Download Negative Elasticity

Empirical Analysis of CO2 Emissions and GDP Relationships
in OECD Countries
by
Jeffrey M. Fang, Advisor
Science & Technology Policy Research and Information Center
National Applied Research Laboratories
16th Fl. No.106, Sec. 2, Heping E. Rd.
Taipei 106, Taiwan
Phone: 886-2-2737-7169/Fax:886-2-2737-7644/Email: [email protected]
and
J. C. Chen, Graduate Student
Department of Mechanical Engineering
National Taiwan University
No. 1, Sec. 4, Roosevelt Rd
Taipei 106, Taiwan
Phone: 886-0968-432-637/Fax:886-2-2363-1755/Email: [email protected]
Abstract
Decoupling CO2 emissions from income growth is an important world issue today. Generally,
decoupling can be represented by the income elasticity of CO2 emissions. There is absolute decoupling
(AD) when the income elasticity is negative or zero (NE or Zero E). As income grows, CO 2 emissions
will either decline or stay at the same level. If the elasticity is positive and less than +1.0, there is
relative decoupling (RD). The percentage growth of CO2 emissions is smaller than that of income
growth. If the elasticity is positive and greater than or equal to +1.0, then CO 2 emissions is directly
coupled with income growth; there is close coupling (CC). In this aspect, understanding how the 30
OECD countries have performed historically would be useful. This paper applies simple and multiple
regression analysis to derive empirical estimates of “apparent” and “net” income elasticity of per capita
CO2 emissions and use them to classify the countries into categories of decoupling status groups.
Additional explanatory variables introduced into multiple regression are fuel share ratio and real energy
prices. The fuel share ratio variable is defined as (share of nuclear energy + share of renewable
energy)/(share of coal +share of oil).
For 14 countries of the 30 OECD countries, using the apparent elasticity derived from simple
regression to classify a country’s decoupling status will do it correctly. Introduction of the fuel share
ratio and real energy price variables will not affect the classification, although the absolute values of
the elasticity estimates may be different. More specifically, Germany, Hungary, Luxembourg, Poland,
Slovak Republic, Sweden and Switzerland would stay in the AD group; Austria, Finland, Ireland, Italy,
Japan and Norway stay in the RD group; Portugal stays with the CC status.
For the other 16 countries, utilizing the apparent elasticity estimate as the guide will prove to be off the
mark. For Belgium, Canada, Czech Republic, Denmark, France, Iceland, Korea, Spain, United
Kingdom and United States, it would be over-optimistic. For Australia, Greece, Mexico, Netherlands,
New Zealand and Turkey, it would tend to be overly pessimistic. Hence, for these latter 16 countries, it
would be necessary to introduce the fuel share ratio and real energy prices variables into the regression.
1. Introduction
As global warming looms increasingly larger in world affairs and the pursuit of economic growth by
nations of the world continues, the relationships between economic growth and emissions of greenhouse
gases (GHGs), particularly carbon dioxides (CO2), is of much interest. On the one hand, Kyoto Protocol
has required the more advanced economies to reduce total GHG emissions by some percentages of their
respective emissions levels in 1990. Further, it is expected that the next round of global negotiations
would lead to requiring even higher levels of GHG emissions reductions and more countries to be subject
to total emissions control regimes. On the other hand, practically all nations, developing and less
developed countries in particular, would like to continue to grow their respective economies at some
respectable rates. Under such circumstances, if GHG or CO2 emissions were closely coupled with gross
domestic products (GDP) growth, then the objective of economic growth would be in conflict with the
objective of reducing CO2 emissions.
What are the empirical relationships between CO2 emissions and GDP growth? Figure 1 depicts three
general types of such relationships on a per capita basis with scatter diagrams. In Part (a), the relationship
is positive for Japan and Australia. Per capita CO2 emissions increase with the growth in per capita GDP,
although it appears that, in Japan’s case, there is a range in the US$15,000-20,000 area with a slight
declining tendency. In the cases of Germany and the U.S, the relationship is negative (Part (b)). As per
capita GDP grows, per capita CO2 emissions fall as a result. The direction of the relationship is more
clearly defined in Germany’s case than in the U.S. case. In Part (c), per capita CO2 emissions appear to
stay relatively unchanged as per capita income increases in the cases of Switzerland and Canada. In
Canada’s case, there seems to be a break in the US$20,000 to US$22,000 range.
The relationships shown in Figures 1 are derived from simply looking at the correlation between
capita CO2 emissions and per capita GDP. No other factors which can influence the quantities of
capita CO2 emissions are taken into account. In reality, many other factors can affect a country’s
capita CO2 emissions from energy use, such as changes in the composition of fuel use, energy
efficiency, energy prices and societal environmental consciousness.
per
per
per
use
In economics, the concept of income elasticity of CO2 emissions can be used to characterize the
relationship between per capita CO2 emissions and economic growth. The income elasticity of per capita
CO2 emissions is represented by the percentage changes in per capita CO2 emissions as per capita GDP
increases by 1%. I.e., income elasticity of per capita CO2 emissions = (percentage changes in per capita
CO2 emissions)/(percentage changes in per capita GDP). It is useful in understanding the performance of
a country in reducing CO2 emissions. If the income elasticity of per capita CO2 emissions of a country is
positive and greater than 1.0, then the growth rate of per capita CO2 emissions is greater than the growth
rate of per capita GDP; income growth and CO2 emissions growth go hand in hand. In other words, there
is “close coupling” between CO2 emissions and GDP growth (CC). If the income elasticity of CO2
emissions per capita is positive but less than 1.0, then per capita CO2 emissions will grow with the
increase in per capita GDP, but at a slower rate than that of per capita GDP. The relationship between the
two is characterized by “relative decoupling” (RD). If the income elasticity of CO2 emissions per capita is
zero or negative, then as per capita GNP increases, per capita CO2 emissions will either not change at all
or decline. That is to say, there is “absolute decoupling” (AD) between per capita CO2 emissions and per
capital GDP.
The purposes of this paper are to present the initial results in applying simple and multiple regression
analysis to derive empirical estimates of the income elasticity of CO2 emissions for the 30 OECD
countries using IEA data for the 1970-2004 period and to use such estimates to classify the countries into
the three categories of close coupling, relative decoupling and absolute decoupling. In the process, the
differences between “apparent income elasticity” and “net income elasticity” are elaborated and the
effects of introducing additional explanatory variables such as fuel share ratio and real energy prices are
explored.
2
Japan
10
Per Capita CO2 Emissions
(Tons/Person)
Per Capita CO2 Emissions
(Tons/Person)
a. positive relationship
8
6
4
2
0
0
5000
10000 15000 20000 25000
Per Capita GDP (PPP) (US$)
Australia
20
15
10
5
0
30000
0
5000
10000 15000 20000 25000 30000 35000
Per Capita GDP (PPP) (US$)
Germany
16
14
12
10
8
6
4
2
0
Per Capita CO2 Emissions
(Tons/Person)
Per Capita CO2 Emissions
(Tons/Person)
b. negative relationship
0
5000
10000 15000 20000
Per Capita GDP (PPP) (US$)
25000
United States
25
20
15
10
5
0
30000
0
10000
20000
30000
Per Capita GDP (PPP) (US$)
40000
Switzerland
8
7
6
5
4
3
2
1
0
Per Capita CO2 Emissions
(Tons/Person)
Per Capita CO2 Emissions
(Tons/Person)
c. constant relationship
0
5000
10000 15000 20000 25000 30000 35000
Per Capita GDP (PPP) (US$)
Canada
20
15
10
5
0
0
5000
10000 15000 20000 25000 30000 35000
Per Capita GDP (PPP) (US$)
Figure 1. Three types of per capita CO2 emissions and per capita GDP relationships
In the next section, theoretical considerations and methodology are explained. Empirical results are then
presented in Section 3. Finally, a summary of the main results and a brief discussion of the implications
and qualifications of the study then conclude this paper.
3
2. Approach and Methodology1
2.1 Theoretical Considerations
In the literature, the Environmental Kuznets Curve (EKC) hypothesis, as applied to the relationship
between per capita CO2 emissions and economic growth, states that such a relationship is characterized
by an inverted U-shaped curve. In the low income range, a country’s energy use will grow with the
growth of income, resulting in growing pollution and CO2 emissions. Thus, initially, there is a positive
relationship between CO2 emissions and income growth in the low income range. However, as economic
development reaches a certain stage and per capita income attains a certain level, greater awareness of
environmentalism is generated and more capability and resources are devoted to develop and implement
environmental protection and regulations. This will allow the pollution level and CO2 emissions to be
maintained at a constant level for sometime and then become reduced as income grows further. Thus, the
overall relationship between income and CO2 emissions will change from positive to constant, and then
