Applied Energy 88 (2011) 1986–1996

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

A note on the environmental Kuznets curve for CO2: A pooled mean group approach
Hiroki Iwata a,1, Keisuke Okada b,⇑, Sovannroeun Samreth b,c
a
Faculty of Socio-Environmental Studies, Fukuoka Institute of Technology, 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka 811-0295, Japan
b
Graduate School of Economics, Kyoto University, Yoshida-honmachi, Sakyo-ku, Kyoto 606-8501, Japan
c
Japan Society for the Promotion of Science (JSPS), Japan

a r t i c l e i n f o a b s t r a c t

Article history: This paper investigates whether the environmental Kuznets curve (EKC) hypothesis for CO2 emissions is
Received 30 April 2010 satisfied using the panel data of 28 countries by taking nuclear energy into account. Using the pooled
Received in revised form 29 October 2010 mean group (PMG) estimation method, our main results indicate that (1) the impacts of nuclear energy
Accepted 6 November 2010
on CO2 emissions are significantly negative, (2) CO2 emissions actually increase monotonically within the
Available online 8 December 2010
sample period in all cases: the full sample, OECD countries, and non-OECD countries, and (3) the growth
rate in CO2 emissions with income is decreasing in OECD countries and increasing in non-OECD countries.
Keywords:
Ó 2010 Elsevier Ltd. All rights reserved.
CO2
Environment
EKC
Nuclear power
Panel data

1. Introduction between CO2 emissions and GDP in China, the world’s largest
CO2 emitter.2
Global warming has attracted a great deal of attention in recent The current study investigates whether the EKC hypothesis for
years. Since carbon dioxide (CO2) emissions are considered the CO2 emissions is satisfied when considering the role of nuclear en-
main source of global warming, the United Nations and individual ergy in the panel data analysis of 28 countries. The pooled mean
countries have attempted to take various measures to reduce CO2 group (PMG) estimator developed by Pesaran et al. [13] is em-
emissions. Therefore, it is interesting to consider the environmen- ployed as the estimation methodology. This approach allows heter-
tal Kuznets curve (EKC) hypothesis for CO2 emissions. The EKC ogeneous dynamics in the short-run but imposes homogenous
hypothesis claims that an inverted U-shaped relationship exists relationships in the long-run. Results using the mean group (MG)
between income and environmental pollutants (e.g., Selden and estimation proposed by Pesaran and Smith [14] and the dynamic
Song [1] and Grossman and Krueger [2]). Although nuclear power fixed effects (DFE) estimation are also reported.
generation requires huge expenditures for safety management,
the adoption of nuclear power is one of the main measures used
to cope with global warming because it generates less CO2 emis- 2. Data and models
sions than other resources such as oil, coal, and natural gas.
Although previous studies have investigated the EKC hypothesis This study focuses on the effects of nuclear energy in addition to
for CO2 emissions based on panel data, there is no clear evidence the standard EKC hypothesis because the production of electricity
for the hypothesis (e.g., Martínez-Zarzoso and Bengochea-Moran- by nuclear power may result in reduced CO2 emissions. We utilize
cho [3], and Richmond and Kaufmann [4]). In more recent studies, the annual data of 28 countries (17 OECD and 11 non-OECD coun-
Apergis and Payne [5,6] point out the EKC for CO2 emissions in the tries) that used nuclear power in electricity production.3 The esti-
cases of Central America and the Commonwealth of Independent mation sample of our study consists of an unbalanced panel.
States, while Narayan and Narayan [7] show the existence of the 2
Another strand of literature is the case study on the relationship between GDP
EKC for CO2 emissions in some selected developing countries. and energy consumption. Payne [9] and Ozturk and Acaravci [10] investigate the
Furthermore, Chang [8] demonstrates bi-directional causality causal relationship between energy consumption and GDP for the case of Eastern
European countries and the United States. Narayan et al. [11] and Apergis and Payne
[12] focus on the relationship between electricity consumption and GDP and between
⇑ Corresponding author. Tel.: +81 75 753 3400; fax: +81 75 753 3492. coal consumption and GDP, respectively.
3
E-mail address: [email protected] (K. Okada). Our study excludes Italy, Armenia, and Romania because nuclear power has not
1
This research was partly conducted when Hiroki Iwata was a research fellow at been used since 1988 in Italy and the sample sizes are extremely small for Armenia
the Graduate School of Global Environmental Studies, Kyoto University, Japan. and Romania.

0306-2619/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.apenergy.2010.11.005
 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996 1987

Table 1 Table 2
List of countries in our sample. Baseline estimation results: ARDL(1,1,1,1), Dependent variable: ln co2 .

CO2 GDP per capita Nuclear (1) (2) (3)

Mean Pooled mean Dynamic fixed
1 Argentina 1960–2003 1975–2003 1974–2003
group group effects
2 Armenia 1960–2003 1990–2003 1995–2003
3 Belgium 1960–2003 1975–2003 1962–2003 Long-run coefficients
4 Brazil 1960–2003 1975–2003 1984–2003 ln y 59.001 3.511*** 2.823**
5 Bulgaria 1960–2003 1980–2003 1974–2003 (68.377) (0.509) (1.172)
6 Canada 1960–2003 1975–2003 1963–2003
ðln yÞ2 3.363 0.156*** 0.119*
7 China 1960–2003 1975–2003 1993–2003
8 Czech Republic 1960–2003 1990–2003 1983–2003 (3.862) (0.028) (0.064)
9 Finland 1960–2003 1975–2003 1977–2003 ln nuc 0.196 0.199*** 0.164***
10 France 1960–2003 1975–2003 1960–2003 (0.287) (0.024) (0.051)
11 Germany 1960–2003 1975–2003 1961–2003
Error correction 0.513*** 0.251*** 0.108***
12 Hungary 1960–2003 1975–2003 1983–2003
coefficient
13 India 1960–2003 1975–2003 1971–2003
(0.066) (0.047) (0.017)
14 Italy 1960–2003 1975–2003 1963–1987
15 Japan 1960–2003 1975–2003 1966–2003
Short-run coefficients
16 Korea 1960–2003 1975–2003 1977–2003
D ln y 11.195 11.214 0.501
17 Lithuania 1960–2003 1990–2003 1992–2003
(40.882) (10.113) (0.904)
18 Mexico 1960–2003 1975–2003 1989–2003
19 Netherlands 1960–2003 1975–2003 1968–2003 Dðln yÞ2 0.233 0.485 0.054
20 Pakistan 1960–2003 1975–2003 1971–2003 (2.274) (0.524) (0.049)
21 Romania 1960–2003 1980–2003 1996–2003
22 Russian Federation 1960–2003 1989–2003 1992–2003 D ln nuc 0.050 0.038 0.005
23 Slovak Republic 1960–2003 1984–2003 1972–2003 (0.069) (0.037) (0.005)
24 Slovenia 1960–2003 1990–2003 1992–2003 Constant 19.673 4.214*** 1.492**
25 South Africa 1960–2003 1975–2003 1984–2003 (39.901) (0.787) (0.687)
26 Spain 1960–2003 1975–2003 1968–2003
27 Sweden 1960–2003 1975–2003 1965–2003 No. of countries 28 28 28
28 Switzerland 1960–2003 1975–2003 1969–2003 Observations 625 625 625
29 Ukraine 1960–2003 1987–2003 1992–2003 Hausman test 0.90
30 United Kingdom 1960–2003 1975–2003 1960–2003 Turning point of y 6438 73,571 138,829
31 United States 1960–2003 1975–2003 1960–2003
Notes: 1. The numbers in parentheses are standard errors.
Notes: 1. Our sample excludes Italy, Armenia, and Romania. 2. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and
2. For Belgium, nuclear data is not available in 1965 and 1971. 10% of significance levels, respectively.

