CO2 Kuznets hypothesis: from cross

CO2 Kuznets hypothesis: from cross-section to panel data
structure
Autores y e-mail de la persona de contacto:
Ana Angulo (aangulo@unizar.es), Majed Atwi, Jesús Mur, Ramón Barberán
Departamento: Análisis Económico
Universidad: Zaragoza
Área Temática: Energía, sostenibilidad, recursos naturales y medio ambiente
Resumen: (máximo 300 palabras)
The Environmental Kuznets Curve (EKC) hypothesis conjectures that environmental
degradation initially intensifies when a country’s per capita income increases and
subsequently subsides after a certain level of income is reached, resulting in an inverted
U-shaped relationship between environmental degradation and per capita income.
There is abundant literature on this topic, produced especially in the last 20 years, that
corroborates the existence of a positive income elasticity for environmental quality.
However, results are controversial, because results also depend on the type of pollutant
under analysis, the methodology or the data used.
In this study, we analyse this last issue, since we compare results obtained for a crosssectional data with the results derived from a panel data set. Furthermore, we may
special attention also to the spatial interaction effect inherent to the problem. Main
results are that while cross-sectional results show evidence in favor of the EKC
hypothesis, when we take into account the unobserved heterogeneity through the
consideration of panel data models, the EKC relationship is not supported anymore.
Palabras Clave: (máximo 6 palabras) Kuznets Environmental Curve; CO2 emissions;
Spatial effects; Nonlinearities; Structural breaks.
Clasificación JEL: Q25, L83
1
1. Introduction
Pollution induced by human activities is a major threat and serious issue to
sustainable growth and development in the world. In this sense, the Fifth Assessment
Report from the Intergovernmental Panel on Climate Change (IPCC, 2013-Working
Group I; IPCC, 2014)) states that “human influence on the climate system is clear. It is
extremely likely that human influence has been the dominant cause of the observed
warming since the mid-20th century”.
We can classify pollutants in terms of the local or global externalities they generate.
Local pollutants cause clear local health effect by affecting water or air local conditions.
Among local pollutants we can quote sulphur oxides, nitrogen oxides, suspended
particulate matter, lead or carbon monoxide, among others. By contrast, the damage of
global pollutants is less immediate and less evident to the society since they are locally
innocuous, but they impact the global environment over the long term. The main global
pollutant is carbon dioxide (C02).
The first set of influential empirical studies were conducted by Grossman and
Krueger (1991, 1993, 1995), Shafik and Bandyopadhay (1992, 1994), Panayotou (1993,
1995), Selden and Song (1994) and Arrow et al. (1995). Since then, over the last two
decades, an explosion of studies concerning the relationship between environmental
degradation and economic growth has appeared in the literature. The vast majority of
papers have focused on the testing of the hypothesized inverted-U shaped between
environmental degradation and economic growth, known as the Environmental Kuznets
Curve (EKC) by the analogy with the income-inequality relationship postulated by
Kuznets (1965, 1966). Specifically, they have focused on the relationship between per
capita income and a variety of environmental indicators: CO2, SO2 and water pollutant,
among others.
2
The EKC hypothesis attempts to explain a long term relationship between
environmental degradation and economic growth. It describes that environmental
degradation tends to increase more rapidly than economic growth in the early stages,
thereafter declines after reaching a certain level of economic growth. Panayotou (1993)
gives the following explanation concerning the EKC: “At low levels of development,
both the quantity and the intensity of environmental degradation are limited to the
impacts of subsistence economic activity on the resource base and to limited quantities
of biodegradable wastes. As agriculture and resource extraction intensify and
industrialization takes off, both resource depletion and waste generation accelerate. At
higher levels of development, structural change towards information-based industries
and services, more efficient technologies, and increased demand for environmental
quality result in levelling-off and a steady decline of environmental degradation”. In the
same line, Arrow et al. (1995) gives a similar explanation that the EKC pattern reflects
the natural progression of economic development, from clean agrarian economies to
polluting industrial economies to clean service economies.
Despite the existence of an important theoretical literature analysing the
Environmental Kuznets Curve, the theoretical framework is still ambiguous. The
inverted-U shaped relationship is mostly considered in the literature as an empirical
phenomenon of mostly ad hoc specifications and estimation of a reduced form model
relating an environmental impact indicator to income per capita. Thus, the robustness of
the EKC relationship is questioned by many in the literature arguing that it is sensitive
to the pollutants type, the data and analysing techniques used. Hence, many studies have
arrived at different results regarding the form of this relationship.
In general terms, EKC relationship has often been confirmed for several local
pollutants such as sulphur oxides, nitrogen oxides, suspended particulate matter, lead,
carbon monoxide, water, and land use such as Grossman and Krueger (1991, 1995);
Shafik and Bandyopadhay (1992); Panayotou (1993, 1995); Shafik (1994); Selden and
Song (1994); Holtz-Eakin and Selden (1995); Tucker (1995); Vincent (1997); Cole et
al. (1997); Carson et al. (1997); Ansuategi et al. (1998); Kaufman et al. (1998);; List
3
and Gallet (1999); Hill and Magnani (2002); List and Gerking (2000); Millimet et al.
(2003); Perrings and Ansuategi (2000); Stern and Common (2001), among others.
Nevertheless, the inverted-U relationship (EKC relationship) is less likely for global
pollutant as CO2 emissions. In this context, many studies have been identified a
monotonically increasing relationship between CO2 emissions and income per capita.
These authors emphasized that and in case that relationship shows a declining stage
with higher income, the turning points are reached at very high incomes (well outside
the range of incomes in the studies’ samples). Hence, most of the countries, especially
