted U-shaped relationship
between emissions (environmental pressure) and
GDP per capita. The inverse relationship between
pollution and per capita income has been explored
sulfur dioxide, suspended particulate matter, car-
bon monoxide, lead, and for deforestation, biolo-
gical oxygen demand and others1. While the EKC
hypothesis has been confirmed */albeit far from
unanimously */for a set of air pollutants, water,
and land use, the empirical evidence is very
inconclusive in the case of greenhouse gas
(GHG) emissions, in particular CO2 emissions.

* Corresponding author. Tel.: /43-463-2700-4124; fax: /

1
43-463-2700-4191. For critical surveys, see, for instance, Stern et al. (1996),
E-mail address: michael.getzner@uni-klu.ac.at (M. Getzner). Ekins (1997), and Panayotou (2000).

0921-8009/03/$ - see front matter # 2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0921-8009(03)00008-9
134 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
Due to the release of enormous amounts of CO2 in
the combustion of fossil energy, CO2 emissions are
classified as one of the main driving forces behind
global warming today. Worldwide, greenhouse gas
emissions are increasing despite common efforts to
implement internationally binding agreements
such as the Kyoto Protocol, and a turning point
in the positive relationship between CO2 emissions
and per capita income has not yet been identified,
at least not on a global scale. The question arises
as to whether the delinking of carbon emissions
and income is possible at high levels of income.
This is the main motivation for testing the EKC
hypothesis for the case of CO2 emissions in a single
open and industrialized country.
From a theoretical point of view, the inverted-U
relationship is less likely for CO2 emissions than
for ‘traditional’ air pollutants such as NOx or SO2.
While these air pollutants have local effects, CO2
emissions cause problems on a global scale, and
the social costs of global warming accrue both
across time and nations. Therefore, free-rider
behavior might lead to a close relationship be-
tween carbon emissions and income at all levels of
per capita income (Arrow et al., 1995). In line with
this argument, a linear relationship for CO2
emissions and GDP per capita was confirmed in
early studies (Shafik and Bandyopadhyay, 1992;
Shafik, 1994). A delinking of growth in carbon
emissions from economic growth does, however,
seem possible, if either carbon emissions fall as a
by-product of other abatement activities (Holtz-
Eakin and Selden, 1995) or institutions such as
environmental regulations are improved with in-
creasing per capita income. Thus, an inverted U-
shaped function has also been identified (De
Bruyn et al., 1998; Heil and Selden, 2001; Holtz-
Eakin and Selden, 1995; Moomaw and Unruh,
1997), followed by N-shaped (cubic) specifications
(e.g. Galeotti and Lanza, 1999). Neither the linear
nor the cubic relationship allows for an optimistic
interpretation of economic growth as being bene-
ficial for the environment. Rather, at high levels of
income, CO2 emissions are found to increase with
income.
One of the major qualifications regarding the
existence of an EKC for all types of pollution is
that economic growth is not sufficient for environ-
mental recovery (see, for example, Stern et al.,
1996). The fact that nations which formerly had or
currently have low per capita income are experi-
encing increasing pollution while industrialized
countries are successful in abating emissions does
not imply that economic development will solve
environmental problems quasi automatically.
Rather, the EKC might be considered as a
descriptive statistic on a par with literacy rates or
health indicators relative to per capita income. The
more promising approach, therefore, seems to be
an investigation of the time-series data of a single
country which may be able to account for historic
experience such as environmental policy, develop-
ment of trade relations, and exogenous shocks
such as the oil crisis (Stern et al., 1996).
The vast majority of investigations regarding the
existence of an EKC concentrate on cross-section
and panel data. To the authors’ knowledge, EKC
studies for single countries most often address
developing countries (e.g. Patel et al., 1995;
Vincent, 1997). Rare exceptions addressing indus-
trialized countries include De Bruyn et al. (1998)
and Moomaw and Unruh (1997). Thus, the
current study contributes to the discussion on the
implementation of the Kyoto Protocol since the
EKC relationship found for CO2 could serve as a
counterfactual scenario reflecting current develop-
ments of emissions in industrialized countries.
As far as Austria is concerned, Schandl et al.
(1999, 2000) present an in-depth case study of
Austrian materials flows. Other time-series inves-
tigations focusing particularly on Austrian CO2
emissions in comparison to other countries have
been carried out by Dijkgraaf and Vollebergh
(1998), Liski and Toppinen (2001) and Panayotou
et al. (2000).
The paper is structured as follows. Section 2
reviews previous efforts to test the EKC for carbon
emissions in more detail. Different functional
specifications are introduced and their economic
implications are explained. Section 3 deals with the
Austrian database and a description of the time
path of CO2 emissions, including environmental
policy efforts to reduce CO2 emissions. In addi-
tion, the empirical EKC model is described,
stationarity and cointegration tests are reported,
and structural breaks within the time series are
 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
dealt with. The paper closes with a discussion of
the results and conclusions for environmental and
economic policy.
2. Lessons from the past
2.1. The relevant set of data
Two of the main points of criticism of the EKC
hypothesis are, first, that the inverted U-shape is
merely a statistical result (a juxtaposition) and not
a common development path holding for the
subsets of industrialized or developing countries
since it is derived from cross-section data (see, for
example, Roberts and Grimes, 1997), and second,
that by abstracting from other causalities besides
income and emissions the EKC hypothesis does
not account properly for crucial historic develop-
ments and unique events such as the oil crisis in the
mid-seventies (e.g. Panayotou, 1997). Therefore,
investigations of the EKC relationship for one
single country could shed some light on the
validity of the claim that economic development
might improve environmental quality.2 Contribu-
tions to the existence of EKC relations in time-
series data can be found in De Bruyn et al. (1998)
for the Netherlands, the United Kingdom and
Western Germany, in Roberts and Grimes (1997)
for three groups of countries differing in their
income, and in Moomaw and Unruh (1997) for all
countries having experienced a negative relation-
ship between emissions and income per capita.
With respect to the second criticism that im-
portant variables are omitted, valuable contribu-
tions have been made by Panayotou (1997) and De
Bruyn et al. (1998). For Austria, as a small open
economy, it follows that the influence of develop-
ments abroad might be of crucial importance, both
regarding the possibility of ‘exporting’ emissions,
and the influence of foreign and international
environmental policy. Empirical indications of
2
It is worth noting that per capita income should be
interpreted as a proxy for other less easily measurable variables
such as economic development, institutions, and environmental
awareness.
135
the former, namely the hypothesis that polluting
industries are displaced to developing countries
(sometimes referred to as the ‘pollution haven’
hypothesis), have been presented by Hettige et al.
(1992), Jänicke et al. (1997), Suri and Chapman
(1998) and Cole (2000).
Both critical aspects are dealt with in the present
study for Austria. The data set for CO2 emissions
is available from 1960 onwards on an annual basis
and was provided by the Austrian Central Me-
teorology and Geodynamics Institute.3 Three
additional variables (besides economic growth)