Although the data of CO2 in each country in consideration are avail-

able from 1960, the data of GDP per capita at purchasing power par- Using the PMG approach, the lag orders of the variables in con-
ity in 2000 international dollars are available only from 1975.4 As for sideration have to be determined. Therefore, we conduct the lag
the data of nuclear power used in electricity production, the sample selection using the Akaike Information Criterion (AIC) and Schwarz
period is different across countries and is shorter than those of CO2 Bayesian Criterion (SBC). By setting one as the maximum lag and
and GDP per capita in many countries, especially in the non-OECD running the estimation with various cases of the autoregressive
group. Table 1 provides detailed descriptions on the sample size of distributed lag (ARDL) model, the results from these two criteria
each variable in each country. One of the contributions of this paper show that the ARDL(1,1,1,1) is the most appropriate form. Eq. (1)
is to adopt the panel data analysis because it enables us to take into can be written in the form of the ARDL(1,1,1,1) model as follows:
account the effects of nuclear power in the framework of the EKC in
non-OECD countries whose sample period is not sufficient for indi- lnðco2 Þit ¼ li þ b10i ln yit þ b11i ln yit1 þ b20i ðln yit Þ2
vidual country analysis.
þ b21i ðln yit1 Þ2 þ b30i ln nucit þ b31i ln nucit1
CO2 emissions (co2 ) are measured in metric tons per capita. Real
GDP (y) is GDP per capita at purchasing power parity in 2000 inter- þ ki lnðco2 Þit1 þ git : ð2Þ
national dollars. Electricity produced from the nuclear source (nuc)
is the percentage of total electricity produced. All data are obtained Following Pesaran et al. [13], the estimation Eq. (1) can be
from the World Development Indicators (WDI) 2007 CD-ROM re- rewritten as an error correction representation of the ARDL model
leased by the World Bank. as follows:
Our estimation model can be written as the following logarith-
mic form: D lnðco2 Þit ¼ /i ½lnðco2 Þit1  a0i  a1i ln yit
 a2i ðln yit Þ2  a3i ln nucit   b11i D ln yit
lnðco2 Þit ¼ a0i þ a1i ln yit þ a2i ðln yit Þ2 þ a3i ln nucit þ eit
; ð1Þ  b21i Dðln yit Þ2  b31i D ln nucit þ git ; ð3Þ
i ¼ 1; 2; . . . ; N; t ¼ 1; 2; . . . ; T:
li
where e is an error term.5 where a0i ¼ 1k
i
þb11i
, a1i ¼ b10i1k
i
þb21i
, a2i ¼ b20i1k
i
þb31i
, a3i ¼ b30i1k
i
, and /i ¼
ð1  ki Þ. The coefficient of the error correction term /i represents
4
For some countries, the available sample period for GDP per capita at purchasing the speed of adjustment of lnðco2 Þ toward its long-run equilibrium
power parity in 2000 international dollars is even shorter. For detailed information on following the shock to the short-run.
the sample size of each variable in each country, see Table 1. The model is estimated for the full sample and then separately
5
Narayan and Narayan [7] verify the EKC hypothesis for CO2 based on the short- for OECD and non-OECD countries. To check the stationary proper-
run and long-run income elasticises to avoid the multicollinearity problems resulting
from simultaneously adding the income and its squared terms into the estimation
ties of the variables, we first conduct the unit root tests in individual
equation. However, following many previous studies on the EKC, we simultaneously countries by using the standard Augmented Dickey–Fuller (ADF),
include the income and its squared terms in our specification of Eq. (1). Phillips–Perron (PP), and Kwiatkowski–Phillips–Schmidt–Shin
1988 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996

Table 3
OECD and non-OECD estimation results: ARDL(1,1,1,1), Dependent variable: ln co2 .

OECD Non-OECD
(1) (2) (3) (4) (5) (6)
Mean group Pooled mean group Dynamic fixed effects Mean group Pooled mean group Dynamic fixed effects
Long-run coefficients
ln y 3.788 2.864*** 7.660*** 156.042 9.733*** 4.721**
(23.353) (0.960) (2.314) (170.977) (2.926) (2.192)

ðln yÞ2 0.127 0.105** 0.361*** 8.758 0.591*** 0.251*

(1.200) (0.050) (0.120) (9.705) (0.173) (0.139)
ln nuc 0.100 0.122*** 0.283*** 0.343 1.267*** 0.094
(0.173) (0.013) (0.061) (0.700) (0.148) (0.067)
Error correction coefficient 0.515*** 0.269*** 0.136*** 0.511*** 0.148* 0.125***
(0.068) (0.051) (0.023) (0.137) (0.082) (0.034)

Short-run coefficients
D ln y 23.975* 40.566** 0.435 8.554 9.675 0.303
(14.409) (17.859) (2.196) (104.448) (15.330) (1.423)

Dðln yÞ2 1.158 2.011** 0.002 1.197 0.675 0.042

(0.731) (0.912) (0.114) (5.821) (0.832) (0.081)
D ln nuc 0.057* 0.042 0.012 0.039 0.047 0.007
(0.033) (0.028) (0 .011) (0.172) (0.088) (0.006)
Constant 26.977 4.197*** 5.110*** 91.769 6.724* 2.593*
(38.907) (0.805) (1.933) (79.644) (3.838) (1.380)

No. of countries 17 17 17 11 11 11
Observations 426 426 426 199 199 199
Hausman test 3.63 1.59
Turning point of y 2,922,646 813,418 39,498 7391 3729 11,802

Notes: 1. The numbers in parentheses are standard errors.

2. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and 10% of significance levels, respectively.

(KPSS) tests, and the unit root test with structural breaks suggested conducted for the homogeneity of long-run coefficients between
by Narayan and Popp [15]. We adopt the Narayan and Popp [15] the PMG and MG estimators. Pesaran et al. [13] indicate that the
approach because it is relevant in the literature for the reason that PMG estimator is consistent and more efficient than the MG esti-
it selects the break dates with greater precision by allowing for two mator under the homogeneity hypothesis.
endogenous structural breaks in the level and trend of the data ser-
ies. The precise selection of breaks is very important since the incor- 3. Empirical results
rect selection of breaks biases the rejection of the null hypothesis.
This application of Narayan and Popp [15] approach for the unit We start with conducting the unit root and cointegration tests.
root test with structural breaks is one of the contributions of this The detailed results of the unit root tests in individual countries,
study in the literature. In addition, we perform the unit root tests the panel unit root tests, and the panel cointegration tests are illus-
in panel form by using Im et al. (IPS) [16], Levin et al. (LLC) [17], trated in Appendices A, B, and C. As reported in Table A1, the
Breitung [18], Fischer-ADF proposed by Maddala and Wu [19] and results of the standard unit root tests such as the ADF, PP, and KPSS
Choi [20], and Hadri [21] tests. Second, to confirm whether a tests in each country indicate that while the orders of integration
long-run relationship exists between variables, this study performs of lnðco2 Þ, ln y, and ðln yÞ2 are one (i.e., I(1)) in most cases, the evi-
the panel cointegration tests without structural breaks proposed by dence showing that lnðnucÞ is I(1) is somewhat weak. Specifically,
Pedroni [22,23] and the panel cointegration tests with structural when only an intercept is considered, lnðnucÞ is integrated of order
breaks suggested by Westerlund [24]. Finally, the model is esti- zero (i.e., I(0)) in 24 out of 28 countries and when both an intercept
mated by employing the PMG method as well as the MG and DFE and a time trend are considered in the test, lnðnucÞ is shown to be
methods for the purpose of comparison. I(0) in 16 out of 28 countries. Hence, the results seem to provide
The PMG estimator constrains the long-run coefficients to be evidence that lnðnucÞ is I(0) in many countries. However, when
identical but allows the short-run coefficients, speed of adjustment we apply the unit root test in individual countries considering
(the coefficient of error correction term), and error variance to dif- the structural breaks developed by Narayan and Popp [15], the re-
fer across countries. Given that the number of countries (N) and the sults in Table A2 suggest that the variables in consideration are I(1)
length of the time series (T) in our analysis are sufficiently large, in most cases.
employing the PMG estimation is appropriate. On the other hand, Based on the IPS, LLC, Breitung, and Fischer-ADF tests for the
the DFE estimator requires all slope coefficients to be homoge- unit root test in panel form, the panel unit root null hypothesis
neous, and allows only intercepts to be different across countries. cannot be rejected for lnðco2 Þ, ln y, and ðln yÞ2 in all cases of the full
However, Pesaran et al. [13] show that the generalized method of sample, OECD countries, and non-OECD countries in the model
moment (GMM) estimation in the dynamic panel model has incon- with an intercept and a time trend, as shown in Tables A3 and
sistent long-run coefficients unless they are identical. The MG esti- A4, whereas the null hypothesis of no panel unit root can be re-
mator has no restriction on the coefficients across countries since it jected in the Hadri test for these variables from Table A5. Hence,
is the average of regression coefficients in each country. The PMG we can conclude that, in a panel form, lnðco2 Þ, ln y, and ðln yÞ2 are
estimator can, therefore, be an intermediate estimator between I(1) processes. For lnðnucÞ, while the panel unit root null hypothe-
the MG and DFE estimators. Furthermore, the Hausman test is sis can be rejected in the IPS, LLC, Breitung, and Fischer-ADF tests,
 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996 1989

