The Journal of Energy and Development

THE JOURNAL OF ENERGY
AND DEVELOPMENT
Nicolas Schneider, Mihaela Simionescu,

and Wadim Strielkowski,
“Searching for Long Equilibrium Behaviors

into the Stochastic Features of
Electricity Series from the
World’s Largest Producers,”
Volume 47, Number 2
Copyright 2022
SEARCHING FOR LONG EQUILIBRIUM
BEHAVIORS INTO THE STOCHASTIC FEATURES
OF ELECTRICITY SERIES FROM THE WORLD’S
LARGEST PRODUCERS
Nicolas Schneider, Mihaela Simionescu, AND Wadim Strielkowski*
Introduction
G lobal warming has emerged as a major concern to be addressed because it

directly threatens food security, biodiversity, and the safety of populations
living in coastal zones.1 For some time, researchers have identified that the anthro-
pogenic release of polluting greenhouse gas (GHG) emissions contribute to trig-
gering the growing frequency of extreme weather events and warming average
surface temperatures more generally. Nobuo Tanaka, Executive Director of the
*Nicolas Schneider is currently a Ph.D. doctoral student in the Department of Earth & Environment
at Boston University (BU), Massachusetts (U.S.). A native of France, Nicolas Schneider holds a B.A.
in Economics from Grenoble-Alps University (France), a M.D. in Development Economics & Applied
Econometrics at Paris Sorbonne University (France), and a M.Sc. in Environmental Economics &
Climate Change from the London School of Economics & Political Science (U.K.), with a major in
resource extraction modelling, energy economics and environmental policy. As part of his doctoral
research, Nicolas combines climate econometrics tools with Computable General Equilibrium
frameworks to model the effect of extreme weather shocks on various socioeconomic outcomes,
including energy demand and crop yields in the United States. Nicolas published several research
articles in peer-reviewed journals, including Waste Management, Applied Energy, Environmental
Research Letters, Environmental Impact Assessment Review, and the International Review of
Environmental and Resource Economics, to name a few. Nicolas is also a Teaching Fellow in the Earth
& Environment Department and was recently appointed co-chair of the Workshops and Panels
Working Group of the BU Science Policy Group in Boston. (continued)
The Journal of Energy and Development, Vol. 47, Nos. 1-2

Copyright # 2022 by the International Research Center for Energy and Economic Development
(ICEED). All rights reserved.
223
224 THE JOURNAL OF ENERGY AND DEVELOPMENT
International Energy Agency, highlighted that “Energy is at the heart of the prob-
lem – and so must form the core of the solution.”2 But to implement changes in
future energy paths, an in-depth understanding of today’s energy situation and poli-
cies are required.
It becomes unlikely that the energy sector, in general, and the ways through
which power is generated and supplied to meet demand, in particular, can be
spared from structurally comprehensive reforms in the future. Thus, shifting from
a power generation system with a high-carbon content to electricity production
based on renewable resources is necessary to curb harmful emissions and mitigate
the social cost of carbon. Put into perspective, the sustainability issue facing devel-
oped countries, as well as emerging economies, is how to reconcile the need to
provide secure and affordable electricity supplies to the industrial and domestic
sectors while avoiding adverse externalities.3 In this context, both renewable and
nuclear sources emerge as a means of providing low-carbon alternatives to power
Mihaela Simionescu (Bratu), who holds a Doctorate in Economics (Cybernetics and Statistics) and a
Habilitation in Economics, is a Ph.D. Supervisor in Economics and Scientific Researcher 2nd degree at
the Institute for Economic Forecasting (Romania). The author is also an Associate Researcher at Prague
Institute for Qualification Enhancement, Center for Macroeconomic Modelling of the Romanian
Academy, and a fellow at the Global Labour Organization. Dr. Simionescu is an Associate Professor at
the University of Bucharest, Faculty of Administration and Business, and at Bucharest University of
Economics, Faculty of Economic Cybernetics, Statistics and Informatics. Her prior positions included:
Full Professor and Principal Investigator at University of Social Sciences, Lodz (Poland); Full Professor
at Mykolas Romeris University of Vilnius (Lithuania); Visiting Researcher at the Institute for Research
on Population and Social Policies in Rome (Italy) and at the Romanian Academy, Iasi Branch,
Gheorghe Zane Institute for Economic and Social Research; and Visiting Professor at the Pan-European
University (Slovakia), Rzeszow University of Technology (Poland), and North-Caucasus Federal
University (Russia). In addition to numerous prizes and memberships, the author is a prolific writer with
her works appearing in Energies, Journal of Competitiveness, Economics & Sociology, Technological
Forecasting and Social Change, and Journal of Economics and Development Studies, to name a few.
Wadim Strielkowski is a Research Fellow at the Department of Trade and Finance, Faculty of
Economics and Management, Czech University of Life Sciences Prague (Czech Republic), Senior
Researcher at the Cambridge Institute for Advanced Studies (U.K.), and a Visiting Professor at the
University of California, Berkeley (U.S.). Previously, the author held the following positions: Research
Associate at the University of Cambridge (U.K.); Assistant Professor at Charles University in Prague
(Czech Republic); Research Fellow at the University of Nottingham (U.K.); and Senior Researcher at
the Czech Academy of Sciences. He also worked as a Vice-Chancellor of the Prague University of
Economics and Management and as a Deputy Director for PR and Marketing at CERGE-EI Prague.
The author received his master’s degrees in economics from Charles University in Prague and
University of Siena (Italy) and his Ph.D. degrees from Charles University in Prague and the National
University of Ireland, Galway (Republic of Ireland). Dr. Strielkowski is a productive writer with over
250 publications in world-leading journals, his illustrious works have been published in Nature and
Science, as a sampling.
The authors would like to acknowledge the comments from the editor and the anonymous referees;
however, the usual disclaimer applies.
ELECTRICITY & THE WORLD’S LARGEST PRODUCERS 225
generation based on fossil fuels without boosting the electricity dependence on for-
eign suppliers and trade agreements.4 Additionally, while industrial and power
structures present dynamic interconnections, energy and power-dependent indus-
tries tend to be more carbon-intensive because electricity also operates as a produc-
tion input, along with all other individual production factors (gross fixed capital
formation, labor force).5 Conversely, the recent endorsement of climate protection
measures targeting the most polluting sectors will undoubtedly constrain govern-
ments to reframe how electricity is produced by oligopolistic-structured sectors
and consumed along each stage of a supply chain involving increasing digitaliza-
tion, Artificial Intelligence (AI), and power-dependent applications. This is where
low-carbon electricity generation systems are expected to find their contribution.
The interest in assessing the stationary properties of electricity consumption
series is due to the far-reaching policy implications that unit root testing results
confer. From a statistical standpoint, a time-series is said to be stationary if its sta-
tistical properties (mean, variance, and autocorrelation) are constant over time. In a
nutshell, if a series is stationary, one can predict that its properties will remain
constant in the future, which is highly useful when it comes to forecasting shocks
and designing policy responses. In other words, a process is considered to be sta-
tionary if its probability distribution is unchanged as the number of consecutive
observations rises. This concept does apply to stochastic processes for which the
data generation process does not vary over time. Naturally, one distinguishes
between strict and weak stationarity (or stationarity), in the sense that the former
refers to a stochastic equilibrium process yt with similar realizations and distribu-
tions over different time intervals; whereas the latter designates processes for
which the covariance between any two observations depends only on the length of
time separating the observations.6 While a weakly stationary process can also indi-
cate covariance stationary or wide-sense stationary time-series, one can measure
how the concept of stationarity has played an important role in shaping the theory
of stochastic processes and time-series analysis. From an empirical standpoint, if
electricity use is found to be stationary, then any shock affecting a country’s power
trends (i.e., emergence of a backstop technology reducing the marginal cost of a
competitive resource, carbon tax set on producers, climate agreements endorsed
internationally with limited carbon leakages) will be temporary, and electricity
consumption is expected to shortly return to its normal and long-run equilibrium
pace.7 Conversely, in the presence of a unit root, a shock occurrence on power use
(i.e., electricity price shock) is more likely to have permanent effects, indicating
that any shock (i.e., input price volatility) will be transmitted back to other macro-
economic variables.8 Regarding the world’s largest electricity producers, given that
these countries—to varying degrees— are shifting toward a low-carbon electricity
sector to meet their climate strategies for 2050 in addition to adopting emissions
reducing policies, a comprehensive understanding of these properties is required to
avoid counter-expected effects on their economies as a whole. A survey of the
energy convergence literature highlights a number of critical points that we will

now address.
First, a wide range of papers have explicitly assessed the stationary properties
of total energy use including the works of P. Chen and C. Lee for 104 countries,
Y.-C. Hsu et al. for 84 countries, V. Mishra et al. for 13 Pacific Island countries,
A. Aslan and H. Kum for 7 sectors in Turkey, H. Kum for 15 East Asia and Pacific
countries, and B. Ozcan for 17 Middle Eastern and North African (MENA) coun-
tries.9 While the authors typically have collected aggregate consumption series,
series accounting for disaggregated energy sources have received much less con-
sideration. Moreover, a recent research focus examining the integration properties
of renewable energy series has emerged (C. Pestana Barros et al. and C.-C. Lee
et al. for the United States; G. Gozgor for Brazil, China, and India; E. Demir and
G. Gozgor for 54 developing and developed countries; E. Fendoglu for Brazil,
Russia, India, China, South Africa, and Turkey (BRICS-T); and C. Saba and N.
Ngepah for a mixed panel of 184 economies).10
However, for both literature research areas, it is clear that the cases studied, data
periods, and methodologies strongly differ, and, just as importantly, their associated
conclusions often are conflicting. Therefore, a trade-off with diverging costs-benefits
seems to have emerged. On the one hand, we do observe a progressive shift toward
more globalized panel assessments with samples containing numerous but structur-
ally heterogeneous economies. This not only prevents the supply of country-specific
empirical insights, but also provides concluding remarks and policy implications
that are less generalizable because they are constrained by the collapsing nature of
the panel. On the other hand, fewer attempts to conduct single-country investigations
can be identified. Sometimes, those studies relied on sectoral or province-level data
series, but not systematically. While this approach offers the advantageous feature
of providing more reliable policy insights, it fails to exploit the cross-cases compara-
bility potential of time-series analyses. Furthermore, this approach departs from the
general scope of standard b-convergence frameworks. Thus, there is a need to rec-
oncile those directions and propose a two-stage approach combining panel unit root
analyses and country-level investigation of stationary properties. Additionally, it is
crucial to reorient the literature toward a more reliable and relevant selection of sam-
ples while overcoming choices mainly driven by data availability constraints.
Instead, gathering together countries sharing a slightly similar structure or exhibiting
a comparable energy trend may deliver more fruitful insights than large panels with
mixed and highly heterogenous slope coefficients.
Second, as we previously underscored regarding the rising frequency of large
panel assessments, nonetheless we highlight that cross-sectional dependence and
spatial correlation among the series remain often uncontrolled for in the methodol-
ogy. Naturally, there are some relevant exceptions, namely, the works by C. Mag-
azzino, M. Destek and S. Sarkodie, and S. Erdogan et al.; however, our overall
observation still holds.11 From a statistical point of view, a cross-sectionally
dependent data structure can emerge in the presence of common shocks, spatial
dependence, and unobserved components, which are captured by the error term but
correlate with our electricity series.12 From an empirical standpoint, inter-dependencies
driven by rising trade and financial integrations of national energy markets and tak-
ing the form of spatial correlations across countries may play out and, thus, can
no longer be neglected by integration properties assessment models.13 Further, the
outcome of these assessments is that they will likely introduce inconsistent spuri-
ous estimates; therefore, methods allowing for those inferences should complement
standard univariate ones.14
Third, although a fruitful research direction recently has emerged in studies that
collect and analyze the stationary properties of electricity use series (F. Kula et al.,
D. Borozan and L. Borozan, M. Herrerias et al., and Y. Wang et al.),15 to the best of
our knowledge, no existing assessment has focused explicitly on the world’s top
largest electricity producers to date. Addressing this empirical gap is critical since
the electricity demand recorded in these ten economies cannot be disconnected from
individual pollution trends, global energy consumption, and trade structures, but also
exhibits patterns of cross-market dependence. Accordingly, this calls for further
inquiry into this topic. Hence, this paper seeks to fill these above-mentioned gaps by
providing four empirically and methodologically unique aspects:
(1) Foremost, this paper is among the first to provide an extensive survey of the
energy convergence literature stratified by the energy source considered.
Additionally, it does not only present a methodological discussion outlining
econometric challenges faced by past studies, but also identifies and locates
empirical gaps and caveats that call for further inquiry into this topic.
(2) Second, this study has a competitive edge as it investigates the stationary
properties of electricity consumption series for China, the United States,
India, Russia, Japan, Brazil, Canada, South Korea, Germany, and France.
This sample of economies, which encompasses the major energy consumers
and emitters of pollution globally, also corresponds to the world’s largest
power producers (ranked from 1st to 10th).
(3) Third, this work contrasts with previous ones as it employs a two-stage
stratified approach: (a) univariate and endogenously determined structural
breaks-accounting unit root tests at the country-level; (b) followed by a set
of cross-sectional dependence tests and panel stationary tests accounting for
heterogeneity and cross-sectional dependence among the series; and (c) pro-
viding a k-means clustering and b-convergence analysis. Combining panel-
scaled results and nationwide evidence is thought to offer a more reliable
picture distinguishing between some general electricity patterns at the sam-
ple level and other country-level specificities and heterogeneities. Overall,
we compare those results with the outputs of first-generation panel unit root
procedures, which neither control for heterogeneity nor for cross-sectional
dependence, and examine the sensitivity of our sample-level stationary

