Ren occhini, F ancesco; Vezzani, An onio; Mon eso , Sand o
Wo king Pape
Walking he g een line: Go e nmen sponso ed R&D and
clean echnologies
JRC Wo king Pape s on Co po a e R&D and Inno a ion (CoRDI), No. 01/2023
P o ided in Coope a ion wi h:
Join Resea ch Cen e (JRC), Eu opean Commission
Sugges ed Ci a ion: Ren occhini, F ancesco; Vezzani, An onio; Mon eso , Sand o (2024) : Walking he
g een line: Go e nmen sponso ed R&D and clean echnologies, JRC Wo king Pape s on Co po a e
R&D and Inno a ion (CoRDI), No. 01/2023, Eu opean Commission, Join Resea ch Cen e (JRC),
Se ille
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/311180
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 ps://c ea i ecommons.o g/licenses/by/4.0/
EUR XXXXX XX
Walking he G een Line: Go e nmen Sponso ed
R&D and Clean Technologies
JRC Wo king Pape s on Co po a e R&D and Inno a ion
No 01/2023
2024
Ren occhini, F., Vezzani, A., Mon eso , S.
This documen has been p oduced wi hin he con ex o he Global Indus ial Resea ch & Inno a ion
Analyses (GLORIA) ac i i ies ha a e join ly ca ied ou by he Eu opean Commission's Join
Resea ch Cen e –Di ec o a e Inno a ion and G ow h and he Di ec o a e Gene al o Resea ch and
Inno a ion-Di ec o a e F, P ospe i y. GLORIA has ecei ed unding om he Eu opean Union's
Ho izon 2020 esea ch and inno a ion p og amme.
Any commen s can be sen by email o: JRC-B6-[email p o ec ed], Mo e in o ma ion,
including ac i i ies and publica ions, is a ailable a : h ps://i i.j c.ec.eu opa.eu/home/.
Con ac in o ma ion
E-mail: JRC-B6-[email p o ec ed]
EU Science Hub
h ps://join - esea ch-cen e.ec.eu opa.eu
JRC133670
Se ille: Eu opean Commission, 2024
© Eu opean Union, 2024
The euse policy o he Eu opean Commission documen s is implemen ed by he Commission
Decision 2011/833/EU o 12 Decembe 2011 on he euse o Commission documen s (OJ L 330,
14.12.2011, p. 39). Unless o he wise no ed, he euse o his documen is au ho ised unde he
C ea i e Commons A ibu ion 4.0 In e na ional (CC BY 4.0) licence
(h ps://c ea i ecommons.o g/licenses/by/4.0/). This means ha euse is allowed p o ided app op ia e
c edi is gi en and any changes a e indica ed.
How o ci e his epo : Eu opean Commission, Join Resea ch Cen e, Ren occhini, F. Vezzani, A.,
Mon eso , S., Walking he G een Line: Go e nmen Sponso ed R&D and Clean Technologies,
Eu opean Commission, Se ille, 2024, JRC133670.
This documen is a publica ion by he Join Resea ch Cen e (JRC), he Eu opean Commission’s
science and knowledge se ice. I aims o p o ide
e idence-
based scien i ic suppo o he Eu opean
policymaking p ocess. The con en s o his publica ion do no necessa ily e lec he posi ion o
opinion o he Eu opean Commission. Nei he he Eu opean Commission no any pe son ac ing on
behal o he C
ommission is esponsible o he use ha migh be made o his publica ion. Fo
in o ma ion on he me hodology and quali y unde lying he da a used in his publica ion o which
he sou ce is nei he Eu os a no o he Commission se ices, use s should co
n ac he e e enced
sou ce. The designa ions employed and he p esen a ion o ma e ial on he maps do no imply he
exp ession o any opinion wha soe e on he pa o he Eu opean Union conce ning he legal
s a us o any coun y, e i o y, ci y o a ea o o i s au ho i ies, o conce ning he delimi a ion o i s
on ie s o bounda ies.
1
Con en s
Abs ac ....................................................................................................................................................................................................................................................................... 2
Acknowledgemen s .......................................................................................................................................................................................................................................... 3
Execu i e summa y .......................................................................................................................................................................................................................................... 4
1 In oduc ion..................................................................................................................................................................................................................................................... 5
2 Da a and desc ip i e e idence ..................................................................................................................................................................................................... 8
3 Es ima ion s a egy and a iables ......................................................................................................................................................................................... 11
4 Resul s .............................................................................................................................................................................................................................................................. 13
4.1 Co e es ima ions ....................................................................................................................................................................................................................... 13
4.2 Disen angling he spillo e s o go e nmen sponso ed R&D ........................................................................................................ 15
4.3 Robus ness es s ....................................................................................................................................................................................................................... 17
4.4 The dis ibu ional e ec s o go e nmen sponso ed R&D ............................................................................................................... 21
5 Concluding ema ks ............................................................................................................................................................................................................................. 24
Re e ences ............................................................................................................................................................................................................................................................. 25
Lis o igu es ..................................................................................................................................................................................................................................................... 28
Lis o ables ........................................................................................................................................................................................................................................................ 29
Annexes .................................................................................................................................................................................................................................................................... 30
A Da a Cons uc ion ...................................................................................................................................................................................................................................... 31
B. Robus ness Checks and Speci ica ions ................................................................................................................................................................................ 33
Misspeci ica ion o ou come and ea men models: doubly obus es ima o s ...................................................................... 33
Resul s om he IPW ea men model and al e na i e weigh s ............................................................................................................ 35
Al e na i e dis ibu ional e ec s o go e nmen sponso ed clean echnologies..................................................................... 41
2
Abs ac
We examine whe he go e nmen sponso ed R&D induces he de elopmen o clean echnologies
wi h a high impac on subsequen echnological de elopmen . The analysis uses in o ma ion on
USPTO pa en s g an ed be ween 2005 and 2015 and combines di e en me hods o con ol o
possible so ing o p ojec s in o public unding and o non- andom (public) ea men . We also
assess he dis ibu ional e ec o go e nmen sponso ed R&D. Resul s show ha pa en s om
public unded p ojec s ha e a signi ican ly highe impac and ha his is pa icula ly ue o highly
ci ed pa en s, hus suppo ing a ole o echnology-push policies in de e mining a clean
echnological ansi ion.
3
Acknowledgemen s
Au ho s a e lis ed in andom o de ollowing he Ame ican Economic Associa ion guidelines on
andom o de o co-au ho s. A consen o m has been signed by all co-au ho s, digi al con i ma ion
code: _5Dc WHF8lZK.
A p e ious e sion o his pape has been p esen ed a he ollowing e en s: GDS17 s udy g oup
semina , Roma T e Uni e si y, 9 h May 2022; EC-JRC IID semina se ies, 16 h May 2022; 1s annual
wo kshop o he eco-inno a ion socie y, Uni e si y o Fe a a, 10-11 No embe 2022; INNOVA
MEASURE V Final Wo kshop, JRC-ISPRA, 18 h Ap il 2023; IV wo kshop on eco-inno a ion, ci cula
economy, and i m pe o mance, Uni e si y o Ba celona, 20-21 Ap il 2023. We also hank wo
anonymous e iewe s and he edi o o he wo king pape se ies o hei help ul sugges ions.