it cannot be rejected in the Hadri test as indicated in Table A5; these countries. However, although this result satisfies the inverse
thus, it is inconclusive whether lnðnucÞ is I(0) or I(1). Despite these U-shaped curve, it should be interpreted with caution since the
mixed results, as pointed out by Kim et al. [25], the PMG estima- turning point of income is outside the sample period in these coun-
tion of an ARDL regression provides consistent estimators irrespec- tries. Because most human activities such as production, consump-
tive of whether the variables in consideration are I(0) or I(1) if tion, and transportation cause CO2 emissions, it is difficult to
there exists a unique vector defining the long-run relationship control such emissions with existing environmental regulations.
among variables with suitable lag orders chosen. Therefore, since The scope of the damage from CO2 emissions is global, and the dam-
no variables are I(2) or higher, the mixed integration order of vari- age they cause will also impact future generations, leading to
ables is not problematic. decreasing incentives to address global warming in the present
Next, we conduct Pedroni panel cointegration tests for the cases generation. Compounding the problem is the fact that the econo-
where only an intercept is included and where both an intercept mies of non-OECD countries are experiencing rapid growth, and
and a time trend are considered. As indicated in Table A6, the re- the share of secondary industries and consumption are growing
sults reject the null of no cointegration in at least four out of seven rapidly. In addition, environmentally friendly technologies are not
tests at the 5% significance level.6 Since panel cointegration tests do actively adopted. These factors will cause CO2 emissions to increase
not provide overwhelming evidence on cointegration relationships, in these countries despite the reduction effects of nuclear power on
we focus on the error correction coefficient to confirm the cointegra- CO2 emissions. There is a possibility that CO2 emissions in develop-
tion relationship. If this coefficient is negative and less than unity in ing countries will not be decreased without further appropriate
absolute value, a long-run relationship exists. measures against global warming, even if advanced technology
Table 2 contains the baseline results for the full sample. From such as nuclear power generation is introduced.
column (2) in the table, a significant and negative value of the error These results suggest the importance of cooperation between
correction coefficient ensures the cointegration relationship be- OECD and non-OECD countries in dealing with global warming.
tween considered variables. We can, therefore, argue that there is The continuous economic development of non-OECD countries will
a cointegration (long-run) relationship among considered variables increase CO2 emissions. Accordingly, OECD countries should be
in the case of the full sample. Furthermore, since the Hausman test more active in transferring advanced environmentally friendly
cannot reject the null hypotheses of the homogeneity of long-run technologies to non-OECD countries, which would improve their
coefficients, the PMG estimator is preferable to the MG estimator. energy efficiency and lead to the reduction in global CO2 emissions.
Column (2) in Table 2 shows that the EKC for CO2 seems to be pro- In the future, nuclear power may be introduced in many develop-
ven from the estimated long-run coefficients of lnðyÞ and ðln yÞ2 . ing countries with financial and technological assistance from
However, since the calculated turning point is outside the sample developed countries. Although our estimation results indicate that
period, CO2 emissions are actually increasing monotonically. nuclear power alone is not necessarily sufficient to decrease CO2
The estimation results of OECD and non-OECD countries are re- emissions, it can play an important role in doing so. However, it
ported in Table 3. As in Table 2, the estimated coefficients of error is also necessary to bear in mind that it is very costly to ensure
correction terms are also significantly negative and smaller than the safety management of nuclear energy generation to prevent
unity in columns (2) and (4), implying the existence of cointegra- any accidents that may be detrimental to both the environment
tion relationships among variables in consideration in both cases. and humans.
Furthermore, the Hausman tests suggest that the PMG estimators
are preferable to the MG estimators in both cases. The PMG estima-
tion results in columns (2) and (4) in Table 3 indicate that CO2 4. Conclusion
emissions are concave in income in the former and convex in the
latter, controlling nuclear energy. CO2 emissions, however, are This study investigates whether the EKC for CO2 emissions is
increasing monotonically within the sample period in OECD coun- proven, considering the role of nuclear energy in the panel data
tries because the calculated turning point is outside the sample analysis of 28 countries. CO2 emissions actually increase monoton-
period. Since the turning point lies on the value where the income ically in all cases: the full sample, OECD countries, and non-OECD
level is quite low inside the sample period, CO2 emissions are countries. Additionally, the effects of nuclear energy on CO2 emis-
roughly increasing monotonically in non-OECD countries. Further- sions are significantly negative in all cases. These results suggest
more, although the effects of nuclear energy in reducing CO2 emis- that the differences between developed and developing countries
sions are larger in non-OECD countries, the increase in nuclear must be considered when taking measures against global warming.
energy in electricity production does not necessarily lead to the
satisfaction of the EKC hypothesis. It is noteworthy that based on
the DFE estimation results, the magnitudes of the long-run coeffi- Acknowledgements
cients are slightly different from those based on the PMG estima-
tion, but their signs are the same in all cases. We are grateful to Akihisa Shibata for many helpful comments
Our estimation shows the inverse U-shaped relation between in- and encouragements, and to Paresh Kumar Narayan and Joakim
come and CO2 emissions in the case of OECD countries. These re- Westerlund for providing us with the Gauss codes used in their pa-
sults may be attributed to the fact that most OECD countries pers. The comments and suggestions from two anonymous review-
employ advanced technologies which have less effect on global ers of this journal are also gratefully acknowledged. Samreth
warming. Furthermore, the share of tertiary industries, which cause wishes to acknowledge the financial support from the Japan Soci-
less CO2 emissions, is higher than that of secondary industries in ety for the Promotion of Science (JSPS). We are, however, solely
responsible for any mistakes in this paper.

6
We also conduct the panel cointegration test with structural breaks developed by
Westerlund [24]. The estimated values of the Lagrange multiplier are 13.287, 27.556, Appendix A. Results of the unit root tests in individual
and 23.364 for the full sample, OECD, and non-OECD cases, respectively. These results countries.
indicate that there is no evidence of cointegration relationship between considered
variables. However, in such circumstances, we focus on the error correction
coefficient to confirm the cointegration relationship, following Kim et al. [25] and See Table A1.
Goswami and Junayed [26]. See Table A2.
1990 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996

Table A1
Unit root test without structural breaks in individual countries.

