The impact of domestic R&D and North–South R&D spillovers on energy intensity in developing countries

He ze , Die k
A icle — Published Ve sion
The impac o domes ic R&D and No h–Sou h R&D
spillo e s on ene gy in ensi y in de eloping coun ies
Economic Change and Res uc u ing
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: He ze , Die k (2024) : The impac o domes ic R&D and No h–Sou h R&D
spillo e s on ene gy in ensi y in de eloping coun ies, Economic Change and Res uc u ing, ISSN
1574-0277, Sp inge US, New Yo k, NY, Vol. 57, Iss. 2,
h ps://doi.o g/10.1007/s10644-024-09591-3
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/315252
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h p://c ea i ecommons.o g/licenses/by/4.0/
Vol.:(0123456789)
Economic Change and Res uc u ing (2024) 57:33
h ps://doi.o g/10.1007/s10644-024-09591-3
1 3
The impac o domes ic R&D andNo h–Sou h R&D
spillo e s onene gy in ensi y inde eloping coun ies
Die kHe ze 1
Recei ed: 29 June 2023 / Accep ed: 1 Decembe 2023 / Published online: 14 Feb ua y 2024
© The Au ho (s) 2024
Abs ac
This s udy u ilizes panel da a be ween 1995 and 2015 o a c oss sec ion o 33
de eloping (low- and middle-income) coun ies o in es iga e he impac on domes-
ic ene gy in ensi y bo h o domes ic R&D and o possible spillo e s om o eign
R&D conduc ed in de eloped (high-income) coun ies. Mo e speci ically, i exam-
ines R&D spillo e s om de eloped coun ies (No h) o domes ic ene gy in ensi y
in de eloping coun ies (Sou h) h ough disembodied channels, o al goods impo s,
and impo s o machine y and equipmen . Ou main indings, based on panel coin-
eg a ion echniques, a e as ollows: Fi s , domes ic R&D in he long un does no
con ibu e o educ ions in ene gy in ensi y in de eloping coun ies; second, he e
is no e idence o sugges ha disembodied No h–Sou h R&D spillo e s a ec he
long- un le el o domes ic ene gy in ensi y; hi d, he e a e ne e heless signi ican
spillo e s om R&D conduc ed in indus ial coun ies ha educe ene gy in ensi y
in de eloping coun ies; and ou h, while many impo ed goods a e no a channel
o No h–Sou h R&D spillo e s, such spillo e s a e ansmi ed h ough impo s o
machine y and equipmen .
Keywo ds Ene gy in ensi y· Domes ic R&D· No h–Sou h R&D spillo e s·
De eloping coun ies· Panel coin eg a ion me hods
JEL Classi ica ion Q43· Q55· F18
1 In oduc ion
Se e al s udies ind posi i e e ec s o domes ic esea ch and de elopmen (R&D)
pe o med in de eloping coun ies and o eign R&D pe o med in indus ial coun-
ies on o al ac o p oduc i i y (TFP) in de eloping coun ies (see, e.g., Coe e al.
* Die k He ze
[email p o ec ed]
1 Depa men o Economics, Helmu -Schmid -Uni e si y Hambu g, Hols enho weg 85,
22043Hambu g, Ge many
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 2 o 31
1997; Madsen e al. 2010; He ze 2022a, b). An inc ease in TFP implies ha a gi en
ou pu can be p oduced wi h ewe s anda d ac o s o p oduc ion, such as labo and
physical capi al, as well as human capi al, o ha mo e ou pu can be p oduced wi h
he same quan i y o ac o s o p oduc ion. Thus, an inc ease in TFP can be in e -
p e ed as ac o -sa ing echnical change. Consequen ly, he a ailable R&D-TFP li -
e a u e sugges s ha bo h domes ic R&D and o eign R&D conduc ed in de eloped
coun ies gene a e new echnologies ha a e ac o sa ing in de eloping economies.
Gi en his implica ion, and gi en ha ene gy is necessa y o he p oduc ion o all
kinds o goods, including ene gy i sel , i is na u al o ask: Do domes ic and o eign
R&D also gene a e ene gy sa ings pe uni o ou pu and hus educe ene gy in en-
si y (i.e., he a io o ene gy use o GDP) in de eloping economies? The answe o
his ques ion is he subjec o his s udy.
A educ ion in ene gy in ensi y in de eloping coun ies means ha hey can aise
hei li ing s anda ds wi hou a p opo ional inc ease in ene gy use, he eby educ-
ing he g ow h o he en i onmen al p oblems associa ed wi h inc easing ene gy
demand—such as ai and wa e pollu ion, land dis u bance, adioac i e was e om
nuclea ene gy p oduc ion, and global clima e change due o g eenhouse gas emis-
sions om ossil- uel i ed powe plan s. Acco ding o da a om he Wo ld De el-
opmen Indica o s, he a io o ene gy use o eal GDP in he g oup o low- and
middle-income coun ies (de ined he e as de eloping coun ies) exceeded ha in he
g oup o high-income coun ies by mo e han ac o 1.4 in 2014 ( he las yea wi h
a ailable da a o hese coun y g oups). The e o e, he answe o he abo e ques ion
is no only o academic in e es bu also di ec ly ele an o policymake s conce ned
wi h bo h economic and sus ainable de elopmen .
Howe e , he answe is heo e ically unclea (as discussed in Sec .2), and he
empi ical e idence is sca ce. The e a e only six s udies on he impac o domes ic
R&D on ene gy in ensi y in de eloping coun ies (Yu 2012; Wang and Han 2017;
Dong e al. 2018; Huang e al. 2018, 2020; Huang and Chen 2020), and only h ee
on he impac o bo h domes ic and o eign R&D on ene gy in ensi y in de elop-
ing coun ies (Huang e al. 2018, 2020; Wang and Han 2017). All hese s udies a e
single-coun y s udies o China, based on p o ince-le el panel da a.1
The e idence om all o hese s udies sugges s ha domes ic R&D has an ene gy
in ensi y educing e ec .2 Wi h espec o he e ec o o eign R&D on domes ic
ene gy in ensi y, wo s udies ind e idence o a nega i e ( educing) e ec o bo h
impo - ela ed and o eign di ec in es men (FDI)- ela ed spillo e s om R&D
1 A ela ed s udy is ha o Godil e al. (2021), who examine, among o he hings, he e ec o R&D
in ensi y (i.e., he a io o R&D expendi u es o GDP) on ene gy consump ion pe capi a (no measu ed
in logs) using ime-se ies da a o India. Thei esul s sugges ha R&D in ensi y has a nega i e e ec on
ene gy consump ion pe capi a. Howe e , hei empi ical model has he coun e ac ual implica ion ha a
doubling o R&D can educe ene gy consump ion pe capi a in an economy wi h only one dolla o R&D
o he same ex en as in an economy whe e R&D expendi u es amoun o one billion dolla s. In addi-
ion, R&D in ensi y su e s om endogenei y because highe ene gy use may esul in highe GDP ( he
denomina o o R&D in ensi y).
2 Huang and Chen (2020) also conside di e en ypes o R&D and ind ha while indus ial R&D con-
ibu es o educ ions in ene gy in ensi y, independen R&D and highe educa ion R&D ha e no signi i-
can e ec on ene gy in ensi y. They also ind ha expe imen al R&D has a nega i e e ec on ene gy
in ensi y, whe eas he e ec o bo h basic R&D and applied R&D is insigni ican .
1 3
Economic Change and Res uc u ing (2024) 57:33 Page 3 o 31 33
pe o med in high-income coun ies on domes ic ene gy in ensi y (Wang and Han
2017; Huang e al. 2018); one s udy inds, somewha su p isingly, ha while he
e ec o o eign R&D spillo e s h ough impo s is insigni ican , and while he
e ec o o eign R&D spillo e s h ough FDI is nega i e, o eign R&D spillo e s