F ancesco Ren occhini is cu en ly an employee o he Eu opean Commission. The iews exp essed
a e pu ely his pe sonal iews and may no in any ci cums ances be ega ded as s a ing an o icial
posi ion o he Eu opean Commission. The usual ca ea s apply.
Au ho s
F ancesco Ren occhini: Eu opean Commission, JRC Se ille and DEMM, Uni e si y o Milan.
Co esponding au ho a : Eu opean Commission, Join Resea ch Cen e (JRC). O ice 00/017, Edi icio
Expo, calle Inca Ga cilaso 3, 41092 Se illa, Spain. E-mail: ancesco. en [email protected] opa.eu
An onio Vezzani: Rennes School o Business, Depa men o S a egy and Inno a ion
Sand o Mon eso : G an Sasso Science Ins i u e.
The JRC Wo king Pape s on Co po a e R&D and Inno a ion a e published unde he edi o ial
supe ision o Alexande Tübke and James Ga igan in collabo a ion wi h So ia Ama al-Ga cia,
Fe nando He ás, Koen Jonke s, F ancesco Ren occhini a he Eu opean Commission – Join
Resea ch Cen e, and in coope a ion wi h Sa a Amo oso (Ge man Ins i u e o Economic Resea ch,
DEU), Michele Cince a (Sol ay B ussels School o Economics and Managemen , Uni e si é Lib e de
B uxelles, BEL), Alex Coad (Waseda Uni e si y, Tokyo, JAP), En ico San a elli (Uni e si y o Bologna,
ITA), Daniel Ve esy (In e na ional Telecommunica ion Union, CHE – and UNU-MERIT, NLD), An onio
Vezzani (Roma T e Uni e si y, ITA); Ma co Vi a elli (Uni e si à Ca olica del Sac o Cuo e, Milan, ITA)
and Zol an Cse al ay (Ma hias Co inus Collegium, HUN).
4
Execu i e summa y
This wo k p esen s a c i ical analysis o he ole o public policy in he de elopmen and adop ion o
clean echnologies, pa icula ly he in luence o go e nmen -sponso ed Resea ch and De elopmen
(R&D) p og ams. The s udy aims o p o ide empi ical e idence o he e ec o public R&D policies
on ad ancing clean echnologies wi h subs an ial knowledge spillo e s.
Despi e he global commi men o educe g eenhouse gas emissions, he le el o public suppo o
clean echnologies has been inconsis en ac oss OECD coun ies since 2011. The de elopmen o
clean echnologies has s agna ed, and he p i a e sec o 's incen i e o inno a e in his a ea seems
o ha e dec eased. This calls o a eassessmen o he e ec i eness o R&D policies in p omo ing
he de elopmen o clean echnologies.
The wo k unde sco es he need o public ac ion in he economics o clima e change. Ma ke -based
policies, such as ca bon- ax, a e no enough as hey ail o in e nalise he long- e m bene i s o
supe io , clean echnologies. The e o e, R&D subsidies a e necessa y o edi ec inno a ion om
di y o clean echnologies.
The s udy posi s ha science-push policies, such as public R&D, a e expec ed o ha e a p o ound
impac on new clean echnologies due o hei no el y and ole as ounda ional elemen s o
subsequen echnological ad ancemen s. Public R&D is c ucial in p omo ing he de elopmen o
high-impac clean echnologies, which o m he basis o a new echnological pa adigm.
In ou empi ical in es iga ion, we use pa en s g an ed by he USPTO be ween 2005 and 2015 linked
o p ocu emen con ac s o esea ch g an s wi h a US unding agency o examine he e ec o
echnology-push policy. The esul s e eal a signi ican impac o go e nmen -suppo ed clean
echnologies on subsequen inno a ions, wi h suppo ed echnologies ecei ing abou 26% mo e
ci a ions han non-suppo ed ones wi hin a 5-yea pe iod. This e ec was no ed among clean
echnologies wi h he highes impac on subsequen echnological de elopmen .
This analysis p o ides wo signi ican implica ions:
Clima e change policy modelling should acknowledge he po en ial in luence o policies on
he knowledge spillo e s o echnologies a he han ea ing hem as exogenous.
In he implemen a ion o clima e change policies, R&D suppo should accompany s anda d
ma ke pull in e en ions o expedi e echnical change owa ds sus ainable g ow h. This
a gumen p o ides a a ionale o e e sing he declining end in echnology suppo
policies obse ed in OECD coun ies since 2011.
In conclusion, his wo k highligh s he c i ical ole o go e nmen -sponso ed R&D in os e ing
impac ul clean echnologies. The indings p o ide a compelling a gumen o policy make s o
ein o ce hei R&D p og ams o achie e sus ainable g ow h and mi iga e clima e change.
5
1 In oduc ion
The educ ion o g eenhouse gas emissions is he mos impo an mean o mi iga e clima e change
(IPCC, 2022). Inno a ion policy packages ha e enabled cos educ ions and suppo ed global
adop ion o low-emission echnologies. Howe e , an impo an pa o emission educ ion will
depend on new clean echnologies ha a e s ill emb yonic and ma ked by high echnological and
ma ke unce ain y (IEA, 2021). This ep esen s an obs acle o p i a e ini ia i es and makes he
suppo o public policy c ucial o hei de elopmen . To his end, some go e nmen s ha e
ein o ced hei Resea ch and De elopmen (R&D) p og ams o os e en i onmen al inno a ion, like
in he US wi h he ARPA-E scheme and in he EU wi h he Inno a ion Fund.
Ne e heless, he public suppo o clean echnologies is less sys ema ic han i may appea . Ac oss
he OECD, om 2011 he (s ingency) le el o echnology suppo policies has declined un il 2016
and has hen expe ienced a sca e ed inc ease, bu wi hou eaching he 2011-peak (K use e al.,
2022). This has occu ed while he de elopmen o clean echnologies, as e ealed by
en i onmen al pa en s, has s opped g owing and emba ked along a con inuous slow down un il he
mos ecen yea s (Dechezlep ê e and K use, 2022; IEA, 2020). P i a e incen i es o de elop new
clean echnologies migh ha e dec eased and e idence abou he e ec i eness o R&D policies in
es o ing hem is hus needed o jus i y hei budge ing.
The ele ance o public ac ion is a well- ecognised in insic ea u e o he economics o clima e
change (S e n, 2008; No dhaus, 2019). Among he di e en le e ages, he ole o public suppo o
R&D has been less sc u inized compa ed o ma ke -based and egula o y app oaches. Despi e i s
asce ained ole in di ec ing echnical change owa ds sus ainable g ow h, a gap emains abou he
s eng h o public R&D in playing his ole: does i acili a e en i onmen al inno a ions ha ac as
s eppings ones o subsequen echnological de elopmen s?
F om a heo e ical poin o iew, a ecen s eam o endogenous g ow h models applied o he
en i onmen ha e shown ha policy is c ucial o he de elopmen o clean echnologies, gi en he
pa h-dependen na u e o echnical change (Acemoglu e al., 2012; 2016; Hémous and Olsen,
2021). A sole ma ke -based policy, such as ca bon- ax, is no enough (Acemoglu e al., 2012). As
he ma ke keeps on alloca ing esou ces o inno a ion by looking a immedia e p o i s, wi hou
e aining he discoun ed bene i s ha supe io echnologies will b ing o e he long un, he di y
echnology sec o may emain he i s bes alloca ion o incumben s e en wi h a ca bon ax. In
o de o edi ec inno a ion om di y o clean echnologies, R&D subsidies a e needed o make he
ma ke in e nalise he highe e u ns, p i a e and social, o clean echnologies in he long un.