ADF-test PP-test KPSS-test

Level First difference Level First difference Level First difference
Sample Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept
and and and and and and
trend trend trend trend trend trend
Argentina
ln co2 1960–2003 2.597 1.732 5.680*** 6.023*** 2.625* 1.658 5.639*** 5.993*** 0.444* 0.178* 0.312 0.083
ln y 1975–2003 1.878 1.951 4.583*** 4.486*** 1.981 2.053 4.553*** 4.450*** 0.168 0.112 0.107 0.103
(ln y)2 1975–2003 1.874 1.947 4.584*** 4.487*** 1.979 2.052 4.554*** 4.451*** 0.169 0.112 0.106 0.103
ln nuc 1974–2003 3.578** 3.259* 8.033*** 7.879*** 3.617** 3.381* 8.668*** 8.715*** 0.297 0.167* 0.303 0.076
Belgium
ln co2 1960–2003 1.399 2.380 5.429*** 5.456*** 1.515 2.383 5.359*** 5.358*** 0.397* 0.135* 0.203 0.512***
ln y 1975–2003 1.137 2.919 5.656*** 5.527*** 1.155 3.091 5.610*** 5.490*** 0.695** 0.049 0.124 0.064
(ln y)2 1975–2003 1.026 2.848 5.599*** 5.469*** 1.039 3.033 5.560*** 5.438*** 0.694** 0.049 0.108 0.064
ln nuc 1962–2003 4.502*** 3.533** 7.160*** 6.817*** 4.631*** 3.512* 7.008*** 6.703*** 0.498** 0.169** 0.404* 0.138*
Brazil
ln co2 1960–2003 2.090 0.952 4.103*** 4.386*** 1.856 1.353 4.182*** 4.447*** 0.715** 0.158** 0.230 0.065
ln y 1975–2003 2.373 3.093 4.518*** 4.415*** 2.388 3.195 4.523*** 4.423*** 0.644** 0.048 0.105 0.061
(ln y)2 1975–2003 2.350 3.076 4.516*** 4.414*** 2.367 3.181 4.521*** 4.422*** 0.644** 0.048 0.103 0.061
ln nuc 1984–2003 2.558 3.302* 6.126*** 6.225*** 2.577 3.260* 6.362*** 6.559*** 0.354* 0.133* 0.156 0.084
Bulgaria
ln co2 1960–2003 3.393** 2.857 6.152*** 7.642*** 3.248** 3.221* 6.188*** 7.645*** 0.243 0.213** 0.527** 0.130*
ln y 1980–2003 1.270 1.257 2.414 2.343 1.713 1.713 2.486 2.403 0.110 0.094 0.132 0.133*
(ln y)2 1980–2004 1.259 1.248 2.407 2.338 1.707 1.708 2.481 2.399 0.109 0.094 0.132 0.132
ln nuc 1974–2003 7.297*** 9.117*** 5.728*** 5.574*** 6.030*** 9.077*** 7.568*** 7.268*** 0.668** 0.177** 0.442* 0.164**
Canada
ln co2 1960–2003 2.298 2.020 6.191*** 6.350*** 2.292 2.026 6.264*** 6.389*** 0.370* 0.156** 0.234 0.162**
ln y 1975–2003 0.235 1.615 3.500** 3.458* 0.350 2.039 3.508** 3.458* 0.680** 0.085 0.082 0.078
(ln y)2 1975–2003 0.159 1.555 3.479** 3.444* 0.288 1.988 3.489** 3.445* 0.680** 0.087 0.084 0.078
ln nuc 1963–2003 2.758* 1.364 6.972*** 8.069*** 3.002** 1.237 6.931*** 8.074*** 0.594** 0.192** 0.390* 0.067
China
ln co2 1960–2003 0.187 3.942** 4.593*** 4.467*** 0.195 3.945** 4.629*** 4.474*** 0.770*** 0.115 0.192 0.124*
ln y 1975–2003 1.206 3.877** 4.601*** 4.317** 1.010 3.769** 4.516*** 4.279** 0.689** 0.099 0.200 0.089
(ln y)2 1975–2003 2.204 3.465* 4.116*** 3.968** 1.830 3.301* 4.077*** 3.983** 0.688** 0.157* 0.369* 0.081
ln nuc 1993–2003 9.241*** 9.474*** 13.065*** 15.269*** 8.267*** 8.436*** 10.838*** 14.787*** 0.423* 0.137* 0.247 0.186**
Czech Republic
ln co2 1960–2003 1.268 1.811 4.108*** 4.627*** 1.465 1.755 4.057*** 4.604*** 0.226 0.213** 0.580** 0.100
ln y 1990–2003 0.120 6.337*** 5.783*** 4.788*** 0.422 5.808*** 6.099*** 4.983*** 0.474** 0.115 0.308 0.115
(ln y)2 1990–2003 0.105 6.310*** 5.763*** 4.780*** 0.406 5.789*** 6.079*** 4.972*** 0.474** 0.116 0.310 0.115
ln nuc 1983–2003 13.847*** 10.861*** 6.321*** 4.870*** 14.927*** 11.438*** 7.310*** 5.389*** 0.435* 0.147** 0.439* 0.184**
Finland
ln co2 1960–2003 3.048** 2.464 5.200*** 5.576*** 3.145** 2.471 5.241*** 5.568*** 0.605** 0.175** 0.337 0.121*
ln y 1975–2003 0.353 1.333 2.524 2.472 0.565 1.891 2.715* 2.661 0.647** 0.085 0.072 0.072
(ln y)2 1975–2003 0.288 1.320 2.523 2.467 0.521 1.891 2.712* 2.656 0.648** 0.084 0.072 0.072
ln nuc 1977–2003 4.172*** 3.260* 4.159*** 5.342*** 4.172*** 3.529* 4.236*** 5.358*** 0.228 0.141* 0.353* 0.126*
France
ln co2 1960–2003 1.300 2.481 6.537*** 6.841*** 1.417 2.483 6.553*** 6.831*** 0.371* 0.159** 0.300 0.130*
ln y 1975–2003 1.915 2.435 3.359** 3.415* 1.628 2.654 3.346** 3.422* 0.695** 0.104 0.190 0.057
(ln y)2 1975–2003 1.763 2.404 3.358** 3.394* 1.507 2.652 3.350** 3.404** 0.695** 0.095 0.169 0.056
ln nuc 1960–2003 4.987*** 1.417 4.601*** 5.904*** 4.686*** 1.422 4.631*** 5.897*** 0.770*** 0.212** 0.716** 0.065
Germany
ln co2 1960–2003 0.308 2.339 5.881*** 6.377*** 0.422 2.340 5.864*** 6.388*** 0.517** 0.211** 0.466** 0.110
ln y 1975–2003 2.232 1.645 3.325** 3.378* 1.936 2.107 3.330** 3.396* 0.691** 0.108 0.255 0.065
(ln y)2 1975–2003 2.116 1.621 3.311** 3.355* 1.843 2.097 3.316** 3.373* 0.691** 0.102 0.239 0.065
ln nuc 1961–2003 4.141*** 2.063 7.286*** 8.783*** 4.725*** 2.111 7.095*** 8.349*** 0.690** 0.206** 0.637** 0.134*
Hungary
ln co2 1960–2003 2.144 2.373 6.096*** 6.644*** 2.172 2.365 6.146*** 6.657*** 0.209 0.204** 0.402* 0.125*
ln y 1975–2003 0.409 0.927 3.229** 3.188 0.928 1.569 3.243** 3.191 0.462* 0.093 0.135 0.131*
(ln y)2 1975–2003 0.341 0.880 3.202** 3.167 0.883 1.541 3.215** 3.168 0.462* 0.093 0.136 0.131*
ln nuc 1983–2003 6.534*** 4.793*** 2.304 3.056 5.951*** 5.574*** 2.137 2.982 0.301 0.155** 0.470** 0.167**
India
ln co2 1960–2003 0.116 1.900 8.188*** 8.108*** 0.065 1.826 8.120*** 8.062*** 0.833** 0.148** 0.126 0.119*
ln y 1975–2003 1.457 2.449 5.570*** 6.032*** 2.117 2.382 5.584*** 6.179*** 0.681** 0.170** 0.428* 0.101
(ln y)2 1975–2003 1.701 2.222 5.363*** 5.996*** 2.427 2.138 5.365*** 6.135*** 0.680** 0.175** 0.459* 0.097
ln nuc 1971–2003 3.048** 3.005 7.725*** 7.589*** 3.122** 3.077 7.730*** 7.568*** 0.115 0.115 0.077 0.079
Japan
ln co2 1960–2003 5.572*** 3.382* 4.571*** 5.030*** 4.774*** 3.058 4.611*** 5.132*** 0.636** 0.166** 0.469** 0.163**
ln y 1975–2003 3.216** 0.024 2.491 3.053 2.519 0.440 2.421 3.014 0.664** 0.163** 0.392* 0.088
 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996 1991