are identified which can be hypothesized to be
relevant for a small open economy: (i) the oil price
shock followed by improvements in energy effi-
ciency and stricter environmental regulations, (ii)
the structural change in the Austrian economy
from a strong emphasis on industrial and agrarian
production in the 1950s to an economy dominated
by the service sector accounting for more than 60%
of GDP at present, and (iii) the worldwide increase
in trade and international policies which have led
to a specialization in more eco-friendly production
in Austria on the one hand and have raised the
pressure on environmental policy (EU legislation,
Kyoto Protocol) on the other.
2.2. The relevant indicator for CO2 emissions
Before considering different functional forms to
explain the environment
/income relationship, we
make some comments on the correct dependent
variable. Four types of indicators are commonly
employed for different pollutants or sorts of
environmental degradation: (i) emissions per ca-
pita, (ii) emissions per gross domestic output
(pollution intensity) or gross product, (iii) ambient
levels of pollution (concentrations; impacts on a
certain area), and (iv) total emissions. In cross-
country studies by far the most frequently used
indicator for CO2 is emissions per capita (see Heil
3
Previous studies by Panayotou et al. (2000) and Liski and
Toppinen (2001) use the data set for the period 1870-current for
17 OECD countries provided by Marland et al. (1996) of the
Carbon Dioxide Information Center (CDIAC) in which a
correction for the enormous territorial and economic
adjustments after WWI is missing.
136 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
and Selden, 2001; Holtz-Eakin and Selden, 1995;
Liski and Toppinen, 2001; Moomaw and Unruh,
1997; Panayotou et al., 2000).
The indicator chosen in the current paper is the
annual level of total CO2 emissions, mainly for the
following reasons. First, Kyoto reduction targets
relate to percentage decreases in annual GHG
emission levels (and not decreases in per capita
emissions or emissions per unit of output). As has
been shown in various studies (e.g. Tisdell, 2001),
total emissions can still increase even when emis-
sions per unit of output decrease. More precisely,
the scale effect of economic growth outweighs the
composition effect (structural change of the econ-
omy) and the technological effect due to higher
productive efficiency (cf. Panayotou, 2000). Sec-
ond, for a single country with roughly constant
population across the previous decades (annual
population growth rates of less than one percent),
dividing total emissions by population merely
results in scaling down the numbers.4 Third, since
one of the main contributors to CO2 emissions is
(freight and private) transport, emissions should
not be related to industrial output. Fourth, con-
centration levels are suitable for local pollutants
(see, for example, Borghesi, 1999) but not for
global pollutants, which cannot be described in
local impact levels.
2.3. Possible functional forms of EKCs for CO2
As mentioned in Section 1, three specifications
for EKCs for CO2 have been described in the
literature: a linear, a quadratic (inverted U) and a
cubic specification (N-shaped or sideways-mir-
rored S-shaped). One general functional form for
the level of CO2 emissions, abstracting from other
explanatory variables (e.g. time and country
specific effects), is:
CO2t  b0 b1 Yt b2 Yt2 b3 Yt3 (1)
In the linear case, b1
/0 and b2 /b3 /0. In the
4
Modeling the dependent variable in CO2 per capita instead
of total CO2 emissions (in million mt) did not lead to any
significant changes in the results obtained in the econometric
estimations (see section Section 3.4).
quadratic case, if b1
/0, b2 B/0, and b3 /0,
emissions exhibit an inverted-U relationship to
per capita income.5 In the cubic case, if b1
/0,
b2 B/0, and b3
/0, an N-shaped relationship
between emissions and income can result6. If the
coefficients are reversed in their sign (b1 B/0, b2
/
0, and b3 B/0), a sideways-mirrored S graph can be
identified. The functional forms are specified as
logarithms in some studies as well, e.g. Cole et al.
(1997), Galeotti and Lanza (1999) and Holtz-
Eakin and Selden (1995).
The linear relationship between GDP and CO2
emissions was confirmed by Shafik and Bandyo-
padhyay (1992) and Shafik (1994) in panel studies,
and by De Bruyn et al. (1998) for four single
countries. The inverted-U pattern was laid out by
Holtz-Eakin and Selden (1995) in a panel investi-
gation albeit basically on explanatory grounds
rather than econometric results (for a discussion
see Ekins 1997, 811). The inverted-U hypothesis
was also confirmed by Heil and Selden (2001) and
Galeotti and Lanza (1999) for worldwide panel
data (IEA database). So far, no country-specific
study has confirmed the existence of an inverted-U
relationship. The only indication along these lines
is the structural change model employed by
Moomaw and Unruh (1997) who find an in-
verted-V shape with discontinuous rather than
gradual change. Their key result is that the cubic
functional form is caused by data aggregation
across countries. De Bruyn et al. (1998), however,
show theoretically that cross-section inverted-U
relationships for different points in time corre-
spond to single country N-shaped curves if poli-
tical adjustments are taken into account.
A review of these studies does not conclusively
reveal which relationship we should expect for the
emission
/income relationship in a single country.
As the evidence is rather diverse for panel data and
the studies on single countries are rather rare,
additional research is called for.
5
For an inverted U to result, the second coefficient has to be
smaller than the first in absolute terms (½b2 ½B/½b1 ½). See Ekins,
1997, 807 f.
6
As Ekins (1997) correctly points out, the N shape is
dependent on the relative size of the coefficients. Only if ½b3 ½ B/
½b2 ½ B/½b1 ½ does this pattern result.
 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
2.4. Explanatory variables other than income
In the present study, two further independent
variables are tested for their significance. The first
is trade (to test for the pollution haven hypothesis)
and the second is structural change to take the
composition effect into account (see Section 3.2
for an operationalization of both potentially
explanatory variables).
For an industrialized country, exports are likely
to be serially correlated with GDP, i.e. if exports
increase, total emissions will fall only if increased
efficiency outweighs the scale effect. On the other
hand, imports should lead to a reduction in
emissions if the pollution haven hypothesis holds.
For the case of a small open economy such
as Austria, the impact of trade on emissions
should be comparatively greater than for a larger
economy. Previous results obtained in cross-coun-
try studies (e.g. Arrow et al., 1995; Cole et al., 1997
for a discussion see Cavlovic et al., 2000) are
mixed.
In previous studies, a country’s openness was
approximated by different trade measures, such
as the ratio of imports (exports) of manufactured
goods to domestic manufactured production,
the sum of imports and exports relative to GDP
(Harbaugh et al., 2000), or export volumes per
capita (Panayotou et al., 2000). In the present
case, the ratio of imports to GDP will be used as
an explanatory variable for the reasons given
above.
When testing for structural change, it turned out
to be difficult to come up with convincing
statistics. Structural change can be the result of
changes in the political system (which is appar-
ently not the case for Austria in the period covered
by the current study), the degree of (environmen-
tal) legislation, or increases in energy prices etc.
Again, as Austria is a small open economy, the
influence of international markets is not negligible.
Over the last 50 years, the service sector has
expanded and the industrial, agricultural and
resource sectors have been observed to shrink.
Therefore, we test for the influence of the increase
in the service share (value added in the service
sector relative to GDP) on CO2 emissions.
137