Following he same backg ound, science-push policies like public R&D can be expec ed o ha e a
deepe impac on new clean echnologies, ela ed o hei no el y and hei ole as basic
componen s o subsequen echnological de elopmen (T aj enbe g e al., 1997). Unlike ma ke -
pull policies, which ac on p i a e incen i es in he sho un, science-push policies inc ease he
expec ed social alue o clean echnologies, whose ime ho izon is longe and p o ide in en o s wi h
incen i es o wo k on mo e adical inno a ions. Simila ly o wha has been ound o o he
echnologies (Acemoglu and Linn, 2004; D ano e e al., 2020; Dubois e al., 2015; Finkels ein,
2004), public R&D is a guably needed o spu he de elopmen o high impac clean echnologies,
which ep esen basic s eppings ones in he un olding o a new echnological pa adigm. Despi e he
in ui ion behind his a gumen , i s suppo ing heo e ical mechanisms ha e no been ully add essed
ye . Fu he mo e, he ex en o which his speci ic impac o R&D policy ac ually happens s ill lacks
sys ema ic empi ical e idence.
To ill his gap, we examine whe he go e nmen sponso ed R&D acili a es he de elopmen o
clean echnologies wi h la ge knowledge spillo e s on subsequen inno a ions. Gi en he pa h-
dependency ha cha ac e ises echnical change (Acemoglu e al., 2012), an impo an pa o he
policy impac in ac passes h ough he (knowledge) alue ha newly de eloped g een echnologies
ha e o he de elopmen o subsequen ones. Consis en ly wi h he pa h-dependency hypo hesis,
6
should policy induced clean echnologies be ma ked by la ge spillo e s, hey could os e he
di usion o clean knowledge h ough hei in luence on subsequen echnological de elopmen s.
We do expec his o happen by e e ing o i ms’ decisions o in es in adically new esea ch
p ojec s (Azoulay e al., 2019), which yield inno a i e ou comes o high impac in e ms o
knowledge spillo e s. These p ojec s a e ypically isky and ea ly-s age, and a e hus ma ked by
ma ginal cos s ha o e come hei ma ginal bene i s o a la ge ex en han lowe impac p ojec s.
This is due o di e en mechanisms. To s a wi h, high impac inno a i e ou comes inc ease he
isk o being imi a ed and his s imula es he inno a o o delay hei ealisa ion o e ime
(Mukhe jee and Pennings, 2004). Fu he mo e, ha ing a la ge impac na u ally inc eases he
ex e nali ies ha ea ly inno a o s can dynamically ha e on la e inno a o s, dec easing he e u ns
ha he o me can app op ia e and hus he incen i es o unde ake he ela i e in es men
(Sco chme , 1991). Fo hese easons, i ms gene ally ind unp o i able o in es in high impac
p ojec s and hei ealisa ion is hus c ucially linked o he public suppo . By he same oken, he
ma ginal echnology ha go e nmen sponso ed R&D suppo s, can be expec ed o ha e highe
knowledge spillo e s han non-suppo ed ones.
This di e en ial e ec has been ound by Azoulay e al. (2019) looking a he impac o scien i ic
g an s on i ms’ pa en ing in he pha maceu ical and bio echnology indus ies. The unde lying
mechanisms leading o hei esul s a e expec ed o hold also wi h espec o clean echnologies.
These a e echnologies whose de elopmen elies on he combina ion o mo e di e se and no el
echnological componen s han non-clean ones, and which hus equi e a la ge and mo e unce ain
cogni i e e o (Ba bie i e al., 2020). Fu he mo e, in es men s in clean echnologies ha e been
p o ed o yield posi i e e u ns o i ms only in he p esence o high ene gy cos s, hus inc easing
hei ma ke unce ain y (Popp, 2002). This u he cons ains he incen i es o i ms o in es in
new, high-impac clean echnologies. Such echnologies, a guably, a e mo e likely o ecei e suppo
om go e nmen -sponso ed R&D.
We in es iga e he ex en o which his is ac ually he case, by illing a gap in he empi ical esea ch
abou he de elopmen o clean echnologies, mainly ocused on en i onmen al policies ha ac on
he (p i a e) ma ke side, like: shocks inducing changes in ene gy p ices (Noailly and Smee s, 2015;
Hassle e al., 2021), emission ading sys ems (Calel and Dechezlep ê e 2016), changes in
emission s anda ds (Rozendaal and Vollebe gh, 2021), in e na ional en i onmen al ag eemen s
(Dugoua, 2021), and ca bon and en i onmen al axes (Aghion e al., 2016). Empi ical analyses o
R&D policies a e ins ead mo e sca e ed. Wi h espec o he au omobile indus y, Aghion e al.
(2016) showed ha empo a y R&D subsidies designed o inc ease ene gy e iciency can a ou he
de elopmen o inc emen al clean (g ey) inno a ions, while adically clean inno a ions emain
una ec ed. Wo king on he R&D g an s issued by he US Depa men o Ene gy, Howell (2017)
shows ha ecipien small businesses in clean ene gy sec o s inc ease hei pa en ing, VC inancing,
and su i al a e, while hese e ec s a e non-signi ican in con en ional (di y) ene gy echnologies
like na u al gas and coal. Addi ional e idence ega ding he impac o o he ypes o echnology
suppo policies, pa icula ly hose ocusing on demand-side s a egies such as public g een
p ocu emen , is limi ed and p ima ily ound in a ew selec wo ks (Ghise i 2017; K iege and
Zippe e , 2022).
We add o his s eam o empi ical esea ch ocusing on go e nmen sponso ed R&D in he US and
p o ide e idence o i s ole in os e ing he de elopmen o clean echnologies wi h a high impac
on subsequen inno a ions. We ely on pa en s g an ed a he USPTO be ween 2005 and 2015 o
applican s linked o a leas one p ocu emen con ac o esea ch g an wi h a US unding agency,
and in es iga e he impac o echnology-push policy h ough a quasi-expe imen al es ima ion
amewo k.
Using ci a ions om o he pa en s o p oxy he impac on subsequen echnological de elopmen ,
we show ha he e ec is ema kable in size: in a 5-yea s window go e nmen suppo ed clean
echnologies ha e abou 26% mo e ci a ions han non-suppo ed ones. The size o he e ec
emains sizeable also when we u he disen angle ci a ions o conside di e en ypes o
13
4 Resul s
4.1 Co e es ima ions
The p ima y objec i e o his pape is o iden i y pa ame e 𝛽𝛽, which ep esen s he e ec o
go e nmen R&D suppo on he de elopmen o impac ul clean echnologies. Howe e , be o e
discussing he main esul s we p esen es ima ions o he whole sample, including clean and non-
clean echnologies. The pu pose o his analysis is o in es iga e whe he go e nmen R&D suppo
has a di e en ia ed e ec be ween he wo in e ms o subsequen echnological de elopmen .