Table A1 (continued)

ADF-test PP-test KPSS-test

Level First difference Level First difference Level First difference
Sample Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept
and and and and and and
trend trend trend trend trend trend
(ln y)2 1975–2003 3.070** 0.015 2.522 3.049 2.413 0.446 2.456 3.009 0.665** 0.161** 0.378* 0.090
ln nuc 1966–2003 3.390** 1.171 5.448*** 7.693*** 4.564*** 0.914 5.521*** 8.871*** 0.622** 0.218*** 0.561** 0.080
Korea
ln co2 1960–2003 3.522** 1.607 5.536*** 6.743*** 4.214*** 1.505 5.521*** 6.851*** 0.830*** 0.193** 0.609** 0.081
ln y 1975–2003 1.218 1.128 4.360*** 4.390*** 1.200 1.343 4.332*** 4.365*** 0.688** 0.118 0.194 0.200**
(ln y)2 1975–2003 0.839 1.372 4.515*** 4.470*** 0.837 1.570 4.491*** 4.446*** 0.688** 0.104 0.518** 0.349***
ln nuc 1977–2003 7.825*** 7.276*** 15.180*** 14.874*** 6.511*** 5.985*** 12.969*** 13.642*** 0.472** 0.177** 0.462* 0.115
Lithuania
ln co2 1960–2003 1.058 0.738 4.277*** 5.480*** 1.305 0.801 4.342*** 5.481*** 0.216 0.212** 0.539** 0.079
ln y 1990–2003 1.294 1.737 1.104 2.304 1.663 1.753 1.103 2.509 0.160 0.149** 0.398* 0.099
(ln y)2 1990–2003 1.305 1.740 1.097 2.324 1.670 1.760 1.086 2.527 0.159 0.149** 0.401* 0.099
ln nuc 1992–2003 2.421 2.856 5.527*** 5.336*** 2.519 2.882 5.143*** 5.020*** 0.176 0.097 0.133 0.129*
Mexico
ln co2 1960–2003 1.522 0.625 5.614*** 5.827*** 1.461 0.985 5.769*** 5.929*** 0.705** 0.185** 0.214 0.092
ln y 1975–2003 1.557 1.986 3.933*** 3.879** 1.643 2.223 3.919*** 3.864** 0.554** 0.082 0.091 0.086
(ln y)2 1975–2003 1.532 1.976 3.944*** 3.888** 1.621 2.216 3.929*** 3.872** 0.554** 0.083 0.090 0.086
ln nuc 1989–2003 9.448*** 8.620*** 7.723*** 6.824*** 8.506*** 7.811*** 9.909*** 8.995*** 0.450* 0.156** 0.362* 0.217***
Netherlands
ln co2 1960–2003 3.051** 2.677 5.906*** 6.127*** 3.046** 2.621 5.868*** 6.133*** 0.193 0.138* 0.282 0.109
ln y 1975–2003 0.089 1.392 2.251 2.213 0.231 1.823 2.534 2.507 0.675** 0.136* 0.086 0.080
(ln y)2 1975–2003 0.024 1.394 2.210 2.172 0.185 1.810 2.502 2.476 0.674** 0.139* 0.090 0.080
ln nuc 1968–2003 6.556*** 5.554*** 8.550*** 8.551*** 6.724*** 5.764*** 8.462*** 8.866*** 0.344 0.177** 0.544** 0.150**
Pakistan
ln co2 1960–2003 0.363 1.362 5.758*** 5.682*** 0.427 1.598 5.825*** 5.756*** 0.797*** 0.133* 0.104 0.103
ln y 1975–2003 2.729* 0.046 4.136*** 5.549*** 2.713 0.087 4.171*** 5.628*** 0.659** 0.180** 0.428* 0.118
(ln y)2 1975–2003 2.581 0.052 4.195*** 5.478*** 2.570* 0.074 4.225*** 5.550*** 0.661** 0.179** 0.412* 0.120*
ln nuc 1971–2003 3.479** 3.455* 7.109*** 6.999*** 3.492** 3.472* 7.482*** 7.390*** 0.150 0.105 0.076 0.063
Russia
ln co2 1960–2003 2.186 1.233 2.606 2.996 1.943 1.279 2.518 3.068 0.262 0.207** 0.428* 0.078
ln y 1989–2003 1.951 0.213 1.206 3.491* 1.846 0.204 1.096 3.687** 0.291 0.165** 0.459* 0.104
(ln y)2 1989–2003 1.985 0.208 1.189 3.519* 1.866 0.207 1.076 3.729** 0.293 0.165** 0.461* 0.103
ln nuc 1992–2003 0.233 2.607 3.542** 3.759** 0.274 2.767 3.709** 4.056** 0.478** 0.138* 0.248 0.169**
Slovakia
ln co2 1960–2003 0.448 1.895 4.173*** 4.473*** 0.870 1.863 4.152*** 4.457*** 0.339 0.209** 0.485** 0.096
ln y 1984–2003 0.275 0.504 1.872 2.084 1.000 1.080 2.043 2.186 0.177 0.134* 0.189 0.088
(ln y)2 1984–2003 0.255 0.489 1.868 2.085 0.986 1.068 2.037 2.185 0.178 0.134* 0.191 0.089
ln nuc 1972–2003 4.129*** 4.423*** 7.197*** 7.030*** 4.063*** 4.382*** 8.311*** 8.164*** 0.586** 0.186** 0.465** 0.152**
Slovenia
ln co2 1960–2003 2.294 1.837 7.241*** 7.550*** 2.332 1.780 7.208*** 7.551*** 0.598** 0.200* 0.282 0.065
ln y 1990–2003 0.571 8.237*** 4.272*** 2.939 0.235 8.643*** 5.899*** 4.433*** 0.515** 0.131* 0.313 0.146**
(ln y)2 1990–2003 0.602 8.394*** 4.304*** 2.946 0.262 8.931*** 6.000*** 4.449*** 0.515** 0.133* 0.318 0.147**
ln nuc 1992–2003 2.526 2.204 2.616 2.561 2.590 2.194 2.569 2.526 0.265 0.104 0.167 0.112
South Africa
ln co2 1960–2003 2.302 1.413 6.457*** 6.919*** 2.338 1.352 6.463*** 6.959*** 0.602** 0.205** 0.360* 0.061
ln y 1975–2003 1.226 1.049 3.541** 3.544* 1.381 1.471 3.523** 3.528* 0.488** 0.105 0.146 0.114
(ln y)2 1975–2003 1.228 1.057 3.551** 3.553* 1.382 1.476 3.532** 3.536* 0.489** 0.105 0.149 0.116
ln nuc 1984–2003 4.351*** 4.091** 5.960*** 6.177*** 4.774*** 4.281** 6.508*** 6.946*** 0.371* 0.110 0.245 0.098
Spain
ln co2 1960–2003 3.083** 2.007 5.472*** 5.970*** 2.860* 1.995 5.594*** 6.046*** 0.727** 0.174** 0.371* 0.128*
ln y 1975–2003 1.409 1.793 2.187 2.265 0.756 1.977 2.316 2.434 0.672** 0.124* 0.194 0.073
(ln y)2 1975–2003 1.507 1.782 2.168 2.259 0.825 1.954 2.299 2.432 0.671** 0.128* 0.205 0.072
ln nuc 1968–2003 6.022*** 5.551*** 9.059*** 9.537*** 5.150*** 4.900*** 8.296*** 8.576*** 0.622** 0.154** 0.387* 0.098
Sweden
ln co2 1960–2003 0.848 2.462 6.978*** 7.205*** 0.938 2.460 6.973*** 7.160*** 0.529** 0.152** 0.240 0.140*
ln y 1975–2003 0.582 1.482 3.062** 3.034 0.276 1.957 3.049** 3.038 0.666** 0.070 0.113 0.069
(ln y)2 1975–2003 0.636 1.457 3.043** 3.021 0.315 1.938 3.031** 3.027 0.666** 0.072 0.118 0.069
ln nuc 1965–2003 2.644* 1.424 6.025*** 6.555*** 2.670* 1.402 6.058*** 6.540*** 0.559** 0.181** 0.360* 0.082
Switzerland
ln co2 1960–2003 4.186*** 3.396* 6.639*** 7.717*** 4.674*** 4.050** 6.637*** 7.783*** 0.278 0.186** 0.644** 0.156**
ln y 1975–2003 1.574 1.462 3.629** 3.852** 1.568 1.560 3.625** 3.828** 0.656** 0.144* 0.161 0.086
(ln y)2 1975–2003 1.553 1.470 3.622** 3.834** 1.545 1.576 3.618** 3.810** 0.657** 0.143* 0.157 0.086
ln nuc 1969–2003 6.300*** 5.372*** 7.500*** 8.472*** 8.633*** 6.975*** 7.303*** 8.313*** 0.637** 0.194** 0.474** 0.148**

