Import competition and firm‐level CO emissions: Evidence from the German manufacturing industry

Leh , Jakob
A icle — Published Ve sion
Impo compe i ion and i m‐le el CO emissions: E idence
om he Ge man manu ac u ing indus y
Canadian Jou nal o Economics/Re ue canadienne d'économique
P o ided in Coope a ion wi h:
John Wiley & Sons
Sugges ed Ci a ion: Leh , Jakob (2025) : Impo compe i ion and i m‐le el CO emissions: E idence
om he Ge man manu ac u ing indus y, Canadian Jou nal o Economics/Re ue canadienne
d'économique, ISSN 1540-5982, Wiley, Hoboken, NJ, Vol. 58, Iss. 2, pp. 747-770,
h ps://doi.o g/10.1111/caje.70003
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/323862
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/
747
Impo compe i ion and fi m-le el CO2
emissions: E idence om he Ge man
manu ac u ing indus y
Jakob Leh
Uni e si y o Mannheim
Abs ac . Using he Ge man census o he manu ac u ing indus y, I analyze he impac o impo com-
pe i ion on ca bon emissions pe uni o defla ed sales (emission in ensi y). I combine p ecise in o ma ion
on fi m-le el CO2emissions wi h sec o -le el ade flows. Looking a he pe iod 1995 un il 2017, I ocus
on he impac o he ise o Eas e n Eu ope and China while add essing he endogenei y o ade flows
wi h an ins umen al a iable app oach. The baseline esul s sugges ha a 1 pp inc ease in he impo
pene a ion a io caused a educ ion o he a e age fi m’s emission in ensi y by app oxima ely 0.3%. This
esul implies ha he ise o he join Eas be ween 1995 and 2017 kep he a e age fi m’s emission
in ensi y 6% below he le el i would ha e had in he absence o he Eas ’s ise. I do no find s ong
indica ion o ealloca ion o p oduc ion owa ds mo e efficien fi ms. Finally, I supplemen he analysis
by examining he effec o expo oppo uni ies due o he Eas ’s ise. The esul s indica e ha expo ing
o he Eas inc eased sales and emissions, wi h a small, i any, nega i e effec on emission in ensi ies.
Résumé. Concu ence des impo a ions e émissions de CO2des en ep ises : l’exemple de l’indus ie
manu ac u iè e allemande. En u ilisan le ecensemen allemand de l’indus ie manu ac u iè e, j’analyse
l’incidence de la concu ence des impo a ions su les émissions de ca bone pa uni é de en es en e mes
éels (in ensi é des émissions). Je combine des données p écises su les émissions de CO2des en ep ises
a ec les flux des échanges comme ciaux sec o iels. En examinan la pé iode allan de 1995 à 2017, je me
concen e su l’effe de la mon ée en puissance de l’Eu ope de l’Es e de la Chine ou en enan comp e
de l’endogénéi é des flux comme ciaux à l’aide de la mé hode des a iables ins umen ales. Les ésul a s
de base suggè en qu’une augmen a ion d’un poin du aux de péné a ion des impo a ions en aîne
une éduc ion de l’in ensi é des émissions de l’en ep ise moyenne d’en i on 0,3 %. Ce ésul a indique
que la mon ée en puissance des pays de l’Es en e 1995 e 2017 a main enu l’in ensi é des émissions
de l’en ep ise moyenne 6 % en dessous du ni eau qu’elle au ai a ein en l’absence de celle-ci. Je ne
cons a e pas d’indica ion o e d’une edis ibu ion de la p oduc ion e s des en ep ises plus efficaces.
Enfin, je complè e l’analyse en examinan l’effe des possibili és d’expo a ion dues à la mon ée en
puissance de l’Es . Les ésul a s indiquen que les expo a ions e s l’Es on augmen é les en es e les
émissions, a ec un effe néga i aible, oi e nul, su l’in ensi é des émissions.
JEL classifica ion: F18, Q54, L60, D22
Co esponding au ho : Jakob Leh , [email p o ec ed]
I hank Ka in Rehdanz and Robe Gold o hei help ul commen s. The pape also benefi ed om
discussions a he 15 h RGS Doc o al Con e ence in Economics and he 56 h Canadian Economics
Associa ion Con e ence. The pape imp o ed significan ly du ing he e iew p ocess hanks o he aluable
eedback om wo anonymous e e ees and he edi o , Je an Che niwchan. I acknowledge financial suppo
p o ided by he Leibniz Associa ion h ough he Kiel Cen e o Globaliza ion (G an SAS-2016-I W-LWC)
and unding by he Ge man Resea ch Founda ion (DFG) h ough CRC TR 224 (P ojec B07). I he eby
decla e ha no ac ual o po en ial compe ing in e es s exis . Any e o s a e my own. Open Access unding
enabled and o ganized by P ojek DEAL.
Canadian Jou nal o Economics / Re ue canadienne d’économique 2025 58(2)
May 2025. / Mai 2025.
25 / pp. 747–770 / DOI: 10.1111/caje.70003
c
The Au ho (s). Canadian Jou nal o Economics/Re ue canadienne d’économique published by Wiley Pe iodicals LLC on
behal o Canadian Economics Associa ion.
This is an open access a icle unde he e ms o he C ea i e Commons A ibu ion License, which pe mi s use, dis ibu ion and
ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly ci ed.
748 J. Leh
1. In oduc ion
How ade and globaliza ion affec en i onmen al pe o mance and, in pa icula ,
clima e change is an impo an and widely discussed opic (see Copeland and
Taylo 2004, Che niwchan e al. 2017). While olde s udies ha e mos ly ocused on coun y
o sec o -le el effec s (see Copeland and Taylo 1994, Cole and Ellio 2003), mo e ecen
wo ks ha e emphasized he impo ance o he unde lying fi m-le el esponse o ade
and globaliza ion (see Che niwchan 2017, Ba ows and Olli ie 2018, Gu ié ez and
Teshima 2018, Fo slid e al. 2018). Fo ins ance, his esea ch has explo ed he ole o
o eign di ec in es men on fi ms’ ene gy use (B ucal e al. 2019) o he effec o fi ms’
expo ing s a us on CO2in ensi y o p oduc ion (Rich e and Schie sch 2017). Howe e ,
li le is known abou how impo compe i ion affec s fi ms’ CO2emissions and emission
in ensi y. My pape add esses his gap by analyzing he effec o impo compe i ion on
CO2emissions pe uni o sales in he Ge man manu ac u ing indus y.
The ole o compe i ion in gene al and impo compe i ion in pa icula on fi m-le el p o-
duc i i y has ecei ed much a en ion in he li e a u e. Fo example, Schmid (1997) a gues
ha inc easing compe i ion h ea ens fi ms’ su i al and hus o ces manage s o educe
slack. Indeed, p e ious empi ical esea ch ha looks a fi ms in Eu ope has es ablished a
posi i e link be ween impo compe i ion and p oduc i i y and inno a ion (see Holmes and
Schmi z 2001, Bloom e al. 2015, Shu and S einwende 2019, Chen and S einwende 2021).
Because he en i onmen al and ene gy economics li e a u e has iden ified ene gy use as
pa icula ly inefficien (“ene gy efficiency pa adox”; see DeCanio 1993), imp o emen s in
ene gy efficiency migh appea as a “low-hanging ui ” om he pe spec i e o a manage
who needs o cu cos s o ensu e he fi m’s su i al. The e o e, I expec fie ce compe i ion
o affec he CO2in ensi y o p oduc ion nega i ely.
Fo he empi ical analysis I combine he Ge man census o he manu ac u ing indus y
wi h sec o -le el ade flows. The census da a span he pe iod om 1995 un il 2017, co e
he uni e se o manu ac u ing plan s wi h mo e han 20 employees (app oxima ely 40,000
plan s annually) and p o ide, among o he hings, de ailed in o ma ion on plan -le el uel
use. This in o ma ion allows calcula ing CO2emissions based on uel-specific con e sion
ac o s.1To iden i y he effec o impo compe i ion on emission in ensi y, I exploi he ise
o China and Eas e n Eu ope (which I will join ly e e o as “ he Eas ”) as majo ac o s in
he wo ld economy. A ising sha e o impo ed manu ac u ing goods in Ge many o igina ing
om hese egions is one mani es a ion o his p ocess.2
To unco e causal effec s, I exploi he ac oss sec o a ia ion in exposu e o impo s om
he Eas , and I add ess he endogenei y o impo s wi h an ins umen al a iable app oach
ollowing Au o e al. (2014) and Au o e al. (2020).
The pape ela es o se e al s ands o li e a u e. Fi s , i con ibu es o he li e a u e
on ade, globaliza ion and he effec s on he en i onmen .3By examining he effec o
1 In ecen yea s, he Ge man manu ac u ing sec o emi ed app oxima ely 200 million ons o
CO2annually, oughly one qua e o o al emissions in Ge many, abso bed mo e han 15% o
Ge many’s labou o ce and con ibu ed app oxima ely one qua e o Ge many’s g oss
domes ic p oduc . These figu es eflec he cen al ole o he manu ac u ing sec o in he
Ge man economy.