h ough expo s inc ease domes ic ene gy in ensi y (Huang e al. 2020).
Howe e , he majo i y o hese s udies (Wang and Han 2017; Huang e al. 2018,
2020; Huang and Chen 2020) do no con ol o (s ong) e o c oss-sec ional
dependence due o unobse ed common ac o s. Consequen ly, he esul s o he
majo i y o s udies may be biased in he p esence o omi ed common ac o s ha
a e co ela ed wi h he included explana o y a iables and he dependen a iable.3
In addi ion, some s udies (Yu 2012; Huang and Chen 2020; Huang e al. 2020) use
me hods ha assume s a iona y da a and hus can p oduce misleading esul s when
he da a a e non-s a iona y. Mo eo e , mos s udies (Yu 2012; Wang and Han 2017;
Huang and Chen 2020; Dong e al. 2018) u ilize es ima o s ha equi e s ic exoge-
nei y o he eg esso s, hus yielding po en ially misleading esul s when he eg es-
so s a e no s ic ly exogenous.4 Since all hese s udies su e om a leas one o
hese sho comings, hei esul s should be iewed wi h some cau ion. In addi ion, i
is well known ha indings om single-coun y s udies canno necessa ily be gen-
e alized. E en i he indings o hese s udies a e alid, i may he e o e be ha hey
apply only o China.
Gi en he lack o gene al c oss-coun y s udies on he impac o domes ic and
o eign R&D on ene gy in ensi y in de eloping coun ies, his s udy aims o ill his
gap. Mo e speci ically, we conduc a c oss-coun y panel analysis using da a om
33 de eloping coun ies (including China) spanning he yea s 1995 o 2015. I is
wo h no ing ha ou s udy is he i s o use panel da a o a c oss sec ion o de el-
oping coun ies.
In addi ion, his s udy di e s om p e ious esea ch by examining disembod-
ied, non- ade- ela ed R&D spillo e s om de eloped o de eloping coun ies, as
well as R&D spillo e s h ough impo s o all goods and impo s o machine y and
equipmen . Fu he mo e, his s udy akes in o accoun all he me hodological p ob-
lems add essed abo e.5 Mo e speci ically, we use panel coin eg a ion me hods o
3 C oss‐sec ional dependence may be due o common ac o s ha a ec all panel uni s and/o spa ial
spillo e e ec s ac oss subse s o panel uni s. C oss‐sec ional dependence due o common ac o s is also
known as s ong c oss-sec ional dependence; c oss-sec ional dependence due o spa ial spillo e s is also
known as weak c oss-sec ional dependence. The p esence o weak c oss-sec ional dependence does no
a ec he consis ency o con en ional panel da a es ima o s, bu he s anda d e o s may be biased. In
con as , s ong c oss-sec ional dependence, i no con olled, can lead o biased coe icien es ima es
(Chudik and Pesa an 2015).
4 I educ ions in ene gy in ensi y imply GDP g ow h due o ene gy-sa ing echnical change, and i i ms
espond o g ow h-induced inc eases in demand o a ie y by engaging in ho izon al R&D o de elop
new a ie ies o exis ing p oduc s, hen i is possible ha educ ions in ene gy in ensi y con ibu e o
inc eased R&D ac i i ies ia inc eases in GDP, a leas in he sho un. The implica ion is ha domes ic
R&D is likely no s ic ly exogenous.
5 We no e wo hings he e. Fi s , i is no possible o measu e R&D spillo e s h ough FDI o e ou
sample pe iod because comple e ime se ies da a on bila e al FDI lows om de eloped sou ce coun ies
a e no a ailable o de eloping hos coun ies o e he pe iod 1995–2015. I is pe haps in e es ing o
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 4 o 31
add ess he non-s a iona y na u e o he da a and analyze he long- un ela ionships
be ween ou a iables o in e es . As discussed in mo e de ail in Sec .4.3, and as
no ed by Coe e al. (2009, p. 724), “[u]nde coin eg a ion, pa ame e es ima es a e
supe consis en , and hence a e obus o p oblems such as omi ed a iables, simul-
anei y, and endogenei y.” While we a e no awa e o heo e ical easons o sugges
ha o eign R&D is endogenous o domes ic ene gy in ensi y, we hus accoun o
he likely endogenei y o domes ic R&D. In addi ion, we con ol and es o e o
c oss-sec ional dependence in he esiduals o ou models.
I should be explici ly no ed he e ha we use he Wo ld Bank classi ica ion o
de eloping coun ies acco ding o which low- and middle-income coun ies a e clas-
si ied as “de eloping coun ies” (Wo ld Bank 2007, 2012). Thus, he e m “de el-
oping coun ies” includes pos -communis coun ies. All coun ies in ou sample
(lis ed in Table2) ha all unde he ca ego y o de eloping coun ies acco ding o
he Wo ld Bank classi ica ion a e classi ied by he IMF as “eme ging ma ke and
de eloping economies,” which also include pos -communis coun ies. As a obus -
ness check, we also use a sample o “de eloping economies” as classi ied by UNC-
TAD ha does no include pos -communis coun ies. We come back o his poin in
Sec .4.2. He e, we no e o comple eness ha , ollowing common p ac ice, we use
he e m “No h” as sho hand o de eloped o indus ial economies and “Sou h” as
sho hand o de eloping coun ies.
An impo an poin is ha ou s udy also di e s om p e ious wo k in ha i also
examines he impac o domes ic R&D conduc ed in de eloped sou ce coun ies o
o eign R&D spillo e s o de eloping coun ies on ene gy in ensi y wi hin hese
sou ce coun ies. I o eign R&D conduc ed in indus ialized coun ies con ibu es
o educ ions in ene gy in ensi y in de eloping coun ies ( h ough in e na ional
R&D spillo e s), i can be plausibly concluded ha R&D conduc ed by indus ial-
ized coun ies ends o esul in ene gy-sa ing echnologies. I his conclusion is co -
ec , hen one should expec o ind a nega i e e ec o domes ic R&D (conduc ed in
de eloped coun ies) on ene gy in ensi y in de eloped coun ies. To ou knowledge,
he e is only one c oss-coun y panel s udy on he impac o R&D on ene gy in en-
si y in de eloped coun ies: Alam e al. (2019). The au ho s analyze i m-le el da a
o he G-6 coun ies (which include Canada, F ance, Ge many, Japan, he UK, and
he US) and ind ha R&D educes ene gy in ensi y in hese na ions. Ou s udy is
he i s bo h o examine he impac o o eign R&D conduc ed in indus ial coun-
ies on ene gy in ensi y in de eloping coun ies and o conduc a plausibili y check
Foo no e 5 (con inued)
no e in his con ex ha he e a e se e al s udies on he impac o FDI on ene gy in ensi y in de eloping
coun ies, which, howe e , do no explici ly examine he e ec o FDI- ela ed o eign R&D spillo e s
on ene gy in ensi y in de eloping coun ies, bu ocus on he b oade impac o FDI. The e idence om
hese s udies is mixed, wi h some indica ing ha FDI educes ene gy in ensi y in de eloping coun ies,
while o he s ind no signi ican e ec . Fo a e iew o his li e a u e see He ze and Schmelme (2022).
Second, we also examined R&D spillo e s h ough expo s o o al goods and R&D spillo e s h ough
expo s o machine y and equipmen , bu ound li le o no e idence o long- un spillo e s om o eign
R&D conduc ed in de eloped o domes ic ene gy in ensi y in de eloping coun ies h ough expo s o
o al goods and expo s o machine y and equipmen ( om de eloping o de eloped coun ies).