to negative. Therefore, in econometric estimation, the income and CO2 emissions relationship can be
positive, zero or negative. It all depends on the characteristics of a country’s economy, the degree of
economic development and the implementation of environmental protection and CO2 emissions
reduction measures.
Since its inception in 1992 in the World Bank Development Report (Shafik and Bandyopadhyay, 1992),
there have been several empirical studies attempting to determine the turning point of the reverse
U-shaped EKC curve, the income level at which the income elasticity of CO2 emissions is zero, utilizing
cross-section (country) data or time-series data of several developed countries. By 1997-98, the literature
appeared to have reached some consensus that the empirically derived EKC would largely depend on the
characteristics of the countries, the models and the variables incorporated into the models (Richmond
and Kaufmann 2006). However, most of the studies did not include real energy prices into the analysis.
Recently, Richmond and Kaufmann (2006) used 1978-1997 data for 16 OECD countries and introduced
energy prices (light fuel oil for industrial use) as well as the shares of electricity, oil and coal in total
energy supply as independent variables. They found that introducing real energy price variable will
reject the null hypothesis that there is a turning point in the relationship between CO 2 emissions and
income. In other words, the relationship between the two is not necessarily reverse U shaped. On another
front, Coondoo and Dinda (2002) argued that, while the EKC hypothesis presumes that the causality
goes from income to CO2 emissions, the causality between the two variables may vary from one country
to the other. Their econometric analysis using data for 88 countries covering the period 1960-1990 found
that the causality runs from emissions to income for the developed countries of North America and
Western Europe (and also for Eastern Europe); runs from income to emissions for country groups of
Central and South America, Oceania and Japan; and is bi-directional for the country groups of Asia and
Africa.
This study attempts to apply the EKC concept to analyze the data in the 30 OECD countries, to utilize
simple and multiple regression to quantify the income elasticity of per capita CO2 emissions in each
country and to explore how introduction of additional explanatory variables such as fuel share ratio and
real energy prices will impact the income elasticity estimates. The analysis presumes that the causality
go from income to CO2 emissions and there is no attempt to determine the turning point in the EKC.
In multiple regression analysis, one of the independent variables is the fuel share ratio. Total CO2
emissions in a country are computed from the quantities of coal, oil, natural gas and other energy types.
Among the various fuels and energy, coal and oil and products are high-carbon content fuels and, hence,
their use involves high CO2 emissions. Natural gas’ carbon content is relatively low and is referred to as
low-carbon fuel. In contrast, nuclear energy and renewable energy such as biomass, photovoltaic, wind
power and hydropower are carbon-free energy sources. The higher the proportion of coal and oil in a
country’s total energy supply, the higher the per capita CO2 emissions. The relationship between the
1
See Fang and Chen (2007) for another application of the same approach.
4
share of coal and oil and per capita CO2 emissions is positive. In contrast, the higher the proportion of
nuclear energy and renewable energy, the lower the per capita CO2 emissions. The relationship between
the two is negative. It can be expected that, as the shares of energy are introduced into regression of per
capita CO2 emissions, the coefficient for coal and oil is positive and that for nuclear energy and
renewable energy is negative. However, there is a practical difficulty in that, as the shares of individual
fuels are all incorporated into the equation, it is likely that there will be correlation among the fuel shares
variables. To avoid such situation, we have in practice introduced an independent variable called “fuel
share ratio,” defined as (share of nuclear energy + share of renewable energy)/(share of coal + share of
oil). The higher the numerator of this independent variable, the smaller is per capita CO2 emissions. The
higher is the value of the denominator itself, the higher is per capita CO2 emissions. However, since it is
in the denominator, the independent variable will be smaller, and per capita CO2 emissions will be
smaller. It is therefore expected that the sign for the fuel share ratio variable is negative.
In theory, energy prices can affect the amount of energy use. The higher the energy prices, the smaller
quantities of energy are consumed and the smaller the quantities of CO2 are emitted. Therefore, the
expected sign of the coefficient of the real energy price variable in the per capita CO 2 emissions
regression is negative. However, there are different categories of energy and different energy prices for
different users of the same category of energy. Ideally, the prices for the aggregate energy should be used.
Yet the data for aggregate energy prices for different countries are not complete. For these reasons, the
practice is usually to represent energy price by a particular fuel type in industrial, residential and
commercial sectors, and in transportation. So there is some gap between theory and practice. For the
purpose of this study, three different price series are tested: price of light fuel oil for industrial use,
industrial electricity prices, and the consumer price index for energy (CPI-energy).
2.2 Methodology
In actual computation, three different methods can be used to derive estimate of the income elasticity of
per capita CO2 emissions: (1) It is computed as the ratio of the percentage change in per capita CO2
emissions for a certain time period over the percentage change in per capita GDP for the same time
period. (2) Run the simple regression using LN(per capita CO2 emissions) as the dependent variable and
LN(per capita GDP) as the independent variable. In this case, LN(x) represents natural logarithm
transformation of x. The estimated coefficient of the independent variable is the income elasticity of per
capita CO2 emissions. (3) Run multiple regression using LN(per capita CO2 emissions) as the dependent
variable and LN(per capita GDP) as the first independent variable, and LN(other related variables) as the
second or additional independent variables. Other related variables may be real energy prices or energy
share ratios. The estimates derived from Approaches (1) and (2) above can be termed “apparent income
elasticity”, because only per capita CO2 emissions and per capita GDP are considered, while other factors
are neglected. When other factors such as fuel share ratio and real energy prices are also considered, as
in Approach (3), the resultant elasticity estimate is “net income elasticity.”
In general, the ratio approach (Approach 1) is suitable for data series whose time period is short and,
hence, regression analysis cannot be applied. Also, since the time period is short, it is subject to the
influences of extreme values. A major advantage of Approaches 2 and 3 is that the elasticity estimate can
be subject to the statistical significance testing with attached confidence levels.
This study runs the following three regression equations:
Model 1.
Model 2.
Model 3.
LN(CO2PC) = a1 + b1*LN(GDPPC)
LN(CO2PC) = a2 + b2*LN(GDPPC) + c2*LN(ESRATIO)
LN(CO2PC) = a3 + b3*LN(GDPPC) + c3*LN(ESRATIO) + d*LN(PEN)
where the variables are defined as follows:
CO2PC
= per capita CO2 emissions
GDPPC
= per capita GDP
ESRATIO
= energy (fuel) share ratio
5
PEN
= real energy prices (deflated by the CPI).
The regression coefficients are:
a1, a2, a3 : intercept term
b1 ：apparent income elasticity of per capita CO2 emissions; the expected sign can be
positive, zero or negative
b2, b3 : net income elasticity of per capita CO2 emissions; the expected sign can be positive,
zero or negative
c1, c2, c3 ：elasticity of per capita CO2 emissions with respect to fuel share ratio; the expected
sign is negative.
d ：price elasticity of per capita CO2 emissions; the expected sign is negative.
2.3 Data
Most data are from the internet databank SourceOECD which includes International Energy Agency
(IEA)’s reports and statistical data, such as the annual report series of CO2 Emissions from Fuel
Combustion (IEA, 2006a), which contains data on total and per capita CO2 emissions, GDP and
population. Another example is IEA’s annual reports on Energy Prices and Taxes (IEA, 2007) which
provides data on energy prices and taxes of different types of energy for different uses. There is also
IEA’s annual report series for World Energy Statistics and Balances (IEA, 2006c), containing data on
the energy supply and balances, imports, total primary energy supply and net usage. Generally, the time
period for the data set used in this analysis is from 1970 through 2004, although some countries have
shorter series due to missing data. For example, some energy price series do not start until 1978. Other
series are even shorter.
3. Results
3.1 Apparent Income Elasticity
(1) 1970-2004 Data Period (Model 1a).
When the 1970-2004 data or available maximum-length data period are used in simple regression with
natural logarithm-transformed variables, apparent income elasticity estimates are derived. Based upon
such estimates, the 30 OECD countries can be grouped into three groups of absolute decoupling, relative
decoupling, and close coupling (Table 1):