(continued on next page)

1992 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996

Table A1 (continued)

ADF-test PP-test KPSS-test

Level First difference Level First difference Level First difference
Sample Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept Intercept
and and and and and and
trend trend trend trend trend trend

Ukraine
ln co2 1960–2003 0.872 0.794 3.290** 3.958** 1.223 0.960 3.199** 3.983** 0.204 0.204** 0.438* 0.078
ln y 1987–2003 1.391 0.701 0.874 1.791 1.349 0.190 1.036 1.738 0.502** 0.150** 0.295 0.150**
(ln y)2 1987–2003 1.402 0.692 0.888 1.820 1.352 0.187 1.049 1.770 0.504** 0.150** 0.296 0.149**
ln nuc 1992–2003 2.684* 1.573 2.661* 2.999 3.165** 1.500 2.651* 3.026 0.440* 0.141* 0.342 0.177**
UK
ln co2 1960–2003 1.618 3.854** 8.387*** 8.285*** 1.375 3.844** 8.754*** 8.640*** 0.749*** 0.067 0.065 0.062
ln y 1975–2003 0.381 1.664 2.696* 2.689 0.189 2.111 2.845* 2.830 0.689** 0.079 0.075 0.044
(ln y)2 1975–2003 0.504 1.609 2.676* 2.683 0.286 2.054 2.830* 2.827 0.689** 0.086 0.085 0.048
ln nuc 1960–2003 4.976*** 3.168 4.069*** 4.992*** 4.482*** 3.048 4.186*** 5.181*** 0.730** 0.152** 0.405* 0.099
US
ln co2 1960–2003 2.491 2.038 4.693*** 4.905*** 2.489 2.102 4.660*** 4.857*** 0.235 0.126* 0.244 0.138*
ln y 1975–2003 1.128 2.500 3.946*** 3.862** 1.075 2.775 3.916*** 3.832** 0.692** 0.050 0.084 0.048
(ln y)2 1975–2003 1.031 2.463 3.938*** 3.851** 0.988 2.749 3.910*** 3.822** 0.692** 0.046 0.075 0.048
ln nuc 1960–2003 5.633*** 2.415 6.858*** 8.275*** 4.920*** 2.294 6.418*** 7.676*** 0.710** 0.214** 0.688** 0.090

Notes: 1. Null hypothesis is nonstationary for ADF and PP test and stationary for KPSS test.
2. Optimal bandwidth selection procedures are conducted based on Newey and West [27].
3. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and 10% of significance levels, respectively.

Table A2
Unit root test with structural breaks in individual countries.

Sample T M1 M2
Test statistic Break year 1 Break year 2 Optimal lag Test statistic Break year 1 Break year 2 Optimal lag
Argentina
ln co2 1960–2003 44 3.214 1968 1984 0 2.660 1980 1986 0
ln y 1975–2003 29 2.966 1988 1991 0 2.509 1988 1991 0
(ln y)2 1975–2003 29 2.966 1988 1991 0 2.509 1988 1991 0
ln nuc 1974–2003 30 4.666** 1983 1990 1 3.965 1981 1989 0
Belgium
ln co2 1960–2003 44 0.825 1982 1988 4 2.931 1974 1982 5
ln y 1975–2003 29 4.072 1987 1992 0 4.518* 1984 1995 3
(ln y)2 1975–2003 29 4.072 1987 1992 0 4.518* 1987 1992 3
ln nuc 1972–2003 32 NAa 1981 1984 4 NA NA NA NA
Brazil
ln co2 1960–2003 44 NAa 1971 1973 0 NA NA NA NA
ln y 1975–2003 29 9.233*** 1984 1988 2 2.768 1982 1989 1
(ln y)2 1975–2003 29 9.233*** 1984 1988 2 2.768 1982 1989 1
ln nuc 1984–2003 20 NAa 1992 1995 3 NA NA NA NA
Bulgaria
ln co2 1960–2003 44 3.508 1990 1992 1 1.805 1990 1992 1
ln y 1980–2003 24 3.564 1989 1995 0 2.888 1988 1995 5
(ln y)2 1980–2004 24 3.564 1989 1995 0 2.888 1988 1995 5
ln nuc 1974–2003 30 0.095 1985 1987 4 7.159*** 1985 1995 3
Canada
ln co2 1960–2003 44 2.929 1981 1993 0 2.249 1981 1994 0
ln y 1975–2003 29 2.047 1983 1990 0 2.405 1982 1990 0
(ln y)2 1975–2003 29 2.047 1983 1990 0 2.405 1982 1990 0
ln nuc 1963–2003 41 NAa 1970 1972 5 NA NA NA NA
China
ln co2 1960–2003 44 NAa 1968 1970 4 NA NA NA NA
ln y 1975–2003 29 2.894 1986 1988 4 4.442* 1985 1988 4
(ln y)2 1975–2003 29 2.894 1986 1988 4 4.442* 1985 1988 4
ln nuc 1993–2003 11 NA NA NA NA NA NA NA NA
Czech Republic
ln co2 1960–2003 44 3.220 1969 1990 0 0.515 1971 1989 5
ln y 1990–2003 14 NA NA NA NA NA NA NA NA
(ln y)2 1990–2003 14 NA NA NA NA NA NA NA NA
ln nuc 1983–2003 21 2.051 1994 1996 1 NA NA NA NA
Finland
ln co2 1960–2003 44 4.558** 1981 1984 5 2.458 1981 1991 5
ln y 1975–2003 29 2.699 1990 1993 0 6.684*** 1987 1990 5
(ln y)2 1975–2003 29 2.700 1990 1993 0 6.684*** 1987 1990 5
ln nuc 1977–2003 27 3.037 1984 1986 3 4.178* 1984 1993 3
 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996 1993

Table A2 (continued)

Sample T M1 M2
Test statistic Break year 1 Break year 2 Optimal lag Test statistic Break year 1 Break year 2 Optimal lag

