the environmental kuznets curve reconsidered from the ... · 1 the environmental kuznets curve...
TRANSCRIPT
THE ENVIRONMENTAL KUZNETS CURVE RECONSIDERED FROM THE PERSPECTIVE OF HETEROGENEITY: INSIGHTS FOR CLIMATE CHANGE AND ENERGY POLICY
Documents de travail GREDEG GREDEG Working Papers Series
Thomas JobertFatih KaranfilAnna Tykhonenko
GREDEG WP No. 2012-15
http://www.gredeg.cnrs.fr/working-papers.html
Les opinions exprimées dans la série des Documents de travail GREDEG sont celles des auteurs et ne reflèlent pas nécessairement celles de l’institution. Les documents n’ont pas été soumis à un rapport formel et sont donc inclus dans cette série pour obtenir des commentaires et encourager la discussion. Les droits sur les documents appartiennent aux auteurs.
The views expressed in the GREDEG Working Paper Series are those of the author(s) and do not necessarily reflect those of the institution. The Working Papers have not undergone formal review and approval. Such papers are included in this series to elicit feedback and to encourage debate. Copyright belongs to the author(s).
Groupe de REcherche en Droit, Economie, GestionUMR CNRS 7321
1
The Environmental Kuznets Curve reconsidered from the perspective
of heterogeneity: insights for climate change and energy policy
Thomas Jobert ba, , Fatih Karanfil ,*,dc , Anna Tykhonenko a
a
Nice Sophia Antipolis University, Nice, France. b CREDEG – CNRS,Valbonne, France. c EconomiX – CNRS, University of Paris Ouest, Nanterre, France.
d GIAM, Department of Economics, Galatasaray University, Istanbul, Turkey.
Abstract
Designing an efficient global climate policy turns out to be a difficult yet crucial task since
there are noteworthy cross-country differences in energy and carbon intensities. In this paper,
the environmental Kuznets curve (EKC) hypothesis is tested for carbon dioxide ( 2CO )
emissions and the iterative Bayesian shrinkage procedure is employed to handle the cross-
country differences. The results suggest that: first, the EKC hypothesis is rejected for 49 out
of the 51 countries considered when the heterogeneity in countries’ energy efficiencies and
cross-country differences in the 2CO emissions trajectories are accounted for; second, a
classification of the results, with respect to the development levels of the countries concerned,
reveals that the emergence of an overall inverted U-shape curve is due to the fact that in the
high-income countries, increase in gross domestic product (GDP) decreases emissions, while
in the low-income countries emissions and GDP are positively correlated.
Keywords: Environmental Kuznets curve; Bayesian shrinkage estimator; Heterogeneity
JEL classification: O13; O44; Q56
*Corresponding author at University of Paris Ouest, EconomiX-CNRS, 200 Avenue de la République,
92001 Nanterre cedex, France. Tel.:+33140977815 Fax:+33140974198.
E-mail addresses: [email protected] (T. Jobert), [email protected] (F. Karanfil),
[email protected] (A. Tykhonenko).
2
1. Introduction and previous work
As an introduction, let us cite a few lines from a number of published articles dealing with the
assessment (and reassessment) of the environmental Kuznets curve (henceforth EKC).
Emphasizing the fact that the EKC literature has made so far several significant contributions
in the field of development and environmental economics, Copeland and Taylor (2004)
concluded that their review of this literature, including both theoretical and empirical studies,
leads them “to be skeptical about the existence of a simple and predictable relationship
between pollution and per-capita income.” (Copeland and Taylor, 2004, p. 8). Indeed, the
EKC hypothesis, which postulates in a general sense that the relationship between per capita
pollutant emissions and per capita income has an inverted U-shape, has been revisited in an
increasing number of studies in this research area. However, revisiting the EKC hypothesis
has led to a large number of inconsistent outcomes. In recent years, scholars begun to question
the necessity of further research on the EKC and to claim that “the literature on the EKC is
very large and why, indeed, do we need another paper?” (Johansson and Kriström, 2007, p.
78). Before we respond to this question, it would be useful to discuss different properties and
aspects of the emissions-income relationship in light of the current knowledge on the EKC.1
A concise and purposive review of the literature reveals that, thanks to recent advances in
theoretical and empirical tools, the bulk of the research has been directed towards extending
the basic framework of Grossman and Krueger’s (1991) path breaking study to include a
number of additional variables and equations. In a recent study, Acaravci and Ozturk (2010)
analyzed the relationship between energy consumption, carbon dioxide ( 2CO ) emissions and
gross domestic product (GDP) for 19 European countries using autoregressive distributed lag
bounds testing approach of cointegration and concluded that the EKC hypothesis is validated
only for Denmark and Italy. Very similarly, Apergis and Payne (2010) included energy
3
consumption with GDP as an explanatory variable of 2CO and validated the EKC hypothesis
for the Common wealth of Independent States over the period 1992-2004 employing a panel
vector error correction model. The main reason for introducing energy consumption into the
GDP-2CO nexus is that, first, energy consumption is a key determinant of
2CO emissions,
and has a strong impact on pollution levels, and, second, there is a direct relationship between
energy consumption and economic development “since more energy consumption leads to
higher economic development through the enhancement of productivity” (Ang, 2007, p.4773).
The present paper proposes a novel approach to analyzing the EKC, within the analytical
framework described above, and it takes into account cross-sectional heterogeneity in the
relationship between economic growth, energy consumption and 2CO emissions.
In fact, in the related literature, a wide range of variables has been suggested as being
associated with the EKC hypothesis. For instance, trade intensity or openness of a country is a
very popular variable involved in the EKC studies (Suri and Chapman, 1998; Cole, 2003;
Copeland and Taylor, 2004). The idea in these studies is that trade liberalization may lead to a
decrease in emission levels as countries become more competitive and thus more efficient in
their use of resources. In parallel, a related strand of literature focuses on the pollution haven
hypothesis. This theory posits that the existence of an EKC in developing countries is a
consequence of free trade, which displaces “dirty” industries from developed to developing
countries that have relatively lax environmental standards, regulations and enforcement
mechanisms. However, the empirical evidence of this hypothesis is, once again, inconclusive
(Janicke et al., 1997; Cole, 2004). Following Stern’s (2002) decomposition model, Auci and
Becchetti (2006) offered an “adjusted” EKC model. Their specification considers the
percentage of coal, gas and oil used in electricity generation, as well as the GDP shares of
manufacturing, agriculture and services sectors, in order to account for the output and input
4
mix and the effects of technology on emission levels. They use a fixed effects estimation
technique on a panel dataset of 173 countries and the results indicate that there is very little
evidence of the existence of an EKC. Furthermore, using fixed effects as indicators of
environmental performance, their model also provides a ranking of countries where Eastern
European countries and high income OECD (Organisation for Economic Co-operation and
Development) countries seem to be the worst performers while, in general, Latin American
and Asian emerging countries have the best environmental performances. Furthermore, in
order to obtain country specific results of the EKC, other studies (e.g. Dijkgraaf and
Vollebergh, 2005) have offered to include country fixed effects in the regression models.
However, fixed-effect model has some disadvantages such as the non-stationarity of the data
or the omission of cross-sectional variations.
In another string of literature, polynomial functional form is criticized and nonparametric
specifications are used to investigate the 2CO -GDP nexus. For example, Azomahou et al.