low income countries, will not be able to reach those levels of incomes, at least, in a
reasonable (short) period. In Panayotou (2000)’s woods: “It may take decades for a lowincome country to cross from the upward to the downward sloping part of the curve, the
accumulated damages in the meanwhile may far exceed the present value of higher
future growth”. Some authors supporting this view are the following: Cole et al. (1997),
Roberts and Grimes (1997), de Bruyn et al. (1998), Hill and Magnani (2002), Carlsson
and Lundström (2001), Talukdar and Meisner (2001), Dijkgraaf and Vollebergh (2001),
Heil and Selden (2001), Roca et al. (2001), Heenrink et al. (2001), Magnani (2001),
Azar (2002), Lindmark (2002), Coondoo and Dinda (2002), Bruvoll and Medin (2003),
Aldy (2006) or Wagner (2008). Within this framework, for instance, Azar (2002)
suggests that CO2 emissions are much more difficult to decouple from income because
the impacts of these emissions are distant in time and space and the political pressure to
do something about them is weak. Similarly, Aldy (2006) concluded that most
environmental Kuznets curve (EKC) theories do not apply to carbon dioxide because it
is an unregulated, invisible, odourless gas with no direct human health effects. Finally,
it is remarkable the work by Wagner (2008) who argues that “We use the important
special case of the relationship between GDP and CO2 (and SO2) emissions to show
and discuss in detail that the seemingly strong evidence for an inverted U-shaped
relationship between these variables obtained with commonly used methods is entirely
spurious and vanishes when resorting to estimation strategies that take the discussed
problems into account.”
4
In the line of previous works, in this paper we will try to offer empirical evidence in
favour or against the EKC theory in the case of CO2, making use of the recent
development in modelling strategies for spatial panel data. More precisely, we propose
to study the EKC relationship from the simpler cross-sectional model to a more
complete spatial panel model. With this exercise, we will show whether the EKC theory
applies in the case of CO2 or it vanishes when we take into account important issues
such as the effect of omitted variables, spatial dependence or technological changes,
among others.
The structure of the paper is as follows. In the next section, a picture of data is
offered. Then, the third section is devoted to the methodology applied in the paper.
Next, we show the obtained results. Finally, paper concludes with a summary of
conclusions and some draws on future research.
2. Data
Data per capita C02 emissions and per capita GDP are gathered for a panel of
182 countries over the period 1992-2011. The sources of data are the following: i) the
webpage
of
the
United
States
Energy
Information
Administration
(EIA),
(http://www.eia.gov), for per capita C02 emissions data set; and ii) the webpage of the
Organization of United Nations (http://www.un.org), for per capita GDP data.
The distribution of per capita C02 emissions and per capita GDP in 2011 is
shown, in terms of the quantiles of each distribution, in Figures 1 and 2, respectively.
As shown in legends, a darker colour indicates a higher value for the respective variable
in the corresponding country.
(Insert Figures 1 and 2)
From both figures, a clear positive correlation between both variables seems to
exist. In fact, USA, Canada, Saudi Arabia, central Europe and Australia are some of the
countries that are located at the fourth quantile of both distributions. On the opposite
side, most of the countries in central Africa locate at the first quantile of both
5
distributions. Finally, the vast majority of the rest of countries are located in similar
central position in both distributions.
The apparent positive spatial autocorrelation deduced from previous maps is
corroborated by means of three global autocorrelation statistics, commonly used in
literature: Moran’s I, Geary’s C and Getis and Ord’s G. Results for all of them,
calculated considering that one country is mainly affected by its five nearest countries in
space, are gathered in Table 1.
(Insert Table 1)
3. Methodology
As indicated previously, our main objective is to obtain empirical evidence in favour
or against the EKC relationship between CO2 emission and per capita income, paying
special attention to the robustness of the results as data and/or econometric strategy
change. To cope with this objective, we start by the estimation of the simpler model
with data referred to the set of countries in 2011 (the last period of the sample):
Y  X  
(1)
where Y is the (Rx1) vector of the logarithm of CO2 per capita emissions (ln ei for
i=1…R); X is an (Rxk) matrix of covariances, together with a vector of ones to account
for a constant term (  ln yi
 ln yi 
2
for i=1…R), being ln yi is the logarithm of per
capita GDP in country i; the vector of parameter are   (, 1 , 2 ) ; and i is the
stochastic error term. From (1), an inverted U shape is present if 1  0 and 2  0 and
the inflexion point with respect to per capita GDP is in the following point:
y  exp(1 / 22 ) .
6
However, from previous section, we concluded on the clear spatial dependence
inherent in our variables. Hence, the next step in our strategy deals with testing the null
of no spatial autocorrelation in the residuals of our OLS regression. If statistics confirm
the presence of spatial dependence, we will conclude on the proper specification for our
data following the general to specific strategy shown in Figure 3. To cope with this
objective, the connectivity matrix W, is defined as the row-standardization of the fivenearest neighbours binary matrix.
(Insert Figure 3)
As shown in Figure 3, our starting point in model selection strategy is the CliffOrd model, defined as follows:
Y  WY  X  WX  u
u  Wu  
(2)
which includes all types of interaction effects: i) the endogenous interaction effect, WY,
to consider the possibility that emission in country i could depend on emission in
another country j and vice versa; ii) the exogenous interaction effect, WX, to account
for the fact that emission in country i could depend on per capita income of its
neighbour countries; and iii) interaction effect among the error terms, since the
determinant of emissions omitted from the model can be spatially autocorrelated or the
unobserved shocks can follow a spatial pattern.
From the general Cliff-Ord model, we will test if some simpler model such as
the SARAR, the Spatial Durbin model, the SLM or the SEM models could be supported
for our data. Finally, from the selected model, the following sequence of derivate will
be calculated:
 E  ln e1 
E  ln e1  