France
ln co2 1960–2003 44 2.325 1974 1980 4 3.733 1974 1980 0
ln y 1975–2003 29 2.832 1992 1995 1 1.965 1987 1992 1
(ln y)2 1975–2003 29 2.833 1992 1995 1 1.966 1987 1992 1
ln nuc 1960–2003 44 NAa 1968 1979 3 NA NA NA NA
Germany
ln co2 1960–2003 44 3.603 1975 1990 0 5.310*** 1974 1990 0
ln y 1975–2003 29 4.987** 1987 1989 1 3.876 1987 1992 3
(ln y)2 1975–2003 29 4.986** 1987 1989 1 3.876 1987 1992 3
ln nuc 1961–2003 43 NAa 1970 1973 2 NA NA NA NA
Hungary
ln co2 1960–2003 44 2.414 1969 1989 0 0.810 1989 1992 0
ln y 1975–2003 29 2.705 1986 1990 5 4.856** 1986 1990 5
(ln y)2 1975–2003 29 2.705 1986 1990 5 4.855** 1986 1990 5
ln nuc 1983–2003 21 3.534 1991 1993 4 10.510*** 1991 1994 5
India
ln co2 1960–2003 44 NAa 1991 1994 2 NA NA NA NA
ln y 1975–2003 29 1.404 1987 1990 0 3.501 1987 1990 3
(ln y)2 1975–2003 29 1.404 1987 1990 0 3.501 1987 1990 3
ln nuc 1971–2003 33 0.183 1981 1994 4 4.162* 1982 1995 0
Japan
ln co2 1960–2003 44 3.254 1973 1987 5 5.541*** 1970 1989 1
ln y 1975–2003 29 3.029 1984 1987 4 2.693 1987 1991 4
(ln y)2 1975–2003 29 3.029 1984 1987 4 2.694 1987 1991 4
ln nuc 1966–2003 38 NAa 1973 1976 2 NA NA NA NA
Korea
ln co2 1960–2003 44 NAa 1969 1973 3 NA NA NA NA
ln y 1975–2003 29 2.088 1982 1986 0 3.548 1986 1992 0
(ln y)2 1975–2003 29 2.088 1982 1986 0 3.548 1986 1992 0
ln nuc 1977–2003 27 4.826 1987 1992 2 NA NA NA NA
Lithuania
ln co2 1960–2003 44 4.051 1990 1992 5 0.767 1984 1992 5
ln y 1990–2003 14 NA NA NA NA NA NA NA NA
(ln y)2 1990–2003 14 NA NA NA NA NA NA NA NA
ln nuc 1992–2003 12 NA NA NA NA NA NA NA NA
Mexico
ln co2 1960–2003 44 2.348 1977 1989 2 7.237*** 1977 1992 4
ln y 1975–2003 29 3.292 1985 1994 1 1.969 1983 1994 0
(ln y)2 1975–2003 29 3.292 1985 1994 1 1.969 1983 1994 0
ln nuc 1989–2003 15 NA NA NA NA NA NA NA NA
Netherlands
ln co2 1960–2003 44 2.114 1981 1988 5 1.056 1981 1988 5
ln y 1975–2003 29 6.168*** 1984 1990 5 5.435*** 1984 1988 5
(ln y)2 1975–2003 29 6.171*** 1984 1990 5 5.435*** 1984 1988 5
ln nuc 1968–2003 36 10.440 *** 1981 1990 0 NA NA NA NA
Pakistan
ln co2 1960–2003 44 2.659 1970 1989 4 NA NA NA NA
ln y 1975–2003 29 3.896 1986 1991 5 3.972 1989 1992 0
(ln y)2 1975–2003 29 3.896 1986 1991 5 3.972 1989 1992 0
ln nuc 1971–2003 33 NAa 1987 1992 0 NA NA NA NA
Russia
ln co2 1960–2003 44 2.609 1984 1994 3 1.280 1984 1994 1
ln y 1989–2003 15 NA NA NA NA NA NA NA NA
(ln y)2 1989–2003 15 NA NA NA NA NA NA NA NA
ln nuc 1992–2003 12 NA NA NA NA NA NA NA NA
Slovakia
ln co2 1960–2003 44 0.560 1969 1994 0 0.407 1989 1994 5
ln y 1984–2003 20 5.417 1991 1993 5 4.657*** 1991 1994 4
(ln y)2 1984–2003 20 5.418 1991 1993 5 4.656*** 1991 1994 4
ln nuc 1972–2003 32 NAa 1979 1983 4 NA NA NA NA
Slovenia
ln co2 1960–2003 44 4.559** 1969 1990 3 5.982*** 1971 1990 3
ln y 1990–2003 14 NA NA NA NA NA NA NA NA
(ln y)2 1990–2003 14 NA NA NA NA NA NA NA NA
ln nuc 1992–2003 12 NA NA NA NA NA NA NA NA
South Africa
ln co2 1960–2003 44 1.601 1970 1991 0 2.929 1989 1991 0
ln y 1975–2003 29 1.029 1984 1995 3 0.128 1983 1987 4

(continued on next page)

1994 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996

Table A2 (continued)

Sample T M1 M2
Test statistic Break year 1 Break year 2 Optimal lag Test statistic Break year 1 Break year 2 Optimal lag
(ln y)2 1975–2003 29 1.029 1984 1995 3 0.128 1983 1987 4
ln nuc 1984–2003 20 3.754 1992 1994 0 3.233 1992 1995 0
Spain
ln co2 1960–2003 44 3.488 1968 1988 3 2.149 1976 1988 5
ln y 1975–2003 29 5.920*** 1984 1993 5 4.051 1986 1993 4
(ln y)2 1975–2003 29 5.920*** 1984 1993 5 4.052 1986 1993 4
ln nuc 1968–2003 36 9.992*** 1979 1983 4 NA NA NA NA
Sweden
ln co2 1960–2003 44 3.881 1981 1989 0 3.301 1979 1984 0
ln y 1975–2003 29 2.663 1983 1993 5 3.586 1990 1993 5
(ln y)2 1975–2003 29 2.662 1983 1993 5 3.586 1990 1993 5
ln nuc 1965–2003 39 NAa 1974 1977 4 NA NA NA NA
Switzerland
ln co2 1960–2003 44 2.205 1973 1994 4 5.976*** 1973 1981 0
ln y 1975–2003 29 2.940 1987 1990 1 4.605** 1982 1990 5
2
(ln y) 1975–2003 29 2.940 1987 1990 1 4.605** 1982 1990 5
ln nuc 1969–2003 35 3.612 1976 1983 3 0.289 1977 1995 5
Ukraine
ln co2 1960–2003 44 2.634 1991 1994 3 0.202 1991 1994 2
ln y 1987–2003 17 0.455 1994 1996 3 0.102 1994 1996 0
(ln y)2 1987–2003 17 0.455 1994 1996 3 0.015 1994 1996 0
ln nuc 1992–2003 14 NA NA NA NA NA NA NA NA
UK
ln co2 1960–2003 44 5.370*** 1979 1990 0 5.322*** 1979 1990 0
ln y 1975–2003 29 4.690** 1987 1990 4 3.983 1987 1990 4
(ln y)2 1975–2003 29 4.691** 1987 1990 4 3.983 1987 1990 4
ln nuc 1960–2003 44 2.085 1969 1973 0 2.379 1973 1977 0
US
ln co2 1960–2003 44 2.899 1977 1981 3 4.662** 1973 1981 1
ln y 1975–2003 29 5.317*** 1983 1985 5 3.236 1982 1990 5
(ln y)2 1975–2003 29 5.317*** 1983 1985 5 3.236 1982 1990 5
ln nuc 1960–2003 44 NAa 1969 1971 4 NA NA NA NA

Notes: 1. Null hypothesis is non-stationary.

2. M1 is the case in which two breaks in the level of the time series are allowed in the model. M2 represents model in which two breaks in the level and slope are considered.
3. aIndicates values with complex number.
4. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and 10% of significance levels, respectively.

Table A3
IPS and LLC panel unit root tests.

IPS-test LLC-test
Level First difference Level First difference
Intercept Intercept and trend Intercept Intercept and Trend Intercept Intercept and Trend Intercept Intercept and Trend
World
ln co2 2.126** 1.008 9.157*** 9.362*** 4.738*** 2.164** 9.203*** 12.303***
ln y 3.437 1.156 8.672*** 7.968*** 0.956 2.533 8.569*** 7.743***
(ln y)2 4.068 1.377 7.782*** 7.853*** 0.188 2.329 7.591*** 7.580***
ln nuc 10.677*** 4.626*** 14.051*** 12.901 10.184*** 2.484*** 8.853*** 6.043
OECD
ln co2 2.461*** 0.769 8.479*** 8.235*** 5.138*** 3.893*** 9.868*** 11.995***
ln y 3.113 1.233 7.354*** 5.874*** 1.177 2.182 6.716*** 5.474***
(ln y)2 3.466 1.330 7.326*** 5.829*** 0.741 2.091 6.652*** 5.378***
ln nuc 8.976*** 3.382*** 8.196*** 12.247*** 8.705*** 1.466* 1.098 13.334
Non-OECD
ln co2 0.347 2.455 4.096*** 4.682*** 1.460* 1.671 2.842*** 4.973***
ln y 1.629 0.338 4.723*** 5.392*** 0.129 1.454 5.315*** 5.311***
(ln y)2 2.193 0.566 3.388*** 5.268*** 1.135 1.248 3.825*** 5.163***
ln nuc 5.925*** 3.161*** 12.059*** 6.460*** 8.125*** 3.783*** 15.563*** 7.147***

Notes: 1. Modified Akaike Information Criteria (MAIC) is used to select the optimal lag-length. Each test has the null hypothesis of non-stationary.
2. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and 10% of significance levels, respectively.
 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996 1995

Table A4
Breitung and Fisher-ADF panel unit root tests.