3. Estimating a CO2 /EKC for Austria7
3.1. CO2 emissions and Austria’s commitments to
reducing CO2 emissions
Initial insight into the development of CO2
emissions in Austria can be gained by describing
the time series of CO2 emissions for the period
1960
/1999. Fig. 1 shows CO2 emissions in Aus-
tria. The starting point for the official statistics for
CO2 is the year 1960, when CO2 emissions
amounted to around 34 million metric tons (mt).
Until the early 1970s, CO2 emissions grew rapidly
to over 61 million mt in 1973. The oil price shock
did not fail to leave its mark on CO2 emissions.
Economic growth slowed down, and due to
changes in energy policies, such as rationalizing
private consumption of fuels by restricting private
car use, CO2 emissions dropped to about 57
million mt in 1975. Between 1973 and 1975 there
is a structural break in CO2 emissions (regarding
the slope of the time path), as general awareness of
the growing scarcity of energy resources increased
dramatically. (A Chow breakpoint test in a simple
autoregressive trend model to test for stationarity
indicated a clear structural break in the year 1974
with a log likelihood ratio of 19.0858 with P B/
0.001; see also Section 3.3.)
Subsequently, CO2 emissions increased slightly
until 1981 and then fell during the recession in the
early 1980s. A rather constant rise in CO2 emis-
sions followed, and at the end of the 1980s, the
next peak was reached with CO2 emissions
amounting to around 66 million mt. The recession
in 1992 and 1993 again reduced CO2 emissions.
Most recently, from 1994 onwards, there has been
constant but slow growth currently amounting to
around 65 million mt (see also Fig. 2 which
presents a scatter plot of CO2 emissions and
Austrian per capita GDP in constant 1995 prices).
While absolute figures for CO2 emissions are
important for assessing their implications for the
world’s climate and ecosystems, a comparison with
the growth in GDP is important when considering
7
All data and statistical and econometric results are
available in detail from the authors on request.
138 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
Fig. 1. CO2 emissions in Austria 1960
/1999.
the EKC relationship. In this respect, Fig. 3
presents three indices (base year 1960 /100):
GDP, total CO2 emissions, and CO2 intensity.
While real GDP more than tripled by 1999, total
CO2 emissions doubled in this period. The index of
‘CO2-intensity of GDP’, expressed as CO2 emis-
sions in kilograms per unit of GDP (in Euro 1000),
followed a constant time path until the mid-1970s,
and decreased from an index of around 100 to
about 55 index points afterwards. However, there
was a rather constant decrease between the 1970s
and the beginning of the 1990s followed by a
leveling off in the 1990s. Consequently, the data
suggests a structural break in the mid-1970s.
The only significant structural break can be
attributed to the oil price shock. Changes in
environmental and energy policies such as, for
instance, the international Bergen, Strasbourg and
Toronto agreements in the 1980s and the Kyoto
protocol in the 1990s, and the government’s
domestic energy policy report in the early 1990s,
had no impact on reducing CO2 emissions. After
1990, there was a sharp increase in CO2 emissions,
followed by a short-term reduction after the
recession in 1993, and subsequently by constant
growth in CO2 emissions.
As will be argued later, the time series up to the
present and the structural models estimating an
EKC for CO2 emissions in Austria give no
indication that there is any mean-reverting ten-
dency in the data. Between 1990 and 1999,
Austrian CO2 emissions grew from 62 to 66
million mt. Thus, according to the Kyoto Proto-
col, Austria has to reduce its emissions to around
53 million mt, corresponding to a reduction of
nearly 20% based on current emissions. What is
more, if there is no significant policy change
leading to a dramatic structural break such as
the one caused by the oil price shock in the mid
1970s, CO2 emissions will grow further, at least in
a ‘business-as-usual’ scenario.
3.2. The empirical model
Based on the theoretical considerations dis-
cussed in Section 2, we apply the following
 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
Fig. 2. CO2 emissions and GDP in Austria.
empirical model to test for the EKC hypothesis:
CO2t  f (Yt ; ydt ; Tt ; it ; st ; Xt ) (2)
where CO2t are CO2 emissions in year t .8 We
hypothesize that these emissions depend on Yt
denoting GDP at constant 1995 prices per capita,
either in a linear, quadratic or cubic relationship.
Furthermore, we test for short-term fluctuations in
CO2 emissions by accounting for short-term dis-
turbances in economic growth as measured by
GDP (Yt ). The variable ydt represents the devia-
tion of Yt from its long-term trend, expressed as a
percentage of trend-Yt . Trend-Yt is calculated by
means of a linear autoregressive model. This
specification includes short-term variations of
GDP and has proven to be the most appropriate
when analyzing Austrian fiscal and economic
policy (see, for example, Getzner et al., 2001;
8
Note that emissions are estimated in absolute numbers. To
compare the results with other countries, divide emissions by
the factor of 8/106 (the approximate population of Austria
which has remained roughly constant during the last decades).
139
Neck and Getzner, 2002). Short-term fluctuations
can be hypothesized to lead to a significantly
positive coefficient of ydt because an actual GDP
above the long-term trend might lead to a short-
term increase in CO2 emissions (all variables are
listed and described in Table 1).
While economic determinants may be the reason
for long- and short-term developments in CO2
emissions, the average temperature in year t might,
of course, be influential. For an approximation of
climatic variations, we take a time series of the
yearly temperature deviation Tt from the long-run
temperature average. Climatic conditions opera-
tionalized by Tt can be hypothesized to exhibit a
negative coefficient: in years with an above-aver-
age temperature, energy consumption (e.g. for
heating) is lower, and thus lower CO2 emissions
can be expected.
In order to test for structural changes in an
Austrian and an international economic context,
we test for the influence of two more variables.
First, the variable it denotes imports as a ratio to
GDP to account for the openness of the Austrian
140 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148