(2006) rejected the EKC hypothesis in favor of a monotonic relationship hypothesis for the
period 1960-1996 on a panel of 100 countries. Halkos and Tsionas (2001) have also reported
a monotonic relationship, on a sample of 61 countries, using switching regime models, which
consider different economic and demographic variables, even very distant ones such as
urbanization and mortality. In a well-written paper, Begun and Eicher (2008) applied the
Bayesian model averaging approach searching for the proper model that could be used to
assess the emissions-income relationship. In addition to the “conventional” regressors, they
also account for some other interesting variables such as education, executive constraints,
precipitation variation and a dummy for nations that signed the Helsinki Protocol in 1985.
Although their overall results do not support the EKC hypothesis in the case of sulphur
dioxide ( 2SO ) emissions, they showed that richer countries, having higher physical and
5
human capital levels and higher trade volumes, also have cleaner environments (and vice
versa for poorer countries). From another but related perspective, Bimonte (2002) used the
number of newspapers (per 1000 people) sold yearly in each country (as a proxy of
information accessibility) and the Gini index (measuring income distribution) as explanatory
variables, along with GDP, and showed that the EKC hypothesis holds true in a sample of 35
European countries. Other interesting studies have looked at extending the EKC hypothesis
examining, for example, the impact of corruption on emissions and on the income level at the
turning point of the EKC (Leitao, 2010), or the relationship between health, environment and
income (Gangadharan et al. 2001).
We provided here-above a set of non-inclusive examples of different model specifications (or
combinations of variables) used in the search for the EKC-type relationships. What we have
understood from the existing literature is that most previous investigators of the EKC
hypothesis attempt to account for other potential variables that may impact (directly or
indirectly) environmental quality, besides the GDP variable. Although these attempts are
somewhat promising, they do not allow to address the main concerns adequately, thus
diverging from the general idea of the EKC analysis. The reasons for including the
aforementioned variables as additional regressors are, to a certain extent, described in these
papers.2 However, these descriptions shift the debate of the form of the environment-income
relationship away from the principal purposes of the EKC hypothesis (i.e. energy and
environmental efficiency), to a vast research field dealing with multiple social, economic and
institutional issues, thus clouding the picture of what is expected to be the shape of this
relationship.
These postulates bring us back to the question that we asked above, that is, whether or not
further analysis of EKC is still needed. Despite the abundance of studies on this topic, this
6
literature seems to be deficient with respect to the heterogeneity of countries in terms of
economic development and energy and environmental efficiency (apart from a few exceptions
using the fixed-effect models, as mentioned above). The reason why little attention has been
given to this issue so far is that commonly used methods in time series and panel data
econometrics do not provide researchers with adequate analytical tools to investigate the EKC
hypothesis in a way that the heterogeneity in countries’ energy efficiencies and cross-country
differences in the 2CO emissions trajectories can be accounted for. This paper attempts to
address this deficiency by using Bayesian shrinkage estimators that can capture this
heterogeneity, revealing global and country-specific information about the economy-energy-
environment relationship. This information is crucial to the effectiveness of global climate
change policies and programs, which depend, evidently, on cross-country differences in the
intensity of 2CO emissions. As addressed in the “common but differentiated convergence”
approach introduced by Hohne et al. (2006), all countries may share the responsibility to
reducing global emissions, but this non-normative obligation may be delayed for non-Annex-I
countries (i.e. mostly developing countries) and it should be conditional to the mitigation
actions of Annex-I countries (i.e. the industrialized countries that were members of the OECD
in 1992, plus countries with economies in transition). Hence, from the EKC hypothesis point
of view, we suggest that countries may follow different emission reduction timetables
depending on the shape of their 2CO -GDP relationship, consistently with the “common but
differentiated responsibilities” principle of the Climate Change Framework Convention
(UNFCCC). As will be shown below, this paper seeks to contribute to the EKC literature by
showing that the shrinkage estimator is an accurate and efficient tool for the analysis of the
differentiated emission profiles in the EKC framework.
7
The outline of the remaining part of this paper is as follows: first, in Section 2 we introduce
the data sets used in the study and perform some descriptive analyses to provide an overview
of energy consumption and 2CO emission trends; some details of the empirical methods
employed are discussed in Section 3; then, the results and their interpretations are presented in
Section 4; finally, in Section 5, we draw our conclusions and further discuss the results.
2. Data and preliminary analysis
2.1. Data description
The variables considered in this study are per capita 2CO emissions, real per capita GDP and
per capita energy consumption. Both 2CO emissions and primary energy consumption data
(in millions tones of 2CO (MtC) and in million tones of oil equivalent, respectively) are taken
from BP (2010).3 Furthermore, data for per capita GDP (in US dollars at constant 1990 prices
and exchange rates) and the data for total population (in thousand) are taken from UNCTAD
(2009). All data are annual, cover the years 1970 to 2008, and extend to 55 countries. The
tables and figures of the present paper use the following country abbreviation: Algeria (ALG),
Argentina (ARG), Australia (AUS), Austria (AUT), Belgium & Luxembourg (BEL), Brazil
(BRZ), Bulgaria (BLG), Canada (CND), Chile (CHL), China (CHN), China Hong Kong SAR
(CHK), Colombia (CLB), Czech Republic (CZE), Denmark (DNK), Ecuador (ECD), Egypt
(EGP), Finland (FIN), France (FRA), Germany (DEU), Greece (GRC), Hungary (HUN),
Iceland (ICL), India (IND), Indonesia (INA), Iran (IRN), Italy (ITL), Japan (JPN), Kuwait
(KUW), Malaysia (MLS), Mexico (MEX), Netherlands (NLD), New Zealand (NZL), Norway
(NRW), Pakistan (PKS), Peru (PER), Philippines (PHI), Poland (POL), Portugal (PRT), Qatar
(QTR), Republic of Ireland (IRL), Romania (ROM), Saudi Arabia (SAR), Singapore (SGP),
South Africa (AFR), South Korea (KOR), Spain (ESP), Sweden (SWE), Switzerland (SWZ),
8
Taiwan (TWN), Thailand (TAI), Turkey (TRK), United Arab Emirates (EMT), United
Kingdom (GBR), United States of America (USA), and Venezuela (VEN).
The sample of 55 countries involved in the analysis covers nearly 90% of global 2CO
emissions. We should mention here that although at present the United Nations has more than
190 member states, we think that it is not necessary to include all of these countries in the
analysis. This is because, on one hand, some countries emit very low levels of 2CO (e.g.
African countries), on the other hand, following the collapse of the USSR in 1991, 15 Soviet
republics became independent countries and if these countries were included in the analysis,
then the study would have to cover the period after 1991. To give some examples of the
magnitude of this exclusion, in 2009, Russian 2CO emissions represented 4.9% of global
2CO emissions while its primary energy consumption was 5.7% of global primary energy
consumption, which is roughly equal to the total primary energy consumed in Middle-Eastern
countries. Similarly, primary energy consumption in both Ukraine and Australia represent 1%
of global consumption, and Ukrainian emissions account for 0.9% of global 2CO emissions
due to fossil fuel combustion.
Some summary statistics on the variables of interest for the countries under analysis are
provided in the Appendix A (Table A.1).
2.2. A first look
Fig. 1 shows the first sign of the existence of an EKC for a sample of 55 countries in the
period considered. Representing per capita 2CO emissions as a function of per capita GDP
seems to create an inverted U-shape curve. Naturally, such a relationship is not surprising, and
it has similar (but not identical) representations in the literature.