2 Fo example, be ween 1995 and 2017, he sha e o Ge man impo s om he Eas ose om
10% o almos 30% (see able A6 in he online appendix).
3 See Che niwchan (2017) and Copeland e al. (2021) o ecen li e a u e e iews.
Impo compe i ion and emissions 749
impo compe i ion on CO2emission in ensi y in Ge many, i closely ela es o Gu ié ez
and Teshima (2018), who ound ha in ensified impo compe i ion due o a iff cu s
imp o ed ene gy efficiency among fi ms in Mexico. I also complemen s ecen wo k by
Leisne e al. (2023), who documen ha impo compe i ion om China lowe ed emissions
and sales among Danish manu ac u ing fi ms bu did no educe emission in ensi ies excep
in he mos emission-in ensi e qua ile. My pape ex ends his li e a u e by ocusing
on manu ac u ing in Ge many—Eu ope’s la ges economy, i s la ges emi e and a
coun y wi h a ela i ely ene gy-in ensi e indus ial s uc u e.4Leisne e al. (2023) also
analyze he effec s o offsho ing, i.e., impo ing in e media e inpu s. Simila o Ake man
e al. (2021) and Dussaux e al. (2023), who examine Swedish and F ench manu ac u ing
fi ms, espec i ely, hey find ha offsho ing leads o lowe emission in ensi ies by inc easing
sales mo e han emissions.5In ela ed wo k on he effec s o ade libe aliza ion h ough
NAFTA, Che niwchan (2017) a ibu es educed emissions o local pollu an s om US
manu ac u ing plan s o he inc eased a ailabili y o emission-in ensi e in e media e inpu s
and new expo oppo uni ies o Ame ican manu ac u e s. The ela ion be ween expo ing
and fi ms’ CO2emission in ensi y has been s udied by Rich e and Schie sch (2017)
and Ba ows and Olli ie (2021) o Ge many and India, while Kong e al. (2022)
and Bomba dini and Li (2020) examine he impac o expo ing on local pollu an s
in China.
The pape u he ela es o he li e a u e on he de e minan s o ene gy efficiency and he
so-called “ene gy efficiency pa adox” (see DeCanio 1993, Jaffe and S a ins 1994, Ge a den
e al. 2017), which e e s o seemingly subop imal ene gy use by fi ms. The li e a u e has
iden ified se e al po en ial explana ions o inefficien ene gy use, such as manage ial inabil-
i y (see Bloom e al. 2010, Ma in e al. 2012), capi al cons ain s (see Le ine e al. 2018,
De Haas e al. 2021) o ma ke condi ions such as size (Fo slid e al. 2018). By linking
changes in he le el o compe i ion o changes in CO2in ensi y, I in es iga e a u he
de e minan o ene gy efficiency.
Mo e b oadly, my pape con ibu es o he li e a u e in es iga ing he effec o impo
compe i ion om China and Eas e n Eu ope on he manu ac u ing sec o in wes e n indus-
ialized coun ies.6Fo ins ance, Bloom e al. (2015) documen “ ade induced echnical
change,” i.e., echnological upg ading, mo e pa en ing and highe o al ac o p oduc i i y
(TFP) e enue among Eu opean fi ms ha ope a e in sec o s mo e exposed o Chinese
impo s. Chen and S einwende (2021) also find posi i e effec s o impo compe i ion on
p oduc i i y among ini ially unp oduc i e amily-owned fi ms in Spain.7In con as , Au o
4 Fo example, du ing he fi s h ee ading pe iods o he EU–ETS (2005–2020)
app oxima ely 20% o e ified emissions om indus ial ins alla ions came om Ge many (see
Eu opean En i onmen al Agency 2023).
5 While his is an a guably impo an channel h ough which ade in eg a ion can affec
Ge man fi ms, I canno s udy offsho ing because I lack ele an in o ma ion on impo s.
6 An abundan li e a u e s udies he labou ma ke consequences, bo h on he egional and
indi idual le el, e.g., Au o e al. (2013), Au o e al. (2014)andAcemoglue al.(2016)look
a he US and Dau h e al. (2014) analyze he case o Ge many. These pape s documen
nega i e effec s o impo compe i ion on employmen . The China shock appea s o be mos
ele an o he US, whe eas o Ge many he ise o Eas e n Eu ope was mo e c i ical.
7 They p o ide e idence ha compe i ion o ces unp oduc i e fi ms o imp o e ma e ial usage
and elimina e wha hey call “X-inefficiencies,” which desc ibes subop imal fi m beha iou
ega ding he objec i e o maximize mone a y p ofi s.
750 J. Leh
e al. (2020) es ima e a nega i e effec o inc easing impo compe i ion om China on R&D
expendi u e and pa en ing among US manu ac u ing fi ms. Indeed, he li e a u e on impo
compe i ion and inno a ion summa ized by Shu and S einwende (2019) finds “la gely pos-
i i e e idence o such [impo compe i ion inc easing inno a ion] in Eu ope and mixed
e idence o such in No he n Ame ica.”
I s a he analysis wi h a decomposi ion o he h ee-digi sec o -le el emission in ensi y
(see Olley and Pakes 1996), showing ha o al sec o -le el emission in ensi y is nega i ely
ela ed o impo s om he Eas , d i en by wi hin-fi m changes. Realloca ion o ma ke
sha es owa ds mo e p oduc i e fi ms, i a all, played only a mino ole.
The main fi m-le el analysis confi ms he esul om he sec o al decomposi ion.
Baseline es ima es imply a dec ease o emission in ensi y in he ange o 0.3% o 0.4%
in esponse oa1ppinc easein hesha eo impo s om heEas ela i e o baseline
abso p ion (impo pene a ion a io). By examining fi ms’ CO2emissions and sales
sepa a ely, I show ha bo h decline due o inc eased impo compe i ion, wi h a s onge
esponse in emissions, esul ing in a all in emission in ensi y. The effec is cen ed on fi ms
wi h abo e-median emission in ensi y—unde lining he ele ance o he effec o agg ega e
emissions—and among fi ms ope a ing in sec o s wi h low impo pene a ion in 1995. The
esul s a e obus o al e na ions o he eg ession specifica ion and o “con olling” o
changes in sec o al expo s o he Eas , o which I cons uc ed an ins umen analogous
o he one o impo s. Respec i e es ima es sugges ha new expo oppo uni ies lead o
a nea ly p opo ional inc ease in emissions and sales. I u he analyze he effec o impo
compe i ion on emissions pe uni o alue added, which can be calcula ed o a subsample
o fi ms yielding quan i a i ely simila es ima es. The baseline esul s a e also obus o
an al e na i e iden ifica ion s a egy based on g a i y esiduals p esen ed in he pape ’s
online appendix.
The emainde o he pape is s uc u ed as ollows. In sec ion 2, I ou line he empi ical
app oach. Sec ion 3in oduces he da a se , shows desc ip i e s a is ics and p o ides fi s
esul s a he sec o al le el o mo i a e he main analysis. The main esul s om he
fi m-le el analysis a e p esen ed in sec ion 4 oge he wi h obus ness checks and effec
he e ogenei ies. Sec ion 5p esen s addi ional esul s o complemen he main analysis, and
finally, sec ion 6discusses he findings and concludes.
2. Empi ical app oach
To es ima e he effec o impo compe i ion on fi m-le el ou comes conside he ollowing
eg ession specifica ion:
yi z =β0+αIPREas
z +νi+i z.(1)
The dependen a iable yi z can be any ou come o fi m iin yea ope a ing in sec o
z. The coefficien o in e es αcap u es he effec o an indus y’s exposu e o impo s
om he Eas defined as o al impo s om he Eas in yea scaled wi h ini ial abso p ion
(see Au o e al. 2020). Conc e ely, he “impo pene a ion a io” (IPR) is defined as ollows:
IPREas
z ≡ImpEas
z
Yz,1995 +Impz,1995 −Expz,1995 .
Finally, i z in equa ion 1is a andom e o e m. I ollow Bloom e al. (2015) by aking long
diffe ences ( ou yea s) which elimina es he fi m fixed effec νi. The diffe enced equa ion
eads as ollows:
Δyi z =β0+αΔIPREas
z +Δi z.(2)

Impo compe i ion and emissions 751
The eg essions I ake o he da a may o may no include addi ional se s o fixed effec s,
e.g., yea dummies o yea by emission in ensi y decile dummies, o con ol o annual shocks
o shocks occu ing along he ene gy-in ensi y dis ibu ion.8
To es ima e a causal effec o he ade exposu e o a sec o on fi m-le el ou comes,
equi es add essing he endogenei y o ade flows. Fo ins ance, demand condi ions in
Ge many a e expec ed o affec impo s and domes ic fi ms’ beha iou simul aneously.9
I add ess he endogenei y simila o Au o e al. (2013) and Dau h e al. (2014)by
ins umen ing he changes in Ge man impo s om he Eas ’s indus y zwi h changes in
ade flows om he same indus y zin he Eas o a se o o he coun ies:
IPRO he ←Eas
z ≡ImpO he ←Eas
z
Yz,1995 +Impz,1995 −Expz,1995 .