Absolute Decoupling: There are 15 countries in the absolute decoupling (AD) group, which can be
divided into two subgroups: negative income elasticity (NE) subgroup and zero income elasticity
(Zero E) subgroup. NE subgroup includes 11 countries: Belgium, Demark, France, Germany,
Hungary, Luxembourg, Poland, Slovak Republic, Sweden, United Kingdom and U.S. The other 4
countries (Canada, Czech Republic, Iceland and Switzerland) are in the Zero E subgroup.

Relative Decoupling: The relative decoupling (RD) group is comprised of 10 countries: Australia,
Austria, Finland, Ireland, Italy, Japan, Korea, the Netherlands, Norway and Spain. Note, however,
Korea’s elasticity is +0.970, which is very close to +1.0.

Close Coupling: Five (5) countries are in the close coupling (CC) group. They are Greece, Mexico,
New Zealand, Portugal and Turkey. They show no decoupling between CO2 emissions and income
growth.
6
Table 1.
Decoupling Groups According to Estimates of Apparent Income Elasticity
of CO2 Emissions Derived Using Different Time Period Data
Decoupling Group
Absolute Decoupling
Classified According to Income Elasticity Estimated from
1970-2004 Data (Model 1a)
1990-2004 Data (Model 1b)
Negative Elasticity
Negative Elasticity
Belgium (-0.202**),
Denmark (-0.163*),
France (-0.654**),
Germany (-0.412**),
Germany (-0.754**)
Hungary (-0.167*),
Hungary (-0.167*)
Luxemburg (-0.688**),
Poland (-0.355**)
Poland (-0.355**)
Slovak Republic (-0.416**),
Slovak Republic (-0.416** )
Sweden (-1.274**),
Sweden (-0.274*)
U.K. (-0.299**),
U.K. (-0.245**)
US (-0.119**)
Zero Elasticity
Canada,
Czech Republic
Iceland,
Switzerland,
Relative Decoupling
Zero Elasticity
Czech Republic
Iceland
Switzerland
Belgium
Denmark
Finland
France
Luxembourg
Netherlands
Australia (0.541*)
Austria (0.784**)
Australia (0.672**)
Austria (0.336**)
Finland (0.341**)
Ireland (0.341**)
Italy (0.421**)
Japan (0.297** )
Korea (0.970**)
Netherlands (0.124** )
Norway (0.288**)
Spain (0.876** )
Ireland (0.323** )
Italy (0.574**)
Korea (0.903** )
Norway (0.429**)
Canada (0.484**)
Greece (0.754**)
Mexico (0.174*)
New Zealand (0.837**)
US (0.152**)
Close Coupling
Greece (1.976**)
Mexico (1.424**)
New Zealand (1.242**)
Portugal (1.542**)
Turkey (1.439** )
Portugal (1.341**)
Turkey (1.085**)
Japan (1.259**)
Spain (1.114**)
Notes:
Significance level: ** 1%; * 5%
Zero elasticity: Estimate which is not statistically significant at 5% level.
Source: Developed in this study.
7
(2) 1990-2004 Data Period (Model 1b)
As the Kyoto Protocol uses 1990 as the base year in setting the CO2 emissions reduction targets, it is
useful to conduct analysis using the 1990-2004 data period. As shown in Table 1, using the shorter period
in regression yields the following results:

Absolute Decoupling: Only six of 11 countries in the NE subgroup in Model 1a remain with the
same subgroup in Model 1b. They are Germany, Hungary, Poland, Slovak Republic, Sweden and
United Kingdom. Belgium, France and Luxemburg move from NE subgroup to the Zero E
subgroup. Finally, Finland and the Netherlands moved from the RD group to the zero elasticity
subgroup. These changes plus the 3 out of the 4 original countries remaining in the Zero E
subgroup (Czech Republic, Iceland and Switzerland) together make a total of 9 countries.

Relative Decoupling: There are 11 countries in the RD group; six remaining in the same group from
the longer-time period (Model 1a) results: Australia, Austria, Ireland, Italy, Korea and Norway.
Canada moves in from the Zero E sub-group. Greece, Mexico and New Zealand move in from the
CC group to the RD group. In addition, U.S. moves from NE subgroup to the RD group.