9
A more interesting point may be made, in Fig.1, by representing the outlier countries with a
diamond shape and representing the data for the rest of the countries with a diamond-on-
square shape. Then we see clearly that an inverted U-shape curve exists for the 2CO -GDP
relationship, both with and without the outliers, although it is much more evident in the first
case. It should be mentioned that the relative share of the outliers’ (the largest emitters of
2CO per capita in the world, that is, Qatar, United Arab Emirates, Kuwait and Singapore)
primary energy consumption and 2CO emissions is not that important since it represents
roughly 1.7% of global energy consumption and emissions.
Fig. 1. Scatter plot of per capita 2CO emissions (in kg of 2CO ) and per capita GDP (in
constant 1990 US dollars): full sample of 55 countries. Data sources: BP (2010), UNCTAD
(2009).
10
To provide a further preliminary analysis, let us now examine this relationship in a more
analytical manner. In the standard EKC hypothesis testing procedure, the equation to be
estimated is in the following form:
tttt yaybce ε+++= 2)( (1)
where te is an indicator of environmental degradation (in general per capita 2CO emissions),
ty denotes income per capita (per capita GDP) and tε and c represent respectively the
stochastic error term and the constant. The shape of the curve is determined by the parameters
b and a . The idea is that the relationship between per capita 2CO emissions and per capita
GDP may have an inverted U-shape curve if 0>b and 0<a . On the other hand, the turning
point income (henceforth TP), where per capita 2CO emissions reach their maximum level,
can simply be calculated by a
byt
2−= .
In the related literature, Eq. (1) is also used to test the same hypothesis in the case of energy
consumption. So on the left-hand side of Eq. (1), one would introduce energy data instead of
2CO data (e.g. Luzzati and Orsini, 2009). However, in general, Eq. (1) is modified by
introducing, as an additional covariate, energy data on the right-hand side (e.g. Apergis and
Payne, 2010; Pao and Tsai, 2010). In our case, per capita primary energy consumption is
included as an additional variable, that is, we have:
ititiitiitiiit NRJdGDPaGDPbcCO ε++++= 2)(2 (2)
where NRJ represents per capita primary energy consumption. Note that other variables are
also in per capita terms.
Table 1 gives the estimation results when an ordinary least squares (OLS) regression is
applied to Eq. (2) using our data set.
11
Table 1. OLS estimation results
With outliers R² = 0.91 Without outliers R² = 0.77
Variables Coefficient Std.-Error T-Stat. Coefficient Std.-Error T-Stat.
Constant 294 101 2.90 635 100 6.32
GDP 187 20 9.35 300 21.1 14.5
GDP2 -9.05 0.58 -15.41 -11.08 0.59 -18.7
NRJ 2.44 0.02 116.8 1.96 0.039 49.9
Both Fig. 1 and the results given in Table 1 give confirmation of the existence of an EKC for
both 55- and 51-country samples, since all variables are found to be significant with
“expected” signs. Furthermore, as predicted from Fig. 1, the EKC hypothesis seems to be
supported more strongly (having greater R² value) when the outliers are included. Moreover,
one may calculate the TP of the EKC from the estimated coefficients, which is 10.33 with the
outliers, and 13.53 without the outliers.
Evidently this analysis ignores two crucial facts: first, the heterogeneity of countries, since it
assumes that all the countries involved in the analysis are homogenous; second, the un-
stationary nature of the data, since the distribution of test statistics generated by the pooled
OLS regression model is based on the assumption that the data is stationary. The issue is that
if either or both of these assumptions do not hold, biased estimates may result. In
consequence, this first look brings us to the research question we address in this paper, that is,
whether or not there is a lack of robustness to heterogeneity in the EKC analysis. In what
follows, we extend the EKC analysis to the Bayesian shrinkage framework, which allows the
question of interest to be addressed rigorously and the heterogeneity between countries to be
accounted for.
12
3. Estimation methodology
3.1. Theoretical background and relation to the literature
Stern (2004) argues that “the research challenge now is to revisit some of the issues addressed
earlier in the EKC literature using the new decomposition and frontier models and rigorous
panel data and time-series statistics” (Stern, 2004, p. 1435). As indicated by Wagner (2008),
the series of per capita GDP and per capita 2CO emissions are often non-stationary, and this
problem has not been sufficiently addressed in the EKC literature. The author surveyed panel
unit root tests, distinguishing between so-called first generation tests designed for cross-
sectionally independent panels and second generation tests which account for cross-sectional
correlation. In fact, although under the alternative hypothesis of stationarity some tests can be
employed to release the constraint on the coefficient homogeneity, their use may have further
shortcomings.
For example, Im et al. (2003) develop several unit root tests for the model with random
coefficients, in which they loosen the homogeneity constraint imposed on the autoregressive
structure under the alternative hypothesis. So far, because the unit root tests developed for
panel data have been based on individual time-series unit root tests, the interpretation of the
unit root test results in panel data should be made cautiously. In other words, rejection of the
null hypothesis of unit root for the whole sample of countries does not imply that all the
variables are stationary, since it is sufficient to have some series that are stationary, and others
that are not (the series contain a unit root) to reject the null hypothesis. Furthermore,
sometimes introducing one atypical country in the sample may be sufficient for the analysis to
fail to assess the stationarity properties of the entire sample of countries. Using recently
developed tests for unit root and cointegration in panel data, some scholars test for
13
cointegration considering that the EKC estimates will be spurious if there is no cointegration
relation.
However, panel cointegration techniques do not take into account the heterogeneity in the
coefficients of the long-term relationship. These coefficients are assumed to be identical for
all countries in the sample, which implies the existence of a common turning point income
(described above). However, this assumption is not reasonable and some recent approaches
attempt to relax it. Although the system fractional integration and cointegration approaches
adopted by Galeotti et al. (2009) have the advantage of explicitly taking into account
individual heterogeneity, further investigation is needed. It is thus necessary to investigate the
EKC hypothesis in a way that accounts for the heterogeneity in countries’ energy efficiencies
and cross-country differences in 2CO emission trajectories.
On the other hand, recent empirical panel studies pointed to the problem of inconsistent
estimators, due not to non-stationary series but to the insufficient consideration of cross-
country heterogeneity (Vollebergh et al., 2009; Baltagi et al., 2008; Baltagi and Kao, 2000;
Maddala et al., 1997). According to Maddala et al. (1997), in the panel data analysis, it is
customary to pool the observations, with or without individual-specific dummies. These
dummy variables are assumed to be fixed (fixed-effects models, named FE models) or
random (random-effects or variance-components models, named RE models). In RE models,
heterogeneity is modeled through the random effects (individual and temporal) absorbed into
the regression residual term. Recently, Stern (2010) used the between estimator, which,
despite the restrictive assumptions associated with its use (including more specifically the
lack of correlation between the specific effects and the explanatory variables), may be seen as
a consistent estimator of the long-run relationship. But still, this specification imposes that the
slope coefficients of this relationship be common to all countries. Besides, Dijkgraaf and
14
Vollebergh (2005) reject, at the 99% significance level, the null hypothesis of homogeneous
country-specific slopes, which, as the authors state, is at the core of the traditional models.
More generally, as recently emphasized by Vollebergh et al. (2009), the current lack of
robustness of EKC estimation is due to under-identification of the reduced-form hypothesis.
More specifically, the results are dependent on imposed restrictions on the independent and
control variables.