ln y1
ln y R 

 E  ln e  E  ln e  

S 







ln
y
ln
y


1
R 

E  ln eR  
 E  ln eR 

 ln y
ln y R 

1
(3)
7
Economic meaning of terms in matrix (3) is the following, for instance in
relation to a 1% change in per capita income in country j [elements in jth column in
matrix defined in (3)]. Firstly, the so-called direct elasticity, which is measured through
the jthjth-element of matrix (3), represents the sensitivity of jth-country per capita
emissions in response to a change in its own per capita incomes. Secondly, the indirect
elasticities are the rest of value in the jth-column and they measure the sensitivity of all
other countries’ emission to such a 1% change in jth-country per capita income. A global
measures for such indirect effect is the sum of all such cross-elasticities. Finally, total
elasticity can be calculated as the sum of the previous two effects, and represents the
percentage change in emission all over the world in response to a change of 1% in one
country per capita income.
Furthermore, following Lesage and Pace (2009), we will calculate, the following
summary measures:
 For the direct elasticities, a summary measure is defined as the mean of the all
main diagonal elements in (3):
tr  S

R
R
 Si,i
i 1
(4)
R
 For the indirect elasticities, the mean of all the elements outside the main
diagonal:
1R S1R tr  S


R
R
  Si, j
i j i
R
(5)
 And finally, for total elasticities, the mean of all elements in S:
1R S1R