Close Coupling. There are 4 countries in the CC group. Among them, only Portugal and Turkey
were originally in this group. Japan and Spain moved from RD group into this group.
(3) Notes on Model 1a and Model 1b Results
Figures 2 and 3 graphically depict the four decoupling status groupings against the 2004 per capita GDP
in PPP terms for Model 1a and Model 1b, respectively. Several aspects can be noted at this point. First,
comparison of the results for the two different data periods show that, for the four countries of Czech
Republic, Hungary, Poland and Slovak Republic, the apparent elasticity estimates are identical for each
country. This is because the per capita GNP in PPP terms are not available for the 1970-1990 period for
these countries, hence the income elasticity estimates for them cover only a much shorter period of 1990
through 2004, or even shorter by one or two years. Thus, the two estimates marked for different time
period cover exactly the same shorter time period, yielding identical results. In addition, these countries
were transitioning from centrally planned economies to market based economies and are called
“Economies in Transition” (EITs). Per capita GDP for these four countries are lower than most other
OECD countries and their respective economic growth were disrupted during the transitioning years of
1989-1991. In both charts, these four countries cluster in the southwest region, with low-income and
negative or zero income elasticity.
Second, statistically, the smaller is the sample size, the more likely is the regression coefficient to be
statistically insignificant. When this happens, the coefficient is treated as equal to zero. In the results
shown in Table 1, the longer time period (Model 1a) represents a sample size of 35, while the shorter time
period (Model 1b) only has a sample size of 15 or less. So the 1990-2004 results has a larger group of
zero-elasticity subgroup than the 1970-2004 results; 9 versus 4. In addition to the original 3 that stay in
this subgroup (Czech Republic, Iceland and Switzerland), Belgium, Denmark, France and Luxembourg
move from the NE subgroup, while Finland and the Netherlands move from the RD group into this group.
Third, for those countries staying with the same group or subgroup as in the longer time period results,
the value of the estimated income elasticity can be either larger or smaller, depending on the country in
question. No obvious pattern is observed.
Fourth, in addition to the changes noted above, the other changes are as follows: U.S changes from the
NE subgroup to the RD group. Japan and Spain move from the RD group to the CC group. In contrast,
Greece and Mexico move from the CC group to the RD group. To some extent, such differing directions
of movement between the different groups may be affected by the different rates of economic growth of
either acceleration or deceleration of economic growth.
8
Model 1a
2.5
Income Elasticity of CO2 Emissions
2.0
Close Coupling
Greece
Turkey
1.5
Relative Decoupling
Absolute Decoupling-zero elasticity
Portugal
Mexico
Absolute Decoupling-negative elasticity
New Zealand
1.0
Korea
Spain
Australia
Italy
0.5
Ireland
Austria Finland
Norway
Netherlands
Iceland
Canada
Switzerland
Belgium
United States
Denmark
United Kingdom
Germany
Japan
0.0
Hungary
Czech Republic
Poland
Slovak Republic
-0.5
France
Luxembourg
-1.0
Sweden
-1.5
0
10000
20000
30000
40000
2004 Per Capita GDP (PPP) (US$)
50000
60000
Source: Compiled in this study from Table 1
Figure 2: Decoupling Status with Model 1a Results
Model 1b
1.500
Portugal
Close Coupling
Japan
Relative Decoupling
Spain
Turkey
Income Elasticity of CO2 Emissions
1.000
Absolute Decoupling-zero elasticity
Korea
New Zealand
Italy
0.500
Absolute Decoupling-negative elasticity
Austria
Greece
Australia
Canada
Finland
Norway
Ireland
France
Mexico
Belgium
0.000
Hungary
Iceland
Netherlands
Switzerland
United Kingdom
Czech Republic
Poland
Slovak Republic
United States
Sweden
-0.500
Denmark
Luxembourg
Germany
-1.000
0
10000
20000
30000
40000
2004 Per Capita GDP (PPP) (US$)
50000
Source: Compiled in this study from Table 1
Figure 3: Decoupling Status with Model 1b Results
9
60000
3.2 Introducing Additional Explanatory Variables
(1) Fuel Share Ratio.
When the fuel share ratio variable is introduced into the regression as the second explanatory variable,
for 28 of the 30 OECD countries, the estimated coefficient of the fuel share ratio variable has the
expected negative sign. Germany and New Zealand are the only two countries with unexpected positive
coefficient for the fuel share ratio variable. Germany’s coefficient is significant at 1% level and that for
New Zealand is not significant. Among the 28 countries with the expected negative sign for the
coefficient for the fuel share ratio variable, seven (7) (Austria, Greece, Ireland, Italy, Korea, Netherlands
and Poland) are not statistically significant, and are accordingly treated as equal to zero. While Mexico
and Switzerland are significant at 5% level, the other 19 countries are significant at the 1% level (Table
2).
Compared to the situation under simple regression with the long time period, 1970-2004, the
introduction of the fuel share ratio variable yields the following results with respect to the income
elasticity of per capita CO2 emissions (Table 3 and Figure 4):

The number of countries in the AD group is 10. There are 4 countries which stayed in place in the
NE subgroup as under Model 1a: Germany, Luxembourg, Poland and Sweden. Of the 6 countries in
the Zero E subgroup, two of which stayed in place (Czech Republic and Switzerland) as under
Model 1a while the other four moved in from the NE subgroup (Belgium, Hungary, Slovak
Republic and the U.K.).

There are 13 countries in the RD group. Among the 13 countries, 8 stayed in place as under Model
1a: Australia, Austria, Finland, Ireland, Italy, Japan, the Netherlands and Norway. Two countries
moved in from the NE subgroup (France and the U.S.). Two moved down from the Zero E
subgroup (Canada and Iceland). And Turkey moved up from the CC group.