In fact, this problem was already discussed by Maddala et al. (1997), who argued that the
reality is probably situated between complete homogeneity and complete heterogeneity. The
parameters are not perfectly identical, but there is a certain similarity between them. One way
to take into account this similarity is to admit that the parameters are assumed to come from a
common distribution, from the same mathematical expectation, and from the non-zero
variance-covariance matrix. The authors show that the resulting parameter estimates are a
weighted average of the overall pooled estimate and the separate time-series estimates based
on each cross-section. Each individual estimator is thus “shrunk” toward the pooled estimator
(i.e., “shrinkage estimators”). The authors also show that the shrinkage estimator gives more
reasonable parameter values. Hsiao et al. (1999) confirmed that in the case of panel data
model with coefficient heterogeneity, the Bayesian approach performs fairly well, even when
the time dimension is small.4 Maddala and Hu (1996) have also presented some Monte Carlo
evidence to suggest that the iterative procedure gave better estimates (in the mean squared
sense) for panel data models. To conclude, in the Bayesian framework, the panel data models
raise other issues than individual time series, such as a correct consideration of cross-country
homogeneity/heterogeneity. This is the reason why the Bayesian shrinkage estimator can be
considered as an alternative estimation method, capturing cross-sectional heterogeneity in the
economy-energy-environment relationship. Indeed, the estimated TP can henceforth differ
15
from country to country. In this way, the solution relies on the use of a random-coefficient
model in which the parameters are assumed to come from a common distribution.
3.2. The Bayesian shrinkage estimator
Let us consider the Bayesian approach of the Eq. (2), which can be rewritten in the framework
of the random-coefficients model, with the following specification:
iiii uXy += γ (3)
where iy contains 2CO time series, X is the matrix with explanatory variables, and iγ slope
coefficients. In the Bayesian framework, the prior distribution of iγ is given by: iγ ∼ ),( ΣµN
where the parameters µ (mean of iγ ), Σ (variance of iγ ) and 2iσ (residual variance) are
unknown. That is why some assumptions have to be made on prior specification of these
parameters. Then we can derive the posterior distribution for the parameters iγ . On the other
hand, if µ , Σ and 2iσ are all known, the posterior distribution of iγ is normal and calculated
by:
Σ+
′
Σ+
′= −
−
− **ˆ*1*
*1* 1
2
1
12
µγσσ
γ iii
i
ii
ii XXXX (4)
where iγ̂ is the OLS estimator of iγ . The posterior distribution mean of iγ and its variance
are shown in Eqs. (5) and (6) respectively.
∑=
=N
i
iN
1
*1* γµ (5)
1
12
**1*][
−
−
Σ+
′= ii
ii XXV
σγ (6)
16
Since in general, Σ and 2iσ are unknown parameters, one needs to specify priors for them.
For this purpose, Smith (1973) suggested using the mode of the joint posterior distribution
given by the following equations:
−
′−+
++= *)(*)(
21*2
iiiiiiii
i
i XyXyT
γγλςς
σ (7)
and
′−−+
+−−=Σ ∑
=
N
i
iiRkT
1
*)**)(*(2
1* µγµγδ
(8)
where the parameters iς , iλ , δ and R arise from the specification of the prior distributions.
Moreover, Smith (1973) proposed the approximation of these parameters by setting 0=iς ,
1=δ and R as a diagonal matrix with small positive entries (e.g., 0.001). By doing so, the
estimators take the following forms:
−
′−
+= *)(*)(
21*2
iiiiiii XyXyT
γγσ (9)
′−−+
−−=Σ ∑
=
N
i
iiRkT
1
*)**)(*(1
1* µγµγ (10)
Σ+
′
Σ+
′= −
−
− **ˆ*1*
*1* 1
2
1
12
µγσσ
γ iii
i
ii
ii XXXX (11)
and
∑=
=N
i
iN
1
*1* γµ (12)
1
12
**1*][
−
−
Σ+
′= ii
ii XXV
σγ (13)
17
Then Eqs. (9-13) should be solved iteratively, with the initial iteration using the OLS
estimator iγ̂ to compute *µ , *Σ and 2*iσ . The second iteration is based on the empirical
iterative Bayes’ estimator *iγ . The third and following iterations are identical to the second
one. The empirical Bayes’ estimator was proposed by Maddala et al. (1997). The only
difference with Smith’s estimator lies in the computation of the parameters 2*iσ and *Σ , that
is, we have:
*)(*)(1*2iiiiiii XyXy
kTγγσ −
′−
−= (14)
′−−+
−=Σ ∑
=
N
i
iiRN
1
*)**)(*(1
1* µγµγ (15)
4. Empirical results and their interpretations
Before we provide the estimation results, let us discuss very briefly the possible shapes that
the 2CO -GDP nexus can take. For this purpose, consider once again Eq. (2). The sign of the
parameter a determines whether the 2CO -GDP nexus has a concave, convex, or linear
relationship. More specifically, we have three possible cases:
• If a<0, we have an inverted U-shape relationship and the curve is concave. Depending
on the TP (i.e. a
b
2− ) the curve may be: increasing (the TP has not yet been reached);
increasing and decreasing (the TP has been reached and passed); or decreasing (the
TP has been passed and increases in per capita GDP decrease per capita emissions).
• If a>0, we have a U-shape relationship and the curve is convex. The curve may be
decreasing; decreasing and increasing; increasing for the three cases of TP given
above, respectively.
18
• If a=0 the relationship is linear. Depending on the sign of the parameter b, the line may
be increasing (b>0); decreasing (b<0); or horizontal (b=0).
On the other hand, the parameter d measures environmental efficiency of energy use. Its
magnitude reflects whether, in a given country, energy consumption is more or less carbon-
intensive.
4.1. Different forms of the 2CO -GDP relationship
The estimated parameters using Bayesian shrinkage estimators for the model given in Eq. (2)
and the corresponding T-Statistics are reported in Table A.2 in Appendix A.
A first analysis of the results from Table A.2 allows us to make the following three
observations. First, we find a non-linear relationship between per capita 2CO emissions and
GDP for nearly half of the countries considered (24 out of 51 countries). We have a standard
inverted U-shape relationship only for China Hong Kong SAR and Denmark. Second, there is
no significant relationship between these two variables for 21 countries. Third, the variable
NRJ was found to be statistically significant for 95% of the countries.
To test the robustness of our results, we estimated two alternative specifications: in the first,
we introduced the cubic term for income in Eq (2) which is found to be statistically
insignificant (i.e. there is no evidence for an N-shaped relationship); in the second, we
excluded the variable NRJ from the model and the form of the 2CO -GDP relationship found
to be invariant for about 90% of the countries in our sample.5
In order to make the estimation results more readable and easier to interpret we also presented
them in a graphical form (see Figs. 2a and 2b).
19
Fig. 2a. Countries with non-linear relationship based on shrinkage estimators
Fig. 2a should be interpreted as follows. The vertical axis reports the values of d̂ (i.e. the
coefficient associated with the variable of primary energy consumption, NRJ) which is always
positive. The two countries on the left hand side have a standard concave (inverted U-shape)
relationship, and countries on the right side are those that have a convex relationship. The
horizontal axis measures the increasing part of the curve as a percentage of the entire curve.