The idea is ha pa o he a ia ion o Ge man impo s om he Eas is due o a ising
compa a i e ad an age o he Eas o lowe ade cos s. The ins umen is ele an o his
pa o he a ia ion as he ise o he Eas also affec s ade flows o he o he coun ies.
The o he (endogenous) pa o he a ia ion is due o domes ic condi ions in Ge many and
needs o be sepa a ed ou . The ins umen succeeds in sepa a ing he exogenous componen
o ade flows o Ge many om he endogenous unde he assump ion ha condi ions
in Ge many a e o hogonal o hose in he chosen se o o he coun ies. The exclusion
es ic ion u he demands ha ade flows be ween he se o o he coun ies and he Eas
ha e no di ec effec on Ge man fi ms. These conside a ions guide he selec ion o an app o-
p ia e se o coun ies o he ins umen g oup. I ollow Dau h e al. (2014), who included
Aus alia, Canada, Japan, No way, New Zealand, Sweden, Singapo e and he Uni ed
Kingdom, all o which a e high-income coun ies bu nei he di ec ly bo de s Ge many no
is any o hem a membe o he Eu opean Mone a y Union (EMU). Dau h e al. (2014)
a gue ha demand condi ions among neighbou ing coun ies a e oo simila and ha he
fixed exchange a e wi hin he EMU migh cause a iola ion o he exclusion es ic ion
i changes in ade flows be ween o he coun ies and he Eas di ec ly affec Ge man
indus ies.10 Finally, o he ins umen o wo k, i needs o be ele an , which, howe e ,
can be es ed and is indeed confi med by he fi s -s age esul s epo ed in sec ion 4.
8 Ideally, I would like o use lagged abso p ion om he pe iod be o e he ise o he Eas in he
denomina o o he ins umen . Howe e , due o da a limi a ions his was no possible. The
s a is ical office p o ides p oduc ion da a a he economic sec o le el, based on he sec o
classifica ion om 1993, only since 1995. Be o e 1995 in o ma ion on sec o al p oduc ion is
a ailable o he sec o classifica ion om 1979. The s a is ical office could no p o ide a
mapping be ween he classifica ions.
9 Suppose he demand o some goods, e.g., hea pumps and sola panels, suddenly inc eases in
Ge many. Tha will simul aneously affec he impo s o espec i e goods and domes ic
p oduce s o hose goods. Fo example, domes ic p oduce s’ p oduc ion migh inc ease, and
hence, hey equi e mo e ene gy inpu s. Mo eo e , hei ene gy in ensi y migh also inc ease
because expanding p oduc ion beyond he efficien le el will s ill be p ofi able gi en ha high
demand d i es up p ices.
10 A u he conce n in he con ex o his applica ion migh ela e o global ene gy p ice shocks
simul aneously affec ing a specific sec o in Ge many and expo s om he same sec o in he
Eas o he ins umen coun ies. Table A8 in he online appendix co ela es changes in
sec o al ene gy p ices in Ge many wi h expo flows om he Eas o he ins umen coun ies.
Condi ional on yea fixed effec s sec o al ene gy p ices in Ge many and expo s om he Eas
o he ins umen coun ies do no co-mo e.
752 J. Leh
3. Da a, desc ip i e s a is ics and mo i a ing exe cises
3.1. Da a
The main da a sou ce is he Ge man census o he manu ac u ing indus y called AFiD
(Am liche Fi menda en ü Deu schland), which co e s he uni e se o Ge man indus ial
plan s wi h mo e han 20 employees. The da a consis s o diffe en “modules” o which
I combine “AFiD Modul Indus iebe iebe” (indus ial plan s module) wi h “AFiD Modul
Ene gie e b auch” (ene gy use module).
The indus ial plan s module con ains economic a iables such as sales, sales ab oad,
numbe o employees and in es men . To defla e hese mone a y alues, I use he ou -digi
sec o -specific p oduce p ice indices published by he Ge man s a is ical office.11 The
ene gy use module de ails plan -le el ene gy use by uel ype. Ene gy use is epo ed in
physical uni s (kWh) and can hus be con e ed o CO2emissions based on uel-specific
con e sion ac o s (see Pe ick e al. 2011, Rich e and Schie sch 2017).12 To accoun o
indi ec emissions esul ing om he gene a ion o elec ici y ha fi ms buy om he g id,
I ake he a e age ca bon con en o he Ge man elec ici y mix, which a ies by yea (see
Umwel bundesam 2018). The sum o di ec and indi ec emissions is o al emissions. A
majo ca ea wi h he ene gy da a is a b eak in he epo ing be ween 2002 and 2003. The
ime se ies be o e and a e 2003 a e bo h in e nally consis en . In my es ima ion, I make
su e o exclude a ia ion ha esul s om he b eak in he epo ing by excluding yea s om
he sample in which sub ac ed lags, used in diffe ence calcula ions, p eda e 2003.13 The
final da ase is an unbalanced panel agg ega ed a he fi m le el, co e ing 1995 un il 2017.
I supplemen he main da a wi h he “cos s uc u e su ey,” an unbalanced panel a he
fi m le el ha ypically includes fi ms o ou consecu i e yea s.14 The cos s uc u e su ey
p o ides—among o he hings—in o ma ion on in e media e inpu expendi u es, allowing o
an es ima ion o he effec o impo compe i ion on he emission in ensi y o alue added.
Un o una ely, mo e g anula in o ma ion on in e media e inpu s, such as by p oduc and
o igin, is una ailable, making i impossible o s udy offsho ing.
I ely on he BACI da abase o in o ma ion on bila e al ade flows, which is cons uc ed
om he Uni ed Na ions Commodi y T ade S a is ics Da abase (Com ade) and p o ided by
CEPII. The da abase epo s ade flows a he six-digi p oduc le el om he Ha monized
Sys em (HS) nomencla u e. To me ge he ade in o ma ion o he fi m-le el da a, I agg e-
ga e om he p oduc le el o he h ee-digi economic sec o le el using he classifica ion
11 See “S a is isches Bundesam , Index de E zeuge p eise gewe bliche P oduk e
(Inlandsabsa z) – Lange Reihen de Fachse ie 17 Reihe 2.”
12 Fo he con e sion o ene gy om p ima y sou ces, I d aw upon he emission ac o s p o ided
by he Umwel bundesam (a ede al agency). A able wi h he ele an in o ma ion is a ailable
a www.umwel bundesam .de/ hemen/klima-ene gie/ eibhausgas-emissionen, las
e ie ed No embe 18, 2020. The able gi es he uel-specific ime- a ying CO2con en pe
e ajoule. This uni can be con e ed o CO2pe kWh. We hen mul iply he uel use in kWh
wi h he espec i e con e sion ac o o ob ain he CO2emissions.
13 In he main specifica ion ha uses ou -yea diffe ences, his implies he omission o he yea s
2003–2006. Fo u he backg ound on he ene gy use module as well as he change in
epo ing, see Pe ick e al. (2011).
14 This sampling p ocedu e also dic a es he choice o diffe encing, making i in easible o ake
diffe ences longe han ou yea s. Only fi ms wi h mo e han 500 employees a e included
annually in he cos s uc u e su ey.
Impo compe i ion and emissions 753
om 1993 (equi alen o NACE indus y codes).15 I fix fi ms’ economic sec o o he sec o
om he fi s yea in which he fi m appea s in AFiD.
The final da a se is cleaned om ou lie s. Specifically, I d op fi ms wi h ei he CO2emis-
sion in ensi y, he ou -yea change in CO2emission in ensi y o sales below he 1s pe cen ile
o abo e he 99 h pe cen ile. I also omi he economic sec o s “manu ac u e o office machin-
e y and compu e s” (WZ 30) and “manu ac u ing o adio, ele ision and communica ion
equipmen and appa a us” (WZ 32). O e he pe iod 1995 o 2017, bo h sec o s ha e been
subjec o apid echnological changes, quali y upg ading and alling p ices. Because o his
de elopmen , he p oduce p ice indices in hese sec o s a e ou lie s making a compa ison
o defla ed sales as measu es o physical ou pu o e ime difficul .16 In a obus ness check
I show esul s ob ained om a sample ha includes bo h sec o s.
3.2. Desc ip i e s a is ics
Table 1shows summa y s a is ics om he es ima ion sample o fi m-le el a iables pooled
o e he pe iod 1995 un il 2017. The a e age fi m in he da a has close o 116 employees and
gene a es app oxima ely 23 million eu os in annual u no e , o which—on a e age—18% a e
expo ed. Addi ional in o ma ion on alue added om he cos s uc u e su ey is a ailable
o abou 40% o he obse a ions. The a e age fi m’s alue added amoun s o 24 million
eu os, mo e han he a e age sales in he p ima y da ase , eflec ing he o e sampling o
la ge fi ms in he cos s uc u e su ey.