Fig. 3. Development of GDP and CO2 emissions relative to GDP (index, 1960/100; 1960
/1999).

(see Section 2 above). If this hypothesis proves to

Table 1 be realistic, higher imports should lead to lower
Variables and data used for estimating Environmental Kuznets
Curves for Austria
CO2 emissions. Thus, we test for the hypothesis
that the coefficient for it is significantly negative
Dependent variables (i.e. higher imports reduce domestic CO2 emis-
CO2t CO2 emissions in Austria in year t (in mt)
rt Residuals of the cointegrating regression (CO2t and Yt )
sions).
Second, we explain structural changes in the
Explanatory variables
Yt GDP (constant 1995 prices), per capita (in Euro 1000)
Austrian domestic economy by accounting for the
y dt Deviation of GDP from trend-GDP (% of trend-GDP) significant growth in the tertiary sector (service
Tt Deviation from long-term mean temperature in Austria sector). Variable st denotes the value added
(8C) produced in the service sector as a share of
it Imports, nominal ratio to GDP (%)
st Value added in the service sector (tertiary sector),
GDP. The share of services (st ) grew from 39%
nominal ratio to GDP (%) in 1960 to over 62% in 2000. As services cause
D75 Dummy variable ( /1 for the period 1975
/2000) relatively lower CO2 emissions than industrial
d (xt) First difference of variable xt production, the coefficient of st can be hypothe-
n Number of observations
sized to be significantly negative as a growing
share of value added from the service sector
reduces pollution-intensive industries and thus
economy. This ratio grew from around 21% in Austrian CO2 emissions.
1960 to over 36% in 2000. A popular hypothesis in Finally, we include variables accounting for
the environmental debate is the ‘export’ of envir- other politico-economic determinants of CO2
onmentally harmful production to other countries emissions. For instance, Xt might include a
 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
dummy variable for a change in the slope of the
regression function, due to the structural break,
from 1974 onwards; Xt might also include vari-
ables denoting policy changes such as the signing
of international agreements to reduce CO2 emis-
sions (e.g. the aforementioned commitments).
The methodological approach in this paper is as
follows: First, we test for stationarity in the time
series of the dependent and independent variables.
Second, we explore whether the main variables
CO2t and Yt are cointegrated. Based on these
results, we then test which functional form is most
appropriate to describe the time path of CO2
emissions in Austria.
3.3. The time series of CO2 emissions: stationarity
and cointegration analysis
3.3.1. Stationarity and structural breaks in the time
path of CO2 emissions
In order to test for stationarity in the time series
of the dependent and independent variables, the
most common stationarity test, namely the aug-
mented Dickey
/Fuller test (ADF test), was ap-
plied. The aim of this procedure is to test for the
characteristics of the time series with CO2t as the
dependent variable and Yt , st , it and Tt as the
explanatory variables.
Without presenting the details of the econo-
metric estimations, it can be concluded that the
hypothesis (H0) of non-stationarity for the time
series of the dependent variable CO2t in levels
cannot be rejected at any reasonable level of
significance. However, if we apply the ADF test
in order to explore the stationarity of the first
difference of the time series, we have to reject H0 at
the 1% level of significance. From a qualitative
viewpoint, we can conclude that the time path of
CO2 emissions in Austria is not independent of
time, i.e. CO2 emissions grew during the relevant
period without a significant mean-reverting trend.
On the contrary, there is some indication that the
time series includes an upward drift eventually
leading to future growth.
As discussed in Section 3.1, the development of
CO2 emissions in Austria might be characterized
by a structural break during the mid-1970s with
more rapid growth before 1975 and slower growth
141
afterwards. Applying a Chow breakpoint test to
the estimations for stationarity strongly indicates
that the structural break appears to have taken
place in 1974 (for that year, the log likelihood ratio
amounts to 19.0858 which is highest compared to
test statistics for other years and significant at P B/
0.001). Thus, we explore the characteristics of the
time series of CO2 emissions for the two periods
separately. First, we apply the ADF test to CO2t
for the period from 1957 to 1974. Again the
estimation strongly implies that the time series is
not stationary in levels. The explanatory power of
the model is much higher than the estimation for
the whole period. Alongside a drift in the time
series, there is also an indication that the time
series is driven by a deterministic trend. For the
time series in its first difference, H0 has to be
rejected. Similar results are obtained when testing
the stationarity of CO2t for the period after 1974.
The time series is again non-stationary in levels but
stationary in its first difference. However, the
structural break in the series can also be seen in
the differing size of the coefficients for the estima-
tions for the two periods. While the drift in the
second period is much larger, the trend is smaller
and only significant at P B/0.1. The mean-revert-
ing tendency is also stronger in the second period
compared to the first one. Summing up, the
variable CO2t exhibits an order of integration of
I(1); this result is valid both for a joint estimation
for the whole period as well as for a separate
estimation for the two periods.
Furthermore, we test the variables Yt , st , it and
Tt for stationarity. Without going too much into
detail here, the explanatory variables of the
empirical model given in Eq. (2) above again
prove to be non-stationary in levels but stationary
in their first differences (I(1)). One exception is the
variable Tt which is stationary in levels. This is a
result, which we would expect; however, Tt also
follows a small but significantly increasing trend
path, indicating a small but significant increase in
temperatures.
The results so far indicate that there is econo-
metric evidence for non-stationarity in the time
series of both the dependent and the explanatory
variables. Simply regressing a non-stationary vari-
able on another non-stationary variable might lead
142 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
to spurious results, i.e. a significant relation might
be indicated where, in fact, there is none. However,
if the time series are cointegrated, i.e. if they ‘run
parallel’ and the residuals of the cointegrating
regression are stationary, OLS is again an ade-
quate estimation procedure. Therefore, in the next
section we test for cointegration between the two
main variables, namely CO2t and Yt .
3.3.2. Cointegration: economic growth and CO2
emissions
Testing for cointegration bilaterally is usually
done by means of the so-called cointegrating
regression. The equation to be estimated is simply:
CO2t  abYt o t (3)
Cointegration is then tested in two steps. First,
the value of the Durbin
/Watson statistic resulting
from the estimation is an indication of cointegra-
tion. If the Durbin
/Watson statistic is signifi-
cantly different from zero, H0 of no cointegration
between the two time series has to be rejected.
However, this test is commonly not considered to
be very decisive. Second, the residuals resulting
from estimating equation (Eq. (3)) are tested for
stationarity, e.g. with the ADF test.
Table 2 presents the results of the bilateral
cointegration test for the time series of CO2t and
Yt . The cointegrating regression exhibits a signifi-
cant constant and a significantly positive coeffi-
cient for Yt with an adj. R2 of about 0.76. With a
Durbin
/Watson statistic of 0.2708, H0 has to be
rejected at a rather ‘weak’ level of significance of
10%. The second column of the table shows the
more important test for cointegration, i.e. the
stationarity test of the residuals of the cointegrat-
ing regression. Well below the significance level of
1%, H0 of no cointegration has to be rejected (the
ADF test statistic amounts to /4.3512 compared
with a critical value of /2.6280 at the 1% level of
significance). The results indicate that there is not
a unit root in the residuals and that the two
variables are indeed cointegrated.
Table 2 also presents the results of the test for
cointegration in the two separate periods (1954
/
1974; 1975
/1999). As discussed above, the two
periods might be different regarding the correla-
tion between CO2 emissions and GDP. The third
column shows the results of the cointegrating
regression for the period 1960 to 1974. The
hypothesis of no cointegration has to be rejected
at the 1% level of significance when we take the
Durbin
/Watson statistic as the main guideline.
Testing the residuals for stationarity, as shown in
the fourth column of the table, we reject the
hypothesis of a unit root in the residuals at the
5% level of significance. The ADF test statistic is
/2.6715 which is smaller than the critical value of
/1.9699 at the 5% level of significance as the
inspection of the graphical representation of the
time series of CO2 emissions shows (Figs. 1 and 2),
CO2 emissions and GDP merely ‘run parallel’ in
the first period. The adj. R2 for the first period is
thus very high with a value of 0.9651. This result
can also be interpreted as a further indication of
the strong links between economic growth and
increasing CO2 emissions between 1960 and 19749.
The fifth column in Table 2 shows the results of
the cointegrating regression for the second period,
1975 to 1999. The Durbin
/Watson statistic again
leads to a rejection of H0 at the 1% level of
significance, and the hypothesis of non-stationar-
ity of the residuals of the cointegrating regression
can also be rejected (the ADF test statistic is
/2.9192 compared to a critical value of /2.6700
at the 1% level of significance). The coefficient for
the explanatory variable Yt is much smaller than
the one for the first period, indicating that the
links between CO2 emissions and GDP are much
weaker in the second period.
Based on the results of the cointegrating regres-
sion, we can conclude that Austrian CO2 emissions
and GDP are cointegrated, i.e. they ‘run parallel’,
hence, OLS is an appropriate estimation tool.
However, the differences in the two periods
(1960
/1974 and 1975
/1999) again become appar-
ent. In order to estimate an EKC, we thus have to
focus on a functional form, which is able to
account for the structural break in the time series.
9
Moomaw and Unruh (1997) derive a similar result for a
subset of OECD countries.
 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148 143