findings.
(4) Fourth and finally, data series span the largest and most available period
(1985-2020) thought to provide accurate economic and policy insights help-
ful for energy and environmental planners.
The rest of this paper is organized as follows. The second section offers a state-
of-the-art review of the literature, followed by the third section which introduces
the data collection, descriptive statistics, and presents in detail the time-series
econometric framework. This is followed by a display of the empirical results and
subsequently a discussion. The last section provides the concluding remarks along
with policy implications and prospects for future research.
Literature Survey
Analyzing energy consumption convergence behaviors and stationary properties

can be traced back to the works of L. Nilsson and J. Goldemberg.16 In their pio-
neering papers, both authors offered evidence that economies tend to exhibit
energy consumption trends, which tend toward a common and general pattern.
Over time, several empirical assessments proliferated in the literature but failed to
generate a consensus on this question. Instead, it opened an intense debate and
fueled a multiplicity of investigations using various aggregate and disaggregate
energy indicators and different scales of study (multi-country, single-country,
regional analysis, and sectoral examination). While failing to provide converging
conclusions, this literature expanded to incorporate another research area—envi-
ronmental economics—as authors started examining the convergence of carbon
emissions using slightly similar methodologies; however, this resulted in a plethora
of sharply diverging conclusions and policy implications. Overall, the use of stan-
dard distribution analyses17 or decomposition analyses18 progressively took center
stage in the field, which seems to have left much for future research and study.
In the past two decades, it is worth noting that many papers have presented cau-
sality investigations on the energy-growth nexus.19 Indeed, when performed on
non-stationary data, regression models (i.e., Ordinary Least Squares (OLS), Fixed
Effects (FE), and Random Effects (RE)), but also time-series-based cointegration
and causality models (Granger (1969, 1988) Causality,20 Johansen and Juselius
(1990),21 and Gregory and Hansen (1996)22 cointegration test), may generate spu-
rious estimates with unreliable general statistics (R-squared, t-statistics).23 None-
theless, given the explicit aim of this paper, the current review does not display
energy-GDP causality examinations. The reason is that those papers broadly tested
the stationary properties of the series at a preliminary stage only, while allocating
more important attention to cointegration, causality, and variance decomposition
features. Instead, we outline here the studies whose specific objective was to
inspect the stochastic convergence of energy consumption using unit root testing
methods. We will now begin our literature overview by summarizing the studies
on the convergence of total energy, fossil-based energy, or electricity consumption,
followed by the research focusing on the convergence of alternative and renewable
energy consumption. To close this literature survey, we discuss the empirical and
methodological gaps in the body of works and a contribution proposal is
formulated.
Studies on Total and Fossil Energy Convergence: The literature on total and
fossil-based energy convergence series is rich and extensive. Generally, previous
papers relied on aggregate energy use data with large and heterogeneous samples
of economies, with panel-level investigations, and constrained their analyses to
multi-country approaches. Among this set of studies are P. Chen and C. Lee for
104 economies, Y. Hsu et al. for 84 countries, V. Mishra et al. for 13 Pacific Island
countries, A. Aslan and H. Kum and I. Ozturk and A. Aslan for 7 sectors in Tur-
key, M. Hasanov and E. Telatar for 178 countries, H. Kum for 15 East Asia and
Pacific countries, B. Ozcan for 17 MENA countries, V. Yilanci and Ç. Tunali for
109 countries, V. Mishra and R. Smyth for 7 sectors in Australia, S. Erdogan et al.
for 6 sectors in Turkey, O. Omoju et al. for 48 Sub-Saharan African countries, and
W. He and H. Chen for 30 Provinces in China.24
On the other hand, we have increasingly seen a growing number of studies
focusing on series for individual fossil energy sources. Regarding fossil fuel use
(all types of sources considered), see H. Lean and R. Smyth along with the work
of M. Destek and S. Sarkodie.25 Regarding both crude oil and NGL production,
interesting conclusions can be found in articles by P. Nayaran et al.26 For petro-
leum consumption, see H. Lean and R. Smyth, N. Apergis and J. Payne, S. Ajlouni
et al., and F. Kızılkaya.27 For coal consumption, one can refer to the article by N.
Apergis et al.28 Finally, for studies related to natural gas, see N. Apergis et al., A.
Aslan, and Y. Cai and C. Magazzino.29
Somewhat connected to this literature, other examinations displayed a signifi-
cant attention toward electricity consumption series. This set of studies is
highlighted by the works of P. Narayan and R. Smyth for 182 countries, L. Gil-
Alana et al. for the United States, N. Apergis and C. Tsouma for 5 sectors in the
United States, F. Kula et al. for 23 high income countries, F. Kula for Turkey, D.
Borozan and L. Borozan for Croatia, M. Herrerias et al. for 29 Chinese regions,
and C. Magazzino for 18 MENA countries.30 In general, previous studies
employed various unit root test procedures over the time domain. Notwithstanding
their relevance, the associated results often conflicted. Table 1 summarizes the
main information on this empirical literature; notice that studies are ordered
chronologically.
Studies on the Convergence of Renewable and Alternative Energy Con-
sumption Series: In this subsection, we highlight the primary features of the more
current studies on the convergence of renewable and alternative energy31
Table 1
230
SUMMARY OF PREVIOUS STUDIES ON TOTAL ENERGY, FOSSIL ENERGY, OR ELECTRICITY USE SERIES CONVERGENCE a
Seg. into
Authors(s) Countries Energy Indicator Data Period Methodology Findings Sub-Series
Narayan and 182 countries Electricity 1979–2000 ADF H0 of non- No
Smyth (2007) consumption per stationarity
capita rejected for 56
countries
Chen and Lee 104 countries Energy 1971–2002 CBL panel unit Non-stationarity No
(2007) consumption per root test
capita
Hsu and Lee 84 countries Energy 1971–2003 Panel SURADF H0 of non- No
(2008) consumption stationarity
rejected
Narayan et al. 6 Australian states Crude oil and NGL 1973–2007 LM (LS) H0 of non- No
(2008a) production stationarity
rejected for 5
out of the 6
states
Narayan et al. 60 countries Crude oil and NGL 1971–2003 LM H0 of non- No
(2008b) production stationarity
rejected with
endogenously
determined
structural break
THE JOURNAL OF ENERGY AND DEVELOPMENT
Lean and Smyth 5 sectors in the Petroleum 1973m1–2008m7 LM (FI) H0 of non- Yes (24
(2009) U.S. consumption stationarity petroleum
rejected for 3 fuels)
out of the 5
sectors
(continued)
Table 1 (continued)
Seg. into
Mishra et al. 13 Pacific Island Energy 1980–2005 CBL panel unit H0 of non- No
(2009) countries consumption per root test stationarity
capita rejected for
8 countries
Apergis and Payne 50 states in the Petroleum 1960–2007 LM (LS, NP) H0 of non- No
(2010) U.S. consumption stationarity
rejected with
endogenously
determined
structural break
Apergis et al. 50 states in the Coal consumption 1982–2007 LM (IPS, CBL, H0 of non- No
(2010a) U.S. CBL and W) stationarity
rejected with
endogenously
determined
structural break
Apergis et al. 50 states in the Natural gas 1980–2007 Panel unit root H0 of non- No
(2010b) U.S. consumption tests stationarity
rejected with
endogenously
determined
structural break
Gil-Alana et al. U.S. Electricity 1973m1–2009m5 FI Non-stationarity Yes (8
ELECTRICITY & THE WORLD’S LARGEST PRODUCERS
(2010) consumption electricity

sources)
(continued)
231
Table 1 (continued)
232
Seg. into
Aslan (2011) 50 states in the Natural gas 1960–2008 LM (KSS) H0 of non- No
U.S. consumption stationarity
rejected for 23
out of the 50
states
Aslan and Kum 7 sectors in Turkey Energy 1970–2006 LM (LS) H0 of non- No
rejected for 4
out of the 7
sectors
Hasanov and 178 countries Primary energy 1980–2006 ADF, LM (KSS H0 of non- No
Telatar (2011) consumption per and S) stationarity
capita rejected
Ozturk and Aslan 7 sectors in Turkey Energy 1970–2006 LM (IPS) H0 of non- No
(2011) consumption per stationarity
out of the 7
sectors
Apergis and 5 sectors in the US Fossil, coal, and 1989m1–2009m12 FI H0 of non- No
Tsouma (2012) electricity stationarity
consumption rejected
Kum (2012) 15 East Asia & Energy 1971–2007 LM H0 of non- No

Pacific countries consumption per stationarity
capita rejected with
endogenously
determined
structural break
(continued)
Table 1 (continued)
Seg. into
Ozcan (2013) 17 MENA Energy 1980–2009 LM (LS and IPS) H0 of non- No
countries consumption per stationarity
capita rejected
Lean and Smyth 4 sectors in Fossil fuel 1978–2010 LM (SP and LS) H0 of non- Yes (10
(2014) Malaysia consumption stationarity fossil
rejected for 7 fuels)
out of the 10
fossil fuels
Kula et al. (2014) 23 high income Electricity 1960–2005 LM (LS) H0 of non- No
countries consumption per stationarity
out of the 23
countries
Kula (2014) Turkey Electricity 1923–2008 LM (LS) H0 of non- No
consumption per stationarity
capita rejected
Yilanci and Tunali 109 countries Energy 1960–2011 LM (FLM) H0 of non- No
(2014) consumption per stationarity
out of the 109
countries
Borozan and Croatia Electricity 2001–2013 LM Non-stationarity No
Borozan (2015) consumption per

capita
(continued)
233
Table 1 (continued)
234
Seg. into
Herrerias et al. 29 Chinese Electricity and coal 1995–2011 CCM Different No
(2017) Regions consumption convergence
between rural
and urban areas
Magazzino (2017) 18 MENA Electricity 1971–2013 CADF H0 of non- No
countries consumption stationarity
rejected without
constant and
trend
Mishra and Smyth 7 sectors in Energy 1973–2014 LM and RALS-LM Evidence of No
(2017) Australia consumption per convergence for
capita 6 out of the 7
sectors
Cai and Magazzino G-7 countries Natural gas 1965–2016 FQUR H0 of non- No
rejected when
modelling breaks
excluded
Destek and 16 OECD Fossil fuel 1970–2018 FKSS H0 of non- Yes (oil,
Sarkodie (2020) countries consumption stationarity coal
rejected for oil and gas)
(4 out of 16
countries), coal
(6 out of 16
countries), and
natural gas (5
out of 16
countries) series
(continued)
Table 1 (continued)
Seg. into
Erdogan et al. 6 sectors in Turkey Energy 1970–2016 F-PKPSS Non-stationarity at No
(2020) consumption per both the sectoral
capita and panel levels
Omoju et al. 48 Sub-Saharan Energy 1980–2011 ADF, CMR, and H0 of non- No
(2020) African consumption LM (LS) stationarity
countries rejected
Ajlouni et al. Jordan Petroleum 1961–2019 LM Non-stationarity No
(2021) consumption
Kızılkaya (2022) 10 OECD Oil consumption 1965–2019 FQUR Non-stationarity Yes
countries
He and Chen 30 Provinces in Energy 1990–2017 CCM Non-stationary/ No
(2022) China consumption non-convergence
a
Seg. 5 segmentation; ADF 5 Augmented Dickey-Fuller unit root test; CADF 5 Cointegration Augmented Dickey-Fuller unit root test; CBL 5
Carrion-i-Silvestre et al. (2005) panel unit root test; CCM 5 Club Convergence Modelling; CMR 5 Clemente, Monta~nes, and Reyes (Clemente et al.,
1998) unit root test; FI 5 Factorial Integration (Nielsen, 2005); FKSS 5 Fourier Kapetanios et al. (2003) unit root test; FLM 5 Fourier type Lagrange
Multiplier (Enders and Lee, 2012); F-PKPSS 5 Fourier Panel Kwiatkowski–Phillips–Schmidt–Shin (Kwiatkowski et al., 1992) unit root test; FQUR 5
Fourier Quantile unit root test; IPS 5 Im et al. (2003) unit root test; KSS 5 Kapetanios et al. (2003) unit root test; LM 5 Lagrange Multiplier; LS 5 Lee
and Strazicich (2003) one break unit root test; NP 5 Narayan and Popp (2010) unit root test; Panel SURADF 5 Panel Seemingly Unrelated Regressions
Augmented Dickey–Fuller (Breuer et al., 2001); RALS-LM 5 Residual Augmented Least Quares-Lagrange Multiplier unit root test; S 5 Sollis (2005)
unit root test with simultaneous structural change and nonlinearity; SP 5 Schmidt and Phillips (1992) unit root test; W 5 Westerlund (2005) unit root test;
and MENA 5 Middle East and North Africa.