For each country separately, this percentage is calculated in the following way: first, from the
estimated parameters a and b (see Table A.2) we calculate the TP; then, taking into account
the sign of the coefficients (in order to determine the form of the curve), we count the number
of per capita GDP data points before and after the TP, which is then used to compute the
proportion of increasing and decreasing parts of the curve.6 Fig. 2a therefore gives an
analytical description of the distribution of countries based on shrinkage estimators. A country
with a value of d̂ close to zero uses primary energy sources that are relatively less carbon
intensive. The countries lying on the vertical axis (BLG, CLB, ESP, FRA, POL, USA) also
have a convex relationship. In consequence, three types of countries can be identified from
Fig. 2a:
20
1. On the left quadrant, we have the countries with a standard (concave) EKC. These
countries can be qualified as having a “carbon-saving growth”7.
2. On the right-central quadrant (close to 0) we have the countries with a decreasing
convex curve (i.e. “carbon-saving growth”).
3. On the right quadrant we have the countries with an exponentially increasing
(convex) relationship. These countries have a “carbon-intensive growth”.
Fig. 2b. Countries with a linear relationship based on shrinkage estimators
Let us now consider Fig. 2b. This time, the horizontal axis reports T-Statistics values (i.e. the
estimated coefficient divided by its standard deviation) of the coefficient associated with the
variable GDP per capita, b̂ . Thus, the sign of the T-Statistics is the same as the coefficient.
Therefore, the countries on the left hand side have a negative relationship and those on the
right hand side have a positive relationship. At a confidence interval of 5%, the tabulated
Student statistics value is equal to 1.96, and the countries positioned in the vertical band
between -1.96 and 1.96 are those for which this coefficient is not significant. This implies that
economic growth does not appear to be an explanatory variable for per capita 2CO emissions.
21
Note that, as in Fig 2a, the vertical axis reports the value of the coefficient associated with the
variable of primary energy consumption. The visual information provided by Fig. 2b leads us
to suggest four types of countries:
1. On the left quadrant we have the countries with a linear negative relationship (i.e.
“carbon-saving growth”).
2. On the south-central quadrant (close to 0), we have the countries having no 2CO -
GDP relationship, but using relatively less polluting energy sources. In this case we have
“carbon-neutral growth” and the economies of these countries are more environmentally
friendly.
3. On the north-central quadrant (away from 0), we have once again the countries
without any 2CO -GDP relationship, but using relatively more carbon intensive energy
sources (i.e. “carbon-neutral growth” and less environmental friendliness).
4. On the right quadrant we have the countries having increasing 2CO emissions with
increasing GDP (i.e. “carbon-intensive growth”).
4.2. A tentative classification
Since the EKC hypothesis is investigated to determine whether environmental degradation
depends on the level of economic development, we reinterpreted the results reported above by
taking into account the development levels of each country. According to the standard
classification of countries by levels of economic development, countries fall into five different
categories: developed countries (group 1), transition economies (group 2), newly
industrialized countries of Asia (group 3), new emerging markets and oil exporting countries
(group 4) and least developed countries (group 5).
22
First, let us consider the countries having a 2CO -GDP relationship indicating an “ecological
trend”. In fact, we cannot qualify these countries as “environmentally friendly” because of the
fact that we have a dynamic vision of the evolution of the 2CO -GDP nexus and that we group
countries according to this dynamic relationship irrespective of their 2CO emission levels.
Not very surprisingly, from our results it appears that the countries in group 1 are found to
have the most environmentally friendly dynamic relationship (i.e. the most ecological trend
independent of the scale of emissions). These countries either diversify their primary energy
sources (Norway, Switzerland, Finland, Sweden, Iceland, Austria, Belgium, Luxembourg,
Germany, Canada, France and United Kingdom), or they consume fossil fuels, but also reduce
their 2CO emissions (Denmark and USA). On the other hand, the transition economies (those
in group 2, i.e., Hungary, Czech Republic, Bulgaria, Poland, and Romania) are the countries
that faced a major transition after the disintegration of the Soviet Union in 1991, which lead to
a decrease in their 2CO emissions. Recently, Jobert et al. (2010) argued that during the
transformation of the economic structure, these countries reduced the industrial share of their
GDP and that therefore, they might be qualified as “ecologists despite themselves”. The
results of the present study give further support to this interpretation.
The countries in group 3 (China Hong Kong SAR and Taiwan), having similar economic
growth paths as those of some European countries in the catch up process (such as Republic
of Ireland and Spain), may be considered as having an “ecological trend” since these countries
have directed their development towards low-polluting industries (high technology, service,
finance and tourism).
Diversification of energy sources allowed the countries of group 4 (Argentina, Venezuela and
Colombia) to be more environmentally friendly. In addition, an unexpected result has been
23
obtained for the case of Pakistan. For this group 5 country, per capita 2CO emissions
decrease linearly with an increasing GDP.
Let us now look at the countries having a “polluting trend”. Those in group 1 have neither
diversified their energy sources nor decreased their 2CO emissions (Netherlands, Australia,
New Zealand, Greece, Portugal and Italy). In group 3, South Korea can be considered as a
polluting country because the steel industry and automobile industry are among the country’s
main economic activities.
In other countries from both group 4 (Saudi Arabia, Chile, Malaysia, Brazil, Mexico, Turkey,
South Africa, Algeria, Thailand, Iran, Peru, China and Ecuador) and group 5 (the Philippines
and India), it seems that the energy mix has been somewhat stable over time. Therefore these
countries appear in our analysis as “pollution trend” countries.8 Finally, for other countries,
such as Japan and Indonesia, the results are somewhat inconclusive as to whether these
countries would be qualified as having an “ecological” or “polluting” trend.
5. Discussion and conclusions
In this paper we have revisited the Environmental Kuznets Curve hypothesis taking into
account heterogeneity in countries’ energy efficiencies and cross-country differences in the
emissions trajectories. To do this, we have employed Bayesian shrinkage estimators, for the
first time in this literature. We hope to have clarified how to interpret the fact that the EKC
hypothesis does not hold for individual countries, but emerges from the overall picture (see
Fig. 1). Keeping in mind the results reported and discussed so far, to make the point concrete,
consider as a final illustration, Fig. 3 which provides per capita 2CO emission trends with
respect to per capita GDP in some selected countries with different levels of development.
24
Fig. 3. Selected countries having different trends in both per capita 2CO emissions
(vertical axis; in thousand tones of 2CO ) and per capita GDP (horizontal axis; in
constant 1990 US dollars). Data sources: See Fig. 1.
From Fig. 3 it is quite clear that depending on the development stage, countries have various
per capita 2CO paths, and that chaining individual paths together shows the emergence of
different EKCs in different per capita 2CO and GDP levels, and combining those provides an
overall EKC. This again confirms what we have obtained from the results depicted in Figs. 2a
and 2b.
We conclude with some potential policy implications of our results. Very recently, during the
United Nations Climate Change Conference (Durban, December, 2011), Ban Ki-Moon,
Secretary-General of the United Nations, stated that “we are nearing the "point of no return"”
and that “we must not forsake our collective vision of a comprehensive, binding climate
25
change agreement that is both effective and fair for all” (Ki-moon, 2011). It is evident that
this notion of fairness cannot be discussed without considering differences among countries in
their relative energy and environmental efficiencies. Considering the heterogeneity in the
coefficients of the long-term relationship between per capita emissions and income, our
analysis has led us to the conclusion that both the dynamics of the relationship between these
two variables and the turning point income of the EKC differ from country to country and that
this divergence may limit the efficiency and fairness of global environmental policies. This
finding is consistent with the generally accepted view that policies and measures that help
mitigate climate change should not apply evenly to all countries as national circumstances
may dictate their relative effectiveness (Groenenberg et al., 2001; Philibert and Pershing,
2001). The same opinion was also expressed in the Copenhagen Accord (Copenhagen,
December, 2009), which reminds in Article 1 that global environmental policies should be
designed by taking into account “the principle of common but differentiated responsibilities
and respective capabilities” (UNFCCC, 2009). Furthermore, recognizing the fact that
adaptation to the adverse effects of climate change is a challenging issue for all countries, the
United Nation’s Framework indicates that developed countries should support developing
countries by providing adequate and sustainable financial resources and technology transfer.