R
  Si, j
i
j
R
(6)
8
The final step of our experiment comes from a higher level of flexibilization
through the consideration of all the panel data set referred to the 182 countries over the
period 1992-2011. Following, Elhorst (2003), we will start with a Fixed-Effect (FE)
general model, which will enable us to take into account unobservable heterogeneity in
data through a new set of parameter,  . That is, the specification on our general model
proposed by Cliff-Ord is the following:
y t  W y t  x t Wx t    u t
u t  W u t   t
 y t1 
1 x1t1
y 
1 x
t2 
1t 2


Where y t 
;x 
   t 

 

 y tR 
1 x1tR
(7)
x kt1 
  t1 
 1 



 
 t2 
 x kt 2 

; 
;   2 
  

  t   

 
 
 x ktR 
 R 
  tR 

Analogously to the cross-sectional case, a general-to-specific model selection
strategy can also be developed from the general model in (7) towards simple panel
model such as panel SARAR, SEM, SLM and Spatial Durbin models. The expression
for all nested models as well as the nested structure is shown in Figure 5.
Finally, the panel data set enable us to account for possible technological
changes occurring during the analysed time period, by introducing the corresponding
trend variable into the model. Within this framework, the model selection strategy as
well as the EKC testing are carried out as previously explained for a cross-sectional data
set.
4. Results
Results for Ordinary Least Square (OLS) estimation of model (1) are gathered in
the first column of Table 1. As shown in the table, sign and significance of the slope
parameters offer empirical evidence in favour of the inverted U relationship between the
9
variables. In other terms, OLS estimation does support the EKC relationship between
CO2 emission and per capita income.
However, as we expected, results for the Moran’s test indicate that the null of no
spatial autocorrelation is rejected by the data. Next, we try to discriminate between the
most common spatial autocorrelation patterns, Spatial Lag Model (SLM) or Spatial
Error Model (SEM). To cope with this objective, we obtain all specific Lagrange
Multiplier (LM) tests, in a non-robust and a robust to misspecification versions. Results
show that all respective null hypotheses are rejected. Consequently, we proceed to select
the spatial specification that better fits the data, following the general to specific
strategy explained in previous section.
Results for the described spatial specifications are shown in Table 2. The selection
process is carried out through the Likelihood Ratio (LR) tests shown at the bottom of
the table. Results indicate that simpler models such as the SARAR, Spatial Durbin,
SLM or SEM models are rejected by the data. Hence, all the estimated parameters for
the Cliff-Ord model are significant at 5% level of significance. Furthermore, the sign
and significance of slope parameters reveals that, after considering all possible spatial
dependence, the EKC relationship between CO2 emission and per capita income is also
supported by our model.
Results for direct, indirect and total effects derived from Cliff-Ord selected
model, are shown in Figure 4. As shown in Figure 4a, a 1% increase in a country per
capita income always generates a positive increase in own CO2 emissions, but
elasticities goes from 0.2 to 1.6. Highest elasticities correspond to the poorest countries
(African and Asian countries) and the lowest elasticities correspond to Western Europe,
United States and Canada. Nevertheless, from legend of Figure 4b, the indirect
elasticities also have a quite range of value from -0.5 to 0.7. That is, for instance, a 1%
increase in per capita income in Western Europe or Australia, decreases per capita
emissions in the rest of countries by percentages within the interval [0.1; 0.5]. Hence, as
deduced from Figure 4c, the total effect of a 1% increase in per capita income in
developed countries does generate a decrease in per capita emission. However, this is
not the case in the case of less developed countries. The obtained results are in
10
accordance with results previously derived in terms of the empirical evidence in favour
of the EKC relationship between CO2 emission and per capita income. Regarding the
summary elasticities proposed in Lesage and Pace (2009) defined in (4) to (6) account
for the following values: 0.890, for a summary of direct elasticities; 0.119, for the
indirect elasticities; and finally, 1.010, as a summary of total elasticities.
Next, we develop the highest degree of flexibilization proposed in this work
which consists on the estimation of the battery of models in the context of our panel
data set referred to the 182 countries over the period 1992-2011. Results for all the
possible specifications (pool, FE and FE-spatial models) are shown in Table 2 and
Table 3. The only difference between the respective models in both tables comes from
the fact that models in Table 3 take into account a possible technological change during
the period through the introduction of a trend variable (Trend).
(Insert Tables 2 and 3)
From results gathered in Tables 2 and 3, we can draw the following conclusions:
i) a FE specification outperforms the simple pooling of data; ii) FE model improves by
accounting for spatial dependence inherit in the problem; iii) among FE-spatial models,
the FE-Cliff-Ord model is best specification for our data; iv) technological change is not
significant in the analysed period; v) at the 5% level of significance, the no significance
of the quadratic term, (ln yi ) 2 , means that, as a parametric model accounts for possible
relevant omitted variables, the empirical evidence doesn’t support the EKC any further.
Results for direct, indirect and total effects derived from panel data Cliff-Ord
selected model, are shown in Figure 6, while he summary elasticities proposed in
Lesage and Pace (2009) defined in (4) to (6) for a panel data models account for the
following values: 0.424, for a summary of direct elasticities; -0.171, for the indirect
elasticities; and finally, 0.253, as a summary of total elasticities.
(Insert Figure 6)
As can be deduced, panel data results differ considerable from the crosssectional ones. The most remarkable issues are the following: i) all panel elasticities are
considerable lower than for the cross-sectional case; ii) in general, panel elasticities are
11
more homogeneous than cross-sectional ones; iii) quantile distribution for panel data
differs considerably from previously obtained, suggesting that the total elasticity of
CO2 emission is higher not only in poor countries, but also in rich countries such the
United States, Canada or Australia.
5. Concluding remarks
The
Environmental
Kuznets
Curve
(EKC)
hypothesis
conjectures
that
environmental degradation initially intensifies when a country’s per capita income
increases and subsequently subsides after a certain level of income is reached, resulting
in an inverted U-shaped relationship between environmental degradation and per capita
income.
There is abundant literature on this topic, produced especially in the last 20
years, that corroborates the existence of a positive income elasticity for environmental
quality. However, results are controversial, because results also depend on the type of
pollutant under analysis, the methodology or the data used.
In this study, we analyse this last issue, since we compare results obtained for a
cross-sectional data with the results derived from a panel data set. Furthermore, we also
pay special attention to the spatial interaction effect inherent to the problem. Main
results are that while cross-sectional results show evidence in favor of the EKC
hypothesis, when we take into account the unobserved heterogeneity through the
consideration of panel data models, the EKC relationship is not supported by the data.
References
Aldy, J. (2006). Per Capita Carbon Dioxide Emissions: Convergence or Divergence?.
Environmental & Resource Economics, 33(4): 533-555.
Ansuategi, A., Barbier, E. and Perrings, C. (1998). The Environmental Kuznets Curve. In: van