Source: Authors’ compilation.
235
consumption series. Recently, this research area, in general, has assessed the inte-
gration properties of clean energy consumption at the aggregate level (see Y.
Wang et al. for Japan and Y. Cai and A. Menegaki for eight emerging countries).32
For renewable energy use, most notably, the reader can refer to the studies by C.
Pestana Barros et al. and C.-C. Lee et al. for the United States, G. Gozgor for Bra-
zil, China, and India, E. Demir and G. Gozgor for 54 countries, and E. Fendoglu
for BRICS-T.33 Concerning research on nuclear power use, we highlight the
articles by C. Pestana Barros et al. for the United States and H. Zhu and P. Guo for
27 countries.34 However, it is worth reporting the few seminal assessments that
have opened a productive research pathway by examining the stationary properties
of individual renewable energy sources; these include N. Apergis and C. Tsouma
for solar, geothermal, and biomass energy consumption and M. Aydin and U. Pata
for the analysis of nine distinct renewable energy sources.35 Another recent
research focus area has been on conducting convergence assessments for global
and relatively wider sample sizes, but composed of structurally heterogeneous
economies (see C. Saba and N. Ngepah for 182 countries).36 Table 2 summarizes
the studies focusing on convergence of renewable and alternative energy
consumption.
Methodological Discussion, Empirical Gaps, and Contribution Proposal: A
survey of the energy convergence literature highlights several major points. First, a
wide range of papers have extensively assessed the stationary properties of total
energy use and focused on various case studies (P. Chen and C. Lee for 104 coun-
tries, Y.-C. Hsu et al. for 84 countries, V. Mishra et al. for 13 Pacific Island coun-
tries, A. Aslan and H. Kum for 7 sectors in Turkey, H. Kum for 15 East Asia and
Pacific countries, and B. Ozcan for 17 MENA countries).37 While they typically
collected aggregate consumption series, disaggregated energy sources have been
considered much less. Moreover, a recent line of investigation in the field has
arisen using integration properties of renewable energy series, which can be seen
in the works of C. Pestana Barros et al. and C.-C. Lee et al. for the United States,
G. Gozgor for Brazil, China, and India, E. Demir and G. Gozgor for 54 countries,
E. Fendo glu for BRICS-T countries, and C. Saba and N. Ngepah for a mixed panel
of 184 economies.38 However, for both lines of research, it is clear that case stud-
ies, data periods, and methodologies differ strongly and the studies’ associated con-
clusions often conflict. Therefore, a trade-off with diverging costs-benefits seems
to emerge. On the one hand, we do observe a progressive shift toward more global-
ized panel assessments with samples containing numerous but structurally hetero-
geneous economies. However, this not only prevents the supply of country-specific
empirical insights, it also offers concluding remarks and policy implications that
are less generalizable because they are constrained by the collapsing nature of the
panel. On the other hand, a few attempts to conduct single-country investigations
can be identified. Sometimes, these relied on sectoral or province-level data series,
but not systematically. While this approach offers the advantageous feature of
Table 2
SUMMARY OF PREVIOUS STUDIES ON CONVERGENCE OF RENEWABLE AND ALTERNATIVE ENERGY CONSUMPTION a
Seg. in
Apergis and 5 sectors in the Solar, geothermal, and 1989m1– FI H0 of non-stationarity No
Tsouma (2011) U.S. biomass energy 2009m12 rejected
consumption
Pestana Barros U.S. Renewable energy 1981m1– FI Non-stationarity No
et al. (2012) consumption 2010m10
Pestana Barros U.S. Nuclear energy 1973m1– FI Non-stationarity No
et al. (2013) consumption 2011m10
Gozgor (2016) Brazil, China, Renewable energy 1971–2014 NP-UR, LS H0 of non-stationarity No
and India consumption rejected for China and
India
Wang et al. (2016) Japan Renewable and nuclear 1965–2011 LM (LS, IPS), H0 of non-stationarity No
energy consumption FLM rejected for renewables
only
Zhu and Guo 27 countries Nuclear energy 1993–2013 CBL panel unit H0 of non-stationarity No
(2016) consumption per capita root test rejected with
endogenously deter-
mined structural break
Demir and Gozgor 54 countries Renewable energy 1971–2016 NP H0 of non-stationarity No
(2018) consumption rejected for 45 out of
the 54 countries
Cai and Menegaki 8 emerging Renewable and nuclear 1990–2012 FQUR H0 of non-stationarity No
(2019) countries energy consumption rejected for China,

Pakistan, and Thailand
(continued)
237
238
Table 2 (continued)
SUMMARY OF PREVIOUS STUDIES ON CONVERGENCE OF RENEWABLE AND ALTERNATIVE ENERGY CONSUMPTION a
Seg. in
Aydin and Pata U.S. Renewable energy 1973m01– Wavelet-based H0 of non-stationarity Yes (9
(2020) consumption 2019m09 ADF rejected for biofuels renewable
and hydropower energy energy
consumption sources)
Fendo
glu (2021) BRICS-T countries Renewable energy 1990–2015 FFFUR H0 of non-stationarity No
consumption rejected for China only
Lee et al. (2021) 50 states in the Renewable energy 1960–2017 FQUR H0 of non-stationarity No
U.S. consumption rejected for 32 out of
the 51 states
Saba and Ngepah 183 countries Renewable energy 2000–2018 CCM/ MLR/ No panel convergence No
(2022) (SSA, MENA, consumption Clustering
ECA, ESAP and Algorithms
America)
Notes: SSA 5 Sub-Saharan Africa (SSA); MENA 5 Middle East and North Africa; ECA 5 Europe and Central Asia; ESAP 5 East and South Asia
and the Pacific; and BRICS-T 5 Brazil, Russia, India, China, South Africa, and Turkey. CBL: Carrion-i-Silvestre et al. (2005) panel unit root test. CCM:
Club Convergence Modelling. FFFUR: Fractional Frequency Fourier unit root test. FI: Factorial Integration (Nielsen, 2005). FQUR: Fourier Quantile unit
root test. IPS: Im et al. (2003) unit root test. LS: Lee and Strazicich (2003) one break unit root test. MLR: Multinomial Logit Regression. NP: Narayan and
Popp (2010) unit root test. Wavelet-based ADF: Wavelet based Augmented Dickey-Fuller unit root test.
providing more reliable policy insights, it fails to exploit the cross-case compara-
bility potential of time-series analyses. Furthermore, it exits from the scope of stan-
dard b-convergence frameworks.
It is clear that integration properties of any given data series are sensitive to the
selection of case studies, definition, and metrics of variables but also choices of
methods. Regarding the former, there is a need to reorient the literature toward a
more reliable and relevant selection of samples and overcome choices only driven
by data availability constraints. Instead, gathering together countries sharing a slig-
htly similar structure or exhibiting a comparable trend may deliver more valuable
insights than large panels with mixed and highly heterogenous slope coefficients.
Looking at the latter, it appears crucial to highlight the diversity of time-series meth-
odologies that have been employed on this topic thus far. Ranging from standard
unit root tests on the time domain to Fourier-transformed procedures on the fre-
quency domain, these strategies rely on econometric frameworks that exhibit dif-
ferent statistical assumptions and scopes. While some are univariate tests with a
constant and trends, others allow for endogenously determined structural breaks in
the series with often the impossibility to transfer from a country to a panel setting
and vice versa using an identical test.
Additionally, as was previously discussed with the rising frequency of studies
conducting large panel assessments, we must highlight that cross-sectional depen-
dence and spatial correlation among the series remain largely uncontrolled for in
these works. While there are some notable exceptions (C. Magazzino, M. Destek and
S. Sarkodie, and S. Erdogan et al.), the outstanding methodological issues still
remain.39 In other words, there is a need to complement conventional panel stationary
tests with more advanced applications accounting for cross-sectional dependence
dynamics. Indeed, over the past decade the Time-Series Cross-Section (TSCS) eco-
nometric literature showed a keen interest in cross-sectional dependence-related bias.
From an empirical standpoint, it is clear that inter-dependencies driven by the eco-
nomic and financial integration of markets and taking the form of spatial correlations
across countries can no longer be neglected by integration properties assessment
models.40 In a globalized context (trade and financial), regional and national linkages
operate through common macroeconomic shocks affecting multiple entities, common
international policy directions pushed by inter-state organizations (i.e., the Bank for
International Settlements (BIS), the Kyoto Protocol and the Paris Agreement, Interna-
tional Labour Organization (ILO)), spillover effects, and technology transfers across
sectors belonging to different industries and countries.41 From a statistical point of
view, a cross-sectionally dependent data structure can emerge in the presence of com-
mon shocks, spatial dependence, and unobserved components, which are captured by
the error term but correlate with our electricity series.42 When not taken into account,
these factors are likely to introduce inconsistent spurious estimates.43
Overall, while a promising research direction has been advanced of late by stud-
ies collecting and analyzing the stationary properties of electricity use series,44 to
the best of our knowledge, no such assessment has explicitly focused on the
world’s top largest electricity producers to date. Such an empirical gap is critical
since the electricity demand recorded in these ten economies cannot be discon-
nected from individual pollution trends, global energy consumption, and trade
structures, in addition to patterns of cross-market dependence. Accordingly, this
calls for further inquiry into this topic.
Therefore, this paper seeks to fill these above-mentioned literature gaps in a sin-
gle manner and provide several distinct and novel aspects—empirically and meth-
odologically. Foremost, this study is the first to investigate the stationary
properties of electricity consumption series for China, the United States, India,
Russia, Japan, Brazil, Canada, South Korea, Germany, and France. This sample of
economies, which is comprised of the major energy-consuming and pollution emit-
ters globally, also corresponds to the top ten largest power producers worldwide
(from 1st to 10th). Second, this paper contrasts to previous ones as it employs two-
stage stratified approach: univariate and endogenously determined structural
breaks-accounting unit root tests at the country-level, followed by a set of cross-
sectional dependence tests, and panel stationary tests accounting for structural
breaks and cross-sectional dependence among the series. Combining panel-scaled
results and nationwide evidence is thought to offer a more reliable picture distin-
guishing between some general electricity patterns at the sample level and other
country’s specificities and heterogeneities. Finally, the data series used in this
study span the largest and most available period (1985-2020), which provides
more accurate economic and policy insights that can be utilized for energy and
environmental planners and policy makers.
Data Collection and Econometric Framework
Data Collection: To implement our methodology, this paper collected data on

electricity consumption for China, the United States, India, Russia, Japan, Brazil,
Canada, South Korea, Germany, and France, all of which are considered to be the
world’s top ten largest electricity consumers (ordered from 1st to 10th). Referring
to C. Magazzino et al. and M. Murshed, electricity power generation can be
employed as a relevant proxy for total electricity demand.45 Spanning the largest
period (1985–2020), our series are expressed in terawatt-hours (TWh) and col-
lected from British Petroleum’s Statistical Review of World Energy.46 Figure 1
presents the per-country time plots of electricity consumption over the period
1985–2020. For a better visualization, data are displayed in level-scale over the
two most recent decades (2000–2020).
A visual inspection of these graphs reveals the existence of a relative upward
trend for all countries. In particular, the consumption of power has dramatically
risen in China and India, whereas flatter slopes are observed for Canada and Japan.
Figure 1
ELECTRICITY CONSUMPTION OF THE WORLD’S TOP TEN LARGEST POWER
PRODUCERS: CHINA, THE UNITED STATES, INDIA, RUSSIA, JAPAN, BRAZIL, CANADA,
SOUTH KOREA, GERMANY, AND FRANCE (2000–2020)
8000
Electricity Power Generaon per Country -Twh
6000
4000
2000
2000 2005 2010 2015 2020

Years
Source: Authors’ elaboration based on British Petroleum’s Statistical Review of World Energy
(2021) data.
One striking observation is that Germany seems to be recording a slightly declining

trend of electricity generation (TWh) over time, whereas the speed and overall
magnitude of this electricity demand appear sharply heterogeneous and sporadic
across economies composing our sample. Besides, one can also depict the exis-
tence of some structural breaks in energy series, which can be linked to external
shocks, internal business cycles, and price fluctuations, but also non-negligible
inter-dependencies among these countries.47 To deal with these inferences, this
study employs logarithm-transformed series and applies a country-level unit root
testing framework allowing for structural breaks, along with panel unit root tests
accounting for cross-sectional dependence and spatial correlation among the series.
We take the natural logarithmic form of each series. The complete time-series
econometric framework is presented in the next subsection.
The descriptive statistics for each country with level-scale data are listed in
table 3. As a preliminary analysis, one sees that the skewness value of renewable
energy consumption series is positive for all countries, indicating that the distribu-
tion is left-skewed, with more observations on the right. Also, the Inter-Quartile
Table 3
DESCRIPTIVE STATISTICS AND CORRELATION ANALYSISa
Country Mean Median Std. Dev. Skew. Kurt. S-W Test IQR
China 2847.407 1782.288 2368.146 20.83193 0.75946 3.409*** 7368.37
(0.00033)
United States 3880.168 4124.877 566.4619 20.50672 20.88775 3.588*** 1804.436
(0.00017)
India 741.0193 623.8907 438.1849 20.75794 0.648809 2.404*** 1417.289
(0.00812)
Russia 992.0327 1013.575 91.00511 21.19565 20.48072 2.596*** 290.9852
(0.00471)
Japan 1013.433 1054.027 139.4059 0.744349 21.22936 3.403*** 511.767
(0.00033)
Brazil 391.9021 356.6304 145.2944 21.40605 0.248037 2.484*** 432.6462
(0.00649)
Canada 585.4512 599.1508 61.4148 20.85677 20.56418 2.190** 204.7292
(0.01426)
South Korea 337.413 339.8886 178.6126 21.44127 20.04825 2.240** 530.2374
(0.01254)
Germany 587.7562 586.5498 44.44705 21.57609 0.004342 2.634*** 130.3922
(0.00422)
France 514.55 544.7425 69.31378 0.197418 21.15938 3.933*** 232.5792
(0.00004)
a
*** 5 p,0.01; ** 5 p,0.05; and * 5 p,0.10. Std. Dev. 5 standard deviation; Skew. 5
skewness; Kurt. 5 Kurtosis; S-W Test 5 Shapiro-Wilk Test; and IQR 5 Inter-Quartile Range.
Source: Authors’ elaborations.
Range (IQR), which measures the difference between the first and third quartile
marks, shows that the middle-half of the data has a high variability in a few coun-
tries only. Finally, the Shapiro-Wilk test (S-W) rejects the null hypothesis of nor-
mal residuals distribution at the 5-percent level for all countries.
Time-Series Econometric Framework: This section presents our stepwise
time-series methodology.
Country-Level Integration Properties—Univariate and Endogenously Deter-
mined Structural Breaks: A time series is said to be stationary if its statistical prop-
erties (mean, variance, and autocorrelation) are constant over time. In a nutshell, if
a series is stationary, one can predict that its properties will remain constant in the
future, which turns out to be highly useful when it comes to forecasting. Con-
versely, most statistical forecasting methods require the validation of this assump-
tion. Here, we employ a stepwise time-series testing methodology involving
univariate country-level unit root tests, as well as country-level stationary tests