Such a cooperation can facilitate the sustainable development of developing countries (i.e. in
our terminology “carbon-saving growth”).
Another issue stems from the fact that some policies and regulations can affect the
competitiveness of States, by encouraging free-riders when other States chose not to adopt
similar climate change policies in order to obtain a competitive advantage. Consequently, the
differentiated efforts may stimulate discussions about unfair competition and carbon leakage
(Azar, 2005; Oikonomou et al., 2006). Finally, turning back to Figs. 2a, 2b and 3, the question
arises whether or not high-income countries reduce their 2CO emissions solely via
26
environmental policies, measures and practices (such as regulations, more efficient use of
energy, investments in abatement technologies, fuel switching or renewable energy facilities)
or by also changing the composition of domestic economic activities by producing high-value
added green products and moving their polluting industries to low-income countries, by
means of pollution haven based investment relocations. To address this question, more focus
is needed on the State level. We hope that further research will continue to explore factors
influencing the shape of the 2CO -GDP relationship as well as climate and energy policy
issues arising from it.
Acknowledgment
We would like to thank Olivia Lauren Dubreuil and two anonymous referees of this journal
for many insightful suggestions that greatly improved the paper.
References
Acaravci, A., Ozturk, I., 2010. On the relationship between energy consumption, 2CO
emissions and economic growth in Europe. Energy 35, 5412-5420.
Ang, J.B., 2CO emissions, energy consumption, and output in France. Energy Policy
35, 4772-4778.
Apergis, N., Payne, J.E., 2010. The emissions, energy consumption and growth nexus:
Evidence from the Common wealth of independent states. Energy Policy 38, 650-655.
Auci, S., Becchetti, L., 2006. The instability of the adjusted and unadjusted
environmental Kuznets curves. Ecological Economics 60, 282-298.
27
Azar, C., 2005. Post-Kyoto climate policy targets: costs and competitiveness
implications. Climate Policy 5, 309-328.
Azomahou, T., Laisney, F., Van, P.N., 2006. Economic development and 2CO
emissions: A nonparametric panel approach. Journal of Public Economics 90, 1347-1363.
Baltagi, B.H., Bresson, G. and Pirotte, A., 2008. To pool or not to pool? in The
Econometrics of Panel Data: Fundamentals and Recent Developments in Theory and Practice,
(L. Mátyás and P. Sevestre eds.). Series: Advanced Studies in Theoretical and Applied
Econometrics 33. Springer-Verlag, New York.
Baltagi, B.H., Kao, C., 2000. Nonstationary Panels, Cointegration in Panels and
Dynamic Panels: a Survey. In: Baltagi, B.H. (Eds.), Advances in Econometrics 15, Elsevier
Science, 7-51.
BP, 2010. Statistical Review of World Energy 2010.http://www.bp.com/statisticalreview
Begun, J., Eicher, T.S., 2008. In search of an environmental Kuznets curve in sulphur
dioxide concentrations: a Bayesian model averaging approach. Environment and
Development Economics 13, 795-822.
Bimonte, S., 2002. Information access, income distribution, and the Environmental
Kuznets Curve. Ecological Economics 41, 145-156.
Cole, M.A., 2003. Development, trade, and the environment: how robust is the
Environmental Kuznets Curve? Environment and Development Economics 8, 557-580.
Cole, M.A., 2004. Trade, the pollution haven hypothesis and the environmental Kuznets
curve: examining the linkages. Ecological Economics 48, 71-81.
Copeland, B.R., Taylor, M.S., 2004. Trade, Growth, and the Environment. Journal of
Economic Literature 42, 7-71.
28
Dijkgraaf, E., Vollebergh, H.R.J., 2005. A Test for Parameter Homogeneity in 2CO
Panel EKC Estimations, Environmental and Resource Economics 32, 229-239.
Galeotti, M., Manera, M., Lanza, A., 2009. On the Robustness Checks of the
Environmental Kuznets Curve Hypothesis. Environmental and Resource Economics 42, 551-
574.
Gangadharan, L., Valenzuela, M.R., 2001. Interrelationships between income, health
and the environment: extending the Environmental Kuznets Curve hypothesis. Ecological
Economics 36, 513-531.
Gerlagh, R., 2008. A climate-change policy induced shift from innovations in carbon-
energy production to carbon-energy savings. Energy Economics 30, 425-448.
Grossmann, G.M., Krueger, A.B., 1991. Environmental impacts of a North American
free trade agreement. NBER Working paper No. 3914.
Groenenberg, H., Phylipsen, D., Blok, K., 2001. Differentiating commitments world
wide: global differentiation of GHG emissions reductions based on the Triptych approach-a
preliminary assessment. Energy Policy 29, 1007-1030.
Halkos, G.E., Tsionas, E.G., 2001. Environmental Kuznets curves: Bayesian evidence
from switching regime models. Energy Economics 23. 191-210.
Hohne, N., Den Elzen, M.G.J., Weiss, M., 2006. Common but differentiated
convergence (CDC), a new conceptual approach to long-term climate policy. Climate Policy
6, 181-199
Hsiao, C., Pesaran, M.H. and Tahmiscioglu, A.K., 1999. Bayes Estimation of Short-Run
Coefficients in Dynamic Panel Data Models, in C. Hsiao, K. Lahiri, L.-F. Lee, and M.H.
29
Pesaran (eds.), Analysis of Panels and Limited Dependent Variables: A Volume in Honour of
G. S. Maddala, Cambridge University Press, pp. 268-296.
Im, K. S., Pesaran, M. H. and Shin, Y. 2003. Testing for unit roots in heterogeneous
panels. Journal of Econometrics 115, 53-74.
International Energy Agency (IEA), 2010a. Energy Balances of OECD Countries, 2010
Edition, Paris.
International Energy Agency (IEA), 2010b. Energy Balances of non-OECD Countries,
2010 Edition, Paris.
International Energy Agency (IEA), 2010c. 2CO emissions from fuel combustion, 2010
Edition, Paris.
Janicke, M., Binder, M., Monch, H., 1997. ‘Dirty industries’: patterns of change in
industrial countries. Environmental and Resource Economics 9, 467-491.
Jobert, T., Karanfil, F., Tykhonenko, A., 2010. Convergence of per capita carbon
dioxide emissions in the EU: Legend or reality? Energy Economics 32, 1364-1373.
Johansson, P.-O., Kriström, B., 2007. On a clear day you might see an environmental
Kuznets curve. Environmental and Resource Economics 37, 77-90.
Kijima, M., Nishide, K., Ohyama, A., 2010. Economic models for the environmental
Kuznets curve: A survey. Journal of Economic Dynamics and Control 34, 1187-1201.
Ki-moon, B., 2011. Statement made at the opening of the High-Level Segment at COP
17/CMP 7. United Nations Climate Change Conference, Durban, South Africa. Available at:
http://unfccc.int/meetings/durban_nov_2011/statements/items/6584.php.
Leitao, A., 2010. Corruption and the environmental Kuznets Curve: Empirical evidence
for sulfur. Ecological Economics 69, 2191-2201.