1 3
Economic Change and Res uc u ing (2024) 57:33 Page 5 o 31 33
o ou main esul s by in es iga ing he impac o domes ic R&D on ene gy in en-
si y in 15 de eloped sou ce coun ies o o eign R&D spillo e s.
To p e iew ou main esul s, we ind ha while domes ic R&D, in he long un,
does no con ibu e o educ ions in ene gy in ensi y in de eloping coun ies, o -
eign R&D pe o med in indus ial coun ies educes ene gy in ensi y in de eloping
coun ies in he long un. Speci ically, ou esul s sugges ha No h–Sou h spillo-
e s occu mainly h ough impo s o machine y and equipmen a he han h ough
impo s o o he goods and ha he impac o o eign R&D a ies wi h he sha e o
machine y and equipmen impo s in GDP. Howe e , we ind no e idence o disem-
bodied spillo e e ec s. An addi ional esul o his s udy is ha he e is e idence
ha domes ic R&D pe o med in indus ial sou ce coun ies o R&D spillo e s
educes ene gy in ensi y in hese coun ies as well.
The emainde o his pape is o ganized as ollows. In Sec .2, we discuss he
heo e ical backg ound. Sec ion 3 p esen s he empi ical model and de ines he
a iables. Sec ion4 desc ibes he da a, including he sample, and discusses some
econome ic issues and he empi ical me hodology. Sec ion5 epo s ou esul s,
and Sec .6 concludes and p o ides some policy implica ions.
2 Theo e ical backg ound
We begin wi h a gene al ene gy-augmen ed agg ega e p oduc ion unc ion o he
o m Y = A (AK, AL, AE). In his equa ion, Y ep esen s agg ega e ou pu , K s ands
o capi al, L o labo , and E o ene gy. The mul iplie A deno es he le el o ech-
nology, which is he ocus o ou discussion he e. In he e ms AK, AL, and AE, A
indica es ha he echnology augmen s capi al, labo , and ene gy, espec i ely. I A
appea s in on o he unc ion , he echnology is ac o neu al. Technical change,

A
, imp o es he p oduc i i y o K, L, and E, espec i ely. In he case o pu ely labo -
o capi al-augmen ing echnical change, wi hou ene gy-sa ing ad ances, echni-
cal change hus educes he labo - o capi al-ou pu a io. I has no e ec on ene gy
in ensi y, he a io o E o Y, p o ided bo h ha he e a e no subs i u ion e ec s
be ween ene gy and labo o capi al and ha he g ow h o income due o echni-
cal change does no induce a shi in consump ion pa e ns owa d ene gy-in ensi e
goods. Since labo - o capi al-sa ing echnical change educes he e ec i e p ice o
labo o capi al, labo o capi al will, howe e , be induced o subs i u e o ene gy. In
addi ion, he educ ion in he e ec i e p ice o labo o capi al should lead o lowe
p ices o labo o capi al-in ensi e p oduc s. The pa e n o demand may he e o e
shi away om ene gy-in ensi e goods, so ha less ene gy-in ensi e sec o s expand
ela i e o ene gy-in ensi e sec o s. Thus, e en pu ely labo o capi al-augmen ing
echnical change may, in he long- un, con ibu e o educ ions in ene gy in ensi y.
I , howe e , i espec i e o ela i e p ices, inc eases in income du ing indus iali-
za ion a e associa ed wi h a shi in consump ion pa e ns owa d ene gy-in ensi e
goods and se ices (such as p i a e ehicles, ai condi ione s, and ligh s), as a gued
and demons a ed by Ha (2018), hen he inc eases in eal income om ising p o-
duc i i y may esul in inc eases in he ela i e size o ene gy-in ensi e sec o s. I is
he e o e also possible, and likely, ha labo o capi al-augmen ing echnical change
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 6 o 31
leads o an inc ease in ene gy in ensi y in he long un, e en i he e is subs i u abil-
i y be ween ene gy and labo o capi al.
Analogously, pu ely ene gy-augmen ing echnical change implies ha he same ou -
pu can be p oduced wi h less ene gy and hus ha he e ec i e p ice o ene gy declines.
The decline in he e ec i e p ice o ene gy induces a subs i u ion in a o o ene gy e -
sus labo o capi al, which o se s o some deg ee he ini ial educ ion in ene gy in ensi y.
The lowe he elas ici y o subs i u ion be ween labo o capi al and ene gy, he smalle
he o se ing e ec .6 In addi ion, he decline in he e ec i e p ice o ene gy implies an
inc ease in eal income. I his ex a income is spend on ene gy-in ensi e goods, he
ela i e size o ene gy-in ensi e sec o s inc eases. Thus, e en ene gy-sa ing echnical
change does, in he long un, no necessa ily lead o educ ions in ene gy in ensi y.
Finally, he concep o ac o -neu al echnical change implies ha he a io o
L o Y, he a io o K o Y, and he a io o E o Y dec ease, a leas ini ially. Like
abo e, inc eases in income as a esul o ac o -neu al echnical change may, how-
e e , induce an inc ease in he ela i e size o ene gy-in ensi e sec o s, and hus an
inc ease in ene gy in ensi y in he long un.
Thus, i can be assumed ha ene gy in ensi y, EI, depends on he le el o ech-
nology (which can be mo e o less ene gy-augmen ing) using a unc ion o he o m
EI = Aβ. Al hough he sign o he elas ici y β is heo e ically inde e mina e, i is eason-
able o assume ha he mo e ene gy sa ing echnical change is, he g ea e he likeli-
hood will be ha echnical change will con ibu e o educ ions in ene gy in ensi y in
he long un. Assuming u he a long- un ela ionship be ween R&D e o and he
le el o echnology o he o m A = R&Dφ, ene gy in ensi y can be exp essed as
whe e α ≡ φ × β is he elas ici y o ene gy in ensi y wi h espec o R&D.7 Based
on his equa ion and on he abo e heo e ical conside a ions, i can be hypo hesized
(1)
EI =R&D𝛼
6 I he elas ici y o subs i u ion o ene gy is less han one, hen imp o emen s in ene gy p oduc i i y
will lead o educ ions in ene gy in ensi y (holding income e ec s cons an ). I he elas ici y o subs i u-
ion is g ea e han one, hen ene gy-augmen ing echnical change will induce inc eases in ene gy in en-
si y. Koe se e al. (2008) ind in a me a-analysis ha he elas ici y o subs i u ion be ween capi al and
ene gy is less han one. S e n and Kande (2012), using his o ical da a o Sweden, ind ha he elas ici y
o subs i u ion be ween a capi al-labo agg ega e and ene gy anges be ween 0.64 and 0.69.
7 The ela ionship be ween R&D e o and he le el o echnology o he o m A = R&Dφ can be de i ed
as ollows. As discussed, o example, in He ze (2022), semi-endogenous g ow h models assume a
knowledge p oduc ion o he o m