allowing endogenously determined structural breaks, slope homogeneity, and
cross-sectional dependence tests, in addition to panel stationary tests excluding/
accounting for those inferences.
The first stage of our stationary testing procedure refers to a set of univariate unit
root tests performed at the country level, namely, the Augmented Dickey and Fuller
(ADF) test, the Phillips and Perron (PP) test, the Dickey and Fuller-Generalized
Least Squares (DF-GLS) test, and the Kwiatkowski–Phillips–Schmidt–Shin (KPSS)
test.48 Moreover, to consider deterministic breaks in the series, the Clemente, Mon-
nes, and Reyes (CMR) test is conducted, followed by the Zivot and Andrews
ta~
(ZA) test.49 While the former allows for additive outliers, the latter accounts for
multiple endogenous breaks in both the intercept and the time trend. When identify-
ing integration properties, they both modify the standard ADF regression and
include a dummy regressor within the specification. In doing so, the period when
the time break TB occurs can be identified. Considering a series yt, the ZA model
endogenies structural breaks as follows:
X
k
yt 5 a1 1 a2 t 1 a3 DUt ðlÞ 1 a4 DTt ðlÞ 1 ryt1 1 wj Dytj 1 «t (1)
j51
Where the dummy variables DUT, capturing a switch in the intercept, equals 1, and
DTt represents a shift in the trend, equals t if t . TB and 0 otherwise. This is the
most common formulation of the ZA model because it allows for changes in both
the intercept and the broken trend. This typically refers to Model ‘C’ in compari-
son to Model ‘A’ (with a4 5 0Þ, which examines whether the series is trend station-
ary with a break in the mean and Model ‘B’ (a3 5 0), which tests for the existence
of a trend shift. The ZA model estimates sequentially these three models: ‘A, ’ ‘B, ’
and ‘C:’ Finally, it ensures that TB is endogenously determined because it allows
this latter to occur in any particular year of the time series, except the first and last
one. Finally, the break fraction “l-ratio of the pre-break sample size to the overall
sample” is chosen to lower the one-sided t-statistic for testing ^r 5 1 and test the
null hypothesis of a unit root process with a break. This is equivalent to:
H0: yt 5 a 1 yt1 1 «t (2)
Panel Integration Properties—Heterogeneous Slope Coefficients and Cross-

Correlations: One important issue in panel analysis is the possible bias that may
be induced by heterogeneous slope coefficients, which is referred to in the litera-
ture as the problem of heterogeneity. Based on the standard framework of the
Swamy’s test50 for relatively small N compared to T, M. Pesaran and T. Yamagata
proposed a test for slope homogeneity for panel data with larger N and T.51
The null hypothesis of the test is of homogenous slopes, which implies that all
slope coefficients are identical across cross-sectional units, against the alternative
hypothesis of heterogeneous slope b coefficients. One competitive feature of this
procedure is that it automatically assumes a vector of heterogeneous constants.
Associated hypotheses are formulated as:
H0 : bi 5 b for some i, against

(3)
H1 : bi Þb for some iÞj
Later, J. Blomquist and J. Westerlund52 offered a more generalized framework

to assess the nature of slope coefficients in the presence of both heteroskedasticity
and serially correlated errors. These latter differ from the straightforward con-
structed versions presented by M. Pesaran and T. Yamagata and which could be
simply removed by demeaning the series. Let us assume a data-generating process
such as:
yi,t 5 ai 1 ji xi,t 1 vi,t (4)
Where i 5 1 . . . N, xi,t corresponds to a vector of regressors, and ji refers to the

slope coefficients for the associated value. The Heteroskedasticity and Autocorrela-
tion Consistent (HAC) version of D found in J. Blomquist and J. Westerlund is
defined as:
8 !
>
> pffiffiffiffi N 1 StHAC k2
>
> D~ HAC 5 N pffiffiffiffiffiffiffi
>
< 2k2
(5)
>
>
> X N 9 1

> ^ 2HAC X^ i,T V
^ 2i G ^ X^ i,Ti G ^ 2HAC
^ 2i G
: StHAC 5
> Ti G i i,Ti
i51
Where G ^ 2HAC is a robust HAC estimator, X^ i,T corresponds to a trajectory matrix

^ 1 is a variance estimator with a
i
composed of the heterogeneous variables, and V i,Ti
kernel function k and bandwidth parameter Bi,T : From a statistical point of view, a
cross-sectionally dependent data structure may emerge in the presence of common
shocks, spatial dependence, and unobserved components, which are captured by the
error term but correlate with our electricity series.53 From an empirical standpoint,
inter-dependencies driven by rising trade and financial integrations of national
energy markets and taking the form of spatial correlations across countries may play
out and, thus, can no longer be neglected by integration properties assessment mod-
els.54 When not evaluated correctly, this could result in inconsistent spurious esti-
mates,55 therefore methods allowing for those inferences should complement
standard univariate ones.56 By contrast to first-generation tests developed on the
assumption that panel units are cross-sectionally independent,57 second-generation

stationary procedures are robust to heterogeneous slope coefficients and cross-
sectional dependence across units.58
To fill this gap, we must first identify whether this inference holds for our sample
or not, considering both level and first difference-transformed variables. Typically,
one would use the Frees’ test of cross-sectional independence,59 the Friedman’s test
of cross-sectional independence,60 or the Breusch Pagan Lagrange Multiplier (LM)
test of independence.61 Additionally, in this paper we also employ the Pesaran
Cross-section Dependence (CD) test, followed by the Pesaran et al. bias-adjusted
LM test of error cross-section independence, which are considered as the most
advanced available statistical techniques to assess such dynamics. Let us consider a
standard panel-data model (Equation 2), uit is assumed to be Independent and Identi-
cally Distributed (IID) over periods and across cross-sectional units. Under the alter-
native, uit may be correlated but the assumption of no serial correlation remains.
Algebraically, we have:
H0 : rij 5 rji 5 corrðuit , ujt Þ 5 0 for iÞj
H1 : rij 5 rji Þ0 for some iÞj (6)
Where rij refer to the sample-level estimation of the pair-wise correlation of the
residuals computed using «it amd the OLS residuals estimate of uit . ^r ij are com-
puted as follows:
PT
« «
t 5 1 it jt
r^ ij 5 r^ ji 5 P 12 P 1 (7)
T T
t51
« it
2
t51
« jt
2 2
Breusch and Pagan proposed the LM statistics valid for fixed N as T ! 1: None-
theless, this is likely to exhibit size distortions when N is large and T is finite. This
situation is commonly encountered in empirical panel approaches and does apply
to the present case study of the world’s top ten largest electricity consumers
(China, the United States, India, Russia, Japan, Brazil, Canada, South Korea, Ger-
many, and France). To fill this gap, Pesaran (2004) proposed the alternative test
statistic, which presents two competitive features: under the null hypothesis of no
cross-sectional dependence and for large T and N ! 1, CD converge to N(0,1) (i)
and, unlike the LM statistic, the CD statistic will display a zero mean for fixed T
and N values under heterogeneous and non-stationary models.62
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2T X X
N 1 N 51
CD 5 r
^ (8)
N ðN 1Þ i 5 1 j 5 i 1 1 ij
Accordingly, we employ the Pesaran (2007) Cross-sectionally Augmented Im,

Pesaran, and Shin (CIPS) panel unit root test, which has the advantage of being
robust to heterogeneity and cross-sectional dependence of panel units. The CIPS
test is based on an extension of the Covariate-Augmented Dickey-Fuller (CADF)
test developed by B. Hansen,63 where the null hypothesis (H0) refers to homoge-
neous non-stationary time series, whereas the alternative hypothesis (H1) corre-
sponds to a stationary process for at least one of the individual series tested.64
Consider a standard Dickey-Fuller65 framework augmented with lagged levels and
first-differences of the cross-section averages of the individual series:
X
p X
p
DYit 5 ai 1 bi Yi,t1 1 ciy t1 1 dij Dy t1 1 dij Dyi,tj 1 «it (9)
j50 j51
Where the Cointegration-Augmented Dickey–Fuller (CADF) statistic for the coun-

try i is here the t-statistic when bi 5 0. This leads us to formulate the CIPS statistic
test of Pesaran (2007), which is computed as the simple arithmetic average of the
CADF statistics.
1 XN
CIPS p 5 CADF i,p (10)
N i51
Moreover, the t-bar test proposed by K. Im et al.66 assumes that all countries con-
verge toward the equilibrium value at heterogeneous speeds under the alternative
hypothesis (H1). However, G. Maddala and S. Wu67 criticized the K. Im et al. test
because many real data applications fail to exhibit cross-correlations that are simi-
lar to the simple version of those effectively eliminated by the K. Im et al. test
while demeaning the data series. Thus, they proposed a panel unit root test based
on Fisher, which combines p-values of the test statistic in each residual cross-
sectional unit. Using the additive property of the chi-squared variable, G. Maddala
and S. Wu68 derived the following test statistic:
X
N
l5 2 loge pi (11)
i51
Where the non-parametric test displays a chi-square distribution with 2N degrees

of freedom and where N is the number of cross-sectional units or countries.
Finally, pi refers to the p-value of the test statistic for unit i. Building upon this, J.
Breitung69 proposed a panel unit root test using the following extended functional
form (later extended in 2005 by J. Breitung and S. Das):70
pX
11
yit 5 ait 1 bik xi,tk 1 «t (12)
k 51
Where the null hypothesis laying under the Breitung71 test statistic is difference
Pp 1 1
stationarity H0 : k 5 1 bik 1 5 0 against the alternative stationary H1 :
Pp 1 1
k 5 1 bik 1,0 for all i: The following transformed vectors are used to con-
struct the test statistic:
(
Yi 5 AYi 5 ½yi1 ,yi2 ,:::,yiT 9
(13)
Xi 5 AXi 5 ½xi1 ,xi2 ,:::,xiT 9
This leads to the following Maddala and Wu-augmented test statistic,72 displaying
the advantageous feature of having a standard normal distribution:
PN
s2 Yi9 Xi9
i51 1
lB 5 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN 2 9 9 9 (14)
i51 1
s Xi A AXi
Overall, juxtaposed to the results of the Breitung73 panel unit root test and the
Pesaran Cross-sectionally Augmented Im, Pesaran, and Shin (CIPS) panel unit root
test, we display the outputs of standard panel unit root procedures, which neither
control for heterogeneity nor for cross-sectional dependence, the Im-Pesaran-Shin
(IPS),74 and the Levin-Lin-Chu (LLC)75 panel unit root tests. Comparing first to
second generation unit root outcomes aims to examine whether our sample-level
stationary findings are sensitive to the presence of those latter inferences within the
data. In sum, our stepwise unit root testing methodology is outlined in figure 2. It
is comprised of the following stages:
Stage 1 - Univariate unit root tests are first performed at the country level,
namely, the Augmented Dickey and Fuller (ADF) test (Dickey and Fuller,
1979), the Phillips and Perron (PP) test (Phillips and Perron, 1988), the
Dickey and Fuller-Generalized Least Squares (DF-GLS) test, and the Kwiat-
kowski-Phillips-Schmidt-Shin (KPSS) test (Kapetanios et al., 2003).76
Stage 2 - We conduct the Clemente, Monta~nes, and Reyes (CMR) test (Cle-
mente et al., 1988), followed by the Zivot and Andrews (ZA) test (Zivot and
Andrews, 2002);77 while the former allows for additive outliers, the latter
accounts for multiple endogenous breaks in both the intercept and the time
trend.
Stage 3 - We assess the main panel statistical properties: heterogeneous slope
coefficients (Pesaran and Yamagata (2008) test of slope homogeneity); cross-
sectional dependence among the series (Friedman (1937) and Frees (1995,
2004) cross-sectional independence tests); Pesaran (2004) cross-section
dependence (CD) test; and Pesaran et al. (2008) bias-adjusted LM test of
error cross-section independence.78
Figure 2
248
STEPWISE UNIT ROOT TESTING METHODOLOGY
Augmented Dickey Fuller (ADF) (Dickey and Fuller, 1979)

Stage 1:
Country-level Phillips and Perron (PP) test (Phillips and Perron, 1988)
univariate unit
root tests Dickey and Fuller-Generalized Least Squares (DF-GLS) test
Kwiatkowski-Phillips-Schmidt-Shin (KPSS) (Kapetanios et al., 2003)

Stage 2: Country-level unit
root tests allowing for Clemente, Montañés, and Reyes (CMR) test for addive outlier (Clemente et al., 1988)
Pesaran and endogenously determined
Yamagata (2008) test structural breaks Zivot and Andrews (ZA) test with structural breaks (Zivot and Andrews, 2002)
of slope homogeneity
Stage 3: Frees (1995, 2004) cross-seconal independence test
Blomquist and
Westerlund (2013) Test of Test of cross- Friedman (1937) cross-seconal independence test
test of slope heterogeneous seconal
homogeneity in large slope + dependence & Pesaran (2004) Cross-seconal Dependence (CD) test
panels with serial coefficients cross-correlaon
correlaon Pesaran et al. (2008) bias-adjusted LM test of error cross-secon
independence
Stage 4:
Levin-Lin-Chu (LLC, Breitung (2000) panel unit root test accounng for cross-correlaons
Levin et al., 2002)
panel unit root test Standard panel Panel unit root Pesaran (2003) CADF panel unit root test based on the mean of DF
unit root tests tests accounng t-stasc of each unit
Im-Pesaran-Shin (for juxtaposion + for cross-seconal
(IPS, Im et al., 2003) purpose only) dependence Pesaran (2007) Cross-seconally augmented Im, Pesaran, and Shin
panel unit root test (CIPS) panel unit root test
Stage 5: Kmeans clustering and β-convergence analysis
Source: Authors’ elaboration.