den Bergh, J.C.J.M., Hofkes, M.W. (Eds.), Theory and Implementation of Sustainable
Development Modelling. Kluwer Academic, Dordrecht.
Arrow, Kenneth, Bert Bolin, Robert Costanza, Partha Dasgupta, Carl Folke, C. S. Holling,
Bengt-Owe Jansson, Simon Levin, Karl-Gôran Mäler, Charles Perrings, David Pimentel.
(1995). Economic Growth, Carry Capacity, and the Environment. Science 268: 520-21.
12
Azar, C., Holmberg, J. and Karlsson, S. (2002). Decoupling - past trends and prospects for the
future. Paper commissioned by the Swedish Environmental Advisory Council, Ministry
of the Environment, Stockholm, Sweden.
http://www.iaea.org/inis/collection/NCLCollectionStore/_Public/34/025/34025734.pdf
Bruvoll, A. and Medin, H. (2003). Factors behind the environmental Kuznets curve, evidence
from Norway. Environmental and Resource Economics, 24(1): 27-48.
Carlsson, F. and Lundström, S. (2001). Political and Economic Freedom and the Environment:
The Case of CO2 Emissions. Working Papers in Economics no 29 Second version August
2001. Department of Economics Göteborg University.
Carson, R.T., Jeon, Y., McCubbin, D. R. (1997). The relationship between air pollution
emissions and income: US Data. Environment and Development Economics, 2(4): 433450.
Cole, M. A., Rayner, A. J. and Bates, J. M. (1997). The Environmental Kuznets Curve: an
empirical analysis. Environment and Development Economics, 2(4): 401-416.
Coondoo, D. and Dinda, S. (2002). Causality between income and emission: a country-groupspecific econometric analysis. Ecological Economics 40(3): 351-367.
De Bruyn, S. M., van den Bergh, J. C. J. M., and Opschoor, J. P. (1998). Economic growth and
emissions: reconsidering the empirical basis of environmental Kuznets curves. Ecological
Economics, 25(2): 161-175.
Dijkgraaf E. and Vollebergh H. R. J. (2001). A note on testing for environmental Kuznets
curves with panel data. Technical Report 63, FEEM (Fondazione Eni Enrico Mattei).
EIA, U.S. Energy Information Administration. Environment. Data.
http://www.eia.gov/environment/data.cfm
Elhorst J.P., 2003. Specification and Estimation of Spatial Panel Data Models, International
Regional Sciences Review, 26,244-268.
Grossman, G. M., and Krueger, A. B. (1991). Environmental impacts of a North American Free
Trade Agreement. National Bureau of Economic Research, Working Paper No. 3914,
NBER, Cambridge MA.
Grossman, G. M., and Krueger, A. B. (1993). Environmental impacts of a North American Free
Trade Agreement. In: Garber, P. (Ed.), The US–Mexico Free Trade Agreement. MIT
Press, Cambridge, MA.
Grossman, G.M., and Krueger, A.B. (1995). Economic growth and the environment. Quarterly
Journal of Economics, 110 (2): 353-378.
Heenrink, N, Mulatu, A and Bulte, E. (2001). Income inequality and the environment:
aggregation bias in environmental Kuznets curves. Ecological Economics, 38: 359-367.
Heil, M., and Selden, T. M. (2001). International Trade Intensity and Carbon Emissions: A
Cross-Country Econometric Analysis. Journal of Environment and Development, 10: 3549.
Hill, R. and Magnani, E. (2002). An exploration of the conceptual and empirical basis of the
environmental Kuznets curve. Australian Economic Papers, 41(2): 239-254.
Holtz-Eakin, D. and Selden, T. M. (1995). Stoking the fires? CO2 emissions and economic
growth. Journal of Public Economics 57, 85–101.
IPCC, (2013). Summary for Policymakers. In: Climate Change 2013: The Physical Science
Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change [Stocker, T.F., D. Qin, G.-K. Plattner, M.
Tignor, S.K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley (eds.)].
Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.
IPCC, (2014). Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II
and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
13
[Core Writing Team, R.K. Pachauri and L.A. Meyer (eds.)]. IPCC, Geneva, Switzerland,
151 pp.
Kaufmann, R.K., Davidsdottir, B., Garnham, S. and Pauly, P. (1998). The determinants of
atmospheric SO2 concentrations: reconsidering the environmental Kuznets curve.
Ecological Economics 25(2): 209– 220.
Kuznets, S. (1955). Economic Growth and Income Inequality. The American Economic Review
45(1): 1–28.
Kuznets, S. (1965). Economic growth and structural change, New York: Norton.
LeSage, J., Pace, R.K. (2009). Introduction to Spatial Econometrics. CRC Press Taylor &
Francis Gropu. Boca Raton.
Lindmark, M (2002). An EKC-pattern in historical perspective: carbon dioxide emissions,
technology, fuel prices and growth in Sweden 1870–1997. Ecological Economics, 42:
333–347.
List, J. A. and Gallet, C. A. (1999). The Environmental Kuznets Curve: does one size fit all?
Ecological Economics 31: 409-423.
List, J. A. and Gerking, S. (2000). Regulatory Federalism and U.S. Environmental Policies,
Journal of Regional Science, 40: 453-471.
Magnani, E. (2001). The Environmental Kuznets Curve: development path or policy result?
Environmental Modelling & Software, 16: 157–165.
Millimet, D. L., List, J. A. and Stengos, T. (2003). The environmental Kuznets curve: real
progress or misspecified models? Review of Economics and Statistics, 85: 1038-1047.
Panayotou, T. (1993). Empirical tests and policy analysis of environmental degradation at
different stages of economic development, International Labor Organization, Technology
and Employment Programme, Geneva.
Panayotou, T. (1995), ‘Environmental Degradation at Different Stages of Economic
Development,’in Beyond Rio (The Environmental Crisis and Sustainable Livelihoods in
the Third World), edited by Iftikhar Ahmed and Jacobus A. Doeleman, International
Labor Organization, Macmillan Press Ltd., London.
Panayotou, T. (2000). Economic Growth and the Environment. CID Working Paper No. 56,
Environment and Development Paper No. 4. Center for International Development at
Harvard University.
Perrings, C. and Ansuategi, A. (2000). Sustainability, growth and development. Journal of
Economic Studies, 27(1/2): 19-54.
Roberts, J. T. and Grimes, P. E. (1997). Carbon Intensity and Economic Development 1962-91:
A Brief Exploration of the Environmental Kuznets Curve, World Development, 25(2):
191-198.
Roca, J., Padilla, E., Farré, M. and Galletto, V. (2001). Economic growth and atmospheric
pollution in Spain: discussing the environmental Kuznets curve hypothesis, Ecological
Economics 39 (1): 85-99.
Selden, T. M. and Song, D. (1994). Environmental quality and development: is there a Kuznets
curve for air pollution? Journal of Environmental Economics and Environmental
Management, 27(2): 147-162.
Shafik, N., and Bandhopadhyay, S. (1992). Economic growth and environmental quality: Timeseries and cross-country evidence. World Bank Working Papers, WPS 904, Washington,
52 pp.
Shafik, N. (1994). Economic development and environmental quality: an econometric analysis.
Oxford Economic Papers, 46: 757-773.
Stern, D. I. and Common, M. S. (2001). Is There an Environmental Kuznets Curve for Sulfur?.
Journal of Environmental Economics and Management, 41: 162-178.
14
Talukdar, D. and Meisner, C. M. (2001). Does the private sector help or hurt the environment?
Evidence from carbon dioxide pollution in developing countries, World Development,
29(5): 827-840.
Tucker, M. (1995). Carbon dioxide emissions and global GDP. Ecological Economics, 15(3):
215-223.
Vincent, J. R., 1997. Testing for environmental Kuznets curves within a developing country.
Environment and Development Economics 2(4): 417-431.
Wagner, M. (2008). The carbon Kuznets curve: a cloudy picture emitted by bad econometrics?.
Resource and Energy Economics, 30: 388-408.
15
Figure 1. Distribution of per capita CO2 emissions in 2011
(7.1,44.4]
(2.6,7.1]
(0.6,2.6]
[0.0,0.6]
Figure 2. Distribution of per capita GDP in 2011
(13728.3,81853.0]
(3738.7,13728.3]
(974.2,3738.7]
[193.6,974.2]
16
Figure 3. Cross-section models framework
Y   WY  X   WX   u
u   Wu  
General:
Cliff-Ord
0
SARAR
Modelo Durbin Spatial
Y  WY  X   u
u   Wu  
0
0
0
0
Y  WY  X   WX   
0
SLM
Y  WY  X   
   