A
= δAϕR&Dλ, whe e

A
is he low o new knowledge o echnical
change; δ is a cons an o p opo ionali y; A ep esen s he s ock o exis ing knowledge o he le el o
echnology; ϕ is a pa ame e ha desc ibes he na u e o he e u ns o he s ock o knowledge, R&D
s ands o R&D e o ; and λ, whe e 0 < λ ≤ 1, is a pa ame e ha cap u es he possibili y o duplica ion in
esea ch (i.e., he possibili y ha a doubling o esea ch e o less han doubles he p oduc ion o new
knowledge because o duplica ion). Assuming ha he s ock o knowledge g ows in he long un a a con-
s an a e g, he abo e equa ion can be sol ed o he s ock o knowledge, yielding
A
=
(
𝛿
g
A)
1
1−𝜙R&D
𝜆
1−
𝜙
.
This equa ion p edic s ha , p o ided he g ow h a e o knowledge is cons an o e he long un,
changes in R&D e o a e posi i ely associa ed wi h changes in he le el o echnology. Fo simplici y,
se ing he e m
(
𝛿
gA)
1
1−
𝜙
, which is cons an , equal o 1, he abo e equa ion co esponds o he equa ion
A = R&Dφ, whe e
𝜑
≡
𝜆
1
−
𝜙
.
1 3
Economic Change and Res uc u ing (2024) 57:33 Page 7 o 31 33
ha i mo e R&D is o ien ed mo e owa d ene gy-sa ing echnologies han owa d
labo - o capi al-sa ing echnologies, i is mo e likely ha R&D will con ibu e o
long- e m educ ions in ene gy in ensi y.
Un o una ely, da a ha allow he cons uc ion o p oxies o R&D in labo - o
capi al-sa ing echnologies and/o R&D in ene gy-sa ing echnologies a e no a ail-
able o a la ge numbe o coun ies, pa icula ly de eloping coun ies.8 I is he e-
o e no possible o quan i y he ela i e amoun s o R&D in ene gy-sa ing ech-
nologies and R&D in labo - o capi al-sa ing echnologies in de eloping coun ies,
and hence o assess a p io i whe he R&D in de eloping coun ies, in gene al, is
o ien ed mo e owa d ene gy-sa ing echnologies han owa d labo - o capi al-sa -
ing echnologies. Wha can be said, howe e , is ha he as majo i y o wo ldwide
R&D ac i i y akes place in indus ial na ions.9 I R&D by indus ial coun ies gen-
e a es echnologies ha sa e mo e ene gy han hose gene a ed by R&D in de elop-
ing coun ies, hen i is possible ha o eign R&D pe o med in indus ial coun ies
con ibu es mo e o educ ions in ene gy in ensi y in de eloping coun ies h ough
in e na ional R&D spillo e s han domes ic R&D. Howe e , o he ex en ha R&D
pe o med in indus ial coun ies gene a es echnologies ha canno be adap ed
o local condi ions in de eloping coun ies, i may con ibu e less o educ ions in
domes ic ene gy in ensi y han domes ically pe o med R&D. Thus, he e ec s o
domes ic and o eign R&D on domes ic ene gy in ensi y in de eloping coun ies a e
an empi ical ques ion.
3 Empi ical model and a iable de ini ions
We begin by aking na u al loga i hms o bo h sides o Eq.1. Then, we in oduce
coun y and ime subsc ip s i and , and add an e o e m εi . Addi ionally, we
include coun y ixed e ec s ci o con ol o any unobse ed ime-in a ian coun y
cha ac e is ics. We also con ol o e ec s o unobse ed ime- a ying common ac-
o s ρF , which, i le uncon olled, can induce c oss-sec ional dependence in he
8 OECD da a (a ailable a h ps:// s a s. oecd. o g/ Index. aspx? Da aS e Code= GERD_ TORD) on o al pub-
lic and p i a e ene gy R&D expendi u es a e a ailable o only 25 coun ies, and all hese coun ies ha e
sho and/o incomple e ime se ies. The In e na ional Ene gy Agency epo s da a on go e nmen spend-
ing on ene gy R&D o 32 coun ies (a ailable a h ps:// www. iea. o g/ da a- and- s a i s ics/ da a- p odu c /
ene gy- echn ology- d- and-d- budge - da ab ase-2), bu da a on o al public and p i a e ene gy R&D expen-
di u es a e epo ed o only h ee coun ies.
9 Acco ding o da a om he UNESCO Ins i u e o S a is ics (a ailable a a ailable a h p:// da a. uis.
unesco. o g/ Index. aspx? Da aS e Code= SCN_ DS), high-income coun ies accoun ed o abou 68% o he
o al wo ldwide R&D in 2015 ( he las yea o ou sample pe iod), whe eas middle- and low-income
coun ies we e esponsible o abou 32% o wo ldwide R&D expendi u es.
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 8 o 31
eg ession e o and lead o inconsis en es ima es. Ou basic empi ical model is
hus gi en by
whe e log EIi is he log o ene gy in ensi y in de eloping coun y i in yea , and log
R&Di ep esen s he log o R&D e o . We es ima e one speci ica ion wi h ( he log
o ) domes ic R&D e o in de eloping coun ies,
log R&Dd
i
, and i e o he speci i-
ca ions wi h o eign R&D, which akes place in de eloped coun ies.
The i s o hese i e speci ica ions is used o examine whe he R&D spillo e s
om de eloped o de eloping coun ies occu h ough disembodied channels such
as scien i ic jou nals, in e na ional con e ences, and he in e ne . Following, among
o he s, Coe e al. (1997) and He ze (2022), we de ine he measu e o o eign R&D
spillo e s in his speci ica ion as he log o he sum o he R&D e o s o N de el-
oped coun ies,
log R&D
,
whe e
R&Dd
j
is he R&D e o o indus ial coun y j. I is pe haps needless o say
ha he wo coun y g oups do no o e lap.
To es ima e he impac o impo - ela ed R&D spillo e s om No h o Sou h on
ene gy in ensi y, we use ou o he speci ica ions wi h u he spillo e a iables.
One o hese spillo e a iables is
log R&D _T
i
, which, ollowing he weigh ing
scheme o Coe and Helpman (1995),10 is he log o he weigh ed a e age o he
domes ic R&D e o s o he N de eloped coun ies, wi h bila e al sha es o ( o al)
impo s as weigh s,
whe e
IMT
ij
s ands o impo s o o al goods o de eloping coun y i om de eloped
coun y j and
IMT
i
deno es impo s o o al goods o coun y i om all N indus ial
coun ies,
IM
T
i
=
N
∑
j
=
1
IMT
ij .
Coe and Helpman (1995) use o al impo s as hei weigh ing ac o and ind e i-
dence o spillo e s om o eign R&D o domes ic TFP in a sample o OECD coun ies.
(2)
log EIi
=
𝛼log R&Di
+
ci
+
𝜌F
+
𝜀i
(3)
log R&D
≡log
N
∑
j
=
1
R&D
d
j
(4)
log R&D
_T
i ≡log
N
∑
j
=
1
IM
T
ij
IMT
i
R&Dd
j
10 Lich enbe g and an Po elsbe ghe de la Po e ie (1998) a gue ha he weigh ing scheme o Coe and
Helpman (1995) is sensi i e o a po en ial me ge be ween coun ies, and sugges an al e na i e weigh -
ing scheme ha is less sensi i e o he le el o agg ega ion. While Lich enbe g and an Po elsbe ghe
de la Po e ie (1998) ind ha hei weigh ing scheme yields somewha be e empi ical esul s han he
Coe and Helpman (1995) weigh ing scheme, Coe e al. (2009) ind ha he weigh ing scheme o Coe and
Helpman (1995) pe o ms somewha be e han he Lich enbe g and an Po elsbe ghe de la Po e ie
(1998) scheme. We epea ed he analysis using he la e scheme and ound quali a i ely simila esul s
(a ailable on eques ).
1 3
Economic Change and Res uc u ing (2024) 57:33 Page 15 o 31 33
Table 2 Sample coun ies and hei classi ica ion du ing he pe iod 1995–2015
Wo ld Bank classi ica ion IMF classi ica ion UNTAD classi ica ion
A gen ina Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
A menia Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
B azil Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Bulga ia Middle-income coun y Eme ging ma ke o de eloping economy
China Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Colombia Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Cos a Rica Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
C oa ia Middle-income coun y Eme ging ma ke o de eloping economy
Czech Republic Middle-income coun y Eme ging ma ke o de eloping economy
Ecuado Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Egyp , A ab Rep Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Es onia Middle-income coun y Eme ging ma ke o de eloping economy
Hunga y Middle-income coun y Eme ging ma ke o de eloping economy
India Low-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
I an, Islamic Rep Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Kazakhs an Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
La ia Middle-income coun y Eme ging ma ke o de eloping economy
Li huania Middle-income coun y Eme ging ma ke o de eloping economy
Mexico Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Mongolia Low-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Panama Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Poland Middle-income coun y Eme ging ma ke o de eloping economy
Romania Middle-income coun y Eme ging ma ke o de eloping economy
Russian Fede a ion Middle-income coun y Eme ging ma ke o de eloping economy
Slo ak Republic Middle-income coun y Eme ging ma ke o de eloping economy
Sou h A ica Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Tajikis an Low-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Thailand Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy

Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 16 o 31
A hi d ad an age associa ed wi h coin eg a ion is ha i implies long- un
G ange causali y in a leas one di ec ion (G ange 1988).16 He e we assume—and
es he assump ion— ha long- un causali y uns om log R&Di o log EIi . Finally,
a ou h ad an age is ha endogenei y does no lead o inconsis ency in he eg es-
sion coe icien s in he p esence o coin eg a ion.
Howe e , al hough e en he s anda d ixed-e ec s es ima o is (supe ) consis en
unde panel coin eg a ion e en when he eg esso s a e endogenous, i su e s om
a second-o de asymp o ic bias due o endogenei y and se ial co ela ion, and, as a
consequence, i s usual s anda d e o s a e no co ec . The e o e, we use he panel
DOLS o es ima o o Kao and Chiang (2001), which allows o endogenous eg es-
so s and which has been shown o pe o m well in samples like he one used he e
(see, e.g., Kao and Chiang 2001; Wagne and Hlousko a 2009).17
A coun y is classi ied as a “middle-income coun y” [“low-income coun y”] i i is o icially ca ego-
ized as such by he Wo ld Bank in i s “his o ical classi ica ion by income” (a ailable a h ps:// da ah
elpde sk. wo ld bank. o g/ knowl edgeb ase/ a ic les/ 906519) o mo e han hal o he calenda yea s be ween
1995 and 2015. The Wo ld Bank classi ies low- and middle-income coun ies as “de eloping coun ies”
(Wo ld Bank 2007). A coun y is classi ied as an “eme ging ma ke o de eloping economy” i i is lis ed
in he ca ego y “eme ging ma ke and de eloping economies” by he IMF in i s Wo ld Economic Ou -
look epo s (a ailable a h ps:// www. im . o g/ en/ Publi ca io ns/ WEO) o he yea s 2004 onwa ds. All
coun ies classi ied as eme ging ma ke s o de eloping economies we e p e iously ca ego ized as ei he
“de eloping coun ies” o “coun ies in ansi ion” in he Wo ld Economic Ou look epo s. A coun y is
classi ied he e as a “de eloping economy” i i is lis ed by UNCTAD as such in i s classi ica ion (a ail-
able a h ps:// unc a ds a . unc ad. o g/ en/ class i ica ions. h ml)
Table 2 (con inued)
Wo ld Bank classi ica ion IMF classi ica ion UNTAD classi ica ion
T inidad and Tobago Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Tunisia Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Tü kiye Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
Uk aine Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
U uguay Middle-income coun y Eme ging ma ke o de elop-
ing economy
De eloping economy
17 The DOLS me hod employs a pa ame ic co ec ion o endogenei y and se ial co ela ion, based on
lead, lag, and cu en alues o he di e enced eg esso s. An al e na i e es ima ion me hod o es ima -
ing coin eg a ing ela ionships is he (panel) FMOLS es ima o , which uses a semi-pa ame ic co ec ion
o endogenei y and se ial co ela ion (based on he OLS esiduals and he i s di e ences o he eg es-
so s). Simula ion e idence sugges s ha he DOLS es ima o pe o ms be e han he FMOLS es ima o
in small samples (see, e.g., Kao and Chiang 2001; Wagne and Hlousko a 2009). The e o e, we p e e he
DOLS es ima o . The esul s (a ailable on eques ) do no change quali a i ely when he FMOLS es ima-
o is used.
16 The concep o long- un (G ange ) causali y is o be dis inguished om he mo e amilia no ion o
“G ange causali y,” which e e s o sho - un o ecas abili y and does no accoun o long- un causali y
h ough he e o co ec ion e m in a coin eg a ed e o -co ec ion model.
1 3
Economic Change and Res uc u ing (2024) 57:33 Page 17 o 31 33
Table 3 Co ela ion ma ix and summa y s a is ics
Log EIi (EIi )
log R&Dd
i
(
R&Dd
i
)
log R&D
log R&D _T
i
log R&D _M
i
mi
T
i
×log R&D
_T
i
mi
M
i
×log R&D
_M
i
A. Sample o 33 de eloping coun ies (acco ding o Wo ld Bank classi ica ion)
log EIi 1.000

log R&Dd
i
0.183 1.000

log R&D
− 0.256 − 0.028 1.000

log R&D _T
i
− 0.412 0.024 0.293 1.000

log R&D _M
i
− 0.365 0.024 0.292 0.986 1.000

mi
T
i
×log R&D
_T
i
− 0.202 − 0.491 0.003 − 0.122 − 0.149 1.000

mi
M
i
×log R&D
_M
i
− 0.049 − 0.386 0.412 − 0.192 − 0.193 0.578 1.000
Mean 11.810 (155,407.5) 7.669 (8899.64) 22.828 20.791 20.807 3.631 0.546
Maximum 13.113 12.740 23.002 22.101 22.093 19.136 3.807
Minimum 10.895 2.555 22.456 19.666 19.865 0.309 0.013
S d. De 0.468 1.818 0.134 0.594 0.545 2.969 0.554
B. Sample o he 15 de eloping coun ies ha we e used as sou ces o R&D spillo e s
Log EIi 1