TABLE 1
Desc ip i e s a is ics – Fi m-le el in o ma ion
Va iable Mean S d. de . p10 p50 (median) p90 N
Numbe o employees 116 188 24 54 255 744,062
Sales 22,923.2 53,710.3 1,729.5 6,705.33 51,559.5 744,062
Expo sha e 0.18 0.24 0 .07 0.56 744,062
Value added 23,726.5 56,173 1,668.8 7,398.2 58,492.91 297,241
Value added sha e 0.60 0.17 0.37 0.61 0.82 297,241
To al ene gy (in MWh) 11,048.7 103,877.5 155.8 933.5 12,400 744,062
To al CO2emissions (in ) 3,722.4 31,634.2 65.9 381.7 4,823.46 744,062
To al elec ici y (in MWh) 3,619.55 29,104.7 64.4 423.3 5,351.5 744,062
Impo pene a ion a io 8.85 9.94 1.04 5.26 20.44 744,062
NOTES:The able shows he a e age o espec i e a iables om he pe iod 1995–2017. Defla ed sales and
alue added a e in 1,000 eu o, ene gy use ( o al and elec ici y) is in MWh and CO2emissions in ons.
SOURCES:Resea ch da a cen es o he Fede al S a is ical Office and he S a is ical Offices o he Lände :
AFiD-Panel Indus iebe iebe, 1995–2017; own calcula ions.
15 To be p ecise, I fi s con e he p oduc -le el in o ma ion om HS92 o SITC3 (con e sion
able was downloaded om h ps://uns a s.un.o g/unsd/ ade/classi ica ions
/co espondence- ables.asp) and hen I map om SITC3 o he h ee-digi indus y
classifica ion using he same mapping as Dau h e al. (2014). The indus y classifica ion om
1993 was in place un il 2008 wi h mino modifica ions in 2003. The e o e, I omi all fi ms om
he analysis ha we e fi s obse ed only a e 2008 because he economic sec o based on he
classifica ion om 1993 is unknown o hese fi ms.
16 The a e age p oduce p ice indices in he manu ac u ing sec o inc eased mode a ely om 94
in 1995 o 107 in 2017 (indexed o 100 in 2010), while he a e age p oduce p ice indices in
WZ 30 collapsed om 307 o 81 and in WZ 32 om 217 o 89 in he same pe iod.
754 J. Leh
In addi ion o p esen ing economic pe o mance indica o s, able 1summa izes o al
ene gy use, elec ici y use and CO2emissions. Fo ins ance, he able shows ha mean
ene gy use amoun s o 11,048 MWh annually, esul ing in 3,722 ons o CO2emissions
pe fi m. Finally, he las ow o he able shows he a e age fi m’s impo pene a ion
a io om he Eas a app oxima ely 9 pp. This a e age masks s ong dynamics o e ime
and diffe ences ac oss sec o s as can be seen om able A6 in he online appendix, which
epo s summa y s a is ics o impo s (sha es) om China and Eas e n Eu ope om 22
wo-digi indus ies o 1995 and 2017. In he ini ially leas -exposed sec o (Manu ac u e o
obacco p oduc s, NACE Re .1 16), impo s om he Eas accoun ed o me ely 1.4% in
1995 bu o 27% in he mos exposed sec o (Manu ac u e o wea ing appa el; d essing and
dyeing o u , NACE Re .1 18). Twen y- wo yea s la e , impo sha es anged om 11%
(manu ac u e o chemicals and chemical p oduc s, NACE Re . 1 24) o 58% (manu ac u ing
o office machine y and compu e s, NACE Re .1 30).17 China’s accession o he WTO and
he EU’s eas wa d enla gemen played an impo an ole in he inc ease in impo sha es,
which indeed inc eased in all 22 sec o s du ing he ime window unde conside a ion.
3.3. Mo i a ing exe cise: Sec o al decomposi ion o emission in ensi y
Be o e u ning o he fi m-le el analysis, I desc ibe he e olu ion o agg ega e sec o al
emission in ensi y om 1995 un il 2017. Following Olley and Pakes (1996) and B ucal
e al. (2019), I decompose agg ega e sec o -le el emission in ensi y in an unweigh ed mean
and a co a iance e m. The la e cap u es he associa ion be ween fi ms’ ma ke sha es and
hei emission in ensi ies. The ollowing exp ession desc ibes he decomposi ion:
Wz =
i∈Z
si lnEi
 
Weigh ed CO 2In ensi y
in Sec o Z
=lnEz

Unweigh ed a g.
In ensi y
+
i∈Z
(si −s )(lnEi −lnE )
 
Co a iance
,(3)
whe e si is he sha e o sales by fi m iin o al sales in sec o z(is ma ke sha e) a ime
,lnEz is he a e age emission in ensi y om all fi ms in sec o zand s is he a e age
ma ke sha e in sec o z.
The agg ega e weigh ed emission in ensi y in sec o z(Wz ) can be e-w i en as he
a e age emission in ensi y om all fi ms in sec o z(lnEz ) and he co a iance e m. A
nega i e co a iance implies highe ma ke sha es o mo e ca bon-efficien fi ms ( he in e se
o ca bon in ensi y) and hus eflec s a mo e efficien alloca ion o economic ac i i y ac oss
fi ms. Changes in his e m cap u e a ealloca ion o ma ke sha es ac oss fi ms wi h diffe -
en ca bon efficiency, e.g., a dec ease implies a ealloca ion o ma ke sha es owa ds fi ms
wi h a lowe ca bon in ensi y (highe ca bon efficiency). Changes in he unweigh ed a e age
emission in ensi y eflec changes wi hin fi ms.
Figu e 1shows he e olu ion o he weigh ed a e age, he unweigh ed a e age and he
co a iance e m a e aged ac oss h ee-digi economic sec o s. One can see ha he weigh ed
a e age dec eased by app oxima ely 40% be ween 1995 and 2017 as indica ed by he solid
17 Agg ega ed ac oss he sec o s, impo s om China amoun ed o app oxima ely US$8 billion
(app oxima ely 22*350 MIO USD) in 1995, co esponding o an impo sha e o 2%. Un il
2017, his sha e ose o 10% (impo s wo h US$89 billion and app oxima ely US$4 billion
om he a e age wo-digi sec o ). Simila ly, impo s om Eas e n Eu ope ose om US$32
billion (app oxima ely 8% o o al impo s) in 1995 o US$200 billion in 2017, co esponding
o an impo sha e o nea ly 20%.
Impo compe i ion and emissions 761
TABLE 4
Effec s o impo compe i ion and expo oppo uni ies
IV OLS
(1) (2) (3) (4)
PanelA.Logo CO
2emission in ensi y
Δimpo s −0.0026*** −0.0029*** −0.0012*** −0.0001
(0.0007) (0.0009) (0.0005) (0.0005)
Δexpo s −0.0008 0.0037*** −0.0034*** −0.0013**
(0.0010) (0.0012) (0.0005) (0.0005)
Numbe o obse a ions 403,367 403,367 403,367 403,367
Kleibe gen–Paap F-s a is ic 106.78 60.91
PanelB.Logo CO
2emissions
Δimpo s −0.0079*** −0.0027*** −0.0024*** 0.0004
(0.0012) (0.0008) (0.0005) (0.0004)
Δ expo s 0.0041*** 0.0033*** 0.0024*** 0.0022***
(0.0009) (0.0010) (0.0005) (0.0005)
Numbe o obse a ions 403,367 403,367 403,367 403,367
Kleibe gen–Paap F-s a is ic 106.78 60.91
Panel C. Log o g oss ou pu
Δimpo s −0.0053*** 0.0001 −0.0011 0.0005
(0.0013) (0.0011) (0.0007) (0.0007)
Δ expo s 0.0049*** −0.0004 0.0058*** 0.0035***
(0.0013) (0.0014) (0.0007) (0.0007)
Numbe o obse a ions 403,576 403,576 403,576 403,576
Kleibe gen–Paap F-s a is ic 106.68 60.90
CO2in ensi y–decile–yea dummy Yes Yes Yes Yes
Sales-decile-yea -dummy Yes Yes Yes Yes
Expo sha e–decile–yea dummy Yes Yes Yes Yes
Sec o dummy Yes Yes
NOTES: *p<0.1, ** p <0.05, *** p <0.01. Columns (1) and (2) show esul s om 2SLS es ima ions
and columns (3) and (4) om an OLS es ima ion. The dependen a iable is he ou -yea change in he log
o fi ms’ CO2emission in ensi y (panel A), he log o fi ms’ CO2emissions (panel B) and he log o fi ms’
defla ed sales (panel C). The explana o y a iables a e he ou -yea changes in he sec o al IPR om he
Eas and changes in sec o al expo s om Ge many o he Eas scaled wi h domes ic abso p ion in 1995. The
ins umen s a e he ou -yea changes o expo s om he Eas o o he coun ies ( o ins umen changes
in Ge man impo s) and he sec o al change in expo s om o he coun ies o he Eas ( o ins umen he
Ge man expo s). S anda d e o s we e clus e ed bo h a he fi m and a he h ee-digi indus y–yea le el.