Table 2 epo s he esul s o equa ion (1) using he ull sample o pa en s, including a dummy o
clean echnologies and i s in e ac ion wi h G RD; ci a ions a e conside ed o a ime span o 5
yea s. Resul s in column 1 do no include applican ixed e ec s, which a e ins ead included in
column 2. Wi h he inclusion o applican ixed e ec s he coe icien a ached o go e nmen
sponso ed R&D change sign, om posi i e o nega i e. Due o he he e ogenei y o he es ima ion
sample in e ms o echnologies and applican s, we do no hink he esul s should be aken as
o e all e idence o he e ec o go e nmen suppo . Ins ead, he esul s sugges ha he
go e nmen migh end o ac o in he applican s’ po en ial when e alua ing p ojec s: his con i ms
ha in a-applican compa isons a e a key elemen o p ope ly assess he e ec o go e nmen
R&D suppo on subsequen in en ions.
Table 2 - E ec o go e nmen sponso ed R&D on subsequen inno a ion – all echnologies.
# i e-yea o wa d ci a ions
(1) (2)
G R&D 0.971** -1.134**
[0.262] [0.336]
Clean ech 4.375** 4.287**
[0.261] [0.261]
G R&D x clean ech 3.184* 2.278*
[1.267] [0.935]
o iginali y 6.775** 5.189**
[0.266] [0.260]
eam size 1.896** 1.804**
[0.071] [0.071]
# o applican s 0.010 -0.074
[0.070] [0.069]
Filing yea FE Yes Yes
14
Technology FE Yes Yes
Applican FE No Yes
F- es 71.030 40.983
R sq 0.047 0.117
N (Pa en s) 464,123 462,745
The uni o obse a ion is pa en applica ion o he ull sample. Dependen a iable in all columns is he numbe o i e-yea o wa d
ci a ions. All es ima es a e OLS and include echnology ixed e ec a he 3-digi CPC le el. Robus s anda d e o s in pa en hesis. +
p<0.1, * p<0.05, ** p<0.01.
Table 2 con i ms p e ious e idence showing ha clean echnologies ha e, on a e age, a highe
impac on subsequen in en ions compa ed o o he echnologies (Ba bie i e al., 2020;
Dechezlep e e e al., 2017). The able also poin s o a highe ci a ion p emium o go e nmen
sponso ed clean echnologies compa ed o non-go e nmen sponso ed ones. O e all, hese
p elimina y es ima es e eal ha clean echnologies beha e di e en ly han non-clean ones, also
when conside ing he ole o public suppo .
We now ocus on clean echnologies and p esen he main esul s o he pape . Table 3 epo s he
es ima es o equa ion (1) wi h espec o clean echnologies, showing he absolu e and ela i e
e ec o go e nmen R&D suppo on subsequen in en ions. In column 1 we epo he esul s o a
linea speci ica ion wi hou using in e se p obabili y weigh ing scheme and applican ixed e ec s.
We hen epo esul s ob ained by applying he in e se-p obabili y weigh ing (column 2) and by
adding applican ixed e ec s (column 3). The compa ison o he ela i e A e age T ea men E ec
(ATE) ac oss columns, epo ed a he bo om o he able, allows us o e alua e he bias educ ion
de i ing om he en ichmen o he es ima ion app oach.
Table 3 - E ec o go e nmen sponso ed R&D on subsequen inno a ion – clean echnologies.
# i e-yea o wa d ci a ions
(1) (2) (3)
G R&D 6.472** 6.107** 4.427**
[1.346] [1.341] [1.109]
o iginali y 26.221** 32.572** 16.635**
[1.409] [2.322] [1.819]
Team size 4.272** 5.103** 3.520**
[0.384] [0.639] [0.420]
# o applican s -0.951** -1.420* -0.864*
[0.362] [0.589] [0.436]
15
Filing yea FE Yes Yes Yes
Technology FE Yes Yes Yes
Applican FE No No Yes
Weigh ing scheme No IPW IPW
F es 29.197 20.829 16.011
Rela i e ATE 0.394** 0.360** 0.258**
[0.084] [0.082] [0.066]
R sq 0.053 0.052 0.421
N (Pa en s) 38,729 36,726 36,245
The uni o obse a ion is pa en applica ion o he sample o clean ech pa en s. Dependen a iable in all columns is he numbe o
i e-yea o wa d ci a ions. Rela i e ATEs a e compu ed as he ela i e di e ence be ween po en ial ou come means o ea ed and
un ea ed g oups (∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑡𝑡
𝑖𝑖𝑁𝑁𝑡𝑡
�-∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑢𝑢
𝑖𝑖𝑁𝑁𝑢𝑢
�)/ ∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑢𝑢
𝑖𝑖𝑁𝑁𝑢𝑢
� whe e 𝑦𝑦𝚤𝚤
� is he p edic ed alue om he ele an eg ession model. Columns 2
and 3 a e es ima ed using 𝑝𝑝(𝑥𝑥𝑖𝑖)/(1 − 𝑝𝑝(𝑥𝑥𝑖𝑖)) o weigh un ea ed obse a ions and 1 o he wise. 𝑝𝑝(𝑥𝑥𝑖𝑖) is he p opensi y sco e
calcula ed as pe Table B.2 in appendix B. Figu es B.1 and B.2 in Appendix B epo s a is ics ela i e o he p opensi y sco e
p ocedu e used o compu e weigh s, showing a good pe o mance in e ms o bias educ ion. All es ima es a e OLS and include
echnology ixed e ec a he 4-digi CPC le el. Robus s anda d e o s in pa en hesis. + p<0.1, * p<0.05, ** p<0.01.
The e ec o go e nmen R&D suppo is high and signi ican in all he speci ica ions. In e es ingly,
he bias educ ion is pa icula ly s ong when including applican ixed e ec s: he use o p opensi y
sco e lowe s he ela i e ATE ( he ATE in e ms o po en ial ou come) o public suppo by 5.7
pe cen age poin s, while when adding he ixed e ec s his educes by an addi ional 10.7
pe cen age poin s.
Acco ding o ou p e e ed speci ica ion (column 3), go e nmen suppo ed clean echnologies ha e
abou 26% mo e ci a ions han non-suppo ed ones wi hin a 5-yea window. In line wi h
expec a ions, pa en s wi h a highe o iginali y and a la ge eam size ecei e on a e age mo e
ci a ions, while somehow unexpec edly pa en s collabo a ed be ween mo e applican s show a lowe
numbe o ci a ions. This las esul may sugges ha applican s end o de elop hei mo e
p omising R&D p ojec s (in e ms o po en ial u u e impac ) alone o in small collabo a i e se ings.
4.2 Disen angling he spillo e s o go e nmen sponso ed R&D
In his sec ion, we u he disen angle he impac o go e nmen suppo in he de elopmen o
impac ul clean echnologies by conside ing di e en ypes o knowledge spillo e s.
Fi s , we conside he knowledge spillo e s gene a ed by public R&D suppo ou side he sphe e o
he ocal applican , by excluding hose gene a ed by i s own ci a ions. To clean ou dependen
a iable om sel -ci a ions we use wo di e en app oaches: i) we elimina e a ci a ion only i i is
comple ely de e mined by he same applican o he ci ed pa en (nosel _nos ic); ii) we elimina e a
ci a ion i a leas one applican in he ci ing pa en is also in he ci ed pa en (nosel _s ic ). In he
o me case, we ule ou only sha p “in a-applican ” spillo e s, bu s ill allow o knowledge
spillo e s de i ing om collabo a ions o he same applican in subsequen p ojec s (i a pa en
wi h applican s A and B ci e a pa en o applican A, we s ill coun he ci a ion). In he second case,
we cap u e pu e knowledge spillo e s, i.e. in e -applican spillo e s, because none o he ci ing
16
applican s should be in ol ed in he de elopmen o he ci ed pa en . The esul s a e epo ed in
columns (1) and (2) o Table 4.