SEM
0
Modelo de error Durbin Spatial
Y  X   WX   u
u   Wu  
0
Y  X  u
SLX
u  Wu  
0
0
0
Y  X   WX   
0
Estático
Y  X  
17
Figure 4. Direct, indirect and total elasticities of per capita emission with respect to
per capita income
Figure 4a: Direct elasticities
Elasticidades directas emisiones pc. con respecto a renta pc
(1.2,1.6]
(0.9,1.2]
(0.6,0.9]
[0.2,0.6]
Figure 4b:Indirect
elasticities
Elasticidades indirectas emisiones pc. con respecto a renta pc
(0.3,0.7]
(0.1,0.3]
(-0.1,0.1]
[-0.5,-0.1]
Figure 4c: Total elasticities
18
Elasticidades totales emisiones pc. con respecto a renta pc
(1.5,2.3]
(1.0,1.5]
(0.5,1.0]
[-0.3,0.5]
19
Figure 5. Panel data models framework
General
Cliff-Ord
y t  W y t  x t  Wx t    u t
u t  W u t   t
0
SARAR
0
y t  W y t  x t    u t
u t  W u t   t
0
SAR
y t   W y t  x t     t
0
0
SEM
y t  x t    u t
u t  W u t   t
0
SLX
y t  x t  Wx t     t
0
Panel
y t  x t     t
20
Figure 6. Direct, indirect and total elasticities of per capita emission with respect to
per capita income
Figure 6a: Direct elasticities
Figure 6b:Indirect elasticities
Elasticidades indirectas emisiones pc. con respecto a renta pc
(-0.1,0.0]
(-0.2,-0.1]
(-0.3,-0.2]
[-0.5,-0.3]
Figure 6c: Total elasticities
21
Elasticidades totales emisiones pc. con respecto a renta pc
(0.3,0.5]
(0.2,0.3]
(0.2,0.2]
[0.0,0.2]
22
23
Table 1. Global autocorrelation measure (z values) for 2011.
Moran’s I
Geary’s C
Getis and Ord’s G
Per capita CO2 emissions
10.11*
-5.54*
9.11*
Per capita GDP
12.14*
-9.88*
10.43*
Table 2. Results for cross-sections models for 2011
Constant
OLS
15.981*
CLIFFORD
-38.364*
SARAR
-13.497*
Spatial
Durbin
-11.397*
SLM
-13.672*
SEM
-13.247*
ln yi
3.188*
2.927*
2.620*
2.395*
2.723*
2.550*
-0.137*
-0.123*
-0.107*
-0.093*
-0.117*
-0.101*
(ln yi )
2
W ln yi
W (ln yi )