SOURCES: Resea ch Da a Cen es o he Fede al S a is ical Office and he S a is ical Offices o he Lände :
AFiD-Panel Indus iebe iebe, 1995–2017, own calcula ions.
emissions (panel B). Su p isingly, he poin es ima e o he effec o expo oppo uni ies
on emission in ensi y is posi i e due o a posi i e effec on emissions bu a null effec o
expo oppo uni ies on ou pu (panel C).22
In addi ion o he main goal o he analysis p esen ed he e— o con ol o expo oppo -
uni ies as a po en ial con ounde — he findings abou he impac o expo oppo uni ies
on fi m-le el emissions a e o significance by hemsel es. Sec ion A3 in he online appendix
22 An al e na i e is o simply con ol o fi ms expo sha es in he baseline IV es ima ion, wi h
he ca ea ha his does no lend o a causal in e p e a ion o he expo sha e coefficien .
F om able A9 in he online appendix, one can see ha a 1 pp inc ease in fi ms’ expo sha e
is associa ed wi h a educ ion in emission in ensi y by 0.23%, d i en by an inc ease in sales
and a less- han-p opo ional inc ease in emissions. The main effec on he effec o impo
compe i ion on emission in ensi y emains unchanged.

762 J. Leh
expands on his analysis and u he analyzes he effec o expo oppo uni ies on emission
in ensi y.
4.2.3. Ene gy p ices as po en ial con ounde
A conce n b iefly ouched on in he in oduc ion o he empi ical app oach e e s o he
exclusion es ic ion in he pa icula con ex o his analysis. I shocks in he global oil
and/o gas ma ke simul aneously affec he ins umen coun ies’ impo s o goods om
he Eas ’s sec o zand he ene gy in ensi y o p oduc ion in sec o zin Ge many, he
exclusion es ic ion would be iola ed. In o ma ion abou o al ene gy expendi u e in he
cos s uc u e su ey combined wi h physical ene gy use allows o cons uc ing a measu e
o sec o al ene gy p ices. To es ima e he effec o impo compe i ion on emission in en-
si y condi ional on ene gy p ices, I include changes in he log o a e age ene gy p ices a
he h ee-digi sec o le el as a con ol a iable. Table A10 in he online appendix epo s
he esul s, which show ha he p ima y effec o impo s on emission in ensi y emains
unchanged condi ional on sec o al p ice changes. The esul s u he indica e a nega i e
co ela ion be ween changes in sec o al p ices and emission in ensi y, as expec ed.
4.3. He e ogenei y
The subsequen pa ag aphs shed ligh on he e ogenei ies. Fi s , I dis inguish be ween
impo s om Eas e n Eu ope and impo s om China. Second, I analyze he e ogenei ies
depending on fi m cha ac e is ics, e.g., fi ms’ emission in ensi y, expo sha e o size.
4.3.1. Eas e n Eu ope s. China
Table 5 epo s sepa a ely he effec s o changes in he IPR om China (columns (1) and
(2)) and Eas e n Eu ope (columns (3) and (4)). As in he main esul s able, he dependen
a iables a e he logs o emission in ensi y (panel A), emissions (panel B) and sales (panel C).
One can see ha he effec o impo s om China closely esembles he main effec
epo ed in able 3. The coefficien in column (1) indica es a dec ease in emission in ensi y
by 0.37% due o a 1 pp inc ease in he IPR. This effec d ops o 0.18% a e accoun ing
o sec o al ends (column (2)). Panels B and C show ha he effec on CO2emissions
is nega i e and significan in bo h specifica ions, while he effec on sales is nega i e and
significan in column (1) bu becomes insignifican in column (2).
Fo Eas e n Eu opean impo s, I find a la ge and highly significan nega i e effec on
emission in ensi ies in bo h specifica ions. Howe e , hese effec s mus be qualified: fi s ,
he fi s s ages a e weak, wi h F-s a is ics below 10. Second, when examining he effec on
sales in panel C, I find posi i e poin es ima es, which become la ge a e condi ioning on
indus y ends.23 Dau h e al. (2014) emphasize he ole o in a-indus y ade be ween
Ge many and Eas e n Eu ope and he co ela ion be ween indus y-le el impo s om and
Ge man expo s o Eas e n Eu ope. The e o e, he posi i e effec on sales, which educes
emission in ensi y, migh pa ly esul om co ela ed Ge man expo s. The in eg a ion
o Eas e n Eu opean economies in o Ge man fi ms’ alue chains, o example, h ough FDI
om Ge many, u he challenges he iden ifica ion. Fo ins ance, p oduc i i y imp o emen s
23 Specifically, he poin es ima e wi hou indus y ends is 0.0015 wi h a s anda d e o o
0.0049. A e condi ioning on indus y ends, he poin es ima e is 0.0085 wi h a
co esponding s anda d e o o 0.0049, making i ma ginally significan . The poin es ima e
wi h CO2emissions as he dependen a iable is −0.0046 wi hou and −0.0002 wi h indus y
ends.
Impo compe i ion and emissions 763
TABLE 5
Eas e n Eu ope and China sepa a ely
China Eas e n Eu ope
(1) (2) (3) (4)
PanelA.Logo CO
2emission in ensi y
Δimpo s −0.0037*** −0.0018* −0.0062*** −0.0088**
(0.0010) (0.0011) (0.0024) (0.0040)
Numbe o obse a ions 403,367 403,367 403,367 403,367
Kleibe gen–Paap F-s a is ic 205.98 182.10 6.88 5.13
PanelB.Logo CO
2emissions
Δimpo s −0.0098*** −0.0026** −0.0046 −0.0002
(0.0015) (0.0009) (0.0040) (0.0046)
Numbe o obse a ions 403,367 403,367 403,367 403,367
Kleibe gen–Paap F-s a is ic 205.98 182.10 6.88 5.13
PanelC.Logo sales
Δimpo s −0.0061*** −0.0008 0.0015 0.0085*
(0.0015) (0.0013) (0.0049) (0.0049)
Numbe o obse a ions 403,576 403,576 403,576 403,576
Kleibe gen–Paap F-s a is ic 206.02 182.13 6.88 5.13
CO2in ensi y–decile–yea dummy Yes Yes Yes Yes
Sales–decile–yea dummy Yes Yes Yes Yes
Expo sha e–decile–yea dummy Yes Yes Yes Yes
Sec o dummy Yes Yes
NOTES: *p<0.1, ** p <0.05, *** p <0.01. The able show esul s om 2SLS es ima ions. The
explana o y a iable is he ou -yea change in he IPR om China (columns (1) and (2)) and Eas e n
Eu ope (columns (3) and (4)) a he h ee-digi indus y le el. The ins umen s a e he espec i e ou -yea
changes o expo s om China/Eas e n Eu ope o o he coun ies. The dependen a iables a e he ou -yea
changes in he logs o fi ms’ CO2emission in ensi y (panel A), CO2emissions (panel B) and sales (panel
C). S anda d e o s we e clus e ed bo h a he fi m and a he h ee-digi indus y–yea le el. Fo each
specifica ion, I epo he Kleibe gen–Paap F-s a is ic.
SOURCES:Resea ch Da a Cen es o he Fede al S a is ical Office and he S a is ical Offices o he Lände :
AFiD-Panel Indus iebe iebe, 1995–2017; own calcula ions.
in Ge many migh inc ease expo s om Ge man o eign affilia es in Eas e n Eu ope. Gi en
he weak fi s s age in he IV specifica ion o Eas e n Eu ope and he limi a ions men ioned
abo e, he magni ude o he effec o impo s om Eas e n Eu ope should be in e p e ed
wi h cau ion.
4.3.2. Fi m cha ac e is ics
Table 6examines how he effec s a y based on fi m cha ac e is ics. To do so, I in e ac
he ade shock wi h abo e-median-dummies o CO2in ensi ies, expo sha es and size.
Addi ionally, I in e ac he ade shock wi h dummy a iables ep esen ing single-p oduc
fi ms and fi ms in sec o s wi h high impo sha es in 1995.24
Column (1) epo s he e ogenei ies by emission in ensi y, showing a s a is ically
insignifican main effec and a nega i e and highly significan in e ac ion effec .25 Despi e
24 All dummy a iables we e de e mined using fi s -pe iod alues o p e en po en ial
endogenei y issues wi h he ade shock. The choice o con ol a iables depends on he specific
in e ac ion o a oid abso bing oo much use ul a ia ion. All models a e “ ully in e ac ed,”
meaning ha all fixed effec s/con ols we e also in e ac ed wi h he abo e-median dummies.
25 I omi he CO2in ensi y–decile fixed effec s because hey would abso b all he a ia ion in he
in e ac ion effec and condi ion on s a -off pe iod alues ins ead.