log R&Dd
i
0.254 1
Mean 11.631 (118,977.4) 9.931 (52,937.06)
Maximum 12.402 13.05
Minimum 10.656 7.140
S d. De 0.332 1.308
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 18 o 31
I should, howe e , be no ed ha he panel DOLS es ima o is based on he assump ion
o e o c oss‐sec ional independence.18 To accoun o weak c oss-sec ional dependence
in he esiduals o he DOLS models, we ollow Bo do e al. (2017) and use D iscoll and
K aay (1998) s anda d e o s; hese s anda d e o s a e obus o he e oskedas ici y, au o-
co ela ion, and spa ial co ela ion. To accoun o s ong c oss-sec ional e o depend-
ence, we demean he da a by sub ac ing he c oss-sec ional a e ages om he da a and
hen use he demeaned da a in place o he o iginal da a (which is equi alen o using he
esiduals om eg essions o each a iable on ime dummies). In addi ion, o ensu e ha
ou esul s do no su e om e o c oss-sec ional dependence due o common ac o s,
we es o s ong c oss-sec ional dependence in he esiduals om ou DOLS eg essions
using he c oss-sec ional dependence (CD) es o Juodis and Reese (2022).19
5 Resul s
5.1 Main esul s
Panel A o Table4 shows he panel DOLS esul s o he ela ionship in ou sample
be ween each o he R&D a iables and he log o ene gy in ensi y. In Panel A, we
also p esen he esul s o he Juodis–Reese es o s ong c oss-sec ional depend-
ence in he esiduals o he DOLS eg essions. The esul s o se e al panel coin e-
g a ion es s a e shown in Panel B.
Rega ding he esul s in Panel B, wo hings should be no ed. Fi s , o he Ped oni
(1999, 2004) panel coin eg a ion es s, which assume e o c oss-sec ional inde-
pendence, we epo es s a is ics based on he demeaned da a o con ol o e o
c oss-sec ional dependence. Fo he Gengenbach e al. (2016) es , which explici ly
accoun s o c oss-sec ional dependence ia he use o c oss-sec ional a e ages o
he a iables, we epo es s a is ics based on he aw da a.
Second, e o -co ec ion-based coin eg a ion es s such as he Gengenbach e al.
(2016) es inco po a e in he al e na i e hypo hesis he assump ion ha he depend-
en a iable is no weakly exogenous wi h espec o he independen a iables.
Rejec ion o he null o no coin eg a ion using an e o -co ec ion model wi h Δlog
EIi as he dependen a iable can he e o e be in e p e ed as e idence ha log EIi
is no weakly exogenous o he independen a iables and hus ha he independen
a iables “cause” log EIi (p o ided ha hey a e signi ican ).
19 We use he Juodis and Reese (2022) es a he han he s anda d Pesa an (2021) es because he la e
has no powe o de ec e o c oss-sec ional dependence when he es ima ed models include ime dum-
mies (o c oss-sec ional a e ages) o a e based on demeaned da a. The Juodis and Reese (2022) es is a
modi ied e sion o he Pesa an (2021) es ha does no su e om his p oblem.
18 We also expe imen ed wi h he pooled common co ela ed e ec s (PCCE) es ima o and he com-
mon co ela ed e ec s mean g oup (CCEMG) o Pesa an (2006). Bo h hese es ima o s a e speci ically
designed o accoun o e o c oss-sec ional dependence. While he esul s om he PCCE es ima o a e
signi ican (and nega i e) only o
log R&D
, he esul s om he CCEMG es ima o a e signi ican (and
nega i e) only o
log R&Dd
i
. Gi en, howe e , ha hese es ima o s a e designed o la ge N and la ge
T and ha hey equi e s ic ly exogenous eg esso s, he PCCE and CCEMG es ima es a e no eliable
he e due o he possibili y o endogenous o weakly exogenous eg esso s and/o he ela i ely small
numbe o coun ies and yea s in ou sample.
1 3
Economic Change and Res uc u ing (2024) 57:33 Page 19 o 31 33
Tu ning o he esul s in column (1) o Table4, we ind a weakly signi ican posi-
i e e ec o domes ic R&D on ene gy in ensi y. This e ec appea s no o be due o
he p esence o s ong c oss-sec ional dependence in he esiduals, as sugges ed by
he Juodis–Reese es . Since, howe e , wo o he coin eg a ion es s do no ejec he
null o no coin eg a ion, i canno be uled ou wi h ce ain y ha he obse ed e ec
is he esul o spu ious eg ession. We come back o his poin when we discuss he
esul s in column (1) o Table6.
In column (2) o Table4, we see ha while ou o he i e es s indica e coin-
eg a ion, he coe icien on he log o he unweigh ed sum o he R&D expendi-
u es o indus ial coun ies is posi i e and s a is ically insigni ican . Howe e , he
Juodis–Reese es indica es he p esence o s ong c oss-sec ional dependence in he
DOLS esiduals, and hus he es ima ion esul s should be iewed wi h cau ion.
In column (3), he coe icien on
log R&D _T
i
is posi i e bu insigni ican , and only
one es ejec s he null hypo hesis o no coin eg a ion. We hus ind no long- un e i-
dence o unin e ac ed spillo e e ec s o impo -weigh ed o eign R&D expendi u es
using o al impo s as weigh s. Fo comple eness, howe e , i should be said ha we
canno ule ou he possibili y ha he insigni ican coe icien is he esul o unob-
se ed common ac o s in he DOLS esiduals (as sugges ed by he Juodis–Reese es ).
Ou e idence also does no suppo he exis ence o an unin e ac ed e ec o
capi al goods impo -weigh ed o eign R&D expendi u es on he long- un le el o
domes ic ene gy in ensi y. Column (4) shows ha
log R&D _M
i
has a signi ican neg-
a i e coe icien and ha he Juodis–Reese es is insigni ican , as in columns (1),
(5), (6), and (7). Howe e , none o he es s ejec s he null hypo hesis o no coin e-
g a ion, sugges ing ha he eg ession esul is spu ious.
Simila ly, he coe icien in column (5) o he a iable
mi
T
i
×log R&D
_T
i
is
nega i e and signi ican , bu only wo es s p o ide clea e idence (a he con en-
ional 5% le el o be e ) o coin eg a ion. Thus, he e is also no clea suppo o an
in e ac ed spillo e e ec o o al impo -weigh ed o eign R&D expendi u es on he
long- un le el o domes ic ene gy in ensi y.
In con as , he esul s in column (6) show clea e idence ha R&D pe o med
in indus ial coun ies weigh ed by he bila e al sha e o machine y and equipmen
impo s om he indus ial coun ies educes ene gy in ensi y in de eloping coun-
ies h ough i s in e ac ion wi h he machine y and equipmen impo sha e in de el-
oping coun ies’ GDP. All es s indica e coin eg a ion be ween
mi
M
i
×log R&D
_M
i
and log EIi a leas a he 5% le el; he Gengenbach e al. (2016) es sugges s ha
log EIi is „caused‟ in he long un by
mi
M
i
×log R&D
_M
i
20; and he coe icien on
mi
M
i
×log R&D
_M
i
is nega i e and s a is ically signi ican wi h a alue o − 0.025.
20 We also compu ed he Gengenbach e  al. (2016) es s a is ic using he e e se eg ession wi h
Δ
mi
M
i
×log R&D
_M
i
on he le -hand side. The alue o he es s a is ic is − 1.813, implying ha he
null o weak exogenei y canno be ejec ed o
mi
M
i
×log R&D
_M
i
. Weak exogenei y implies long- un
G ange non-causali y (see, e.g., Hall and Milne 1994). Thus, he e idence ha
mi
M
i
×log R&D
_M
i
is
weakly exogenous and log EIi is no weakly exogenous means ha
mi
M
i
×log R&D
_M
i
has a long- un
(causal) e ec on log EIi , whe eas log EIi has no long- un e ec on
mi
M
i
×log R&D
_M
i
.
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 20 o 31
Table 4 Main esul s
(1) (2) (3) (4) (5) (6)
A. DOLS es ima es

log R&Dd
i
0.030* (0.016)

log R&D
0.017 (0.029)

log R&D _T
i
0.028 (0.080)

log R&D _M
i
− 0.312*** (0.073)

miT
i
×log R&D
_T
i
− 0.027** (0.008)

mi
M
i
×log R&D
_
M
i
− 0.025*** (0.008)
Juodis–Reese (p alue) 0.234 0.000 0.050 0.338 0.776 0.866
No. o coun ies 33 33 33 33 33 33
No. o obs 531 532 532 532 532 532
Adjus ed R20.951 0.951 0.951 0.954 0.951 0.951
B. Panel coin eg a ion es s
Ped oni (1999, 2004)
Panel PP -s a is ic − 1.045 − 5.496*** − 1.125 0.073 − 1.385* − 2.159**
Panel ADF -s a is ic − 1.270 − 6.540*** − 0.214 − 0.000 − 1.460* − 1.671**
G oup PP -s a is ic − 1.955** − 5.360*** − 1.120 0.411 − 3.645*** − 2.312**
G oup ADF -s a is ic − 4.058*** − 7.755*** − 0.988 − 0.686 − 3.197*** − 2.806***
Gengenbach e al. (2016)
ECM -s a is ic − 4.107*** − 1.764 − 4.202*** − 2.105 − 2.428 − 3.558***