2
2
5.172*
0.016
-0.265*
-0.747*
0.795*
0.365*
-187.34
-0.022
0.481*
0.140*
0.386*
0.480*
-194.063
Log Ver
DIAGNOSTICS ON SPATIAL AUTOCORRELATION
H0: no autocorrelation
6.59*
Moran's
H0: no autocorrelation SEM
37.39*
LM test
H0: no autocorrelation SEM
19.18*
Robust LM test
H0: no autocorrelation SLM
23.06*
LM test
H0: no autocorrelation SLM
4.85*
Robust LM test
LR:H0: SARAR; HA: Cliff13.45*
Ord
LR: H0: SLM; HA: SARAR
9.696*
LR: H0: SEM; HA: SARAR
3.074
LR: H0: SLM; HA: SDM
LR: H0: SEM; HA: SDM
0.464*
-192.273
0.250*
0.516*
0.478*
-195.6
0.516*
-198.911
13.276*
6.654*
24
Table 3. Panel data model results, assuming no technological change.
Constant
ln yi
(ln yi )
2
FESpatial FE-SLM FE-SEM
Durbin
FE
-3.268*
2.602*
0.557
0.238*
0.272*
0.355
0.484*
0.422*
-0.105*
-0.009
0.012
0.010
0.005
-0.008
0.000
2