Stage 4 - We investigate panel integration properties using panel unit root

tests accounting for cross-sectional dependence and heterogeneity: Breitung
(2000) panel unit root test robust to cross-correlations; Pesaran (2007) Cross-
sectionally Augmented Im, Pesaran and Shin (CIPS) panel unit root test; and
Pesaran (2003) CADF panel unit root based on the mean of individual DF (or
ADF) t-statistics of each unit in the sample). Then we compare our results
with outputs from standard panel unit root procedures, which neither control
for heterogeneity, nor for cross-sectional dependence: the Im-Pesaran-Shin
(IPS) panel unit root test (Im et al., 2003), and the Levin-Lin-Chu (LLC)
panel unit root test (Levin et al., 2002).
Stage 5 – We implement k-means clustering techniques and conduct a
b-convergence analysis.
Empirical Results
Country-Level Time-Series Analysis: Results of the univariate Augmented

Dickey and Fuller (ADF) test, the Phillips and Perron (PP) test, the Dickey and
Fuller-Generalized Least Squares (DF-GLS) test, and the Kwiatkowski-Phillips-
Schmidt-Shin unit root tests excluding structural breaks in the series are insightful
in several ways.79 Above all, the ADF test shows that electricity consumption
series for all of the world’s top ten power producers are stationary in the first dif-
ference at the10-percent significance level, except for China. Regarding the PP
test, the series for India, Japan, South Korea, Canada, and France reject the null
hypothesis of a unit root in level, whereas all countries present stationary series
when transformed in first difference. Interestingly, the outcome of the DF-GLS test
fails to reject the null hypothesis of no stationarity for all series in levels. However,
all countries, once again, show stationary properties when first-difference data are
used, except for China, India, and Germany. The unit root is rejected for data in
the first difference for one lag. Finally, a similar output is observed when looking
at KPSS findings. Again, we remind the reader that the lag length selection has
been chosen based on the information given by the Final Prediction Error (FPE),
the Akaike’s Information Criterion (AIC), the Schwarz’s Bayesian Information
Criterion (SBIC), and the Hannan and Quinn Information Criterion (HQIC). In
general, we do observe heterogeneous patterns across countries, although this uni-
variate unit root testing analysis claim evidence that our series are integrated of
order 1 (i.e., Ið1Þ); their first-difference transformation present stationary proper-
ties, whereas level data fail to do so.
Following this, we perform stationary tests accounting for endogenously deter-
mined structural breaks in the series for level (table 5) and first-difference-trans-
formed series (table 6): the Clemente, Monta~ nes, and Reyes (CMR) test (breaks in
intercepts and/or trends) and the Zivot and Andrews (ZA) test (additive outlier and
Table 4
RESULTS OF UNIVARIATE AUGMENTED DICKEY AND FULLER (ADF), PHILLIPS AND PERRON (PP), DICKEY AND
250
FULLER-GENERALIZED LEAST SQUARES (DF-GLS), AND KWIATKOWSKI-PHILLIPS-SCHMIDT-SHIN (KPSS) UNIT ROOT TESTS
EXCLUDING STRUCTURAL BREAKS IN THE SERIESa
Level D Operator
DF-GLS KPSS DF-GLS KPSS
Country ADF Stat. PP Stat. Stat. Stat. ADF Stat. PP Stat. Stat. Stat.
China
20.303 (0.9894) 20.985 (0.7585) 21.668 0.177 23.001 (0.1317) 22.898** (0.0456) 22.554 0.171*
United States
20.948 (0.9507) 24.997 (0.0000) 0.107 0.45 26.914* (0.0000) 24.669* (0.0001) 25.056* 0.0291*
India
21.861 (0.6747) 23.244** (0.0176) 20.837 0.282 25.423* (0.0000) 24.411* (0.0003) 22.670 0.113*
Russia
20.994 (0.9449) 21.236 (0.6581) 21.404 0.339 23.268*** (0.0717) 23.209** (0.0194) 23.346*** 0.127*
Japan
21.376 (0.8678) 24.017* (0.0013) 20.108 0.44 26.823* (0.0000) 24.013* (0.0013) 23.710** 0.0612*
Brazil
21.016 (0.9419) 21.249 (0.6520) 21.323 0.185 25.800* (0.0000) 25.715* (0.0000) 23.330*** 0.107*
Canada
22.247 (0.4635) 22.645*** (0.0840) 20.937 0.347 26.827* (0.0000) 26.532* (0.0000) 24.944* 0.0278*
South Korea
20.849 (0.9613) 29.791* (0.0000) 20.094 0.448 26.093* (0.0000) 22.104 (0.2428) 24.833* 0.0429*
Germany
20.550 (0.9813) 21.659 (0.4523) 21.065 0.189 25.632* (0.0000) 25.420* (0.0000) 22.165 0.135*
France
21.214 (0.9077) 25.238* (0.0000) 0.096 0.446 27.163* (0.0000) 24.617* (0.0001) 24.005* 0.0448*
a
***p,0.01, **p,0.05, *p,0.10. lag length selected k 5 1. Stat. 5 Statistics. For DF-GLS, the critical values are as follows: 23.77 (1% significance
level), 22.756 (5% significance level) and 22.371 (10% level). For KPSS, the critical values are as follows: 0.119 (10% significance level), 0.146 (5%
significance level), 0.176 (2.5% significance level), and 0.216 (1% significance level). P-values are reported in brackets.
Table 5
RESULTS FOR UNIT ROOT TESTS WITH STRUCTURAL BREAKS (IN INTERCEPT OR IN
TREND) AND FOR ADDITIVE OUTLIER UNIT ROOT TESTS (SINGLE STRUCTURAL
BREAK) – FOR LEVEL SERIESa
Zivot and Andrews (ZA) Tests – Level Series
(a) (b)
Variable Tb k tmin Tb k tmin
China 2003 1 22.899 2014 1 22.512
United States 2008 1 21.740 2003 1 22.997
India 1991 1 22.055 2015 1 22.391
Russia 1992 1 24.714** 1995 1 24.540**
Japan 2009 1 22.296 1991 1 22.859
Brazil 2015 1 23.126 2014 1 23.579
Canada 1994 1 23.892 2001 1 23.962
South Korea 2013 1 22.130 2011 1 22.195
Germany 2000 1 21.742 2015 1 21.931
France 2000 1 22.030 2003 1 22.305
nes, and Reyes (CMR) Tests – Level Series
Clemente, Monta~
Optimal t-statistic Optimal t-statistic
Variable break point k (additive outlier) break point k (innovational outlier)
China 1997, 2007 1 7.331* 1989, 2000 1 1.901***
(0.000) (0.067)
United States 1994, 2001 1 7.65* 2016, 2018 1 1.54
(0.000) (0.134)
India 1996, 2010 1 8.98** 2006, 2009 1 0.458
(0.012) (0.656)
Russia 1995, 2009 1 23.151* 1990, 2003 1 25.566*
(0.004) (0.000)
Japan 1991, 1996 1 6.030* 2009, 2018 1 23.969*
(0.000) (0.001)
Brazil 1996, 2007 1 9.199* 1993, 2002 1 2.280**
(0.000) (0.03)
Canada 1995, 2008 1 8.962* 1989, 2010 1 2.025***
(0.000) (0.052)
South Korea 1996, 2007 1 8.532* 1992, 1997 1 20.357
(0.000) (0.729)
Germany 2000, 2013 1 10.18* 1998, 2011 1 2.127**
(0.000) (0.042)
France 1992, 2000 1 8.187* 1989, 1998 1 2.680**
(0.000) (0.012)
a
(a) corresponds to the model allowing for break in intercept. (b) refers to the model allowing for
break in trend. Tb is the break date endogenously selected. tmin is the minimum t-statistic. k denotes
the lag length selected according to the AIC criterion. 5% critical values are provided in parentheses.
***p , 0.01, **p , 0.05, *p , 0.1.
Table 6
RESULTS FOR UNIT ROOT TESTS WITH STRUCTURAL BREAKS (IN INTERCEPT OR IN
TREND) AND FOR ADDITIVE OUTLIER UNIT ROOT TESTS (SINGLE STRUCTURAL
BREAK) – FOR FIRST-DIFFERENCE SERIESa
Zivot and Andrews (ZA) Tests – First-Difference Series D
(a) (b)
Variable Tb k tmin Tb k tmin
China 2000 1 24.336 2007 1 23.783
United States 1991 1 27.443* 1993 1 27.012*
India 2007 1 26.510* 1997 1 25.636*
Russia 1999 1 24.267* 1992 1 23.529
Japan 1992 1 27.850* 2012 1 27.263*
Brazil 2015 1 26.856* 2011 1 26.529*
Canada 2011 1 27.043* 1991 1 26.937*
South Korea 1998 1 26.628* 1999 1 26.370*
Germany 1995 1 26.390* 2015 1 26.843*
France 2000 1 24.336 2007 1 23.783
nes, and Reyes (CMR) Tests – First-Difference Series D
Clemente, Monta~
Optimal break t-statistic Optimal break t-statistic
Variable point k (additive outlier) point k (innovational outlier)
China 2001, 2009 1 2.65** 2001, 2006 1 4.034*
(0.012) (0.001)
United States 1999, 2007 1 21.898*** 1990, 2006 1 24.368*
(0.068) (0.000)
India 1997, 2009 1 22.129** 1994, 2010 1 22.457**
(0.041) (0.02)
Russia 1992, 1996 1 23.119 1990, 1993 1 28.146*
*(0.004) (0.000)
Japan 1988, 2007 1 23.810* 1989, 2007 1 23.191*
(0.001) (0.003)
Brazil 1999, 2012 1 21.994*** 2000, 2013 1 24.067*
(0.056) (0.000)
Canada 1988, 1996 1 21.008 1989, 1996 1 22.531**
(0.332) (0.017)
South Korea 1995, 2012 1 25.410* 1996, 2011 1 23.366*
(0.000) (0.008)
Germany 1994, 2007 1 22.217** 1992, 2008 1 21.802***
(0.036) (0.082)
France 2001, 2009 1 2.655** 2001, 2006 1 4.034*
(0.012) (0.001)
a
(a) corresponds to the model allowing for break in intercept. (b) refers to the model allowing for
break in trend. Tb is the break date endogenously selected. tmin is the minimum t-statistic. k denotes the
lag length selected according to the AIC criterion. 5% critical values are provided in parentheses.
***p , 0.01, **p , 0.05, *p , 0.1.
innovational outlier).80 For both ZA and CMR procedures, notice that the lag
length selection has been chosen based on the information given by the AIC. For
level series, the null hypothesis of non-stationarity fails to be rejected for all coun-
tries except Russia, regardless of the inclusion of a break in the intercept or the
trend. This contradicts with the CMR output, which shows evidence of stationary
properties for all countries but with heterogeneous optimal break points. Interest-
ingly, these findings are constant over CMR specifications (additive outlier and
innovational outlier) and in line with CMR’s results on first-difference series,
which unanimously reject the null hypothesis of a unit root for all countries. Also,
it is worth noticing that ZA- and CMR-derived optimal break points are sensitive
to the model’s specification of breaks in the intercept/trend and/or the use of an
additive/innovational outlier. All in all, ZA results also do provide ample evidence
of stationary properties when using first difference-transformed series. Therefore,
one can conclude that, when we allow for the presence of structural breaks in the
electricity series, they remain integrated of order 1 (i.e., I(1)).
Panel Stationary Properties Assessment: In the previous subsection the time-
series unit root evidence was provided, but only at the country level. Therefore, we
go beyond country-specific heterogeneities, and hypothesize that there may exist a
general trend shared by panel members at the sample level. However, when pooling
series all together within a panel framework, a standard but constraining assumption
of those models is that the error terms are independent across cross-sections. In
table 7 the empirical results show that the null hypothesis of cross-sectional indepen-
dence is rejected at the 1-percent significance level by all cross-section dependence
testing procedures. Thus, our panel of electricity consumption series is likely to suf-
fer from this inference.
Besides the cross-country correlations of electricity consumption series, we
expect non-homogeneous slope coefficients due to structural economic differences
Table 7
RESULTS OF PANEL CROSS-SECTION DEPENDENCE TESTSa
Test t-Statistics P-Value
1. Pesaran’s CD test (2004) 29.34*** 0.000
2. Frees’ test (2004) 0.944*** 0.000
3. Friedman’s test 110.996*** 0.000
4. Pesaran et al. (2008) test 276.15*** 0.000
a
(1) Pesaran (2004) CD test for cross-section dependence in panel time-series data; (2) Frees (2004)
for cross-sectional dependence by using Frees’ Q distribution (T-asymptotically distributed); (3)
Friedman (1937) test for cross-sectional dependence by using Friedman’s x2 distributed statistic; and
(4) Pesaran et al. (2008) bias-adjusted LM. Tests include the intercept. ***p,0.01, **p,0.05,
*p,0.10.
Source: Author’s elaboration.
among countries. It has been shown that avoiding identifying and controlling for
this inference may generate misleading estimations.81 This hypothesis is checked
using two tests: the Pesaran and Yamagata (2008) test of slope homogeneity, fol-
lowed by the Blomquist and Westerlund (2013) test for slope heterogeneity. The
HAC robust standard errors for slope heterogeneity Kernel are bartlett with average
bandwidth of 3. Looking at the p-values associated with standardized and bias-
adjusted variances in table 8, the null hypothesis of homogenous slope coefficients
can be rejected at the 5-percent significance level. Thus, we have evidence of
heterogeneity.
For balanced panels, under cross-sectional dependence and heterogeneity, the
Breitung panel unit root test can be applied.82 Then, we juxtapose the results of the
Hansen CADF test panel unit root test, which is more adapted to unbalanced panel
structures.83 Moreover, we perform the Pesaran84 Cross-sectionally Augmented
Im, Pesaran and Shin (CIPS) panel unit root test and compare our results with out-
puts from standard panel unit root procedures, which neither control for heteroge-
neity nor for cross-sectional dependence (the Im-Pesaran-Shin (IPS) and the
Levin-Lin-Chu (LLC) panel unit root tests). 85Since the Breitung86 test is sensitive
to the lag length selection, we consider no lag and one lag in the trend-inclusive
equation. Additionally, because the CADF test is also sensitive to the number of
lags selected, we consider an equation with trend and constant for data in level and
an equation with constant only (no trend) for first difference transformed data, aug-
mented by one and two lags, respectively. According to the Breitung test results
displayed in table 9, electricity consumption series from the world’s top ten largest
power producers, when transformed in first differences, reject the null hypothesis
of panel non-stationarity at the 1-percent level; regardless of the use of 1 or 2 lags.
However, this test detects the presence of a unit root when data are used in levels.
Similarly, the Pesaran CIPS and Pesaran CADF panel unit root tests reject the null
of non-stationarity for first difference transformed series and one lag, whereas they
fail to do so for level series. Interestingly, both the IPS and LLC outcomes failed
Table 8
RESULTS OF PANEL SLOPE HOMOGENEITY TESTSa
Pesaran and Yamagata Blomquist and Westerlund
Variance (2008) Test (2013) Test
^
Standardized variance D 3.159*** 2.085**
(0.002) (0.037)
Biased-adjusted variance 3.304*** 2.184**
~ HAC
D adj / D (0.001) (0.029)
a
The statistic of the test is displayed. The p-value are reported in brackets. ***p,0.01, **p,0.05,
*p,0.10.
Table 9
RESULTS OF PANEL UNIT ROOT TESTSa
Panel Unit Root Test Level D Operator
1st Generation Procedures – Standard
1. Levin et al. (2002) LLC test 25.6644*** (0.0000) 22.9935*** (0.0014)
2. Im et al. (2003) IPS test 24.0753*** (0.0000) 27.7236*** (0.0000)
2nd Generation Procedures – Cross-Sectionally Augmented
3. (a) Breitung (2001) test – (0) 4.6268 (1.0000) 24.4738*** (0.0000)
3. (b) Breitung (2001) test – (1) 4.3648 (1.0000) 21.9274** (0.0270)
4. (a) Pesaran (2003) CADF test – (1) 21.341 (1.0000) 23.145*** (0.001)
4. (b) Pesaran (2003) CADF test – (2) 1.700 (1.0000) 1.700 (1.0000)
5. Pesaran (2007) CIPS test 1.2334 (1.0000) 25.5547*** (0.0000)
a
D refers to the first-difference operator. The statistics of the tests are displayed. (1) Levin-Lin-Chu
(LLC, Levin et al., 2002) panel unit root test. (2) Im-Pesaran-Shin (IPS, Im et al., 2003) panel unit root
test. (4 – a,b) Pesaran (2003) CADF panel unit root based on the mean of individual DF (or ADF)
t-statistics of each unit in the sample - Z-t-bar or t-bar statistics are reported - deterministic chosen:
constant and trend (lag 1); constant only (no trend – lag 2). (5) Pesaran (2007) Cross-sectionally
augmented Im, Pesaran and Shin (CIPS) panel unit root test. (x) refers to the lag length selected based
on the information given by the Final Prediction Error (FPE), the Akaike’s Information Criterion (AIC),
the Schwarz’s Bayesian Information Criterion (SBIC) and the Hannan and Quinn Information Criterion
(HQIC). The p-values are reported in brackets. ***p,0.01, **p,0.05, *p,0.10.
to adequately align with the Breitung and Pesaran CADF results, which indicates
that panel stationary findings are sensitive to the presence and control of cross-
sectional dependence across units. Overall, our cross-sectionally dependent series
can be said to be integrated of order 1 (i.e., Ið1Þ); thus, their first-difference trans-
formation present stationary properties, whereas level data fail to do so.
K-means Clustering and b-Convergence Analysis: X. Sala-i-Martin and R.
Barro87 emphasized that a b-convergence dynamics may arise if the partial correla-
tion between growth in a variable over time and its initial level takes a negative
sign. Thus, it is of great interest to assess whether we observe a global converging
or diverging trend among electricity consumption series in our sample.
To do so, we implement a clustering method related to partition procedures,
which break the observations into a distinct number of nonoverlapping groups. One
extensively employed partition clustering method corresponds to k-means cluster
analysis in which each observation is assigned to a clustered group whose arithmetic
average is the closest. Based on this classification, new group means are identified
and created until the group’s structure becomes independent to the treatment of
additional observations. Naturally, an identical procedure is applied in k-medians
cluster methods with the difference being that medians are computed instead of
means. Empirical findings identify two distinct clusters of countries: China, the
Table 10
RESULTS OF b-CONVERGENCE ANALYSISa
Measure Overall Sample Cluster 1 Cluster 2
b-Convergence 0.9683795 0.9822005 0.9454811
a
Cluster 1: 1st to 5th global ranked electricity consumers. Cluster 2: 6th to 10th global ranked
electricity consumers.
United States, India, Russia, and Japan are placed in the first cluster being the larg-
est electricity consumers, while Brazil, Canada, South Korea, Germany, and France
are placed in the second group. Interestingly, the k-means clustering outcome ade-
quately reflects electricity consumption structures as the first cluster gathers the
world’s top five electricity consumers, whereas the second group encompasses 6th
to 10th top power producers. As shown in table 10, the b-convergence measure for
the sample as a whole is 0.9683795, which fails to support the existence of con-
verging dynamics in terms of electricity consumption within the panel. Also, the
b-convergence estimate for the first cluster is 0.9822005, while being 0.9454811
for the second. This indicates general diverging forces, although the second cluster
seems to display a relatively lower diverging trend compared to the first one.
Concluding Remarks, Policy Implications, and Prospects for Future Research
Concluding Remarks: This paper explores the integration properties of electric-