Table 2
Testing for cointegration between GDP and CO2 emissions

Explanatory variables Dependent variable

1960
/1999 1960
/1974 1975
/1999

CO2t d(rt ) CO2t d(rt ) CO2t d(rt )

Constant 28.1188*** /1.7615 47.2344***

(11.0427) (/0.7107) (13.5127)
Yt 1.7572*** 4.5547*** 0.7389***
(11.2095) (19.7301) (3.9723)
rt-1 /1.0607*** /0.9771** /0.6051***
(/4.3512) (/2.6715) (/2.9192)
d (rt-1) 0.0689 0.5385 0.1987
(0.3967) 1.3720 (0.9873)
Adj R2 0.7617 0.4843 0.9651 0.3347 0.3811 0.2634
S.E. of regression 4.5359 2.3156 1.6760 1.5752 2.4980 2.1039
Akaike info criterion 5.9106 4.5712 3.9942 3.8873 4.7455 4.4084
Schwarz criterion 5.9951 4.6592 4.0887 3.9742 4.8430 4.5071
F -statistic 125.6527*** 33.8748*** 389.2776*** 7.0396** 15.7791*** 8.8673***
log likelihood /116.2128 /80.2821 /27.9572 /23.2676 /57.3193 /48.6972
Durbin
/Watson statis- 0.2708* 1.2708*** 1.0413***
tic
Breusch
/Godfrey LM 0.3463 0.5263 0.5924
test
ADF Test statistic /4.3512*** /2.6715** /2.9192***
Critical 1%-ADF test /2.6280 /2.7760 /2.6700
statistic
Critical 5%-ADF test /1.9504 /1.9699 /1.9566
statistic
N 40 36 15 13 25 25
H0a rejected YES* YES*** YES*** YES*** YES*** YES***

OLS estimation; ***P B/0.01, **P B/0.05, *P B/0.1.

a
H0: No cointegration between CO2t and Yt .