Second, we ocus on he knowledge spillo e s ha low om he applican s ha ha e ecei ed
go e nmen R&D suppo o he es o he wo ld. To do his, we exclude om ou dependen
a iable all he ci a ions om pa en s egis e ed by applican s ha ha e ecei ed a leas one US
public con ac as eco ded by he 3PFL. Like in he case o sel -ci a ions, we compu e hese
spillo e s using a non-s ic and a s ic de ini ion; esul s a e epo ed in columns (3) and (4).
Finally, we assess whe he go e nmen R&D suppo helps gene a e clean echnologies wi h highe
in e na ional knowledge spillo e s, hus ha ing a highe impac on he subsequen echnological
de elopmen in o he economies. To do so, we build wo a iables: i) ci _US, a bina y a iable aking
alue 1 is all ci a ions o a gi en pa en a e gene a ed only by US applican s (column 5); ii)
geo_b ead h, coun ing he numbe o applican ’s coun ies o he ci ing pa en s (column 6).
The esul s epo ed in Table 4 sugges ha spillo e s e ec s a e a he s ong and no localized
among he applican s ecei ing public suppo . In ac , he coe icien a ached o go e nmen
sponso ed R&D is s a is ically signi ican a he usual le el in columns 1 o 4. The magni ude o he
coe icien dec eases when using he s ic e de ini ion o spillo e s, which is consis en wi h he
educed numbe o ci a ions conside ed. Howe e , he ela i e a e age ea men e ec s do no
a y subs an ially om he main es ima ions (see Table 3, column 3), con i ming he sizeable e ec
o go e nmen suppo on ollow up in en ions.
Table 4 - E ec o go e nmen sponso ed R&D on di e en kinds o knowledge spillo e s, clean echnologies
(1) (2) (3) (4) (5) (6)
nosel
nos ic
nosel
s ic
ou
nos ic
ou
s ic
ci US geo
b ead h
G R&D 4.376** 3.151** 4.350** 2.176** 0.006 0.137+
[1.017] [0.866] [1.051] [0.672] [0.007] [0.073]
o iginali y 15.795** 12.856** 16.104** 9.132** 0.171** 1.453**
[1.764] [1.655] [1.800] [1.241] [0.018] [0.119]
Team size 3.049** 2.773** 3.259** 2.110** 0.010** 0.238**
[0.370] [0.360] [0.389] [0.262] [0.002] [0.019]
# o applican s -0.692+ -0.824* -0.783+ -0.562* -
0.003*
-0.051**
[0.380] [0.372] [0.401] [0.268] [0.002] [0.019]
Filing yea FE Yes Yes Yes Yes Yes Yes
Technology FE Yes Yes Yes Yes Yes Yes
Applican FE Yes Yes Yes Yes Yes Yes
17
Weigh ing scheme IPW IPW IPW IPW IPW IPW
Rela i e ATE 0.286** 0.256** 0.274** 0.230** 0.007** 0.041+
[0.067] [0.071] [0.067] [0.072] [0.009] [0.022]
N (Pa en s) 36,245 36,245 36,245 36,245 36,245 36,245
The uni o obse a ion is pa en applica ion o he sample o clean ech pa en s. Rela i e ATEs a e compu ed as ela i e di e ence
be ween po en ial ou come means o ea ed and un ea ed g oups (∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑡𝑡
𝑖𝑖𝑁𝑁𝑡𝑡
�-∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑢𝑢
𝑖𝑖𝑁𝑁𝑢𝑢
�)/ ∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑢𝑢
𝑖𝑖𝑁𝑁𝑢𝑢
� whe e 𝑦𝑦𝚤𝚤
� is he p edic ed
alue om he ele an eg ession model. All columns a e es ima ed using 𝑝𝑝(𝑥𝑥𝑖𝑖)/(1 − 𝑝𝑝(𝑥𝑥𝑖𝑖)) o weigh un ea ed obse a ions and
1 o he wise. 𝑝𝑝(𝑥𝑥𝑖𝑖) is he p opensi y sco e calcula ed as pe Table B2 in Appendix B. Fo he eg ession adjus men ia p opensi y
sco e, we en o ce a common suppo by emo ing he 5% o he ea men obse a ions a which he p opensi y sco e densi y o
he con ol obse a ions is a a minimum. All es ima es a e OLS and include echnology ixed e ec a he 4-digi CPC le el. Robus
s anda d e o s in pa en hesis. + p<0.1, * p<0.05, ** p<0.01.
Resul s epo ed in columns 5 and 6 show ha he spillo e s gene a ed by go e nmen suppo ed
clean pa en s a e no mo e localized wi hin he US han p i a ely unded ones bu ha e a highe
geog aphical b ead h: he echnology is u he de eloped by applican s om a la ge baske o
coun ies. This sugges s ha spillo e s om go e nmen suppo ed clean echnologies may di e
om hose in o he a eas o in e en ion. Mo eo e , in line wi h models o g ow h emphasizing he
s a egic complemen a i ies in clean esea ch ac oss coun ies (Aghion e al., 2015; Dechezlep e e
e al., 2017), he la ge geog aphical scope o ci a ions is consis en wi h he idea ha policies
suppo ing he de elopmen o clean echnologies can ha e an e ec in os e ing he global
gene a ion o such echnologies.
4.3 Robus ness es s
In his sec ion we p esen a ba e y o obus ness checks o u he co obo a e ou main esul s.
Fi s , while in ou analysis we con ol o di e en ypes o spillo e s and o obse able di e ences
in po en ial ou comes, wi h he da a a s ake we a e no able o alloca e budge a he pa en le el.
The budge o go e nmen suppo ed p ojec s is no homogeneous and can be alloca ed o di e en
ac i i ies o he han R&D o echnological de elopmen ; mo eo e , he e a e cases o mul iple
pa en s linked o he same go e nmen suppo ed p ojec . This mean ha he ea men migh be
no homogeneous, and we canno di ec ly model i . To assess whe he he he e ogenei y in he
go e nmen R&D suppo can be an issue when ying o iden i y i s e ec on ollow up ci a ions,
we un wo ancilla y eg essions a he applican le el. In he i s , we assess whe he highe
amoun s o go e nmen suppo induce mo e clean ech pa en s. In he second, we assess whe he
highe amoun s o go e nmen suppo lead o mo e ci a ions o he o e all po olio o suppo ed
pa en s once con olling o hei numbe . I he coe icien a ached o go e nmen suppo is
s a is ically signi ican in he o me bu no in he la e , his would be indi ec e idence ha he
he e ogenei y in he ea men is no a majo issue when assessing i s e ec on he impac o
suppo ed clean ech on ollow up inno a ions.
The esul s epo ed in Table 5 sugges ha his is he case. Highe amoun s o go e nmen R&D
suppo lead o a highe numbe o clean ech pa en s, bu once con olling o clean ech pa en s i
has no e ec on he numbe o ci a ions ecei ed.