ity consumption series for the world’s top ten largest power generators: China, the
United States, India, Russia, Japan, Brazil, Canada, South Korea, Germany, and
France. After an extensive survey of the energy convergence literature, we offer a
methodological note and highlight the challenges and empirical gaps that previous
studies have attempted to address thus far. Then, a stepwise two-stage (country-level
and panel) integration property testing framework is combined with data spanning
the largest and most recent period (1985–2020). It includes a set of univariate unit
root tests (Augmented Dickey and Fuller, Phillips and Perron, Dickey and Fuller-
Generalized Least Squares, and Kwiatkowski–Phillips–Schmidt–Shin tests), station-
ary procedures allowing for endogenously determined structural breaks in both the
intercept and the time-trends (the Clemente, Monta~nes, and Reyes and the Zivot-
Andrews tests), as well as panel unit root tests robust to heterogeneous slope coeffi-
cients and cross-sectional dependence inferences among the series (Breitung panel
unit root test robust to cross-correlations and Pesaran Cross-sectionally Augmented
Im, Pesaran and Shin (CIPS) panel unit root test), combined with first generation
panel unit root tests for comparison purpose. While panel integration properties
derived at the sample level highlight mixed results sensitive to the choice of the
deterministic component (constant versus constant and trend), cross-country results

differ whereas country-specific stationarity features remain stable across tests. In
general, we observed that pooled panel series and country-level series are integrated
of order 1, which indicates that those series are stationary in first difference but
not in level. Overall, results derived from k-means clustering techniques and
b-convergence analysis show that diverging dynamics dominate over converging
ones, at both the panel and sub-group levels, although the second cluster seems to
display a relatively lower diverging trend compared to the first one.
Policy Implications: Based upon the aforementioned outcome, we can provide
policy suggestions to enhance a more secure and less volatile supply of electricity.
Since electricity consumption series are found to be stationary when first-
difference transformed, then any shock affecting a country’s power trends (i.e.,
emergence of a backstop technology reducing the marginal cost of a competitive
resource, carbon tax set on producers, climate agreements endorsed internationally
with limited carbon leakages) is likely to be temporary and electricity consumption
is expected to shortly return to its normal and long-run equilibrium pace.88 How-
ever, our results do not exclude that a shock occurrence on power use (i.e., energy
price shock) could absolutely exclude permanent effects. Indeed, while distortions
affecting the electricity supply (in particular when derived from low-carbon energy
resources and input price volatilities due to seasonal fluctuations and barriers to
technology diffusion) are less likely to be transmitted back to other macroeconomic
variables, spillover mechanisms may remain.89 Therefore, it is of the utmost
importance to strengthen the deployment of low-carbon electricity generation
through massive facilities and technological investments and, thus, reinforce the
electricity supply’s resilience to exogenous shocks with potential temporary effects
on the economy. The reason is that, unlike natural gas-fired plants, the production
of power-based renewables (especially hydropower, solar, and wind, with the
exception being geothermal energy as shown in U. Shahzad et al.90) is particularly
volatile across years and seasons and remains highly sensitive to meteorological
factors.91 This limits the energy security advantages of some renewables, which
explains why the temporary introduction of a “bridging fossil fuel” (often natural
gas) has been defended in the recent literature as a “smoothing” tool ensuring a
secure deployment of low-carbon energies (C. Magazzino and N. Schneider).92 On
the one hand, gas-fired plants are more efficient, have easier storage issues, and
provide a better operational flexibility and safety for power production purpose,93
but lower time and capital cost requirements compared with renewable energy
facilities, which eases investment decisions for many firms.94 Displaying the low-
est carbon content compared to other non-renewable resources, natural gas may
operate as a non-negligible temporary input to complement more volatile decar-
bonized sources95 and help them reach a secure share in the electricity supply mix.
Prospects for Future Research: Along with policy suggestions, some pros-
pects for future research can be offered. While extensively studied on the time
domain, the use of tools finding applications in the frequency domain remains rela-
tively overlooked. Recently, Y. Cai and A. Menegaki96 offered the first empirical
application of the innovative Fourier augmented Quantile Unit Root (FQUR) test
with Fourier type deterministic trends, as theoretically proposed in Bahmani-
Oskooee et al. This latter procedure, based on the frequency domain, displays
unexploited competitive edges, including approximating smooth breaks, capturing
heterogeneous stochastic convergences that may differ across quantiles, estimating
a unique mean-reverting speed for each quantile of electricity consumption, and
thus providing integration evidence over the whole quantiles, but also at the quan-
tile level. Therefore, a potentially fruitful direction exists in developing economet-
ric tools that can overcome the conditional central tendency limits of conventional
unit root tests. Undoubtedly, promoting empirical analysis over the frequency
domain analysis may maximize the amount of information extracted from each
time series with potential inclusive knowledge and valuable policy insights gained
on this topic.
NOTES
1
Intergovernmental Panel on Climate Change (IPCC), IPCC Sixth Assessment Report (Geneva:
IPCC, 2022), available at https://www.ipcc.ch/report/ar6/wg2/.
2
International Energy Agency (IEA), World Energy Outlook 2009 (Paris: IEA, 2009).
3
K. Menyah and Y. Wolde-Rufael, “CO2 Emissions, Nuclear Energy, Renewable Energy and
Economic Growth in the US,” Energy Policy, vol. 38, no. 6 (2010), pp. 2911–915, and A. G.
Olabi and M.A. Abdelkareem, “Renewable Energy and Climate Change,” Renewable and Sustain-
able Energy Reviews, vol. 158 (2022).
4
N. Apergis, J. E. Payne, K. Menyah, and Y. Wolde-Rufael, “On the Causal Dynamics
between Emissions, Nuclear Energy, Renewable Energy, and Economic Growth,” Ecological Eco-
nomics, vol. 69, no. 11 (2010a), pp. 2255–260, and W. Shen, X. Chen, J. Qiu, J. A. Hayward, S.
Sayeef, P. Osman and Z. Dong, “A Comprehensive Review of Variable Renewable Energy Level-
ized Cost of Electricity,” Renewable and Sustainable Energy Reviews, vol. 133, issue C (2020).
5
Y. Cai and A. N. Menegaki, “Fourier Quantile Unit Root Test for the Integrational Properties
of Clean Energy Consumption in Emerging Economies,” Energy Economics, vol. 78, issue C
(2019), pp. 324–34.
6
M. H. Pesaran, Time Series and Panel Data Econometrics (Oxford: Oxford University Press,
2015).
N. Apergis and C. Tsoumas, “Integration Properties of Disaggregated Solar, Geothermal and