1 3
Economic Change and Res uc u ing (2024) 57:33 Page 21 o 31 33
The dependen a iable in he DOLS eg essions and he Ped oni (1999, 2004) es s is log EIi . The dependen a iable in he Gengenbach e al. (2016) es s is Δlog EIi .
All eg essions (and es s) include coun y ixed e ec s. The DOLS eg essions we e es ima ed wi h one lead and one lag o he i s -di e enced eg esso s. The da a on
log EIi ,
log R&Dd
i
,
log R&D _T
i
,
mi
T
i
×log R&D
_T
i
,
log R&D _M
i
, and
mi
M
i
×log R&D
_M
i
o he DOLS eg essions and he Ped oni (1999, 2004) es s we e demeaned o
accoun o (s ong) e o c oss-sec ional dependence due o unobse ed common ac o s;
log R&D
is he same o each coun y and can be conside ed as an obse ed
common ac o ( ha canno be demeaned). The Gengenbach e al. (2016) es accoun s o s ong e o c oss-sec ional dependence ia he use o c oss-sec ional a e ages.
Juodis–Reese is he es o s ong c oss-sec ional dependence o Juodis and Reese (2022) applied o he esiduals om he DOLS eg essions. The numbe o lags in he
Ped oni (1999, 2004) (PP and ADF) and Gengenbach e al. (2016) es s was de e mined using he gene al- o-speci ic app oach wi h a maximum o wo lags. Two lags o
he c oss-sec ional a e ages we e included in he Gengenbach e al. (2016) es s. All es s ejec o la ge nega i e alues. The Ped oni (1999) es s a is ics a e dis ibu ed
as s anda d no mal. The Gengenbach e al. (2016) c i ical alue o one eg esso a he 1% (5%) [10%] signi icance le el is − 2.735 (− 2.601) [− 2.530] o N = 30. Num-
be s in pa en heses a e D iscoll and K aay (1998) he e oskedas ici y au oco ela ion spa ial co ela ion obus s anda d e o s. *** (**) [*] indica e signi icance a he 1%
(5%) [10%] le el
Table 4 (con inued)
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 22 o 31
To p o ide a sense o he magni ude o he e ec implied by his coe icien , con-
side ha a one s anda d de ia ion inc ease in
mi
M
i
×log R&D
_M
i
is associa ed wi h
a dec ease o 10.12 pe cen o a s anda d de ia ion in he ene gy in ensi y a iable
(− 0.025 × 1.977/0.4885), an e ec ha is economically signi ican .
5.2 Robus ness checks
In columns (1)–(6) o Table5, we check he obus ness o ou esul s wi h espec
o he use o he sample o de eloping coun ies classi ied by UNCTAD. The
esul s a e e y simila o hose in Table4. The only wo h men ioning di e ences
a e ha he coe icien s on
log R&Dd
i
and
log R&D _M
i
a e now insigni ican , and
ha he coe icien on
mi
T
i
×log R&D
_T
i
is signi ican only a he 10% le el. Thus,
he esul s in columns (1)–(6) o Table 5 once again sugges ha domes ic R&D
does no con ibu e o educ ions in ene gy in ensi y in de eloping coun ies. Fu -
he mo e, o eign R&D does no appea o a ec domes ic ene gy in ensi y h ough
disembodied channels. Ins ead, we again ind ha o eign R&D conduc ed in de el-
oped coun ies educes ene gy in ensi y in de eloping coun ies h ough impo s,
pa icula ly impo s o machine y and equipmen , and ha his e ec depends on he
sha e o machine y and equipmen impo s in de eloping coun ies’ GDP.
In column (7) o Table5, we p esen esul s o he ela ionship be ween domes ic
R&D conduc ed in he 15 indus ial sou ce coun ies o R&D spillo e s and ene gy
in ensi y in hose coun ies, as a plausibili y check. Fou o he i e coin eg a ion
es s sugges ha he e is a long- un ela ionship be ween
log R&Dd
i
and log EIi , and
he DOLS coe icien on
log R&Dd
i
is nega i e and highly signi ican . Thus, we ind
e idence ha domes ic R&D con ibu es o educ ions in ene gy in ensi y in de el-
oped sou ce coun ies o o eign R&D spillo e s, which suppo s he plausibili y o
ou inding ha he e a e signi ican spillo e s om R&D conduc ed in indus ial
coun ies ha educe ene gy in ensi y in de eloping coun ies.
In Table 6, we once again use he sample o 33 de eloping coun ies, classi-
ied acco ding o he Wo ld Bank, and assess he obus ness o ou esul s o a i-
ous speci ica ions in ol ing mul iple R&D a iables in columns (1)–(6). In col-
umn (1), we epo DOLS esul s o a eg ession ha includes bo h
log R&Dd
i
and
mi
M
i
×log R&D
_M
i
. The coe icien on
mi
M
i
×log R&D
_M
i
emains nega i e and
s a is ically signi ican , and he coe icien on
log R&Dd
i
is posi i e and signi ican
a he 10% le el, like in column (1) o Table4. Howe e , he e idence o coin eg a-
ion be ween
log R&Dd
i
,
mi
M
i
×log R&D
_M
i
, and log EIi in column (1) o Table6
is weake han he e idence o coin eg a ion be ween
mi
M
i
×log R&D
_M
i
and log
EIi in column (6) o Table4. I (as discussed in Sec .4.3) he e is an in eg a ed
eg esso ha is no coin eg a ed wi h o he coin eg a ed a iables in an equa ion,
he esiduals o such an equa ion will end o be non-s a iona y, and he e idence
o coin eg a ion may he e o e be weak (o e en absen ). Thus, he esul s o he
coin eg a ion es s in column (1) o Table6 oge he wi h hose in column (6) o
Table4 can be in e p e ed as an indica ion ha while he e is a long- un ela ionship
be ween
mi
M
i
×log R&D
_M
i
and log EIi , he e is no long- un ela ionship be ween
1 3
Economic Change and Res uc u ing (2024) 57:33 Page 23 o 31 33
Table 5 Resul s based on he subsample o de eloping coun ies classi ied by UNCTAD (columns (1) – (6)) and esul s using he sou ce coun ies o R&D spillo e s as
he sample (column (7))
(1) (2) (3) (4) (5) (6) (7)
A. DOLS es ima es

log R&Dd
i
0.010 (0.012) − 0.136*** (0.033)

log R&D
0.016 (0.033)

log R&D _T
i
0.027 (0.073)

log R&D _M
i
− 0.042 (0.059)

mi
T
i
×log R&D
_T
i
− 0.022* (0.012)