18
Table 5: The e ec o go e nmen sponso ed R&D on he quan i y and quali y o in en ions - clean ech sample
(1)
# g een
G pa en s
(2)
# ci a ions o
g een pa en s
Amoun o G R&D
0.082** -1.497
[0.028] [0.996]
# g een G pa en s 31.500**
[11.378]
A g. eam size 0.128 37.426**
[0.123] [11.398]
A g. # o applican s 0.098 -0.014
[0.089] [10.099]
A g. o iginali y -0.165 170.284
[1.185] [88.706]
Cons an 2.483* -260.113*
[1.197] [103.976]
R sq 0.320 0.328
N (applican s) 868 868
Reg essions a e based on he 868 applican s ha ecei ed go e nmen sponso ed R&D and de eloped g een echnologies. The dependen
a iable in column 1 is he numbe o g een pa en s sponso ed by public R&D and in column 2 he numbe o ci a ions ecei ed by
g een go e nmen R&D sponso ed pa en s. G R&D amoun is he o al amoun ecei ed by he applican . Es ima es in bo h
columns include con ols o he a e age numbe o in en o s, a e age numbe o applican s, he a e age o iginali y index and he
sha e o pa en s in di e en clean ech echnological classes. Column 2 includes also he numbe o numbe o g een pa en s
sponso ed by public R&D. + p<0.1, * p<0.05, ** p<0.01
Second, we un a andomised alsi ica ion es by andomly assigning ea men ac oss ou sample
o clean ech pa en s while keeping he sha e o ea ed pa en s cons an . We build 100 di e en
eplica ions o ou a ou i e speci ica ion (Column 3 o Table 3) o assess whe he a placebo
go e nmen suppo would s ill exe a posi i e e ec on o wa d ci a ions. Figu e 2 displays he
coe icien s a ached o he placebo go e nmen R&D suppo o he 100 eplica ions: he igu e
shows ha we canno ejec he hypo hesis ha he coe icien is equal o ze o, p o iding u he
e idence suppo ing ou esul s.
19
Figu e 2: Randomised alsi ica ion es
In he igu e a e epo ed he coe icien s a ached o GVT R&D (equa ion 1) when ea men is andomly dis ibu ed ac oss clean
pa en s. As in ou p e e ed speci ica ion (Table 3, column 3), he es ima ions include applican ixed e ec s and p obabili y weigh s.
Thi d, we e- un ou a ou i e speci ica ion by weigh ing obse a ions by he in e se o he size o
applican s’ clean ech pa en po olios. In ou sample, he numbe o applican s’ obse a ions is
p opo ional o he numbe o hei clean ech pa en s; gi ing he same weigh o each applican ,
his eweigh ing scheme ensu es ha esul s a e no d i en by applican s wi h la ge po olios. The
esul s, epo ed in Table 6 column 1, con i m he posi i e e ec o go e nmen suppo .
Table 2: E ec o g een go e nmen sponso ed R&D on subsequen inno a ion – obus ness checks
# i e-yea o wa d
ci a ions
# se en-yea
o wa d
ci a ions
# i e-yea o wa d ci a ions
(1) (2) (3) (4) (5) (6)
G sponso ed R&D 6.593* 6.257** 6.411** 0.233** 3.733** 5.467*
[2.872] [1.232] [1.441] [0.048] [1.091] [2.436]
o iginali y index 32.189** 9.774** 20.327** 1.435** 16.433** 18.574**
[3.647] [1.287] [2.112] [0.134] [1.805] [4.651]
20
Team size 6.222** 1.961** 4.491** 0.127** 3.446** 1.671**
[1.071] [0.254] [0.471] [0.009] [0.411] [0.349]
# o applican s 0.056 -0.296 -1.078* -0.014 -0.900* 0.236
[1.193] [0.271] [0.478] [0.010] [0.424] [0.316]
Filing yea FE Yes Yes Yes Yes Yes Yes
Technology FE Yes Yes Yes Yes Yes Yes
Applican FE No Yes Yes Yes Yes Yes
Filing yea X Tech FE No No No No Yes No
Model OLS OLS OLS Poisson OLS OLS
Weigh ing scheme GPAT IPW IPW IPW IPW IPW
SEs Robus Robus Robus Robus Robus CPC-Yea
F es 5.826 15.888 17.994 . 52.906 5.103
Rel ATE 0.358* 0.445** 0.284** 0.262** 0.216** 0.345*
Rel ATE SE [0.160] [0.090] [0.065] [0.060] [0.065] [0.155]
R sq 0.085 0.287 0.423 . 0.425 0.424
N (Pa en s) 36,245 25,025 36,245 36,238 36,244 31,058
The uni o obse a ion is pa en applica ion o he sample o clean ech pa en s. Dependen a iables a e he numbe o i e-yea
o wa d ci a ions in columns 1, 2, 4, 5 and 6 and he numbe o se en-yea o wa d ci a ions in column 3. Rela i e ATEs a e
compu ed as ela i e di e ence be ween po en ial ou come means o ea ed and un ea ed g oups (∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑡𝑡
𝑖𝑖𝑁𝑁𝑡𝑡
�-
∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑢𝑢
𝑖𝑖𝑁𝑁𝑢𝑢
�)/ ∑𝑦𝑦𝚤𝚤
�
𝑁𝑁𝑢𝑢
𝑖𝑖𝑁𝑁𝑢𝑢
� whe e 𝑦𝑦𝚤𝚤
� is he p edic ed alue om he ele an eg ession model. Column 1 eg ession is weigh ed by he
in e se o he numbe o clea ech pa en s in he applican po olio. Columns 2 and 3 a e es ima ed using 𝑝𝑝(𝑥𝑥𝑖𝑖)/(1 − 𝑝𝑝(𝑥𝑥𝑖𝑖)) o
weigh un ea ed obse a ions and 1 o he wise. 𝑝𝑝(𝑥𝑥𝑖𝑖) is he p opensi y sco e calcula ed as pe Table B2 in Appendix B. Fo he
eg ession adjus men ia p opensi y sco e, we en o ce a common suppo by emo ing he 5% o he ea men obse a ions a
which he p opensi y sco e densi y o he con ol obse a ions is a a minimum. All es ima es a e OLS and include echnology ixed
e ec a he 4-digi CPC le el. Robus s anda d e o s in pa en hesis. + p<0.1, * p<0.05, ** p<0.01.
Fou h, we es he obus ness o ou esul s agains a mo e conse a i e de ini ion o clean
echnologies. We build upon Dechezlep ê e e al. (2021) and de ine clean pa en s using 4-digi CPC
echnological classes Y02B, Y02C, Y02E and Y02T.2 Resul s om he es ima ion on his sub-se o
clean echnologies, epo ed in Table 6 column 2, s ill con i m ou main esul s and possibly
magni ies he e ec o go e nmen suppo in e ms o ela i e ATE. Column 3 in Table 6 u he
shows he esul s when using a 7-yea s window in he compu a ion o he numbe o o wa d
ci a ions. While he coe icien a ached o go e nmen suppo is highe han ha o he 5-yea s
window, he e ec o go e nmen suppo in e ms o ela i e ATE is in line wi h ou main esul s.
2 These ou CPC codes g oups echnologies ela ed o buildings, o GHG cap u e and s o age, o educ ion o GHG in ene gy p oduc ion
and dis ibu ion, and o anspo a ion. In o he wo ds, om ou de ini ion o clean ech we d op pa en s wi h codes: Y02A, Y02D,
Y02P, Y02W and Y04S (see igu e 1 o sho labels).