764 J. Leh
TABLE 6
In e ac ion effec s
(1) (2) (3) (4) (5)
Main effec −0.00049 −0.00332*** −0.00258*** −0.00303*** −0.00373***
(0.00074) (0.00095) (0.00084) (0.00080) (0.00120)
In e ac ion – CO2in ensi y −0.00395***
(0.00079)
In e ac ion – Expo in ensi y 0.00080
(0.00095)
In e ac ion – Size −0.00114
(0.00083)
In e ac ion – Single p oduc 0.00105
(0.00091)
In e ac ion – High impo sha es 0.00329**
(0.00151)
Numbe o obse a ions 403,367 403,367 403,367 370,981 403,367
Kleibe gen–Paap F-s a is ic 56.19 77.07 116.19 70.60 70.77
CO2in ensi y–decile–yea dummy Yes Yes Yes Yes
Sales–decile–yea dummy Yes Yes Yes Yes
Expo sha e–decile–yea dummy Yes Yes Yes Yes
NOTES: *p<0.1, ** p <0.05, *** p <0.01. The able epo s main and in e ac ion effec s om 2SLS
es ima ions. The dependen a iable is he ou -yea change in he log o fi ms’ CO2emission in ensi y. The
fixed effec s a e epo ed a he bo om o he able and a y depending on he in e ac ion a iable. All
fixed effec s we e also in e ac ed wi h he “abo e-median dummy.” The dummies o he in e ac ion effec s
we e de e mined based on fi s -yea alues. The eg ession in column (1) u he includes he s a -off pe iod
alue o he dependen a iable. The explana o y a iable is he ou -yea change in he sec o al IPR om
he Eas ins umen ed wi h he espec i e ou -yea change o expo s om he Eas o o he coun ies.
Only single plan fi ms we e used in column (4). S anda d e o s we e clus e ed bo h a he fi m and a he
h ee-digi indus y–yea le el.
SOURCES:Resea ch Da a Cen es o he Fede al S a is ical Office and he S a is ical Offices o he Lände :
AFiD-Panel Indus iebe iebe, 1995–2017; own calcula ions.
conside able a ia ion in fi ms’ emission in ensi y abo e he median (see able 1), his esul
s ill sugges s ha he effec o impo compe i ion is quan i a i ely ele an o agg ega e
emissions. Column (2) shows he in e ac ion be ween impo compe i ion and fi ms’
in eg a ion in he global economy, measu ed by hei expo sha es. The es ima ed main
effec is nega i e, and he in e ac ion effec is posi i e bu small and insignifican . Fo a
Ge man fi m selling p ima ily domes ically, an inc ease in IPR migh be mo e consequen ial
compa ed o one selling in he compe i i e wo ld ma ke , p o iding a plausible a ionale o
he s onge esponse o fi ms wi h below-median expo sha es. The hi d column epo s
he in e ac ion effec be ween impo compe i ion and fi m size (measu ed by he numbe
o employees), indica ing a la ge effec among la ge fi ms. Column (4) in e ac s he
ade shock wi h a dummy o single p oduc fi ms.26 The co esponding es ima e p o ides
sugges i e e idence o a la ge efficiency imp o emen among mul i-p oduc fi ms. Finally,
column (5) in e ac s he change in impo exposu e wi h a dummy o h ee-digi sec o s
wi h high impo sha es in 1995. One migh expec ha he compe i i e en i onmen is less
affec ed by inc easing impo s om he Eas in sec o s ha we e al eady ela i ely open.
Consis en wi h his idea, he main effec is nega i e and significan , while he in e ac ion
effec is posi i e. This implies ha he educ ion in emissions was concen a ed among
fi ms in sec o s wi h lowe impo exposu e in 1995.
26 I am using single plan fi ms only he e because I do no know he numbe o p oduc s o
mul i-p oduc fi ms.
Impo compe i ion and emissions 765
When in e p e ing he he e ogenei y p esen ed abo e, accoun should be aken o he
ac ha pa o he a ia ion is d i en by ac oss sec o diffe ences. Fo ins ance, he
mos ene gy-efficien s eel p oduce will s ill emi mo e CO2pe uni o sales han he
leas -efficien ex ile company. Thus, he he e ogenei y ega ding emission in ensi y speaks
mo e o he ele ance o he effec o impo compe i ion on emissions o agg ega e emissions
and less o diffe en ial effec s depending on fi ms’ “CO2p oduc i i y.”
To analyze he e ogenei ies wi hin h ee-digi sec o s, I spli he sample in o wi hin-sec o
quin iles o emission in ensi y, expo sha es and size. Figu e A1 in he online appendix plo s
esul s o he baseline specifica ion (le subfigu e) and he specifica ion ha accoun s o
indus y ends ( igh subfigu e). S a ing wi h he subsamples o med based on diffe en
emission in ensi y quin iles, one can see ha he effec is ela i ely s able. All poin es i-
ma es in he le Subfigu e a e nega i e and o simila magni ude; only he effec among he
fi ms in he mos emission-in ensi e quin ile is sligh ly la ge han he es . Resul s om
specifica ions ha accoun o sec o al ends also hin a a sligh inc ease in he effec ’s
size o mo e emission-in ensi e fi ms— he ones ha p esumably ha e he mos oom o
imp o emen .27
When examining diffe ences in fi m size wi hin sec o s, he e exis s some indica ion o a
mo e significan decline in emission in ensi ies among la ge fi ms, aligning wi h he esul s
p esen ed in able 6. As o he expo sha e quin iles wi hin sec o s, he poin es ima es a e
less s able, possibly due o a ying numbe s o obse a ions wi hin quin iles, e.g., fi ms wi h
lowe expo sha es appea o ha e a lowe su i al a e. Resul s in figu e A1 sugges ha
fi ms wi h highe expo sha es ela i e o o he fi ms in hei h ee-digi sec o espond
s onge o inc eased impo compe i ion. In con as , he emission in ensi y o fi ms wi h
an abo e-median expo in ensi y showed a mu ed esponse (see able 6).
5. Addi ional esul s
5.1. Leakage
An ob ious conce n wi h his analysis is “ca bon leakage.” Fo ins ance, he inc ease in
impo s om he Eas could be high ca bon con en inpu s in domes ic fi ms’ p oduc ion
p ocess. I hese inpu s used o be p oduced by he domes ic fi ms hemsel es and ou sou ced
once he opening up o he Eas allowed fi ms o do so, he emission in ensi y o sales would
decline. To (pa ially) add ess his conce n, I es ima e he effec o impo compe i ion on he
emission in ensi y o alue added. In he scena io desc ibed abo e, alue added would decline,
so he emission in ensi y o alue added would emain unchanged o e en inc ease despi e a
dec ease in he emission in ensi y o sales. Recall ha his analysis is easible only o fi ms
included in he cos s uc u e su ey, which o e samples la ge fi ms (i.e., i includes fi ms
wi h >500 employees yea ly and addi ional fi ms in ou -yea blocks). Because he diffe ence
specifica ion u he dis o s he sample owa ds la ge fi ms, able 7includes es ima es om
a specifica ion ha con ols o ime-in a ian fi m-le el unobse ables h ough fi m fixed
effec s (simila o columns (5) and (6) in able A2 in he online appendix), in addi ion o
esul s om he baseline diffe ence specifica ions.
The baseline esul s in panel A o able 7indica e ha a 1 pp inc ease in he IPR
induces an app oxima e 0.3% o 0.23% dec ease in fi ms’ emission in ensi y o alue added.
In con as o he main esul s, condi ioning on indus y ends leads o a s a is ically
27 These esul s should be in e p e ed cau iously as a la ge effec among fi ms wi h highe
in ensi ies could also be ela ed o mean e e sion.
766 J. Leh
TABLE 7
Addi ional esul s – Value added
IV IV–FE OLS OLS–FE
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Panel A. CO2in ensi y o alue added
Δimpo s −0.0029*** −0.0025*** −0.0023** −0.0008 −0.0037*** −0.0018** −0.0026*** −0.0003 −0.0034*** −0.0002
(0.0009) (0.0009) (0.0009) (0.0010) (0.0007) (0.0008) (0.0006) (0.0006) (0.0005) (0.0005)
Numbe o obse a ions 75,499 75,499 75,499 75,499 285,581 285,581 75,499 75,499 285,581 285,581
Kleibe gen–Paap F-s a is ic 198.96 201.61 195.03 144.00 568.76 418.40
Panel B. Sha e o alue added
Δimpo s −0.0111 −0.0182 −0.0120 −0.0140 0.0054 0.0075 −0.0292*** −0.0300** −0.0222** −0.0211*
(0.0152) (0.0151) (0.0148) (0.0163) (0.0125) (0.0154) (0.0106) (0.0121) (0.0092) (0.0110)
Numbe o obse a ions 75,514 75,514 75,514 75,514 285,581 285,581 75,514 75,514 285,581 285,581
Kleibe gen–Paap F-s a is ic 198.97 201.62 195.08 144.00 568.95 418.24
Panel C. Log ma e ial in ensi y
Δ impo s 0.0005 0.0006 0.0005 0.0007 −0.0001 −0.0001 0.0009*** 0.0009*** 0.0007** 0.0007*
(0.0005) (0.0005) (0.0004) (0.0005) (0.0004) (0.0005) (0.0003) (0.0004) (0.0003) (0.0004)
Numbe o obse a ions 75,419 75,419 75,419 75,419 285,360 285,360 75,419 75,419 285,360 285,360
Kleibe gen–Paap F-s a is ic 198.91 201.56 194.98 143.90 568.97 418.02
CO2in ensi y–decile–yea dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Sales–decile–yea dummy Yes Yes Yes Yes Yes Yes Yes Yes Yes
Expo sha e–decile–yea dummy Yes Yes Yes Yes Yes Yes Yes Yes
Sec o dummy Yes Yes Yes Yes
NOTES: *p<0.1, ** p <0.05, *** p <0.01. Columns (1) o (6) show esul s om 2SLS es ima ions and columns (7) o (10) om an OLS es ima ion.