The other 7 countries are in the CC group. Among them, 4 countries (Greece, Mexico, New
Zealand and Portugal) stayed in place in the same group as under simple regression (Model 1a).
Two countries (Korea and Spain) moved in from the RD group under Model 1a. Finally,
Denmark moved in all the way from the NE group. With the estimated value of income elasticity
changing from -0.163 to +1.086, this change is the most dramatic. However, note that the two
explanatory variables ― per capita GDP and energy share ratio ― combined only accounted for
about 47% of the variation in per capita CO2 emissions in Denmark (Table 2).
It is observed that introducing the fuel share ratio variable into the regression will result in 18 countries
staying in the same decoupling status as under Model 1a and 12 countries changing status. As shown
above, the 18 staying-in-place countries are composed of 4 NE-, 2 Zero E-, 8 RD- and 4 CC-group
countries. In the latter case, there is more of a tendency to downgrade than to upgrade the decoupling
status of OECD countries. The following comparison shows that there are 11 downgrading versus only
one upgrading:
Downgrading
From Negative Elasticity to Zero Elasticity (4):
Belgium, Hungary,
Slovak Republic and U.K.
From Negative Elasticity to Relative Decoupling (2): France and U.S.
From Negative Elasticity to Close Coupling (1):
Denmark
From Zero Elasticity to Relative Decoupling (2):
Canada and Iceland
From Relative Elasticity to Close Coupling (2):
Korea and Spain
Upgrading
From Close Coupling to Relative Decoupling (1):
10
Turkey
Table 2:
Country
Regression Results for Model 2, with Per Capita GDP
and Fuel Share Ratio as Explanatory Variables
Intercept
Income Elasticity
Energy Share Ratio
Value
t value
Value
t value
Australia
0.119
0.551**
9.862
-0.351** -2.729
Austria
-0.56
0.732**
15.306
-0.264
-0.917
Belgium
2.295
0.033
0.36
-0.028** -3.15
Canada
1.516
0.353**
7.806
-0.218** -8.057
Czech Rep.
0.586
0.52
1.72
-0.268** -3.477
Denmark
-1.84
1.086**
3.848
-0.309** -4.523
Finland
-0.09
0.774**
8.695
-0.286** -5.663
France
0.24
0.508**
7.686
-0.252** -18.638
Germany
5.872
-1.024** -8.028
0.16**
5.006
Greece
-3.9
1.948**
13.205
-0.169
-1.513
Hungary
1.033
0.099
1.025
-0.472** -3.285
Iceland
0.41
0.561**
7.072
-0.29**
-6.616
Ireland
1.103
0.356**
9.89
-0.014
0.46
Italy
0.504
0.437**
14.859
-0.034
-0.895
Japan
-1.05
0.913**
13.223
-0.212** -9.445
Korea
-0.69
1.021**
20.288
-0.023
-1.118
Luxembourg
3.488
-0.294** -2.94
-0.179** -5.084
Mexico
-1.8
1.275**
9.87
-0.199*
-2.161
Netherlands
1.969
0.132*
2.183
-0.002
-0.159
New Zealand
-1.58
1.194**
12.093
0.062
1.372
Norway
0.462
0.456**
7.88
-0.251** -3.379
Poland
2.776
-0.328** -4.34
-0.021
-0.472
Portugal
-2.93
1.546**
52.775
-0.127** -2.485
Slovak Rep.
2.382
-0.196*
-1.879
-0.14**
-2.471
Spain
-1.85
1.194**
24.414
-0.147** -7.712
Sweden
2.879
-0.29*
-2.224
-0.251** -8.437
Switzerland
1.198
0.175
1.335
-0.066*
-2.246
Turkey
-1.24
0.82**
7.065
-0.456** -5.526
United Kingdom
1.558
0.126*
1.814
-0.159** -6.377
United States
2.117
0.175**
3.021
-0.168** -5.645
Notes:
Significance level: ** 1% ; * 5%. d.f.: degrees of freedom
Source: Developed in this study.
R2
0.925
0.916
0.426
0.685
0.533
0.469
0.733
0.975
0.828
0.852
0.653
0.61
0.935
0.909
0.89
0.986
0.818
0.849
0.245
0.851
0.765
0.805
0.989
0.867
0.973
0.945
0.179
0.986
0.891
0.63
d.f.
F-Value
2/31
2/31
2/31
2/31
2/11
2/31
2/31
2/31
2/31
2/31
2/10
2/31
2/31
2/31
2/31
2/30
2/31
2/30
2/31
2/31
2/31
2/11
2/31
2/9
2/31
2/31
2/31
2/31
2/31
2/31
197.35**
175.31**
11.86**
34.75**
7.42**
14.12**
43.97**
614.30**
77.17**
92.33**
10.33**
25.01**
228.76**
159.36**
129.44**
1084.47**
71.85**
86.91**
5.19*
91.31**
52.10**
24.78**
1392.98**
32.47**
587..08**
275.95**
3.49*
1122.52**
131.2**
27.24**
(2) Real Energy Prices
In this study, three different price series are used to represent energy prices: industrial light fuel oil prices,
industrial electricity prices and the CPI-Energy. To represent the “real” energy prices, all three series are
deflated by the CPI. Thus, the relative energy CPI is the consumer price index for energy deflated by the
overall CPI. The results with respect to the price variable’s own coefficient and the decoupling status are
explained in sequence below.
a.
Coefficient of Real Energy Price Variable
Table 4 summarizes the regression results with respect to the coefficient of the real energy price variable
when both fuel share ratio (as defined in this study) and real energy prices are included in the regression
analysis as the second and third explanatory variables.

As the sign of the regression coefficient for the price variable is expected to be negative, the more
significantly negative results the better. From this perspective, the relative energy CPI is the best
among the three price series used. Among the 30 OECD countries, 15 are with the expected sign and
11
Table 3.
Decoupling Group
Absolute Decoupling
Changes in Decoupling Groups due to the Introduction
of the Fuel Share Ratio Variable in the Regression
Classified According to Income Elasticity Estimated from
Apparent Elasticity (Model 1a)
Net Elasticity (Model 2)
Negative Elasticity
Negative Elasticity
Belgium (-0.202**),
Denmark (-0.163*),
France (-0.654**),
Germany (-0.412**),
Germany (-1.024**)
Hungary (-0.167*),
Luxemburg (-0.688**),
Luxembourg (-0.294**)
Poland (-0.355**)
Poland (-0.328**)
Slovak Republic (-0.416**),
Sweden (-1.274**),
Sweden (-0.29*)
U.K. (-0.299**),
US (-0.119**)
Zero Elasticity
Canada,
Czech Republic
Iceland,
Switzerland,
Relative Decoupling
Close Coupling
Zero Elasticity
Czech Republic
Switzerland
Belgium
Hungary
Slovak Republic
United Kingdom
Australia (0.551**)
Austria (0.732**)
Finland (0.774**)
Ireland (0.356** )
Italy (0.437**)
Japan (0.913**)
Australia (0.672**)
Austria (0.336**)
Finland (0.341**)
Ireland (0.341**)
Italy (0.421**)
Japan (0.297** )
Korea (0.970**)
Netherlands (0.124** )
Norway (0.288**)
Spain (0.876** )
Netherlands (0.132*)
Norway (0.456**)
Canada (0.353**)
France (0.508**)
Iceland (0.561**)
Turkey (0.82**)
US (0.175**)
Greece (1.948**)
Mexico (1.275**)
New Zealand (1.194**)
Portugal (1.546**)
Greece (1.976**)
Mexico (1.424**)
New Zealand (1.242**)
Portugal (1.542**)
Turkey (1.439** )
Denmark (1.086**)
Korea (1.021**)
Spain (1.194**)
Notes:
Source:
Significance level: ** 1%; * 5%
Zero elasticity: Estimate which is not statistically significant at 5% level.
Developed in this study.
12
Model 2
2.5
Close Coupling
Income Elasticity of CO2 Emissions
2
Greece
Relative Decoupling
Absolute Decoupling-zero elasticity
Portugal
1.5
Spain
Mexico
Korea
1
Turkey
Absolute Decoupling-negative elasticity
New Zealand
Denmark
Japan
Finland
Austria
France
Australia Iceland
Norway
Italy
Ireland
Canada
Switzerland
United Kingdom
United States
Netherlands
Belgium
Czech Republic
0.5
Hungary
0
Slovak Republic
Poland
Sweden
Luxembourg
-0.5
-1
Germany
-1.5
0
10000
20000
30000
40000
2004 Per Capita GDP (PPP) (US$)
50000
60000
Source: Compiled in this study from Table 3.
Figure 4: Decoupling Status with Model 2 Results
significant at 5% level or better for the relative energy CPI, compared to only 6 or 5, respectively,
for the industrial light fuel oil prices or industrial electricity prices.