30
Luzzati, T., Orsini, M., 2009. Investigating the energy-environmental Kuznets curve.
Energy 34, 291-300.
Maddala, G. S., Hu, W., 1996. The Pooling Problem. In: Matyas, L., Sevestre, P. (Eds.),
The Econometrics of Panel Data: a Handbook of Theory with Applications, Kluwer Academic
Publishers, 2nd Ed., Boston, 307-322.
Maddala, G. S, Trost, R. P., Li, H., Joutz, F., 1997. Estimation of Short-Run and Long-
Run Elasticities of Energy Demand From Panel Data Using Shrinkage Estimators. Journal of
Business and Economic Statistics 15, 90-100.
Oikonomou, V., Patel, M., Worrell, E., 2006. Climate policy: Bucket or drainer? Energy
Policy 34, 3656-3668.
Pao, H.-T., Tsai, C.-M., 2010. 2CO emissions, energy consumption and economic
growth in BRIC countries. Energy Policy 38, 7850-7860.
Philibert, C., Pershing, J., 2001. Considering the Options: Climate Targets for All
Countries. Climate Policy 1, 211-227.
Smith, A. F., 1973. A General Bayesian Linear Model. Journal of the Royal Statistical
Society, Ser. B, 35, 67-75.
Stern, D.I., 2002. Explaining changes in global sulphur emissions: an econometric
decomposition approach. Ecological Economics 42, 201-220.
Stern, D.I., 2004. The Rise and Fall of the Environmental Kuznets Curve. World
Development 32,1419-1439.
Stern, D.I., 2010. Between estimates of the emissions-income elasticity. Ecological
Economics 69, 2173-2182.
31
Suri, V., Chapman, D., 1998. Economic growth, trade and energy: implications for the
environmental Kuznets curve. Ecological Economics 25, 195-208.
United Nations Conference on trade and Development (UNCTAD), 2009. Handbook of
Statistics. http://stats.unctad.org/Handbook/ReportFolders/reportFolders.aspx
United Nations Framework Convention on Climate Change (UNFCCC), 2009.
Copenhagen accord. Available at: http://unfccc.int/resource/docs/2009/cop15/eng/l07.pdf.
Vollebergh, H.R.J., Melenberg, B., Dijkgraaf, E., 2009. Identifying reduced-form
relations with panel data: The case of pollution and income. Journal of Environmental
Economics and Management 58, 27-42.
Wagner, M., 2008. The carbon Kuznets curve: A cloudy picture emitted by bad
econometrics? Resource and Energy Economics 30, 388-408.
Footnotes
1 The emphasis in this review is on understanding how the EKC literature triggered a chain of analysis during the
last two decades. A number of excellent surveys have appeared on the subject in the last years and they provided
useful overviews of the wide range of methods and modeling techniques used in this field. For example, Stern
(2004) and Kijima et al. (2010) should be considered for a more complete review.
2 To keep the review brief, we do not discuss in detail the reasons why these variables are chosen to be
controlled for in the EKC analysis. The reader is referred to the studies cited here for further discussion.
3 BP (2010) uses standard global average conversion factors to estimate carbon emissions. The International
Energy Agency (IEA) provides also data for 2CO emissions from fuel combustion, which are calculated using
the intergovernmental panel on climate change (IPCC) method. Consequently, these two data sets have very
similar trends and magnitudes; therefore, working with either BP or IEA data set does not have a significant
impact on the estimation results of this study.
32
4 In our study the individual dimension (N=51) is more important than the time dimension (T=39).
5 For the purpose of brevity, these estimation results are not reported in this paper, but they are available upon
request from the authors.
6 At this point we note that this method works well for all countries but one, Egypt, for which the TP is found to
be negative. Since such a result is inconsistent with the nature of the relationship, Egypt is excluded from the
later analysis.
7 We employ the terminology used in the literature on technical change (for instance, see Gerlagh, 2008): by the
term “carbon-saving growth” (“carbon-intensive growth”) it is meant that increases in GDP are associated with
decreases (increases) in 2CO emissions; the term “carbon-neutral growth” is used to indicate that GDP growth
has not a significant impact on the carbon intensity of the economy.
8 The reports of the International Energy Agency constitute a very useful source of information about energy
indicators and emission trends. For detailed statistics and further analysis see IEA (2010a, b, c).
33
Appendix A.
Table A.1. Summary statistics of the full sample of 55 countries
Years 1970 1990 2008
Percentage of the world population 76.8 75.6 73.8
Percentage of the world GDP 92.6 94.1 97.6
Percentage of global 2CO emissions 80.8 78.9 87.5
Percentage of global primary energy consumption 81.5 78.6 86.3
Data sources: BP (2010), UNCTAD (2009)
Variable Country Coeff. T-Stat Country Coeff. T-Stat Country Coeff. T-Stat
Const ALG 169.09 0.98 FIN 9654.79 7.29 PER -448.40 -1.74 GDP -126.56 -0.88 -953.38 -5.53 -95.78 -0.34 GDP^2 35.24 1.22 13.68 5.28 -39.07 -0.55 NRJ 2.54 165.86 3.47 7.10 3.90 7.31 Const ARG 2189.53 2.27 FRA 15331.47 6.24 PHI -37.48 -1.05 GDP 182.88 0.57 -1212.55 -4.85 -116.75 -1.26 GDP^2 16.07 0.57 18.95 3.29 54.86 0.95 NRJ -0.08 -0.41 2.42 4.00 3.29 69.70 Const AUS -2362.51 -2.19 DEU 13585.95 5.74 POL 355.49 3.78 GDP 153.68 1.00 -1580.19 -6.24 -660.05 -8.20 GDP^2 -2.50 -0.80 29.74 5.23 79.32 4.49 NRJ 3.26 20.81 4.60 19.65 3.88 289.51 Const AUT 5454.15 7.54 GRC 380.00 0.75 PRT -1197.87 -2.54 GDP -580.42 -6.49 -110.89 -1.00 278.84 1.84 GDP^2 11.73 6.56 -1.39 -0.31 -9.83 -1.03 NRJ 2.59 8.32 3.58 41.48 2.57 7.53 Const BEL 11573.89 5.89 HUN 2325.32 2.54 IRL 26.41 0.19 GDP -1141.59 -9.16 -1164.19 -2.34 -63.75 -2.07 GDP^2 20.03 5.80 57.79 0.91 0.64 1.13 NRJ 3.24 13.31 3.21 7.77 3.31 22.99 Const BRZ -19.54 -0.10 ICL 5075.51 1.69 ROM -603.72 -3.65 GDP 327.64 2.43 -136.36 -0.66 -380.27 -1.94 GDP^2 -36.29 -1.61 6.04 1.35 64.21 1.07 NRJ 1.23 8.23 0.48 3.29 3.16 84.93 Const BLG 1712.44 4.97 IND -6.79 -0.85 SAR -1928.97 -2.38 GDP -2564.81 -9.28 -27.69 -0.33 704.56 4.24 GDP^2 365.30 6.08 6.41 0.14 -33.62 -4.08 NRJ 3.55 31.95 3.31 33.57 2.55 84.40 Const CND 17368.41 6.80 INA 98.94 5.83 AFR 146.29 0.19 GDP -1553.96 -4.29 -481.03 -6.99 -20.10 -0.04 GDP^2 32.22 4.38 247.42 8.38 20.40 0.28 NRJ 2.08 4.75 3.22 40.89 3.57 90.20