Columns (1) o (4) and (7) and (8) epo es ima es om he ou -yea diffe ence specifica ion, while columns (5), (6), (9) and (10) show esul s om a fixed
effec s es ima ion in which each fi m ecei ed a sepa a e fixed effec o 1995 o 2002 and 2003 o 2017 o accommoda e he b eak in he epo ing o ene gy
a iables. The dependen a iable in panel A is he log o CO2emissions pe uni o defla ed alue added, in panel B he sha e o alue added (i.e., alue added
di ided by sales) and in panel C he log o ma e ial in ensi y. The explana o y a iable is he IPR om he Eas in each h ee-digi indus y ins umen ed wi h
espec i e expo s om he Eas o o he coun ies. Fo each specifica ion, I epo he Kleibe gen–Paap F-s a is ic. S anda d e o s we e clus e ed bo h a he
fi m and a he h ee-digi indus y–yea le el.
SOURCES:Resea ch Da a Cen es o he Fede al S a is ical Office and he S a is ical Offices o he Lände : AFiD-Panel Indus iebe iebe, 1995–2017; own
calcula ions.

Impo compe i ion and emissions 767
insignifican coefficien o −0.0008 (column (4)). The es ima es in columns (5) and (6) based
on he wi hin es ima o yield somewha la ge poin es ima es and ma ch he main esul s
closely conce ning effec size. No only a e hese es ima es based on a la ge bu also he
sample is also a guably less dis o ed owa ds la ge fi ms. To u he in es iga e he ole o
impo compe i ion on he dep h o alue added, I ela e he sha e o alue added o e o al
sales o he IPR. Resul s a e shown in panel B o able 7. A declining sha e could poin
o a agmen a ion o fi ms’ alue chains. Howe e , he es ima es in panel B o able 7do
no indica e ha impo compe i ion om he Eas con ibu es o his because all poin
es ima es a e small and indis inguishable om ze o ac oss specifica ions.28 Taken oge he ,
he esul s p esen ed in panels A and B do no suppo he hypo hesis ha he effec o
impo compe i ion on he emission in ensi y o sales esul s om ca bon leakage. O cou se,
his analysis does no imply ha he impo s om he Eas do no also include in e media e
inpu s. Fo example, impo s om he Eas could displace impo s om o he coun ies.
Also impo s om he Eas in he h ee-digi sec o xcould s ill be inpu s in he h ee-digi
sec o y. Finally, panel C o he able epo s he effec on he log o ma e ial in ensi y, which
is closely ela ed o he esul s in panel B. Again, I do no find e idence o an inc ease in
ma e ial in ensi y, which one would expec i ups eam p oduc ion s eps we e sou ced ou .
5.2. Addi ional ou comes
Table A11 in he online appendix shows esul s o addi ional ou comes such as emissions
pe wo ke , numbe o employees, in es men and measu es o fi ms’ p oduc choice. This
analysis aims o unde s and be e how companies adjus o impo compe i ion. Based on
his, specula ions can also be made abou he mechanisms unde lying he main esul s on
emission in ensi y.
Fi s , and in line wi h Dau h e al. (2014), I documen a nega i e effec o inc easing
compe i i e p essu e on he numbe o employees, e.g., he baseline es ima e implies ha
he a e age fi m’s numbe o employees dec eased by app oxima ely 0.35% in esponse o a
1 pp inc ease in he IPR. Like ene gy inpu s, he OLS es ima es ha cap u e he associa ion
be ween he numbe o employees and impo s show a posi i e bias. Despi e a nega i e effec
on bo h p oduc ion inpu s—labou and ene gy—I find ha emissions pe employee all
significan ly in he baseline specifica ion (see panel B), which p o ides sugges i e e idence
o an imp o emen o ene gy efficiency ela i e o po en ial efficiency imp o emen s in he
use o o he inpu s.
Panel C epo s he es ima es wi h in e se hype bolic sine ans o med in es men as
he dependen a iable.29 On he one hand, in es men o upg ade echnology appea s as
a na u al means o fi ms o inc ease hei ene gy p oduc i i y. On he o he hand, in an
en i onmen in which compe i ion is becoming mo e in ense and ma ke sha es a e alling,
a gene al decline in in es men s would be expec ed. The poin es ima es in panel C a he
align wi h he second hypo hesis, sugges ing a subs an ial educ ion in in es men by 1.4%
in esponse o a 1 pp inc ease in he IPR ( ela i ely imp ecisely es ima ed hough). Again,
he OLS es ima es show a posi i e bias, e.g., fi ms in es mo e when hei p oduc s a e in
high demand.
Finally, panels D and E ocus on fi ms’ p oduc mix esponses. The dependen a iables
a e he numbe o p oduc s (panel D) and an indica o o single-p oduc fi ms (panel E).
28 The dependen a iable is on a scale be ween 0 and 100.
29 In es men is qui e lumpy; o he han he log, he in e se hype bolic sine ans o ma ion is
defined o ze o alues.
768 J. Leh
While changes in he ma ke en i onmen ha e been shown o affec fi ms’ p oduc mix
(see Goldbe g e al. 2010), Ba ows and Olli ie (2018) highligh he p oduc mix’s ole as
a de e minan o fi m-le el emission in ensi y. Fo example, when fi ms concen a e hei
economic ac i i y u he on hei co e compe ency, his could dec ease he fi ms’ emission
in ensi y. Panels D and E p o ide some sugges i e e idence in line wi h his idea: The
baseline esul sinpanelDsugges ha a1ppinc easein heIPRdec eases he a e age
fi m’s numbe o p oduc s by 0.005 and inc eases he chance o being a single-p oduc fi m
by 0.1%.
6. Discussion and conclusion
In his pape I analyze he effec o impo compe i ion on fi m-le el emission in ensi y. To
do so, I combine comp ehensi e fi m-le el da a om he Ge man manu ac u ing indus y
wi h sec o -le el ade flow. I ocus on he ise o China and Eas e n Eu ope be ween 1995
and 2017. Using an ins umen al a iable s a egy, I p o ide e idence ha inc easing impo
compe i ion is associa ed wi h ewe CO2emissions pe uni o sales. Wi hin-fi m changes
d i e his effec . I do no find s ong indica ion o be ween-fi m ealloca ion.
The baseline eg ession specifica ion implies ha fi ms’ emission in ensi y o p oduc ion
dec eases by 0.3% in esponse o a 1 pp inc ease in he impo pene a ion a io. Be ween
1995 and 2017 he impo pene a ion om he Eas inc eased by app oxima ely 20 pp.
Hence he poin es ima e sugges s ha he inc ease in compe i ion kep emission in ensi y
abou 6% below he le el i would ha e had in he absence o he ise o he Eas . The
esul s a e obus o examining emission in ensi y o alue added and accoun ing o new
expo oppo uni ies o he Eas . The la e analysis indica es ha new expo oppo uni ies
inc eased sales and emissions, i.e., fi ms scaled up, wi h only a small effec on emission
in ensi y.
The main esul s align wi h pa s o he in e na ional ade li e a u e, which ends
o find posi i e effec s o impo compe i ion on Eu opean fi ms’ p oduc i i y (see Shu
and S einwende 2019 and Chen and S einwende 2021). Gi en ha ene gy consump ion
is o en iewed inefficien (“ene gy-efficiency pa adox”), i appea s plausible ha imp o -
ing ene gy efficiency is an easy cos -cu ing measu e o s ay compe i i e unde oughe
condi ions.
The esul s u he align wi h findings by Gu ié ez and Teshima (2018), who show ha
Mexican manu ac u ing fi ms’ ene gy in ensi y dec eased in esponse o in ensi ying impo
compe i ion. Mo e ecen ly, Leisne e al. (2023) find e idence o a nega i e impac o impo
compe i ion om China on emissions in ensi ies among Danish fi ms, bu only among he
mos emission in ensi e ones. This esul fi s my he e ogenei y analysis, which also showed
ha he mo e emissions-in ensi e companies d i e he effec . Agains he backg ound o a
compa a i ely emissions-in ensi e indus y in Ge many, diffe en a e age effec s can be well
explained.