21
Column 4 in Table 6 p esen s he esul s o a (pseudo-)Poisson eg ession model wi h mul iple
high-dimensional ixed e ec s (Co eia, 2020), ensu ing he obus ness o ou indings o he coun
da a o ma o ou dependen a iable (numbe o ci a ions). Column 5 inco po a es ixed e ec s
o each combina ion o pa en iling yea and echnological class (CPC a he subclass le el, e.g.
Y02A). Finally, Column 6 epo s esul s wi h s anda d e o s clus e ed a he same le el (pa en
iling yea - echnological class le el).3 Reassu ingly, all obus ness checks p oduce ou comes
consis en wi h ou baseline esul s.
4.4 The dis ibu ional e ec s o go e nmen sponso ed R&D
In his sec ion, we assess he dis ibu ional e ec s o R&D go e nmen suppo on ollow up
inno a ions. To do so, we implemen he ecen e ed in luence unc ion (RIF) o he uncondi ional
quan ile (Fi po e al., 2009), o e alua e he impac o changes in he non- ea ed o ea ed
(go e nmen suppo ed) s a us on quan iles o he ma ginal dis ibu ion o o wa d ci a ions.
P e ious esul s could be e en ually in e p e ed as ollows: he go e nmen is able o sys ema ically
de ine/selec echnological needs/solu ions wi h a po en ial highe impac on ollow up inno a ions,
compa ed o o he ac o s in he economy. Despi e lock-in e ec s and he possible local sea ch
pe o med by p i a e ac o s, his is a a he s ong esul o be pu o wa d. One can ins ead expec
ha in mos cases he go e nmen suppo will no ha e a eal e ec . Con e sely, i can be
expec ed o incen i ise esea ch in esea ch a eas wi h lowe expec ed p i a e e u ns in he sho
e m and hus inc ease he p obabili y ha some key echnologies a e de eloped. In o he wo ds,
he (a e age) impac o go e nmen R&D suppo discussed abo e may be d i en by he highly
ci ed pa en s in he sample.
The e o e, he ele ance o assessing he dis ibu ional aspec s o he e ec ound in he p e ious
sec ions de i es om he assump ions unde lying a e age e ec s and he in e p e a ion o he
ela i e esul s. Figu e 3 shows he dis ibu ion o ci a ions o clean ech pa en s compa ing he
g oup o pa en s esul ing om p ojec s suppo ed by he go e nmen wi h ha no o igina ed by
go e nmen suppo ed R&D p ojec s. The igu e shows ha he dis ibu ion o pa en s ci a ions is
a he skewed and sugges s ha di e ences be ween go e nmen suppo ed and no suppo ed
pa en s a e no cons an along he dis ibu ion. A he bo om o he ci a ion dis ibu ion suppo ed
and no suppo ed pa en s do no di e much, wi h he la e showing a sligh ly highe ac ion o
pa en s. Ins ead, go e nmen sponso ed clean ech pa en s a e mo e equen in he uppe ail o
he ci a ion dis ibu ion, sugges ing ha he e ec o go e nmen suppo ope a es h ough he
de elopmen o mos impac ul clean echnologies.
3 Recen discussions in he econome ics li e a u e sugges ha clus e ed s anda d e o s may be o e ly conse a i e, and he
app op ia e le el o clus e ing should be ca e ully chosen (Abadie e al., 2023). In ou s udy, i is no ad isable o clus e s anda d
e o s a he applican le el, due o he high numbe o clus e s (1,400), many o which ha e ew obse a ions. Consequen ly, we
ha e decided o clus e s anda d e o s a he echnology-yea le el, aligning wi h he e idence o a non- andom dis ibu ion o
go e nmen suppo a he echnology le el (see Figu e 1) and po en ial yea ly di e ences, esul ing in 144 clus e s. Howe e , we
had o exclude 5,187 pa en s because hey we e assigned o mo e han one 4-digi CPC code. Fo his eason, we do no apply
clus e ed s anda d e o s in all speci ica ions.
22
Figu e 3: F ac ion o clean ech pa en s o go e nmen sponso ed and non-sponso ed pa en s, by numbe o ci a ions.
Table 7 epo s he esul s o he RIF es ima ions on he quan iles o he ci a ions dis ibu ions.
Consis en ly wi h he a gumen and he desc ip i e e idence abo e, he e ec o he go e nmen
R&D suppo is isible only among he hi d and ou h quin ile o he ci a ions’ dis ibu ion, whe e i
is quan i iable in abou 8.1% and 8.6% ci a ions wi h espec o he espec i e po en ial ou come.
In e es ingly, also he coe icien s a ached o o iginali y sizably inc eases along he quin ile o he
dis ibu ions, con i ming he gene al esul ha mo e o iginal (and mo e isky) clean pa en s may
ha e a highe impac on ollow up inno a ion.
Table 3: E ec o go e nmen sponso ed R&D on subsequen inno a ion, dis ibu ional e ec s ia RIF eg essions o
clean ech
(1) (2) (3) (4)
20 h quan ile 40 h quan ile 60 h quan ile 80 h quan ile
G R&D 0.160 0.264 0.722* 1.610*
[0.101] [0.177] [0.343] [0.821]
o iginali y 1.827** 3.120** 5.982** 15.097**
[0.220] [0.311] [0.538] [1.320]
29
Lis o ables
Table 1: Summa y s a is ics .......................................................................................... 8
Table 2 - E ec o go e nmen sponso ed R&D on subsequen inno a ion – all echnologies. ................13
Table 3 - E ec o go e nmen sponso ed R&D on subsequen inno a ion – clean echnologies. ............14
Table 4 - E ec o go e nmen sponso ed R&D on di e en kinds o knowledge spillo e s, clean echnologies
..........................................................................................................................16
Table 5: The e ec o go e nmen sponso ed R&D on he quan i y and quali y o in en ions - clean ech
sample ..................................................................................................................18
Table 6: E ec o g een go e nmen sponso ed R&D on subsequen inno a ion – obus ness checks .......19
Table 7: E ec o go e nmen sponso ed R&D on subsequen inno a ion, dis ibu ional e ec s ia RIF
eg essions o clean ech ............................................................................................22
Table A.1: Co espondence be ween numbe o con ac o s in 3PFL and numbe o applican s in Pa s a o
clean ech pa en s. ....................................................................................................31
Table B.1: E ec o g een go e nmen sponso ed R&D on subsequen clean inno a ion – doubly obus
es ima o s (AIPW and IPWRA) ........................................................................................33
Table B.2: Selec ion in o g een go e nmen sponso ed R&D – p opensi y sco e. ...............................35
Table B.3: E ec o g een go e nmen sponso ed R&D on subsequen inno a ion – al e na i e weigh s o
eg ession adjus men ................................................................................................40
30
Annexes
31
A Da a Cons uc ion
F om he 3PFL da abase (de Rassen osse e al., 2019) we e ie e in o ma ion on p ocu emen
con ac s and esea ch g an s signed by he US go e nmen , and on he pa en s iled o p o ec he
esul ing in en ions. The 3PFL da abase comp ises in o ma ion o 37,925 pa en s g an ed by he
USPTO be ween 2005 and 2015.
In o de o c ea e he con ol g oup o non- ea ed pa en s we p oceeded as ollows:
i) The 3PFL does no p o ide a code ha can be di ec ly used in Pa s a o iden i y he pa en
applican s. The e o e, we ha e used he pa en _n ield ( he numbe ha iden i ies he publica ion
o he g an ed applica ion a he USPTO) om he 3PFL da abase o e ie e all he pe son
iden i ie s associa ed o 3PFL pa en s om Pa s a . This educes he sample o pa en s o 37,003
as we excluded pa en s whe e he in en o was also he applican , and no o he applican was
epo ed in PATSTAT.