Using one of the three energy price series separately in a regression yields consistent results for only
13 countries, which can be grouped into two categories:

There are only 3 countries with all three coefficients having the expected sign and are
statistically significant: Greece, Mexico and the U.S.

Ten (10) countries have neither negative nor positive sign that are statistically significant:
Australia, Denmark, Finland, France, Germany, Iceland, Luxembourg, Poland, Switzerland
and Turkey.
The other 17 countries have mixed results in the sense that using one price series will yield the
correct expected sign for the coefficient while using another price series may derive the unexpected
sign. I.e., depending upon the energy price series used, there will be conflicting results in terms of
the coefficient for the energy price variable.
b. Decoupling Status
The sign and magnitude of the estimated value of income elasticity of CO2 emissions form a hierarchy of
decoupling status. From the perspective of reducing CO2 emissions and thereby lowering the impacts
on the environment, the most desirable status is absolute decoupling (AD), for which per capita CO 2
emissions will either decline or stay the same as per capita GDP increases. The AD group is further
divided into two subgroups: negative elasticity subgroup (NE) and zero elasticity subgroup (Zero E),
with NE more desirable than Zero E. The next desirable status is relative decoupling (RD) which is
characterized by income elasticity of CO2 emissions being positive but less than unity (+1.0). Under this
status, a country would see her per capita CO2 emissions grow as per capita GDP increase at less than the
percentage increases in per capita GDP. Finally, the close coupling (CC) status is represented by income
13
Table 4: Summary of Regression Results with respect to
the Coefficient of the Real Energy Price Variable
Coefficient of
Real Industrial Light
Real Industrial Electricity Relative Energy CPI
Fuel Oil Price
Price (Model 3b)
(Model 3c)
(Model 3a)
Australia
+
+
Austria
^
^
Belgium
**
^
**
Canada
+
**
Czech Republic
^^
^^
**
Denmark
+
+
+
Finland
France
+
+
Germany
+
Greece
**
**
**
Hungary
^^
^^
Iceland
No data
No data
Ireland
*
*
Italy
^^
^^
**
Japan
**
**
Korea
+
^^
Luxembourg
+
+
+
Mexico
**
**
*
Netherlands
+
^
**
New Zealand
**
**
Norway
**
Poland
+
+
Portugal
No data
+
**
Slovak Republic
^^
^
+
Spain
^^
+
^
Sweden
+
**
Switzerland
+
Turkey
No data
U.K.
**
^^
**
U.S.
**
**
**
Notes: * Expected sign and significant at the 5% level.
**
Expected sign and significant at the 1% level.
Expected sign, but is not significant at the 5% level.
+
Unexpected sign, but is not significant at the 5% level.
^
The sign of the coefficient is not expected and is significant at the 5% level.
^^
The sign of the coefficient is not expected and is significant at the 1% level.
Source: compiled in this study.
elasticity being positive and greater than +1.0. As per capita GDP increases, per capita CO2 emissions
will increase by more than the percentage increases in per capita GDP.
In other words, there is a continuum of decoupling status in the relationship between per capita CO2
emissions and per capita GDP. Starting from the most desirable status, this continuum is represented by
(absolute decoupling ― relative decoupling ― close coupling), or (AD ― RD ― CC). When
absolute decoupling is further separated into negative elasticity and zero elasticity, the continuum is
represented by (NE ― Zero E ― RD ― CC) (Figure 5).
14
(a) 3-way classification
(b) 4-way classification
Close Coupling
Close Coupling
(CC)
(CC)
Relative Decoupling
Relative Decoupling
(RD)
(RD)
Less
Desirable
Zero Elasticity
Absolute
(Zero E)
Decoupling
(AD)
Negative Elasticity
(NE)
More
Desirable
Figure 5. Continuum of Decoupling Status
Table 5 utilizes this continuum concept and estimates of income elasticity of CO2 emissions to
summarize the results of the regression analysis conducted in this study. Results of simple regression
between per capita CO2 emissions and per capita GDP is shown in the Model 1a column. This reflects
classifying according to the apparent income elasticity of CO2 emissions. There are 15 countries in the
AD group, including 11 NE countries and 4 Zero E countries. In addition, there are 10 RDs and 5 CCs.
This is the starting point.
As additional relevant variables are introduced into the regression analysis, there are basically three
ways in which the decoupling status of a country can change. It can stay unchanged in the same group or
subgroup, deteriorate by moving from a more desirable status into a less desirable status, or improve by
moving from a less desirable status to a more desirable status. In cases of deterioration or improvement,
if the change involves two steps or more, the change can be labeled as a “major” change. For example, if
a country moves from the NE subgroup to RD group, skipping over the Zero E subgroup, a major
deterioration has occurred. As another example, changing the CC status to the Zero E status, skipping
over RD, is a major improvement. For the purpose of this study, deterioration in the decoupling status
can also be termed “downgrading;” and an improvement, “upgrading.”
As the fuel share ratio variable is included in the regression analysis, the results are shown in the Model
2 column, classifying according to the net income of elasticity of CO2 emissions. The number of
countries in the absolute decoupling group is reduced from 15 to 10. Out of the 11 NE subgroup
countries in Model 1a, only 4 stay with NE; 4 change to Zero E; 2 change to RD; and one changes to CC.
Out of the 4 Zero E subgroup countries in Model 1a, 2 stay the same; 2 change to RD. Out of the 10 RD
group countries in Model 1a, 8 stay the same and 2 change to CC. Finally, 4 of the original 5 countries in
the CC group stay in the same group, while the other one changes to the RD group. In sum, there are 10
in the absolute decoupling group, 13 in the relative decoupling group and 7 in the close decoupling
group.
When the real energy price variable is introduced into the regression analysis as the third explanatory
variable, the results are shown in the columns marked as Model 3a, Model 3b and Model 3c. The number
of countries in the most desirable NE subgroup tends to decline and the number of countries in the Zero
E subgroup tends to increase (Table 6, Column labeled AD (NE, Zero E)). In Model 3a, only Germany,
Poland and Slovak Republic stay in the NE subgroup. Sweden, Luxembourg and Hungary move into
Zero E subgroup. Belgium, France, UK and US move from the NE subgroup into the RD group. As in
the case of Model 2, Denmark “deteriorates” in a major manner directly from NE subgroup into CC
group. Switzerland and Czech Republic are the two countries staying unchanged in the Zero E subgroup,
15
Table 5: Decoupling Status According to Different Models
(Non-PPP Energy Prices)
Decoupling Status
Net elasticity w/ Net elasticity w/ fuel share ratio
fuel share ratio
and real energy prices (non-PPP)
Model 2
Model 3a
Model 3b
Model 3c
NE
NE
NE
NE
NE
NE
NE
Zero E
NE
Zero E
Zero E
Zero E
NE
Zero E
Zero E
Zero E
Zero E
NE
Zero E
Zero E
Country
Direction of Change
Germany
Poland
Sweden
Luxembourg
Slovak
Republic
Hungary
U.K.
Belgium
U.S.
France
Denmark
Switzerland
Canada
Czech
Republic
Iceland
Netherlands
Australia
Austria
Finland
Italy
Ireland
Japan
Norway
Korea
Spain
Mexico
Turkey
Greece
New
Zealand
Portugal
Absolute Decoupling
Absolute Decoupling
Absolute Decoupling
Absolute Decoupling
Absolute Decoupling
Apparent
elasticity
Model 1a
NE
NE
NE
NE
NE
Absolute Decoupling
Deteriorates
Deteriorates
Deteriorates: Major
Deteriorates: Major
Deteriorates: Major
Absolute Decoupling
Deteriorates
Deteriorates: Major
NE
NE
NE
NE
NE
NE
Zero E
Zero E
Zero E
Zero E
Zero E
Zero E
RD
RD
CC
Zero E
RD