Table A.2. Shrinkage estimators state by state (number of iterations: 5)
34
Variable Country Coeff. T-Stat Country Coeff. T-Stat Country Coeff. T-Stat
Const CHL 173.45 0.63 IRN 415.51 2.11 KOR 458.81 4.07 GDP -161.64 -0.63 -64.65 -0.34 -49.50 -0.63 GDP^2 8.19 0.29 18.69 0.43 -2.95 -1.45 NRJ 2.67 5.91 2.50 153.73 2.90 20.22 Const CHN -16.94 -1.31 ITL -172.27 -0.32 ESP 2528.60 4.50 GDP -165.65 -6.07 49.82 1.44 -499.39 -4.43 GDP^2 -49.32 -4.74 -1.88 -1.72 10.55 3.53 NRJ 3.75 102.90 2.74 18.48 3.55 13.00 Const CHK -739.65 -2.86 JPN 5177.09 6.29 SWE 34241.49 13.07 GDP 148.96 3.52 -380.17 -6.83 -1901.88 -7.24 GDP^2 -4.57 -2.91 7.30 5.05 27.29 6.15 NRJ 3.14 13.75 2.59 14.08 0.89 2.02 Const CLB 804.36 6.72 MLS -54.31 -0.57 SWZ 6548.44 1.99 GDP -1009.19 -4.91 185.48 1.81 -26.39 -0.12 GDP^2 130.35 1.99 -2.78 -0.26 -0.80 -0.25 NRJ 2.86 19.41 2.38 24.29 0.41 1.82 Const CZE -859.07 -0.97 MEX -735.46 -3.05 TWN 323.73 3.63 GDP -1206.07 -3.29 430.94 2.53 -414.54 -8.46 GDP^2 -2.36 -0.05 -75.36 -3.48 13.88 12.33 NRJ 4.44 48.36 2.84 30.79 3.63 22.50 Const DNK -5381.28 -5.26 NLD 2415.15 3.10 TAI -53.11 -2.55 GDP 505.78 7.25 -126.35 -1.92 112.99 3.28 GDP^2 -9.83 -7.63 4.09 2.58 -5.17 -0.45 NRJ 3.01 26.69 2.52 26.08 2.75 43.33 Const ECD -69.92 -0.75 NZL 5556.06 1.83 TRK 200.90 0.71 GDP 291.69 1.86 -721.12 -1.53 -26.67 -0.11 GDP^2 18.23 0.30 32.09 2.05 -10.14 -0.37 NRJ 1.98 24.52 1.51 8.95 2.88 11.27 Const EGP -29.35 -1.50 NRW 5101.85 6.50 GBR 7570.06 6.25 GDP -92.56 -0.70 173.21 1.88 -769.75 -8.62 GDP^2 -148.24 -3.15 -1.41 -1.05 16.64 6.73 NRJ 2.97 25.42 -0.08 -0.53 3.04 11.94 Const USA 2882.47 3.27 PKS 59.39 2.75 VEN 1297.52 1.64 GDP -185.04 -3.77 -385.80 -2.86 405.13 0.91 GDP^2 3.06 2.81 88.76 1.58 34.74 0.43 NRJ 2.76 35.15 2.85 17.07 1.15 7.47
Table A.2. (Continued) Shrinkage estimators state by state (number of iterations: 5)
DOCUMENTS DE TRAVAIL GREDEG PARUS EN 2012GREDEG Working Papers Released in 2012
2012-01 Frédéric Marty Les clauses environnementales dans les marchés publics : perspectives économiques2012-02 Christophe Charlier Distrust and Barriers to International Trade in Food Products: An Analysis of the US- Poultry Dispute2012-03 Gérard Mondello The Equivalence of Strict Liability and Negligence Rule: A Trompe-l’Œil Perspective2012-04 Agnès Festré & Pierre Garrouste Somebody May Scold You! A Dictator Experiment2012-05 Dorian Jullien & Nicolas Vallois A Probabilistic Ghost in the Experimental Machine2012-06 Gérard Mondello Ambiguity, Agency Relationships, and Adverse Selection2012-07 Flora Bellone, Kozo Kiyota, Toshiyuki Matsuura, Patrick Musso & Lionel Nesta Comparative Advantage, Trade Costs, and International Productivity Gap: Evidence Built from French and Japanese Firm-level Data2012-08 Tiziana Assenza & Domenico Delli Gatti E. Pluribus Unum: Macroeconomic Modelling for Multi-Agent Economies2012-09 Catherine Guillemineau Financial Reforms, International Financial Flows, and Growth in Advanced Economies2012-10 Dino Borie Social Decision Theory and Non-Strategic Behaviour2012-11 Edward Lorenz Social Capital and Enterprise Innovative Performance: A Multi-Level Analysis of Developing Nations2012-12 Agnès Festré & Pierre Garrouste The ‘Economics of Attention’: A New Avenue of Research in Cognitive Economics2012-13 Harald Hagemann Capitalist Development, Innovations, Business Cycles and Unemployment: Joseph Alois Schumpeter and Emil Hans Lederer2012-14 Sophie Pommet The Survival of Venture Capital Backed Companies: An Analysis of the French Case2012-15 Thomas Jobert, Fatih Karanfil & Anna Tykhonenko The Environmental Kuznets Curve Reconsidered from the Perspective of Heterogeneity: Insights for Climate Change and Energy Policy2012-16 Francesco Quatraro The Co-Evolution of Knowledge and Economic Structure: Evidence from European Regions
DOCUMENTS DE TRAVAIL GREDEG PARUS EN 2012GREDEG Working Papers Released in 2012
2012-01 Frédéric Marty Les clauses environnementales dans les marchés publics : perspectives économiques2012-02 Christophe Charlier Distrust and Barriers to International Trade in Food Products: An Analysis of the US- Poultry Dispute2012-03 Gérard Mondello The Equivalence of Strict Liability and Negligence Rule: A Trompe-l’Œil Perspective2012-04 Agnès Festré & Pierre Garrouste Somebody May Scold You! A Dictator Experiment2012-05 Dorian Jullien & Nicolas Vallois A Probabilistic Ghost in the Experimental Machine2012-06 Gérard Mondello Ambiguity, Agency Relationships, and Adverse Selection2012-07 Flora Bellone, Kozo Kiyota, Toshiyuki Matsuura, Patrick Musso & Lionel Nesta International Productivity Gaps and the Export Status of Firms: Evidence from France and Japan2012-08 Tiziana Assenza & Domenico Delli Gatti E. Pluribus Unum: Macroeconomic Modelling for Multi-Agent Economies2012-09 Catherine Guillemineau Financial Reforms, International Financial Flows, and Growth in Advanced Economies2012-10 Dino Borie Social Decision Theory and Non-Strategic Behaviour2012-11 Edward Lorenz Social Capital and Enterprise Innovative Performance: A Multi-Level Analysis of Developing Nations2012-12 Agnès Festré & Pierre Garrouste The ‘Economics of Attention’: A New Avenue of Research in Cognitive Economics2012-13 Harald Hagemann Capitalist Development, Innovations, Business Cycles and Unemployment: Joseph Alois Schumpeter and Emil Hans Lederer2012-14 Sophie Pommet The Survival of Venture Capital Backed Companies: An Analysis of the French Case2012-15 Thomas Jobert, Fatih Karanfil & Anna Tykhonenko The Environmental Kuznets Curve Reconsidered from the Perspective of Heterogeneity: Insights for Climate Change and Energy Policy