Table A.1 shows, o each 3PFL pa en he numbe o applican s in Pa s a and he numbe o
con ac o s in 3PFL. In mos cases he co espondence was 1:1 and he applican -con ac o pai
di ec ly iden i ied.
Table A.1: Co espondence be ween numbe o con ac o s in 3PFL and numbe o applican s in Pa s a o clean ech
pa en s.
Numbe o con ac o s in 3PFL
1 2 3 4 5 6 7 8 9
Numbe o applican s
in PATSTAT
1
30,213
3,002 332 40 13 2 1
1
2
2,309 483 93 24 1
1
3
303 70 28 2 1
4
55 13 2 1
5
10 3
6
7
1
No e: he selec ion excluded pa en s whe e he in en o was also he applican and no o he applican was epo ed.
ii) In all he cases whe e he ma ching did no esul in a 1:1 o 1:many co espondence we ha e
disambigua ed he ma ching using he names o applican s and con ac o s in he wo da abases.
32
In all he cases whe e he numbe o applican s and con ac o s was he same ( i s ow and i s
column o Table A.1) we associa ed he ela i e en ies using he Le ensh ein dis ance, which
p o ided a measu e o simila i y be ween he names epo ed in he wo da abases. All o he cases
ha e been manually disambigua ed. As a esul , we c ea ed a co espondence able be ween
applican codes (pe son_id) in Pa s a and he con ac o iden i ie epo ed in 3PFL.
iii) We used he disambigua ed lis o pe son_id o e ie e all he pa en s g an ed o he same
applican o e he 2005-2015 pe iod. These pa en s ep esen he con ol g oup.
33
B. Robus ness Checks and Speci ica ions
Misspeci ica ion o ou come and ea men models: doubly obus es ima o s
To assess he obus ness o ou indings o possible issues ela ed o misspeci ica ion o he
selec ion in o ea men model we ely on he augmen ed in e se-p obabili y-weigh ed (AIPW) and
he in e se-p obabili y-weigh ed eg ession-adjus men (IPWRA) es ima o s.
While he IPW es ima o used in he main analysis models only he ea men p obabili y, he AIPW
es ima o model bo h he ou come and he ea men p obabili y. The ad an age o he AIPW is ha
i is enough ha only one o he wo models is co ec ly speci ied o consis en ly es ima e he
ea men e ec ; o his eason, his ype o es ima o is known as a p ope y known as being. The
AIPW es ima o includes an augmen a ion e m ha co ec s he es ima o when he ea men
model is inco ec . This augmen a ion e m anishes when he ea men is p ope ly speci ied, and
he sample size is la ge.
Simila ly, in e se-p obabili y-weigh ed eg ession-adjus men (IPWRA) es ima o s in eg a e models
o he ou come and ea men s a us and possess he double obus ness p ope y. IPWRA
es ima o s u ilize he in e se o he es ima ed ea men -p obabili y weigh s o es ima e eg ession
coe icien s ha co ec o missing da a. These coe icien s a e hen used o calcula e po en ial
ou come means (Woold idge, 2010).
To he bes o ou knowledge, he e is no li e a u e ha compa es he ela i e e iciency o AIPW
and IPWRA es ima o s, so we epo esul s om bo h app oaches in he able below. Resul s show
high and signi ican coe icien s. As expec ed, coe icien s a e highe han in ou a o i e
speci ica ion (Column 3 in Table 3 in he main ex ) as AIPW and IPWRA es ima o s do no allow o
con ol o he applican ixed e ec s.
Table B.1: E ec o g een go e nmen sponso ed R&D on subsequen clean inno a ion – doubly obus es ima o s (AIPW
and IPWRA)
(1) (2)
G R&D 9.074** 9.250**
[1.801] [1.761]
Con ols Yes Yes
Filing yea FE Yes Yes
Technology FE Yes Yes
Applican FE No No
Es ima o AIPW IPWRA
34
Rela i e ATE 0.566** 0.577**
[0.114] [0.111]
N (Pa en s) 36,726 36,726
+ p<0.1, * p<0.05, ** p<0.01
35
Resul s om he IPW ea men model and al e na i e weigh s
Table B.2: Selec ion in o g een go e nmen sponso ed R&D – p opensi y sco e.
(1)
o iginali y 0.458**
[0.065]
Team size 0.015**
[0.006]
# o applican s 0.014*
[0.006]
CC adap a ion (Y02A) 0.591**
[0.033]
CCMT buildings (Y02B) -0.158**
[0.047]
GHG cap u e (Y02C) 0.040
[0.057]
CCMT ICT (Y02D) -0.763**
[0.051]
GHG ene gy (Y02E) 0.297**
[0.027]
CCMT p oduc ion (Y02P) -0.007
[0.027]
CCMT anspo (Y02T) -0.281**
[0.030]
CCMT was e (Y02W) 0.135+
[0.072]
ICT o ene gy (Y04S) -0.320**
36
[0.069]
Filing yea FE Yes
Technology FE Yes
Chi2 es 1643.711
McFadden's R sq 0.070
N (Pa en s) 38657
+ p<0.1, * p<0.05, ** p<0.01
37
Figu e B.1: Va iance a io o esiduals s bias be o e and a e ma ching
38
Figu e B.2: Co a ia e bias be o e and a e ma ching
We es he obus ness o ou esul s o al e na i e weigh s used o eg ession adjus men . Fi s ,
we implemen wo di e en ly de ined weigh s coming om he p opensi y sco e calcula ion. The
i s weigh ebalances he ea ed g oup only: i akes alue (1 − 𝑝𝑝𝚤𝚤
�)/𝑝𝑝 o he pa en s ha ing
ecei ed go e nmen R&D suppo and 1 o he wise (wi h 𝑝𝑝𝚤𝚤
�being he i ed alue om Table B2
abo e. The second weigh is a s anda d in e se p obabili y weigh aking alue 1/𝑝𝑝 o clean ech
pa en s ecei ing go e nmen R&D suppo and 1/(1 − 𝑝𝑝) o he wise. Following exis ing wo k
(Hi ano el al., 2003; B unell and Di Na do, 2004), bo h weigh s a e compu ed p ese ing p opo ions
be ween he ea ed and un ea ed g oup. Finally, we compu e weigh s om a coa sened ma ching
p ocedu e (Iacus e al., 2012). Figu es B.3 epo s compa ison be ween ea ed and un ea ed
pa en s in ela ion o global and local imbalance measu es. The global imbalance s a is ic is
calcula ed as he local imbalance measu es di e ence be ween he mul idimensional his og am o
p e ea men co a ia es in he ea ed g oup and he same in he con ol g oup. In ou speci ic case,
he alue o 0.672 is he e e ence poin o he unma ched da a, and a dec ease in he alue a e
ma ching (0.597) indica es a educ ion in he le el o imbalance. Simila educ ions in he local
imbalance measu es a e ound o he indi idual a iables. Figu e B.4 p o ides a compa ison o
a iable means be o e and a e ma ching and shows an impo an educ ion in bias ollowing he
ma ching p ocedu e. O e all, he wo igu es eassu e us abou he abili y o he chosen app oach
o educe bias om obse ables.