Zero E
Zero E
RD
RD
RD
RD
CC
Zero E
RD
Zero E
Zero E
RD
RD
Zero E
RD
CC
Zero E
RD
RD
Zero E
Zero E
RD
RD
RD
CC
Zero E
RD
CC
Deteriorates
Improves
Improves
Relative Decoupling
Relative Decoupling
Relative Decoupling
Relative Decoupling
Relative Decoupling
Relative Decoupling
Deteriorates
Deteriorates
Improves: Major
Improves
Improves
Improves
Zero E
RD
RD
RD
RD
RD
RD
RD
RD
RD
RD
CC
CC
CC
CC
RD
RD
RD
RD
RD
RD
RD
RD
RD
CC
CC
CC
RD
CC
CC
Zero E
Zero E
RD
RD
RD
RD
RD
RD
Zero E
CC
Zero E
RD
RD
Zero E
RD
RD
RD
RD
RD
RD
RD
CC
CC
RD
RD
RD
CC
RD
Zero E
RD
RD
RD
RD
RD
RD
RD
RD
CC
RD
RD
RD
RD
CC
CC
-
CC
CC
Notes:
Close Coupling
NE: Negative Elasticity Zero E: Zero Elasticity
RD: Relative Decoupling
CC: Close Coupling
-: Not computed due to lack of energy price data
Model 1a: Simple regression with logarithm transformation
Model 2: Multiple regression with logarithm transformation, including the fuel share ratio variable
Model 3a: Multiple regression with logarithm transformation, including the fuel share ratio and real industrial
light fuel oil price variables
Model 3b: Multiple log-linear regression with logarithm transformation, including the fuel share ratio and real
industrial electricity price variables
Model 3c: Multiple regression with logarithm transformation, including the fuel share ratio and relative
energy CPI variables
16
while Canada is downgraded into the RD group. Iceland is unknown as data do not allow computation.
For the 10 countries originally in the RD group in Model 1a, 6 countries including Austria, Finland, Italy,
Ireland, Japan and Norway stay in the same group; 2 countries (Netherlands and Australia) improve by
moving into the Zero E subgroup; Korea is upgraded to Zero E group; and Spain deteriorates into the CC
group. For the 5 countries in the CC group under Model 1a, Mexico improves in a “major” manner into
Zero E group; Greece and New Zealand are upgraded into the RD group. Turkey and Portugal are
without data (Table 5).
For Model 3b, the results can be viewed in a similar manner as noted above in relation to Model 3a.
For Model 3c, only Germany stays the same in the NE subgroup; Switzerland stays in the Zero E
subgroup; Australia, Austria, Finland, Italy, Ireland, Japan, Korea and Norway stay in the RD group; and
Portugal stays in the CC group. If the NE and Zero E subgroups are combined into the absolute
decoupling group, altogether there are 9 countries in the AD group. In addition to Germany and
Switzerland noted above, there are also Poland, Sweden, Luxembourg, Slovak Republic, Hungary and
United Kingdom; they are downgraded from the NE subgroup to the Zero E subgroup, as well as the
Netherlands having been upgraded from the RD group into the Zero E subgroup. There are 17 countries
in the RD group. As noted above, 8 countries stay unchanged in this group. Five (5) countries are
downgraded, respectively, from the NE subgroup (Belgium, U.S. and France) or from the Zero E
subgroup (Canada and Iceland). Four (4) countries moved in from the CC group into the RD group:
Mexico, Turkey, Greece and New Zealand. There are 4 countries in the CC group. In addition to Portugal
noted above, Denmark moves in from the NE subgroup; Czech Republic moves in from the Zero E
subgroup; and Spain moves in from the RD group. The scatter diagram of Model 3c results with 2004
per capita GDP in PPP terms is shown in Figure 6.
Model 3c
2.0
1.5
Denmark
Income Elasticity of CO2 Emissions
Czech Republic
Portugal
Absolute Decoupling-zero elasticity
Absolute Decoupling-negative elasticity
Spain
1.0
Korea
Turkey
0.5
Mexico
Hungary
0.0
Close Coupling
Relative Decoupling
New Zealand
Austria
Finland
Norway
Japan
Greece
Australia
France
Iceland
Canada
Italy
Ireland
Belgium
Switzerland
Netherlands
United States
Sweden
United Kingdom
Luxembourg
Slovak Republic
-0.5
Germany
Poland
-1.0
-1.5
0
10000
20000
30000
40000
2004 Per Capita GDP (PPP) (US$)
Source: compiled in this study.
Figure 6. Decoupling Status with Model 3c Results
17
50000
60000
Notable cases of decoupling status change can be noted here. Denmark and France deserve special
mention because the status change is “major” and consistent. Once additional variables of fuel share
ratio and real energy prices are introduced into the regression analysis, Denmark’s status changes from
NE to CC, and France’s changes from NE to RD, regardless of which alternative model is adopted.
The change in decoupling status is also consistent for 4 other countries: Canada, Iceland, Spain and
Turkey. Canada and Iceland consistently change from Zero E to RD; Spain is downgraded from RD to
CC; and Turkey is upgraded from CC to RD. (In these cases, lack of data for Turkey and Iceland in some
cases is treated as neutral; i.e., it does not affect the classification of decoupling status.)
Table 6 summarizes the above results with respect to the number of countries in accordance the five
models derived in this study. Using Model 3c as an example, the RD group is the largest group with 17
countries including Belgium, US, France, Canada, Iceland, Australia, Austria, Finland, Italy, Ireland,
Japan, Norway, Korea, Turkey, Mexico, Greece and New Zealand. There are 9 countries in the AD group,
including 1 in the NE subgroup (Germany) and 8 in the Zero E group (Poland, Sweden, Luxembourg,
Slovak Republic, Hungary, UK, Switzerland and the Netherlands). Finally, there are 4 countries in the
CC group: Denmark, Czech Republic, Spain and Portugal.
Using the simple regression results (Model 1a) as the starting point, it is observed that the number of
countries in the AD group tends to decrease, and the number of countries in the RD group tend to
increase, as additional explanatory variables such as fuel share ratio and real energy prices are
introduced into the regression analysis. The AD group changes from 15 in Model 1a to 9 in Model 3c.
This change reflects the different directions of change in the two subgroups within this group. The NE
subgroup decreases from 11 to 1, while the Zero E subgroup increases from 4 to 8. Correspondingly, the
number of countries in the RD group increases from 10 to 17. Finally, the number of countries in the CC
group changes from 5 to 4.
Table 6.
Model
1a
2
3a
3b
3c
Number of Countries in Different Decoupling Status, By Model
AD (NE, Zero E)
RD
CC
No Data
15 (11, 4)
10
5
0
10 (4,6)
13
7
0
12 (3, 9)
13
2
3
9 (2,7)
15
5
1
9 (1,8)
17
4
0
Notes:
AD: Absolute Decoupling
NE: Negative Elasticity Zero E: Zero Elasticity
RD: Relative Decoupling
CC: Close Coupling
No Data: Not computed due to lack of energy price data
Source: Compiled from Table 5.
4. Conclusions
Relationships between CO2 emissions and GDP can be absolutely decoupled, relatively decoupled, or
closely coupled. Such relationships can conceptually be represented by the income elasticity of per
capita CO2 emissions being less than or equal to zero, or positive but less than unity (+1.0), or positive
and greater than unity, respectively. Thus, in terms of decreasing desirability, the continuum of
decoupling status goes from absolute decoupling (AD) to relative decoupling (RD), and then to close
coupling (CC). The AD status can be further divided into two areas, negative elasticity (NE) and zero
elasticity (Zero E), with NE more desirable than Zero E. This paper has applied simple and multiple
regressions to 1970-2004 data to derive estimates of income elasticity of per capita CO2 emissions for
the 30 OECD countries, with fuel share ratio and real electricity prices as additional explanatory
variables, and then used such estimates to classify the countries into different decoupling status groups.
18
Simple regression (Model 1a and 1b) yields “apparent” income elasticity and multiple regression (Model
2, 3a, 3b and 3c) derives “net” income elasticity of per capita CO2 emissions. The main results are as
follows:

For 14 of the 30 countries, the decoupling status is not affected by the choice of models.

Seven countries (Germany, Hungary, Luxembourg, Poland, Slovak Republic, Sweden and
Switzerland) would stay in the AD group, regardless of the model in question. Among them,
Germany maintains the NE status under Model 1a and Switzerland maintains the Zero E
status. The other 5 countries downgrade from NE status to Zero E status when the additional
variables are included in the regression.

Six countries (Austria, Finland, Ireland, Italy, Japan and Norway) stay in the RD group.

Portugal stays with the CC status.

For 8 countries (Belgium, Canada, Czech Republic, Demark, France, Iceland, United Kingdom,
and United States), the decoupling status deteriorates from AD in Model 1a to RD in Model 2 or
the three versions of Model 3 as fuel share ratio and real energy prices variables are introduced into
the regression. Among them, Czech Republic, Denmark, France and U.S. show “major”
deterioration, moving from NE to CC or RD, or from Zero E to CC, skipping at least one of the
status segment in the four-way classification of the decoupling status continuum.

Korea and Spain also show deterioration from RD to CC.

For the other 6 countries (Australia, Greece, New Zealand, Netherlands, Mexico and Turkey), the
decoupling status improves as fuel share ratio and real energy prices variables are introduced.
Netherlands and Australia improve from RD to Zero E while the other four countries improve from
CC to RD. Mexico shows major improvement as industrial fuel oil prices are introduced.

Among the three energy price series tested in this study, the relative CPI-Energy is the best price
data series for inclusion in multiple regression analysis because it yields price variable coefficient
that is with the expected sign and is statistically significant for the largest number of countries, 15
versus 5 or 6 for the other two price series.
Given the data and methods used in this analysis, it appears that for a group of 14 countries, using the
apparent elasticity derived from simple regression to classify a country’s decoupling status will do it
correctly for the three-way classification. Introduction of the fuel share ratio and real energy price
variables will not affect the classification, although the absolute values of the elasticity estimates may be
different. More specifically, Germany, Hungary, Luxembourg, Poland, Slovak Republic, Sweden and
Switzerland would stay in the AD group; Austria, Finland, Ireland, Italy, Japan and Norway stay in the
RD group; Portugal stays with the CC status.
For the other 16 countries, utilizing the apparent elasticity estimate as the guide will prove to be off the
mark. For Belgium, Canada, Czech Republic, Denmark, France, Iceland, Korea, Spain, United Kingdom
and United States, it would be over-optimistic. For Australia, Greece, Mexico, Netherlands, New
Zealand and Turkey, it would tend to be overly pessimistic. Hence, for these latter 16 countries, it would
be necessary to introduce the fuel share ratio and real energy prices variables into the regression. In a
sense, this finding is consistent with the consensus, previously noted in Section 2.1, that the empirically
derived EKC would largely depend on the characteristics of the countries, the models and the variables
incorporated into the models.
Several qualifications to the above results should be noted. First, given that the models adopted are
simple and multiple regressions of natural logarithm-transformed variables, constant elasticity is
assumed. Also, the direction of causality is presumed to go from GDP or income to CO2 emissions; there
19
is no attempt to test the reverse direction of causality as were done by Coondoo and Dinda (2002). In this
context, it would be useful to further explore the factors and conditions contributing to the results
derived in this study. Why do some countries remain unchanged in their decoupling status while others
change status in the directions they did change? Moreover, at this stage, time-series data of individual
countries are used in each regression for a country separately; there is no pooling of data series. Thus, it
would be possible to pooling the time-series data of the countries together to do additional hypothesis
testing regarding different types of relationships.
5. References
Coondoo, D., and Dinda, S. 2002. Causality between income and emission: a country group-specific
econometric analysis, Ecological Economics 40 3. March 2002. pp. 351-367
Fang, J.M., and Chen, J.C. 2007. Comparative analysis of CO2 emissions and economic grow
relationships in Taiwan and Sweden, Sci-Tech Policy Review – International Journal, Volume 1 Issue 1,
June 2007. pp.43-60.
International Energy Agency (IEA) (2006a). CO2 Emissions from Fuel Combustion 1971-2004 -- 2006
Edition. OECD/IEA. Paris, France.
IEA (2006b). Key World Energy Statistics 2006. STEDI MEDIA. France.
IEA (2006c). World Energy Statistics and Balances, 2006. Online Database. http://titania.sourceoecd.org
IEA (2007). Energy Prices and Taxes - Online Database. http://titania.sourceoecd.org
Richmond, A.K., and Kaufmann, R.K. (2006). Energy prices and turning points: the relationship between
income and energy use/carbon emissions, The Energy Journal, 27 4, 2006. pp. 157-180.
Shafik, N.,and Bandyopadhyay, S. (1992). Economic growth and environmental quality: time series and
cross section evidence. Working Paper. World Development Report 1992, Oxford University Press,
New York.
20