7
Biomass Energy Consumption in the US,” Energy Policy, vol. 39, no. 9 (2011), pp. 5474–479.
8
P. K. Narayan and R. Smyth, “Are Shocks to Energy Consumption Permanent or Temporary?
Evidence from 182 Countries,” Energy Policy, vol. 35, no. 1 (2007), pp. 333–41.
P. F. Chen and C. C. Lee, “Is Energy Consumption Per Capita Broken Stationary?
9
New Evidence from Regional-based Panels,” Energy Policy, vol. 35 (2007), pp. 3526–540; Y.-C.
Hsu, C.-C. Lee, and C.-C. Lee, “Revisited: Are Shocks to Energy Consumption Permanent or Tem-
porary? New Evidence from a Panel SURADF Approach,” Energy Economics, vol. 30, no. 5 (2008),
pp. 2314–330; V. Mishra, S. Sharma, and R. Smyth, “Are Fluctuations in Energy Consumption Per
Capita Transitory? Evidence from a Panel of Pacific Island Countries,” Energy Policy, vol. 37, no. 6
(2009), pp. 2318–326; A. Aslan and H. Kum, “The Stationary of Energy Consumption for Turkish
Disaggregate Data by Employing Linear and Nonlinear Unit Root Tests,” Energy, vol. 36, no. 7
(2011), pp. 4256–258; H. Kum, “Are Fluctuations in Energy Consumption Transitory or Permanent?
Evidence from a Panel of East Asia & Pacific Countries,” International Journal of Energy Economics
and Policy, vol. 2, no. 3 (2012), pp. 92–6; and B. Ozcan, “Are Shocks to Energy Consumption
Permanent or Temporary? The Case of 17 Middle East Countries,” Energy Exploration and
Exploitation, vol. 31, no. 4 (2013), pp. 589–605.
10
C. P. Barros, L. Gil-Alana, and J. E. Payne, “Evidence of Long Memory Behavior in U.S.
Renewable Energy Consumption,” Energy Policy, vol. 41, issue C (2012), pp. 822–26; C.-C. Lee,
O. Ranjbar, and C.-C. Lee, “Testing the Persistence of Shocks on Renewable Energy Consump-
tion: Evidence from a Quantile Unit-Root Test with Smooth Breaks,” Energy, vol. 215 (2021); G.
Gozgor, “Are Shocks to Renewable Energy Consumption Permanent or Transitory? An Empirical
Investigation for Brazil, China, and India,” Renewable and Sustainable Energy Reviews, vol. 66
(2016), pp. 913–19; E. Demir and G. Gozgor, “Are Shocks to Renewable Energy Consumption
Permanent or Temporary? Evidence from 54 Developing and Developed Countries,” Environmen-
glu, “Stationarity Test
tal Science and Pollution Research, vol. 4 (2018), pp. 3785–792; E. Fendo
of Renewable Energy Consumption with Fractional Frequency Fourier Unit Root Test: Evidence
from BRICS-T Countries,” Alphanumeric Journal, vol. 9, no. 1 (2021), 99–110; and C. S. Saba
and N. Ngepah, “Convergence in Renewable Energy Consumption and Their Influencing Factors
across Regions: Evidence from Convergence Algorithm Approach,” Environmental Science and
Pollution Research (2022), pp. 1–34.
11
C. Magazzino, “Stationarity of Electricity Series in MENA Countries,” The Electricity Jour-
nal, vol. 30 (2017), pp. 16–22; M. A. Destek and S. A. Sarkodie, “Are Fluctuations in Coal, Oil
and Natural Gas Consumption Permanent or Transitory? Evidence from OECD Countries,” Heli-
yon, vol. 6, no. 2 (2020); and S. Erdogan, G. Akalin, and O. Oypan, “Are Shocks to Disaggregated
Energy Consumption Transitory or Permanent in Turkey? New Evidence from Fourier Panel
KPSS Test,” Energy, vol. 197, no. 15 (2020).
R. E. De Hoyos and V. Sarafidis, “Testing for Cross-Sectional Dependence in Panel-Data

12
Models,” The Stata Journal, vol. 6, no. 4 (2006), pp. 482–96.

13
M. Eberhardt, “Estimating Panel Time-Series Models with Heterogeneous Slopes,” The Stata
Journal, vol. 12, no. 1 (2012), pp. 61–71.
D. Weinhold, “A Dynamic Fixed Effects Model for Heterogeneous Panel Data,” London,
14
London School of Economics, Mimeo, 1999; D. W. Andrews, “Cross-Section Regression with

Common Shocks,” Econometrica, vol. 73, no. 5 (2005), pp. 1551–585; and M. Eberhardt and S.
Bond, “Cross-Section Dependence in Nonstationary Panel Models: A Novel Estimator,” MPRA
Paper no. 17692, University Library of Munich, Munich, Germany, 2009.
15
F. Kula, “Is Per Capita Electricity Consumption Non-Stationary? A Long-Span Study for
Turkey,” Energy Sources, Part B: Economics, Planning, and Policy, vol. 9, no. 2 (2014), pp.
161–64; D. Borozan and L. Borozan, “The Stationarity of Per Capita Electricity Consumption in
Croatia Allowing for Structural Break(s),” in Proceedings of the 13th International Symposium on
Operational Research, SOR (2015), pp. 337–42; M. J. Herrerias, C. Aller, and J. Ord o~
nez,
“Residential Energy Consumption: A Convergence Analysis across Chinese Regions,” Energy

Economics, vol. 62, issue C (2017), pp. 371–81; and Y. Wang, R. Shi, C. Zhang, Y. He, H. Jiang,
and J. Kubota, “Structural Changes and Trends in China’s Renewable Electricity Production in the
Policy Evolution Process,” Renewable Energy, vol. 182, issue C (2022), pp. 879–86.
16
L. J. Nilsson, “Energy Intensity Trends in 31 Industrial and Developing Countries
1950–1988,” Energy, vol. 18, no. 4 (1993), pp. 309–22, and J. Goldemberg, “A Note on the
Energy Intensity of Developing Countries,” Energy Policy, vol. 24, no. 8 (1996), pp. 759–61.
17
This corresponds to analyses relying on beta convergence (catch up process from high to low
convergence), gamma convergence (intra generation mobility), or sigma convergence (spread of
distribution) estimations.
18
This refers to the use of Theil Decomposition Analysis (TDA) or Fisher Index Decomposi-
tion (FID) Analysis, which aim at conducting within and between-group decomposition or estima-
tion of the responsiveness of energy transformation index to energy changes.
19
See, for instance, L. Dagher and T. Yacoubian, “The Causal Relationship between Energy
Consumption and Economic Growth in Lebanon,” Energy Policy, vol. 50, issue C (2012), pp.
795–801; M. Shahbaz, H. H. Lean, and A. Farooq, “Natural Gas Consumption and Economic
Growth in Pakistan,” Renewable and Sustainable Energy Reviews, vol. 18, issue C (2013), pp.
87–94; and M. Shakeel, “Analyses of Energy-GDP-Export Nexus: The Way-Forward,” Energy,
vol. 216, issue C (2021). Also, an exhaustive state-of-the-art review of this topic can be found by
Nilsson, Magazzino, and Schneider (2020).
20
C. W. J. Granger, “Investigating Causal Relations by Econometric Models and Cross-
Spectral Methods,” Econometrica, vol. 37, no. 3 (1969), pp. 424–38, and “Some Recent Develop-
ments in a Concept of Causality,” Journal of Econometrics, vol. 39, nos. 1–2 (1988),
pp. 199–211.
21
S. Johansen and K. Juselius, “Maximum Likelihood Estimation and Inference on Cointegra-
tion with Applications to the Demand for Money,” Oxford Bulletin of Statistics, vol. 52, no. 2
(1990), pp. 169–210.
22
A. W. Gregory and B. E. Hansen, “Practitioners Corner: Tests for Cointegration in Models
with Regime and Trend Shifts,” Oxford Bulletin of Economics and Statistics, vol. 58, no. 3
(1996), pp. 555–60.
23
C. Kao, “Spurious Regression and Residual-based Tests for Cointegration in Panel Data,”
Journal of Econometrics, vol. 90, no. 1 (1999), pp. 1–40, and N. Beck, “Time-Series-Cross-Sec-
tion Methods,” in Oxford Handbook of Political Methodology, eds. J. M. Box-Steffensmeier, H.
E. Brady, and D. Collier (Oxford: Oxford University Press, 2008), pp. 475–93.
24
P. F. Chen and C. C. Lee, op. cit.; Y.-C. Hsu et al., op. cit.; V. Mishra et al., op. cit.; A.
Aslan and H. Kum, op. cit.; I. Ozturk and A. Aslan, “Are Fluctuations in Energy Consumption
Per Capita Transitory? Evidence from Turkey,” Energy Exploration and Exploitation, vol. 29, no.
2 (2011), pp. 161–67; M. Hasanov and E. Telatar, “A Re-Examination of Stationarity of Energy
Consumption: Evidence from New Unit Root Tests,” Energy Policy, vol. 39, no. 12 (2011), pp.
7726–738; H. Kum, op. cit.; B. Ozcan, op. cit.; V. Yilanci and Ç. B. Tunali, “Are Fluctuations in
Energy Consumption Transitory or Permanent? Evidence from a Fourier LM Unit Root Test,”
Renewable and Sustainable Energy Reviews, vol. 36, issue C (2014), pp. 20–5; V. Mishra and R.
Smyth, “Conditional Convergence in Australia’s Energy Consumption at the Sector Level,”
Energy Economics, vol. 62, issue C (2017), pp. 396–403; S. Erdogan et al., op. cit.; O. E. Omoju,
J. Li, J. Zhang, A. Rauf, and V. Sosoo, “Implications of Shocks in Energy Consumption for
Energy Policy in Sub-Saharan Africa,” Energy & Environment, vol. 31, no. 6 (2019), pp.
1077–097; and W. He and H. Chen, “Will China’s Provincial Per Capita Energy Consumption
Converge to a Common Level over 1990–2017? Evidence from a Club Convergence Approach,”
Energy, vol. 249, issue C (2022).
25
H. H. Lean and R. Smyth, “Long Memory in US Disaggregated Petroleum Consumption:
Evidence from Univariate and Multivariate LM Tests for Fractional Integration,” Energy Policy,
vol. 37, no. 8 (2009), pp. 3205–211, and M. A. Destek and S. A. Sarkodie, op. cit.
26
P. K. Narayan, S. Narayan, and R. Smyth, “Are Oil Shocks Permanent or Temporary? Panel
Data Evidence from Crude Oil and NGL Production in 60 Countries,” Energy Economics, vol. 30,
no. 3 (2008), pp. 919–36.
27
H. H. Lean and R. Smyth, op. cit.; N. Apergis and J. E. Payne, “Structural Breaks and Petro-
leum Consumption in US States: Are Shocks Transitory or Permanent?” Energy Policy, vol. 38,
no. 10 (2010), pp. 6375–378; S. A. Ajlouni, A. M. Ghazo, and Z. M. Abu-Lila, “Are Shocks to
Petroleum Products Consumption Permanent or Temporary: A New Evidence from LM Unit Root
Tests with Structural Breaks,” Emerging Science Journal, vol. 5, no. 3 (2021), pp. 322–29; and F.
Kızılkaya, “Investigating the Stationarity Properties of Oil Consumption: An Empirical Analysis
for OECD Countries,” Journal of Economic Policy Researchers, vol. 9, no. 1 (2022), pp. 121–37.
28
N. Apergis, D. Loomis, and J. E. Payne, “Are Fluctuations in Coal Consumption Transitory
or Permanent? Evidence from a Panel of US States,” Applied Energy, vol. 87, no. 7 (2010b),
pp. 2424–426.
29
N. Apergis et al., op. cit.; A. Aslan, “Does Natural Gas Consumption Follow a Nonlinear
Path over Time? Evidence from 50 US States,” Renewable and Sustainable Energy Reviews, vol.
15, no. 9 (2011), pp. 4466–469; and Y. Cai and C. Magazzino, “Are Shocks to Natural Gas Con-
sumption Transitory or Permanent? A More Powerful Panel Unit Root Test on the G7 Countries,”
Natural Resources Forum, vol. 43, no. 2 (2019), pp. 111–20.
30
P. K. Narayan and R. Smyth, op. cit.; L. A. Gil-Alana, D. Loomis, and J. E. Payne, “Does
Energy Consumption by the US Electric Power Sector Exhibit Long Memory Behavior?” Energy
Policy, vol. 38, no. 11 (2010), 7512–518; N. Apergis and C. Tsoumas, “Long Memory and Disag-
gregated Energy Consumption: Evidence from Fossils, Coal and Electricity Retail in the US,”
Energy Economics, vol. 34, no. 4 (2012), pp. 1082–087; F. Kula, A. Aslan, and I. Ozturk, “Is Per
Capita Electricity Consumption Stationary? Time Series Evidence from OECD Countries,” Renew-
able and Sustainable Energy Review, vol. 16, no. 1 (2012), pp. 501–03; F. Kula, op. cit.; D. Boro-
zan and L. Borozan, op. cit.; M. J. Herrerias et al., op. cit.; and C. Magazzino, op. cit.
31
Here, we refer to the definition of “alternative and renewable energy” offered by the World Bank
(2020) comprising solar, wind, biomass, geothermal, biofuels, hydropower, and nuclear sources.
Y. Wang, L. Li, J. Kubota, X. Zhu, and G. Lu, “Are Fluctuations in Japan’s Consumption of
32
Non-Fossil Energy Permanent or Transitory?” Applied Energy, vol. 169 (2016), pp. 187–96, and
Y. Cai and A. N. Menegaki, op. cit.
C. P. Barros, et al., “Evidence of Long Memory Behavior in U.S. Renewable Energy Con-
33
sumption”; C.-C. Lee et al., op. cit.; G. Gozgor, op. cit.; E. Demir and G. Gozgor, op. cit.; and E.
Fendoglu, op. cit.
34
C. P. Barros, L. Gil-Alana, and J. E. Payne, “Evidence of Long Memory Behavior in U.S.
Nuclear Electricity Net Generation,” Energy Systems, vol. 4, no. 1 (2013), pp. 99–107, and H.
Zhu and P. Guo, “Are Shocks to Nuclear Energy Consumption Per Capita Permanent or Tempo-
rary? A Global Perspective,” Progress in Nuclear Energy, vol. 88 (2016), pp. 156–64.
35
N. Apergis and C. Tsoumas, “Integration Properties of Disaggregated Solar, Geothermal and
Biomass Energy Consumption in the US,” and M. Aydin and U. K. Pata, “Are Shocks to Disag-
gregated Renewable Energy Consumption Permanent or Temporary for the USA? Wavelet Based
Unit Root Test with Smooth Structural Shifts,” Energy, 207 (2020).
36
C. S. Saba and N. Ngepah, op. cit.
37
P. F. Chen and C. C. Lee, op. cit.; Y.-C. Hsu et al., op. cit.; V. Mishra et al., op. cit.; A.
Aslan and H. Kum, op. cit.; H. Kum, op. cit.; and B. Ozcan, op. cit.
38
C. P. Barros, et al., “Evidence of Long Memory Behavior in U.S. Renewable Energy Con-
sumption”; C.-C. Lee et al., op. cit.; G. Gozgor, op. cit.; E. Demir and G. Gozgor, op. cit.; E. Fen-
doglu, op. cit.; and C. S. Saba and N. Ngepah, op. cit.
39
C. Magazzino, op. cit.; M. A. Destek and S. A. Sarkodie, op. cit.; and S. Erdogan et al., op. cit.
40
M. Eberhardt, op. cit.
41
B. Liddle, “What Are the Carbon Emissions Elasticities for Income and Population? Bridging
STIRPAT and EKC via Robust Heterogeneous Panel Estimates,” Global Environmental Change,
vol. 31 (2015), pp. 62–73.
42
R. E. De Hoyos and V. Sarafidis, op. cit.
43
D. Weinhold, op. cit.; D. W. Andrews, op. cit.; and M. Eberhardt and S. Bond, op. cit.
44
F. Kula et al., op cit.; D. Borozan and L. Borozan, op. cit.; M. J. Herrerias et al., op. cit.; and
Y. Wang et al., “Structural Changes and Trends in China’s Renewable Electricity Production in
the Policy Evolution Process.”
45
C. Magazzino, M. Mele, and N. Schneider, “A D2C Algorithm on the Natural Gas Consump-
tion and Economic Growth: Challenges Faced by Germany and Japan,” Energy, vol. 219 (2021);
C. Magazzino, M. Mele, and N. Schneider, “A Machine Learning Approach on the Relationship
among Solar and Wind Energy Production, Coal Consumption, GDP, and CO2 Emissions,”
Renewable Energy, vol. 167, issue C (2021b), pp. 99–115; and M. Murshed, “Modeling Primary
Energy and Electricity Demands in Bangladesh: An Autoregressive Distributed Lag Approach,”
Sustainable Production and Consumption, vol. 27 (2021), pp. 698–712.
46
Series are available at https://www.bp.com/en/global/corporate/energy-economics/statistical-
review-of-world-energy.html. British Petroleum (BP), Statistical Review of World Energy, 2022.
47
M. A. Destek and S. A. Sarkodie, op. cit.
48
D. Dickey and W. Fuller, “Distribution of the Estimators for Autoregressive Time Series
with A Unit Root,” Journal of the American Statistical Association, vol. 74, no. 366 (1979), pp.
427–31; P. C. B. Phillips and P. Perron, “Testing for Unit Root in Time Series Regression,” Bio-
metrika, vol. 75, no. 2 (1988), pp. 335–46; and G. Kapetanios, Y. Shin, and A. Snell, “Testing for
a Unit Root in the Nonlinear STAR Framework,” Journal of Econometrics, vol. 112, no. 2 (2003),
pp. 359–79.
49
J. Clemente, A. Monta~nes, and M. Reyes, “Testing for a Unit Root in Variables with a Dou-
ble Change in the Mean,” Economics Letters, vol. 59, no. 2 (1998), pp. 175–82, and E. Zivot and
D. Andrews, “Further Evidence of the Great Crash, the Oil Price Shock and Unit Root Hypoth-
esis,” Journal of Business and Economic Statistics, vol. 10, no. 3 (1992), pp. 251–70.
P. A. V. B. Swamy, “Efficient Inference in a Random Coefficient Regression Model,” Econ-