3.4. Structural models: estimating an EKC for
Austria
After having explored the characteristics of the
different time series, we now turn to an empirical
estimation of the EKC model described above in
Eq. (2). Model 1 (Table 3) is a simple linear model,
which takes the structural break in 1974 into
account. The graph (Fig. 1) shows that for the
first period in particular, a linear model might be
appropriate while in the second period, GDP as a
determinant of CO2 emissions might be weaker.
Thus, we estimate a model, which enables us to
differentiate between the slope of the regression
function between these two periods by including
the dummy variable D75 which takes the value of
1 for the period 1975
/1999. The first column of
Table 3 shows the estimation results (excluding the
variables it and st whose influence we want to test
separately). The coefficient for the variable Yt is
highly significant as are the coefficients for ydt and
Tt which all have the ‘correct’ (theoretically
expected) sign. Accounting for the dummy vari-
able D75 , the estimation shows that the slope in
the second period might be much smaller than in
the first one. However, the coefficient for Yt xD75
is not significant. Estimation of a linear specifica-
tion shows that the dependent variable is mainly
144 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148

Table 3
Estimation of an Environmental Kuznets Curve for Austrian CO2 emissions

Explanatory variables Carbon Dioxide Emissions 1960
/1999

Model 1 (linear) Model 2 (quadratic) Model 3 (cubic) Model 4 (cubic extended)

Constant 5.5915** 3.6924 /68.6600*** /60.1008***

(2.1197) (0.9070) (/4.1447) (/3.5378)
Yt 0.7596*** 1.6320* 18.7849*** 21.1495***
(3.0664) (1.6935) (4.7834) (4.9449)
Y2t 10
2 /0.0367 /1.0791*** /1.1150***
(/1.1852) (/4.5678) (/4.6573)
Y3t /10
5 0.0211*** 0.0219***
(4.4549) (4.4967)
Yt /D75 /0.1280
(/1.3540)
ydt 0.6283*** 0.7029*** 0.7430*** 0.7052***
(2.7706) (3.1466) (4.1415) (3.5594)
Tt /2.1547*** /2.0328*** /2.3404*** /2.4897***
(/3.5577) (/3.3443) (/4.7529) (/5.0236)
it /0.3090*
(/1.6525)
st /0.3613[*]
(/1.7300)
CO2t1 0.7315*** 0.6385*** 0.2969** 0.2117
(10.3888) (5.9182) (2.5675) (1.5792)
Adj R2 0.9575 0.9570 0.9744 0.9794
S.E. of regression 1.9138 1.9254 1.5444 1.4942
Akaike info criterion 4.2736 4.2856 3.8648 3.8361
Schwarz criterion 4.5269 4.5389 4.1603 4.2161
F -statistic 177.0471*** 174.8480*** 229.7808*** 184.6474***
log likelihood /79.4723 /79.7130 /70.2954 /67.7226
Durbin
/Watson stat. 1.9857 1.7325 1.8869 2.0021
n 40 40 40 40

OLS estimation; *** P B/0.01, **P B/0.05, *P B/0.1, [*] P B/0.11.