mi
M
i
×log R&D
_M
i
− 0.023*** (0.007)
Juodis–Reese (p alue) 0.199 0.312 0.002 0.153 0.456 0.766 0.466
No. o coun ies 21 21 21 21 21 21 15
No. o obs 324 325 325 325 325 325 270
Adjus ed R20.950 0.950 0.950 0.950 0.951 0.951 0.977
B. Panel coin eg a ion es s
Ped oni (1999, 2004)
Panel PP -s a is ic − 1.574* − 4.261*** − 0.896 − 0.039 − 0.671 − 3.902*** − 0.816
Panel ADF -s a is ic − 1.772** − 4.823*** − 1.237 − 0.139 − 1.450* − 2.505*** − 2.447***
G oup PP -s a is ic − 1.099 − 3.979*** − 1.093 − 1.041 − 2.361*** − 4.287*** − 2.911***
G oup ADF -s a is ic − 2.463*** − 5.375*** − 0.375 − 1.138 − 1.943** − 2.808*** − 6.153***
Gengenbach e al. (2016)
ECM -s a is ic − 1.869 − 1.886 − 1.745 0.320 − 2.172 − 3.394*** − 2.805**
Economic Change and Res uc u ing (2024) 57:33
1 3
33 Page 24 o 31
Table 5 (con inued)
The dependen a iable in he DOLS eg essions and he Ped oni (1999, 2004) es s is log EIi , The dependen a iable in he Gengenbach e al. (2016) es s is Δlog EIi .
All eg essions (and es s) include coun y ixed e ec s. The DOLS eg essions we e es ima ed wi h one lead and one lag o he i s -di e enced eg esso s. The da a on
log EIi ,
log R&Dd
i
,
log R&D _T
i
,
mi
T
i
×log R&D
_T
i
,
log R&D _M
i
, and
mi
M
i
×log R&D
_M
i
o he DOLS eg essions and he Ped oni (1999, 2004) es s we e demeaned o
accoun o (s ong) e o c oss-sec ional dependence due o unobse ed common ac o s;
log R&D
is he same o each coun y and can be conside ed as an obse ed
common ac o ( ha canno be demeaned). The Gengenbach e al. (2016) es accoun s o s ong e o c oss-sec ional dependence ia he use o c oss-sec ional a e ages.
Juodis–Reese is he es o s ong c oss-sec ional dependence o Juodis and Reese (2022) applied o he esiduals om he DOLS eg essions. The numbe o lags in he
Ped oni (1999, 2004) (PP and ADF) and Gengenbach e al. (2016) es s was de e mined using he gene al- o-speci ic app oach wi h a maximum o wo lags. Two lags o
he c oss-sec ional a e ages we e included in he Gengenbach e al. (2016) es s. All es s ejec o la ge nega i e alues. The Ped oni (1999) es s a is ics a e dis ibu ed
as s anda d no mal. The Gengenbach e al. (2016) c i ical alue o one eg esso a he 1% (5%) [10%] signi icance le el is − 2.796 (− 2.653) [− 2.530] o N = 20. The
Gengenbach e al. (2016) c i ical alue o one eg esso a he 5% signi icance le el is − 2.698 o N = 15. Numbe s in pa en heses a e D iscoll and K aay (1998) he e -
oskedas ici y au oco ela ion spa ial co ela ion obus s anda d e o s. *** (**) [*] indica e signi icance a he 1% (5%) [10%] le el
1 3
Economic Change and Res uc u ing (2024) 57:33 Page 31 o 31 33
Huang J, Chen X (2020) Domes ic R&D ac i i ies, echnology abso p ion abili y, and ene gy in ensi y
in China. Ene gy Policy 138:111184
Huang J, Hao Y, Lei H (2018) Indigenous e sus o eign inno a ion and ene gy in ensi y in China.
Renew Sus ain Ene gy Re 81:1721–1729
Huang J, Lai Y, Hu H (2020) The e ec o echnological ac o s and s uc u al change on China’s
ene gy in ensi y: e idence om dynamic panel models. China Econ Re 64:101518
Juodis A, Reese S (2022) The inciden al pa ame e s p oblem in es ing o emaining c oss-sec ion
co ela ion. J Bus Econ S a 40:1191–1203
Juselius K (2006) The coin eg a ed VAR model: me hodology and applica ions. Ox o d Uni e si y P ess,
Ox o d
Kao C (1999) Spu ious eg ession and esidual-based es s o coin ega ion in panel da a. J Econom
90:1–44
Kao C, Chiang M (2001) On he es ima ion and in e ence o a coin eg a ed eg ession in panel da a. In:
Bal agi BH, Fomby TB, Ca e Hill R (eds) Nons a iona y panels, panel coin eg a ion, and dynamic
panels (ad ances in econome ics), ol 15. Eme ald G oup Publishing Limi ed, Bingley, pp 179–222
Koe se MJ, de G oo HLF, Flo ax RJGM (2008) Capi al-ene gy subs i u ion and shi s in ac o demand:
a me a-analysis. Ene gy Econ 30:2236–2251
Lich enbe g FR, an Po elsbe ghe de la Po e ie BP (1998) In e na ional R&D spillo e s: a commen .
Eu Econ Re 42:1483–1491
Madsen JB, Saxena S, Ang JB (2010) The Indian g ow h mi acle and endogenous g ow h. J De Econ
93:37–48
Me cal GE (2008) An empi ical analysis o ene gy in ensi y and i s de e minan s a he s a e le el.
Ene gy J 29:1–26
Mimouni K, Temimi A (2018) Wha d i es ene gy e iciency? New e idence om inancial c ises.
Ene gy Policy 122:332–348
Ped oni P (1999) C i ical alues o coin eg a ion es s in he e ogeneous panels wi h mul iple eg esso s.
Ox Bull Econ S a 61:653–670
Ped oni P (2004) Panel coin eg a ion: asymp o ic and ini e sample p ope ies o pooled ime se ies es s
wi h an applica ion o he PPP hypo hesis. Econom Theo y 20:597–625
Ped oni P (2007) Social capi al, ba ie s o p oduc ion and capi al sha es: implica ions o he impo ance
o pa ame e he e ogenei y om a nons a iona y panel app oach. J Appl Econom 22:429–451
Pesa an MH (2006) Es ima ion and in e ence in la ge he e ogeneous panels wi h a mul i ac o e o s uc-
u e. Econome ica 74:967–1012
Pesa an MH (2007) A simple panel uni oo es in he p esence o c oss-sec ion dependence. J Appl
Econom 22:265–312
Pesa an MH (2021) Gene al diagnos ic es s o c oss-sec ional dependence in panels. Empi Econ
60:13–50
Sado sky P (2013) Do u baniza ion and indus ializa ion a ec ene gy in ensi y in de eloping coun ies?
Ene gy Econ 37:52–59
S e n DI, Kande A (2012) The ole o ene gy in he indus ial e olu ion and mode n economic g ow h.
Ene gy J 33:125–152
UNCTAD (2023) Classi ica ions. Re ie ed Sep embe 28, 2023, om Classi ica ions h ps:// unc a ds a .
unc ad. o g/ en/ class i ica ions. h ml
Wagne M, Hlousko a J (2009) The pe o mance o panel coin eg a ion me hods: esul s om a la ge
scale simula ion s udy. Econom Re 29:182–223
Wang D, Han B (2017) E ec s o indigenous R&D and o eign spillo e on ene gy in ensi y in China. J
Renew Sus ain Ene gy 9:03590
Wo ld Bank (2007) Wo ld de elopmen epo 2007: de elopmen and he nex gene a ion. Wo ld Bank,
Washing on DC
Wo ld Bank (2012) Wo ld de elopmen epo 2012: gende equali y and de elopmen . Wo ld Bank,
Washing on DC
Yu H (2012) The in luen ial ac o s o China’s egional ene gy in ensi y and i s spa iall inkages: 1988–
2007. Ene gy Policy 45:583–593
Publishe ’s No e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps
and ins i u ional a ilia ions.