50
ometrica, vol. 38, no. 2 (1970), pp. 311–23.
M. H. Pesaran and T. Yamagata, “Testing Slope Homogeneity in Large Panels,” Journal of

51
Econometrics, vol. 142, no. 1 (2008), pp. 50–93.

52
J. Blomquist and J. Westerlund, “Testing Slope Homogeneity in Large Panels with Serial
Correlation,” Economics Letters, vol. 121, no. 3 (2013), pp. 374–78.
53
54
M. Eberhardt, op. cit.
55
For a discussion on the interpretation of individual effects in non-stationary panel data, see
H. R. Moon, B. Perron, and P. C. Phillips, “Incidental Trends and the Power of Panel Unit Root
Tests,” Journal of Econometrics, vol. 141, no. 2 (2007), pp. 416–59.
56
D. Weinhold, op. cit.; D. W. Andrews, op. cit.; and M. Eberhardt and S. Bond, op. cit.
57
G. S. Maddala and S. Wu, “A Comparative Study of Unit Root Tests with Panel Data and a
New Simple Test,” Oxford Bulletin of Economics and Statistics, no. 61 (1999), pp. 631–52; A.
Levin, C. Lin, and C. Chu, “Unit Root Test in Panel Data: Asymptotic and Finite Sample Proper-
ties,” Journal of Econometrics, vol. 108, no. 1 (2002), pp. 1–24; and K. Im, M. Pesaran, and Y.
Shin, “Testing for Unit Roots in Heterogeneous Panels,” Journal of Econometrics, vol. 115, no. 1
(2003), pp. 53–74.
58
L. V. Smith, S. Leybourne, T.-H. Kim, and P. Newbold, “More Powerful Panel Unit Root
Tests with an Application to the Mean Reversion in Real Exchange Rates,” Journal of Applied
Econometrics, vol. 19, no. 2 (2004), pp. 147–70; H. R. Moon and B. Perron, “Testing for a Unit
Root in Panels with Dynamic Factors,” Journal of Econometrics, vol. 122, no. 1 (2004), pp.
81–126; and M. H. Pesaran, “A Simple Panel Unit Root Test in the Presence of Cross-Section
Dependence,” Journal of Applied Economics, vol. 22, no. 2 (2007), pp. 265–312.
59
E. W. Frees, “Assessing Cross-Sectional Correlations in Panel Data,” Journal of Economet-
rics, vol. 64 (1995), pp. 393–414, and E. W. Frees, Longitudinal and Panel Data: Analysis in the
Social Sciences (Cambridge: Cambridge University Press, 2004).
60
M. Friedman, “The Use of Ranks to Avoid the Assumption of Normality Implicit in the
Analysis of Variance,” Journal of the American Statistical Association, vol. 32, no. 200 (1937),
pp. 675–701.
61
For a complete discussion on cross-sectional dependence tests in panel-data models and their
respective properties, see R. E. De Hoyos and V. Sarafidis, op. cit.
62
63
B. E. Hansen, “Rethinking the Univariate Approach to Unit Root Testing: Using Covariates
to Increase Power,” Econometric Theory, vol. 11, no. 5 (1995), pp. 1148–171.
64
D. O. Cushman and N. Michael, “Nonlinear Trends in Real Exchange Rates: A Panel Unit
Root Test Approach,” Journal of International Money and Finance, vol. 30, no. 8 (2011),
pp. 1619–637.
65
D. Dickey and W. Fuller, op. cit.
66
K. Im et al., op. cit.
67
G. S. Maddala and S. Wu, op. cit.
68
Ibid.
69
J. Breitung, “The Local Power of Some Unit Root Tests for Panel Data,” in Advances in
Econometrics, Volume 15: Nonstationary Panels, Panel Cointegration, and Dynamic Panels, ed.
B. H. Baltagi (Amsterdam: JAY Press, 2000), pp. 161–78.
70
J. Breitung and S. Das, “Panel Unit Root Tests under Cross-Sectional Dependence,” Statis-
tica Neerlandica, vol. 59, no. 4 (2005), pp. 414–33.
71
J. Breitung, op. cit.
72
G. S. Maddala and S. Wu, op. cit.
73
74
K. Im et al., op. cit.
75
A. Levin et al. op. cit.
76
D. Dickey and W. Fuller, op. cit.; P. C. B. Phillips and P. Perron, op. cit.; and G. Kapetanios
et al., op. cit.
77
J. Clemente et al., op. cit., and E. Zivot and D. Andrews, op. cit.
78
M. H. Pesaran and T. Yamagata, op. cit.; M. Friedman, op. cit.; E. W. Frees, “Assessing
Cross-Sectional Correlations in Panel Data”; E. W. Frees, Longitudinal and Panel Data: Analysis
in the Social Sciences; M. H. Pesaran, “General Diagnostic Tests for Cross Section Dependence in
Panels,” Cambridge Working Papers in Economics No. 0435, Faculty of Economics, University
of Cambridge, U.K., 2004; and M. H. Pesaran, A. Ullah, and T. Yamagata, “A Bias-Adjusted LM
Test of Error Cross-Section Independence,” The Econometrics Journal, vol. 11, no. 1 (2008),
pp. 105–27.
79
D. Dickey and W. Fuller, op. cit.; P. C. B. Phillips and P. Perron, op. cit.; and G. Kapetanios
et al., op. cit.
80
J. Clemente et al., op. cit., and E. Zivot and D. Andrews, op. cit.
81
82
Ibid.
83
B. E. Hansen, op. cit.
84
M. H. Pesaran, op. cit.
85
K. Im et al., op. cit., and A. Levin et al. op. cit.
86
87
X. X. Sala-i-Martin and R. J. Barro, “Technological Diffusion, Convergence, and Growth,”
Journal of Economic Growth, vol. 2, no. 1 (1997), pp. 1–26.
88
N. Apergis and C. Tsoumas, op. cit.
89
P. K. Narayan and R. Smyth, op. cit.
U. Shahzad, N. Schneider, and M. B. Jebli, “How Coal and Geothermal Energies Interact
90
with Industrial Development and Carbon Emissions? An Autoregressive Distributed Lags

Approach to the Philippines,” Resources Policy, vol. 74, issue C (2021).
N. Schneider and G. Vallet, “The Green Light or The Green Line? Challenges Facing the
91
Energy Choices for Switzerland,” SSRN Electronic Journal (2019).

92
C. Magazzino and N. Schneider, “The Causal Relationship between Primary Energy Con-
sumption and Economic Growth in Israel: A Multivariate Approach,” International Review of
Environmental and Resource Economics, vol. 14, no. 4 (2020), pp. 417–91.
M. A. Destek, “Natural Gas Consumption and Economic Growth: Panel Evidence from
93
OECD Countries,” Energy, vol. 114, issue C (2016) pp. 1007–015, and C. Magazzino et al., “A
D2C Algorithm on the Natural Gas Consumption and Economic Growth: Challenges Faced by
Germany and Japan.”
94
N. Zhao, Y. Zhang, B. Li, J. Hao, D. Chen, Y. Zhou, and R. Dong, “Natural Gas and Elec-
tricity: Two Perspective Technologies of Substituting Coal-Burning Stoves for Rural Heating and
Cooking in Hebei Province of China,” Energy Science & Engineering, vol. 7, no. 1 (2019),
pp. 120–31.
95
G. Fadiran, A. T. Adebusuyi, and D. Fadiran, “Natural Gas Consumption and Economic
Growth: Evidence from Selected Natural Gas Vehicle Markets in Europe,” Energy, vol. 169, issue
C (2019), pp. 467–77.
96
Y. Cai and A. N. Menegaki, op. cit.

The Journal of Energy and Development

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Journal of Energy and Development

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF ENERGY

Nicolas Schneider, Mihaela Simionescu,

“Searching for Long Equilibrium Behaviors

Nicolas Schneider, Mihaela Simionescu, AND Wadim Strielkowski*

G lobal warming has emerged as a major concern to be addressed because it

The Journal of Energy and Development, Vol. 47, Nos. 1-2

energy convergence literature highlights a number of critical points that we will

dependence, and examine the sensitivity of our sample-level stationary

Analyzing energy consumption convergence behaviors and stationary properties

(2010) consumption electricity

Kum (2012) 15 East Asia & Energy 1971–2007 LM H0 of non- No

Borozan (2015) consumption per

and MENA 5 Middle East and North Africa.

(2019) countries energy consumption rejected for China,

Data Collection and Econometric Framework

Data Collection: To implement our methodology, this paper collected data on

2000 2005 2010 2015 2020

One striking observation is that Germany seems to be recording a slightly declining

univariate country-level unit root tests, as well as country-level stationary tests

H0: yt 5 a 1 yt1 1 «t (2)

Panel Integration Properties—Heterogeneous Slope Coefﬁcients and Cross-

H0 : bi 5 b for some i, against

Later, J. Blomquist and J. Westerlund52 offered a more generalized framework

yi,t 5 ai 1 ji xi,t 1 vi,t (4)

Where i 5 1 . . . N, xi,t corresponds to a vector of regressors, and ji refers to the

Where G ^ 2HAC is a robust HAC estimator, X^ i,T corresponds to a trajectory matrix

assumption that panel units are cross-sectionally independent,57 second-generation

Accordingly, we employ the Pesaran (2007) Cross-sectionally Augmented Im,

Where the Cointegration-Augmented Dickey–Fuller (CADF) statistic for the coun-

Where the non-parametric test displays a chi-square distribution with 2N degrees

STEPWISE UNIT ROOT TESTING METHODOLOGY

Augmented Dickey Fuller (ADF) (Dickey and Fuller, 1979)

Kwiatkowski-Phillips-Schmidt-Shin (KPSS) (Kapetanios et al., 2003)

Stage 5: Kmeans clustering and β-convergence analysis

Source: Authors’ elaboration.

 Stage 4 - We investigate panel integration properties using panel unit root

Country-Level Time-Series Analysis: Results of the univariate Augmented

Concluding Remarks, Policy Implications, and Prospects for Future Research

Concluding Remarks: This paper explores the integration properties of electric-

deterministic component (constant versus constant and trend), cross-country results

N. Apergis and C. Tsoumas, “Integration Properties of Disaggregated Solar, Geothermal and

R. E. De Hoyos and V. Sarafidis, “Testing for Cross-Sectional Dependence in Panel-Data

Models,” The Stata Journal, vol. 6, no. 4 (2006), pp. 482–96.

London School of Economics, Mimeo, 1999; D. W. Andrews, “Cross-Section Regression with

“Residential Energy Consumption: A Convergence Analysis across Chinese Regions,” Energy

P. A. V. B. Swamy, “Efficient Inference in a Random Coefficient Regression Model,” Econ-

ometrica, vol. 38, no. 2 (1970), pp. 311–23.

M. H. Pesaran and T. Yamagata, “Testing Slope Homogeneity in Large Panels,” Journal of

Econometrics, vol. 142, no. 1 (2008), pp. 50–93.

with Industrial Development and Carbon Emissions? An Autoregressive Distributed Lags

Energy Choices for Switzerland,” SSRN Electronic Journal (2019).

You might also like

Stage 4 - We investigate panel integration properties using panel unit root