driven by its own autoregressive term, which is not
very satisfactory for an exploration of the relation-
ship between CO2 and GDP emissions.
Model 2 (Table 3, column 2) presents a direct
test of the EKC hypothesis, i.e. an inverted U-
shaped relation between CO2 and economic
growth (GDP), by assuming a quadratic func-
tional form. The estimation of a quadratic func-
tional form is clearly rejected due to the
insignificant coefficient for the quadratic term
(Yt )2. The estimation is weaker than the linear
model with a coefficient for Yt which is only
significant at the 10% level of significance. Again
the autoregressive term plays an important role in
explaining the dependent variable. The coefficients
for the variables ydt and Tt stay roughly the same,
indicating that these coefficients are stable regard-
less of the functional form chosen.
As the linear and quadratic model are not
satisfactory from an econometric point of view, a
cubic relation between CO2 and GDP is tested in
model 3. This functional form suggests an N-
shaped relation, with increasing CO2 emissions at
the beginning of the period, stabilization (or
reduction) in CO2 emissions in the middle of the
period, and an increase in emissions at the end.
The third column in Table 3 presents the results of
the cubic functional form. The statistical quality of
the estimation is better in terms of the highly
significant coefficients for the Yt -terms, an in-
creased adj. R2 as well as an improved F-statistic
and log likelihood. The size and sign of the
 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
coefficients suggest that a functional form mirror-
ing an N-shaped relation between CO2 emissions
and GDP is appropriate.
Finally, in model 4 two more variables are
included to take account for structural changes
in the Austrian economy. The fourth column of
the table includes an estimation of a cubic form
which also accounts for the import share variable
it and the service intensity variable st . While the
size of the other coefficients stays about the same,
the two variables additionally exhibit significant
influence on the dependent variable. The coeffi-
cient of it is negative and significant at the 10%
level of significance indicating that CO2 emissions
are indeed ‘exported’ by importing goods. How-
ever, this ‘export’ of CO2 emissions is compara-
tively small and only seems to slightly mitigate the
positive effect of GDP on CO2 emissions. One
reason for this surprisingly small impact might be
that Germany, Austria’s main trading partner,
pursues similar environmental objectives to Aus-
tria (and, therefore, does not produce very pollu-
tion-intensive commodities). Likewise, the
structural change in the Austrian economy, i.e.
the growth of the service sector, is only weakly
influential. While the sign of the coefficient of st is
negative as expected, the significance level is only
11% which is only a minor indication that CO2
emissions are also determined by the structural
change. This can partly be traced back to the fact
that carbon emissions are increasingly caused by
(public and private) transport, and freight and
passenger transport are classified as services in the
national accounts. Therefore, although an increas-
ing share of the service sector tends to reduce
emissions in general, the growing contribution of
the transport sector works in the opposite direc-
tion. Summing up, both the influence of the
import share and the service intensity are overlaid
by the main driving forces of GDP growth. But
still, the estimation of a cubic functional form is
the most appropriate one10 in terms of forecasting
10
We also tried other functional forms such as log-linear
and logarithmic, as well as an error-correction model. However,
none of the results proved to be as good as the cubic functional
form presented in Table 3.
145
CO2 emissions as the actual and fitted values are
very close (see Fig. 4).
4. Discussion, summary and conclusions
The first insight gained from exploring the time
series of Austrian CO2 emissions is that the
development of CO2 emissions has to be split
into two different periods. In the first period,
before the oil price shock in the mid-1970s, CO2
emissions followed a strong growth path, reflect-
ing a proportional increase in CO2 emissions with
economic growth. For the second period, from
1975 to 1999, growth of CO2 emissions was
significantly smaller, indicating that the oil price
shock led to significantly altered energy and
environmental policies. However, for the whole
period as well as for the two periods separately, the
time series of CO2 emissions are non-stationary in
the sense that they incorporate growth (i.e. their
mean-reverting tendency is too small to result in a
stationary time series). The structural break in the
mid-1970s is also indicated econometrically by
testing for a breakpoint, resulting in 1974 as the
crucial year. Further tests resulted in a rejection of
the hypothesis of no cointegration between CO2
emissions and GDP. Thus, both variables can be
considered to be cointegrated.
A number of functional forms of the relation
between CO2 emissions and GDP were tested. A
linear functional form including a dummy variable
for the period after 1974 to account for the
structural break in the mid-1970s was unsatisfac-
tory as was the choice of a quadratic form. The
insignificant coefficient for the quadratic term is a
clear rejection of the hypothesis of an inverted U-
shaped EKC for CO2 emissions in Austria. How-
ever, a cubic functional form proved to be the
most appropriate one in terms of econometric
quality. This result suggests that Austrian CO2
emissions are exemplified by an N-shaped relation-
ship with GDP; at least, CO2 emissions of the past
can be described most adequately by such a
functional form.
Two possible explanations can be given for the
N-shaped curve. First, it could merely be a
statistical result since CO2 emissions seem to
146 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
Fig. 4. Actual and fitted CO2 emissions.
have stabilized or at least have been growing only
slowly in recent years.11 This would be true if the
sample happens to fall between the two ‘kinks’
(more mathematically, the two local extreme
values) of the N-shaped relationship (which in
our case might be a very flat section). Therefore, a
projection based on the statistically significant N-
shape would lead to an unrealistic result. This
argument must be underlined since in our estima-
tion the minimum value is well outside current
GDP values (in fact, not only the second ‘kink’ but
also the mathematical turning point lies outside
the data set). Second, the N-shape could be the
result of a ‘recovery effect’ because the initial
shock of the oil crisis in the mid-1970s might have
been reduced after one decade. In other words,
since oil prices are not undergoing permanent
11
Cole et al. (1997) argue in a similar way for cross-country
studies on global pollutants, predicting that the second turning
point (implying the upward slope at high income) is outside the
income range reached so far by the countries under
investigation.
increases nowadays, the pressure to increase en-
ergy efficiency and to implement more stringent
environmental policies is gradually falling. This
second point particularly needs further research.
CO2 emissions do not only depend on economic
growth but are determined by short-term fluctua-
tions in GDP and by climatic conditions. Further-
more, structural changes in a small open economy
such as the Austrian one have regularly been
hypothesized to crucially determine CO2 emis-
sions. We find that the growth of the tertiary
sector (service sector) ceteris paribus reduces CO2
emissions. Imports as an auxiliary measure for
‘exporting’ CO2-emitting industries also reduce
CO2 emissions. However, the main driving force
behind CO2 emissions in Austria is economic
growth. Structural changes can only weaken the
tendency for growth but cannot contribute suffi-
ciently to a mean-reversion process of CO2 emis-
sions, let alone reduce CO2 emissions in times of
economic growth.