Breitung-test Fisher-ADF-test
Level First difference Level First difference
Intercept and trend Intercept and trend Intercept Intercept and trend Intercept Intercept and trend
World
ln co2 4.452 8.091*** 86.614*** 49.517 237.611*** 277.178***
ln y 2.632 7.267*** 27.360 53.396 195.876*** 165.871***
(ln y)2 2.702 7.293*** 25.799 50.375 176.179*** 163.517***
ln nuc 0.795 2.410*** 241.360*** 133.583*** 290.466*** 348.983***
OECD
ln co2 2.021 7.436*** 60.613*** 39.090 170.044*** 188.937***
ln y 1.866 7.003*** 9.984 27.460 119.249*** 91.932***
(ln y)2 1.613 6.959*** 9.006 26.663 118.545*** 91.412***
ln nuc 1.467 1.864** 161.739*** 88.165*** 140.767*** 213.62***
Non-OECD
ln co2 4.032 3.619*** 26.000 10.426 67.567*** 88.241***
ln y 1.916 3.088*** 17.376 25.936 76.627*** 73.938***
(ln y)2 2.227 3.146*** 16.792 23.711 57.633*** 72.104***
ln nuc 2.686*** 4.025*** 79.621*** 45.416*** 149.699*** 135.358***

Notes: 1. Modified Akaike Information Criteria (MAIC) is used to select the optimal lag-length. Each test has the null hypothesis of non-stationary.
2. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and 10% of significance levels, respectively.

Table A5 Appendix B. Results of the unit root tests in the panel form
Hadri panel unit root test.

Hadri-test See Table A3.

See Table A4.
Level
See Table A5.
Intercept Intercept and trend
World
ln co2 17.986*** 15.518*** Appendix C. Results of the cointegration tests
ln y 16.784*** 8.865***
(ln y)2 16.639*** 8.799*** See Table A6.
ln nuc 16.508*** 15.339***
OECD
ln co2 14.835*** 11.188*** References
ln y 13.883*** 4.821***
(ln y)2 13.897*** 4.736*** [1] Selden TM, Song D. Environmental quality and development: Is there a
ln nuc 13.197*** 12.434*** Kuznets curve for air pollution emissions? J Environ Econ Manage
Non-OECD 1994;27:147–62.
ln co2 10.361*** 10.400*** [2] Grossman GM, Krueger AB. Economic growth and the environment. Quart J
ln y 9.998*** 6.207*** Econ 1995;110:353–77.
(ln y)2 9.512*** 6.263*** [3] Martínez-Zarzoso I, Bengochea-Morancho A. Pooled mean group estimation of
an environmental Kuznets curve for CO2. Econ Lett 2004;82:121–6.
ln nuc 3.421*** 5.226***
[4] Richmond AK, Kaufmann RK. Is there a turning point in the relationship
Notes: 1. Modified Akaike Information Criteria (MAIC) is used to select the optimal between income and energy use and/or carbon emissions? Ecol Econ
lag-length. 2006;56:176–89.
2. Hadri test has the null hypothesis of stationary. [5] Apergis N, Payne JE. CO2 emissions, energy usage, and output in Central
America. Energy Policy 2009;37:3282–6.
3. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and
[6] Apergis N, Payne JE. The emissions, energy consumption, and growth nexus:
10% of significance levels, respectively.
Evidence from the commonwealth of independent states. Energy Policy
2010;38:650–5.
[7] Narayan PK, Narayan S. Carbon dioxide emissions and economic growth: Panel
data evidence from developing countries. Energy Policy 2010;38:661–6.

Table A6
Pedroni panel cointegration tests.

World OECD Non-OECD

Intercept Intercept and Trend Intercept Intercept and Trend Intercept Intercept and Trend
Pedroni test
Panel m-statistic 2.247 4.617 1.861 3.107*** 1.315 3.319***
Panel rho-statistic 4.690 6.056 3.579 4.407 2.224 2.846
Panel PP-statistic 1.252 5.517*** 0.124 3.021*** 2.060** 5.914***
Panel ADF-statistic 3.432*** 8.366*** 2.338*** 5.284*** 2.701*** 7.074***
Group rho-statistic 6.752 7.809 4.658 5.480 3.957 4.542
Group PP-statistic 3.911*** 9.163*** 2.871*** 7.448*** 2.671*** 5.359***
Group ADF-statistic 6.251*** 8.096*** 4.945*** 5.891*** 3.824*** 5.594***

Notes: 1. The null hypothesis is no cointegration.

2. The asterisks ***, **, and * indicate the rejection of null hypothesis at 1%, 5%, and 10% of significance levels, respectively.
1996 H. Iwata et al. / Applied Energy 88 (2011) 1986–1996

[8] Chang C-C. A multivariate causality test of carbon dioxide emissions, energy [18] Breitung J. The local power of some unit root tests for panel data. Adv Econom
consumption and economic growth in China. Appl Energy 2010;87:3533–7. 2000;15:161–77.
[9] Payne JE. On the dynamics of energy consumption and output in the US. Appl [19] Maddala GS, Wu S. A comparative study of unit root tests with panel data and
Energy 2009;86:575–7. a new simple test. Oxford Bull Econ Stat 1999;61:631–52.
[10] Ozturk I, Acaravci A. The causal relationship between energy consumption and [20] Choi I. Unit root tests for panel data. J Int Money Finance 2001;20:249–72.
GDP in Albania, Bulgaria, Hungary and Romania: evidence from ARDL bound [21] Hadri K. Testing for stationarity in heterogeneous panel data. Econom J
testing approach. Appl Energy 2010;87:1938–43. 2000;3:148–61.
[11] Narayan PK, Narayan S, Popp S. Does electricity consumption panel Granger [22] Pedroni P. Fully modified OLS for heterogeneous cointegrated panels. Adv
cause GDP? A new global evidence. Appl Energy 2010;87:3294–898. Econom 2000;15:93–130.
[12] Apergis N, Payne JE. The causal dynamics between coal consumption and [23] Pedroni P. Panel cointegration: Asymptotic and finite sample properties of
growth: Evidence from emerging market economies. Appl Energy pooled time series tests with an application to the PPP hypothesis. Econom
2010;87:1972–7. Theory 2004;20:597–625.
[13] Pesaran MH, Shin Y, Smith RP. Pooled mean group estimation of dynamic [24] Westerlund J. Testing for panel cointegration with multiple structural breaks.
heterogeneous panels. J Am Stat Assoc 1999;94:621–34. Oxford Bull Econ Stat 2006;68:101–32.
[14] Pesaran MH, Smith R. Estimating long-run relationships from dynamic [25] Kim D-H, Lin S-C, Suen Y-B. Dynamic effects of trade openness on financial
heterogeneous panels. J Econom 1995;68:79–113. development. Econ Modell 2010;27:254–61.
[15] Narayan PK, Popp S. A new unit root test with two structural breaks in level [26] Goswami GG, Junayed SH. Pooled mean group estimation of the bilateral trade
and slope at unknown time. J Appl Stat 2010;37:1425–38. balance equation: USA vis-à-vis her trading partners. Int Rev Appl Econ
[16] Im KS, Pesaran MH, Shin Y. Testing for unit roots in heterogeneous panels. J 2006;20:515–26.
Econom 2003;115:53–74. [27] Newey WK, West KD. Automatic lag selection in covariance matrix estimation.
[17] Levin A, Lin C-F, Chu C-SJ. Unit root tests in panel data: Asymptotic and finite- Rev Econ Stud 1994;61:631–53.
sample properties. J Econom 2002;108:1–24.