2
Log ver
F: H0: No fixed effects
Pesaran test: H0: No spatial
correlation
LR:H0: SARAR; HA: CliffOrd
LR: H0: SLM; HA: SARAR
LR: H0: SEM; HA: SARAR
LR: H0: SLM; HA: SDM
LR: H0: SEM; HA: SDM
FESARAR
POOL
-13.246*
W ln yi
W (ln yi )
FECLIFFORD
0.177
0.405
-0.030*
0.610*
-0.566*
-0.513*
0.601*
-0.035
0.271*
0.061*
-213.368
0.063*
-221.10
0.252*
0.278*
0.066*
-232.49
0.066*
-244.39
176.02*
5.947*
15.46*
46.59*
31.56*
23.80*
8.76*
25
0.066*
-236.87
Table 4. Panel data model results, considering a possible technological change.
FECLIFFORD
FESARAR
FESpatial
Durbin
-4.195
0.000
0.001
0.005*
0.002
-0.002
0.000
2.614*
0.557
0.236*
0.277*
0.347
0.481*
0.422*
-0.105*
-0.010
0.012
0.008
0.005
-0.006
0.000
POOL
FE
Constant
Trend
11.945*
-0.013*
ln yi
(ln yi )
2
W ln yi
W (ln yi )
2


2
log ver
F: H0:No fixed effect
Pesaran test: H0: No
spatial correlation
LR:H0: SARAR; HA:
Cliff-Ord
LR: H0: SLM; HA:
SARAR
LR: H0: SEM; HA:
SARAR
LR: H0: SLM; HA:
SDM
LR: H0: SEM; HA:
SDM
0.152
0.379
-0.030*
0.610*
-0.563*
-0.545*
0.610*
-0.037
0.271*
0.062*
-212.12
0.062*
-217.42
FE-SLM FE-SEM
0.260*
0.278*
0.066*
-231.26
0.066*
-242.55
0.066*
-236.87
175.98*
5.52*
10.60*
50.25*
38.90*
22.57*
11.22*
26