This main driving force behind CO2 emissions is
also highlighted by forecasting CO2 emissions
 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
(based on trend-GDP) for the period of 2000
/
2010 (see Fig. 4). If we take the cubic functional
form (upper dotted line) as the appropriate form,
CO2 emissions will grow significantly in the next
few years. However, assuming a linear model
(lower dotted line), CO2 emissions will increase
less. Both projections clearly show that there is no
indication from past energy and environmental
policies that CO2 emissions will decrease*/a
policy target to which Austria has committed itself
on numerous occasions. From a methodological
point of view, the estimated model describes the
time path of the analyzed period most appropri-
ately. But two qualifications should be borne in
mind. First, as indicated above, the data set falls
within the flat part of the curve, between the two
turning points. Therefore, the estimated model
cannot be used for precise quantitative forecasts
for future emissions. Second, the Lucas critique
(Lucas, 1976) is relevant in the sense that, parti-
cularly with partial models such as the one
presented in the current paper, projections of
policies and their outcomes based on past trends
are problematic. However, we clearly find no
indication for a reduction in CO2 emissions in
the future, and at the very least a slight upward
trend.
The main policy conclusion of this paper is that
in order to fulfill international agreements to
reduce CO2 emissions, Austria needs a radically
revised policy framework to achieve a structural
break in the time series of CO2 emissions equiva-
lent to the policy changes following the oil price
shock in the mid-1970s. There is apparently no
‘automatism’ in the time series, which would
suggest that CO2 emissions would fall with higher
GDP. Structural changes like the increasing im-
portance of the service sector or higher import
quotas do not exhibit sufficient influence to
stabilize or reduce CO2 emissions. Thus, policy
changes aiming at ‘policy induced price shocks’,
e.g. by means of an ecological tax reform, seem to
be the most appropriate environmental policy
instruments to cut CO2 emissions since other
instruments applied by Austrian environmental
policies in the past have had no clear growth-
reversing effects on CO2 emissions. In order to
fulfill Austria’s obligations under the Kyoto Pro-
147
tocol, a significant shift in climate policy will have
to take place.
Acknowledgements
We would like to thank Heinz Schandl (Depart-
ment of Social Ecology, IFF, University of Kla-
genfurt) and anonymous referees for helpful
comments on an earlier draft of the paper. Thanks
are also due to the Austrian Central Meteorology
and Geodynamics Institute for providing data on
climate conditions. All errors are, of course, our
responsibility.
References
Arrow, K., Bolin, B., Costanza, R., Dasgupta, P., Folke, C.,
Holling, C.S., Jansson, B.-O., Levin, S., Mäler, K.-G.,
Perrings, C., Pimentel, D., 1995. Economic growth, carrying
capacity and the environment. Science 268, 520
/521.
Borghesi, S., 1999. The Environmental Kuznets Curve: a survey
of the literature. Working Paper 85-99, Fondazione Eni
Enrico Mattei.
Cavlovic, T.A., Baker, K.H., Berrens, R.P., Gawande, K.,
2000. A meta-analysis of environmental Kuznets curve
studies. Agric. Resource Econ. Rev. 29 (1), 32
/42.
Cole, M.A., 2000. Air pollution and ‘dirty’ industries: how and
why does the composition of manufacturing output change
with economic development. Environ. Resource Econ. 17,
109
/123.
Cole, M.A., Rayner, A.J., Bates, J.M., 1997. The environmental
Kuznets curve: an empirical analysis. Environ. Devel. Econ.
2 (4), 401
/416.
De Bruyn, S.M., van den Bergh, J.C.J.M., Opschoor, J.B.,
1998. Economic growth and emissions: reconsidering the
empirical basis of the Environmental Kuznets Curves. Ecol.
Econ. 25 (2), 161
/175.
Dijkgraaf, E., Vollebergh, H.R.J., 1998. Environmental Kuz-
nets revisited. Time-series versus panel estimation: the CO2
case, OcfEB Research Memorandum 9806, Erasmus Uni-
versity Rotterdam.
Ekins, P., 1997. The Kuznets curve for the environment and
economic growth: examining the evidence. Environ. Plan-
ning A 29, 805
/830.
Galeotti, M., Lanza, A., 1999. Desperately seeking (environ-
mental) Kuznets. Mimeo, International Energy Agency,
Paris.
Getzner, M., Glatzer, E., Neck, R., 2001. On the sustainability
of Austrian budgetary policies. Empirica 28 (1), 21
/40.
Harbaugh, W., Levinson, A., Wilson, D., 2000. Re-examining
the empirical basis for the environmental Kuznets curve,
NBER Working Paper, Cambridge, MA.
148 B. Friedl, M. Getzner / Ecological Economics 45 (2003) 133
/148
Heil, M.T., Selden, T.M., 2001. Carbon Emissions and
economic development: future trajectories based on histor-
ical experience. Environ. Dev. Econ. 6 (1), 63
/68.
Hettige, H., Lucas, R.E.B., Wheeler, D., 1992. The toxic
intensity of industrial production: global patterns, trends,
and trade policy. Am. Econ. Rev. 82 (2), 478
/481.
Holtz-Eakin, D., Selden, T.M., 1995. Stoking the fires? CO2
emissions and economic growth. J. Public Econ. 57, 85
/101.
Jänicke, M., Binder, M., Mönch, H., 1997. ‘Dirty industries’:
patterns of change in industrial countries. Environ. Re-
source Econ. 9, 467
/491.
Liski, M., Toppinen, A., 2001. Stochastic versus Deterministic
Trends in Carbon Dioxide Time Series, Working Paper,
Helsinki School of Economics.
Lucas, R., 1976. Econometric policy evaluation: a critique, the
Phillips Curve and Labor Markets. Carnegie Rochester
Conf. 1, 19
/46.
Marland, G., Andres, R.J., Boden, T.A., Johnson, C., Bernkert,
A., 1996. Global, Regional, and National CO2 Estimates
from Fossil Fuel Burning, Cement Production, and Gas
Flaring: 1751
/1996. Carbon Dioxide Information Center,
Oak Ridge, TN.
Moomaw, W.R., Unruh, G.C., 1997. Are Environmental
Kuznets Curves misleading us? The case of CO2 emissions.
Environ. Dev. Econ. 2 (4), 451
/463.
Neck, R., Getzner, M., 2002. Politico-economic determinants
of public debt growth: a case study for Austria. Public
Choice 109 (3
/4), 243
/268.
Panayotou, T., 1997. Demystifying the Environmental Kuznets
Curve: turning a black box into a policy tool. Environ. Dev.
Econ. 2 (4), 465
/484.
Panayotou, T., 2000. Economic growth and the environment.
CID Working Paper 56, Harvard.
Panayotou, T., Peterson, A., Sachs, J., 2000. Is the Environ-
mental Kuznets Curve driven by structural change? What
extended time series may imply for developing countries.
CAER II Discussion Paper 80.
Patel, S.H., Pickney, T.C., Jaeger, W.K., 1995. Smallholder
wood production and population pressure in East Africa:
evidence of an Environmental Kuznets Curve. Land Econ.
71 (4), 516
/530.
Roberts, J.T., Grimes, P.E., 1997. Carbon intensity and
economic development 1962
/91: a brief exploration of the
Environmental Kuznets Curve. World Dev. 25 (2), 191
/
198.
Schandl, H., Hüttler, W., Payer, H., 1999. Delinking of
economic growth and materials turnover. Innovation 12
(1), 31
/45.
Schandl, H., Weisz, H., Petrovic, B., 2000. Materialflussrech-
nung für Österreich 1960 bis 1997. (Material Flow Analysis
for Austria 1960
/1997) Statistische Nachrichten 55 (2),
128
/137 (in German).
Shafik, N., 1994. Economic development and environmental
quality: an econometric analysis. Oxford Econ. Pap. 46,
757
/773.
Shafik, N., Bandyopadhyay, S., 1992. Economic growth and
environmental quality. Background Paper for the 1992
World Development Report, The World Bank, Washing-
ton, DC.
Stern, D.I., Common, M.S., Barbier, E.B., 1996. Economic
growth and environmental degradation: the Environmental
Kuznets Curve and sustainable development. World Dev.
24 (7), 1151
/1160.
Suri, V., Chapman, D., 1998. Economic growth, trade and
energy: implications for the environmental Kuznets curve.
Ecol. Econ. 25 (2), 195
/208.
Tisdell, C., 2001. Globalisation and sustainability: environ-
mental Kuznets curve and the WTO. Ecol. Econ. 39, 185
/
196.
Vincent, J.R., 1997. Testing for Environmental Kuznets Curves
within a developing country. Environ. Dev. Econ. 2 (4),
417
/431.