The ques ion abou po en ial mechanisms na u ally eme ges om my esul s. Th oughou
he analysis, I find consis en e idence o a nega i e effec o impo compe i ion on sales
and employmen . Mo eo e , Subsec ion 5.2 p o ides sugges i e e idence o fi ms adjus ing
hei p oduc mix, e.g., a sligh inc ease in he numbe o single-p oduc fi ms and a dec ease
in he numbe o p oduc s in esponse o an inc ease in he IPR. These adjus men s could
eflec fi ms concen a ing on hei p oduc i e co e ac i i ies—a plausible explana ion o
he efficiency imp o emen s (see Ba ows and Olli ie 2018).
While he es ima es on he effec o he ade shock on he emission in ensi y o alue
added and he alue added sha e do no suppo he hypo hesis ha ca bon leakage d i es
he esul s, u u e wo k using mo e g anula inpu da a could examine his po en ial channel
Impo compe i ion and emissions 769
mo e closely. Simila ly, imp o ed access o in e media es could enhance he o e all p oduc-
i i y o downs eam fi ms. Indeed, a small numbe o ecen pape s (see Ake man e al. 2021,
Dussaux e al. 2023, Leisne e al. 2023) ha e examined he effec o imp o ed access o in e -
media e inpu s on emission in ensi ies and ound consis en e idence o a nega i e effec
due o a disp opo iona e inc ease in ou pu .
F om a policy pe spec i e, one key message o his pape is ha p o-compe i i e policy
can lead o en i onmen al benefi s ia mo e efficien ene gy use and hence ewe emissions,
a message likely o be ele an beyond ade policy.
Suppo ing in o ma ion
Supplemen a y ma e ial accompanies his a icle. The da a and code ha suppo he
findings o his s udy a e a ailable in he Canadian Jou nal o Economics Da a e se a
h ps://doi.o g/10.5683/SP3/TAJMH0.
Re e ences
Acemoglu, D., D. Au o , D. Do n, G. H. Hanson, and B. P ice (2016) “Impo compe i ion and
he g ea US employmen sag o he 2000s,” Jou nal o Labo Economics 34(S1), S141–S198
Ake man, A., R. Fo slid, and O. P ane (2021) “Impo s and he CO2emissions o fi ms,” CEPR
discussion pape no. 6090
Au o , D., D. Do n, G. H. Hanson, G. Pisano, and P. Shu (2020) “Fo eign compe i ion and
domes ic inno a ion: E idence om US pa en s,” Ame ican Economic Re iew: Insigh s 2(3),
357–74
Au o , D.H., D. Do n, and G. H. Hanson (2013) “The China synd ome: Local labo ma ke effec s
o impo compe i ion in he Uni ed S a es,” Ame ican Economic Re iew 103(6), 2121–68
Au o , D.H., D. Do n, G. H. Hanson, and J. Song (2014) “ T ade adjus men : Wo ke -le el
e idence,” Qua e ly Jou nal o Economics 129(4), 1799–860
Ba ows, G., and H. Olli ie (2018) “Cleane fi ms o cleane p oduc s? How p oduc mix shapes
emission in ensi y om manu ac u ing,” Jou nal o En i onmen al Economics and
Managemen 88(C), 134–58
(2021) “Fo eign demand, de eloping coun y expo s, and CO2emissions: Fi m-le el
e idence om India,” Jou nal o De elopmen Economics 149, 102587
Bloom, N., M. D aca, and J. Van Reenen (2015) “T ade induced echnical change? The impac o
Chinese impo s on inno a ion, IT and p oduc i i y,” Re iew o Economic S udies 83(1),
87–117
Bloom, N., C. Genakos, R. Ma in, and R. Sadun (2010) “Mode n managemen : Good o he
en i onmen o jus ho ai ?,” Economic Jou nal 120(544), 551–72
Bomba dini, M., and B. Li (2020) “T ade, pollu ion and mo ali y in China,” Jou nal o
In e na ional Economics 125, 103321
B ucal, A., B. Ja o cik, and I. Lo e (2019) “Good o he en i onmen , good o business: Fo eign
acquisi ions and ene gy in ensi y,” Jou nal o In e na ional Economics 121, 103247
Chen, C., and C. S einwende (2021) “Impo compe i ion, he e ogeneous p e e ences o
manage s, and p oduc i i y,” Jou nal o In e na ional Economics 133, 103533
Che niwchan, J. (2017) “T ade libe aliza ion and he en i onmen : E idence om NAFTA and
US manu ac u ing,” Jou nal o In e na ional Economics 105, 130–49
Che niwchan, J., B. R. Copeland, and M. S. Taylo (2017) “T ade and he en i onmen : New
me hods, measu emen s, and esul s,” Annual Re iew o Economics 9(1), 59–85
Cole, M.A., and R. J. Ellio (2003) “De e mining he ade–en i onmen composi ion effec : The
ole o capi al, labo and en i onmen al egula ions,” Jou nal o En i onmen al Economics
and Managemen 46(3), 363–83
Copeland, B.R., J. S. Shapi o, and M. S. Taylo (2021) “Globaliza ion and he en i onmen ,”
NBER wo king pape no. 28797
770 J. Leh
Copeland, B.R., and M. S. Taylo (1994) “No h-sou h ade and he en i onmen ,” Qua e ly
Jou nal o Economics 109(3), 755–87
(2004) “T ade, g ow h, and he en i onmen ,” Jou nal o Economic Li e a u e 42(1), 7–71
Dau h, W., S. Findeisen, and J. Suedekum (2014) “The ise o he Eas and he a Eas : Ge man
labo ma ke s and ade in eg a ion,” Jou nal o he Eu opean Economic Associa ion 12(6),
1643–75
De Haas, R., R. Ma in, M. Muûls, and H. Schweige (2021) “Manage ial and financial ba ie s o
he ne -ze o ansi ion,” BOFIT Discussion Pape s, no. 6/2021
DeCanio, S.J. (1993) “Ba ie s wi hin fi ms o ene gy-efficien in es men s,” Ene gy Policy 21(9),
906–14
Dippel, C., R. Gold, S. Heblich, and R. Pin o (2021) “The effec o ade on wo ke s and o e s,”
Economic Jou nal 132(641), 199–217
Dussaux, D., F. Vona, and A. Dechezlep ê e (2023) “Impo ed ca bon emissions: E idence om
F ench manu ac u ing companies,” Canadian Jou nal o Economics 56(2), 593–621
Eu opean En i onmen al Agency (2023) “Eu opean Union Emissions T ading Sys em (EU ETS)
da a om EUTL”
Fo slid, R., T. Okubo, and K. H. Ull ei -Moe (2018) “Why a e fi ms ha expo cleane ?
In e na ional ade, aba emen and en i onmen al emissions,” Jou nal o En i onmen al
Economics and Managemen 91, 166–83
Ge a den, T.D., R. G. Newell, and R. N. S a ins (2017) “Assessing he ene gy-efficiency gap,”
Jou nal o Economic Li e a u e 55(4), 1486–525
Goldbe g, P.K., A. K. Khandelwal, N. Pa cnik, and P. Topalo a (2010) “Mul ip oduc fi ms and
p oduc u no e in he de eloping wo ld: E idence om India,” Re iew o Economics and
S a is ics 92(4), 1042–49
Gu ié ez, E., and K. Teshima (2018) “Aba emen expendi u es, echnology choice, and
en i onmen al pe o mance: E idence om fi m esponses o impo compe i ion in Mexico,”
Jou nal o De elopmen Economics 133, 264–74
Holmes, T., and J. Schmi z (2001) “A gain om ade om unp oduc i e o p oduc i e
en ep eneu ship,” Jou nal o Mone a y Economics 47, 417–46
Jaffe, A.B., and R. N. S a ins (1994) “The ene gy pa adox and he diffusion o conse a ion
echnology,” Resou ce and Ene gy Economics 16(2), 91–122
Kong, D., G. Ma, and N. Qin (2022) “T ade libe aliza ion and fi m oxic emissions,” Re iew o
In e na ional Economics 30(4), 971–1002
Leisne , J., J. R. Munch, A. T. Nielsen, and G. Schau (2023) “The impac o offsho ing and
impo compe i ion on fi m-le el ca bon emissions,” IZA discussion pape no. 16556
Le ine, R., C. Lin, Z. Wang, and W. Xie (2018) “Bank liquidi y, c edi supply, and he
en i onmen ,” NBER wo king pape no. 24375
Ma in, R., M. Muûls, L. B. de P eux, and U. J. Wagne (2012) “Ana omy o a pa adox:
Managemen p ac ices, o ganiza ional s uc u e and ene gy efficiency,” Jou nal o
En i onmen al Economics and Managemen 63(2), 208–23
Olley, G.S., and A. Pakes (1996) “The dynamics o p oduc i i y in he elecommunica ions
equipmen indus y,” Econome ica 64(6), 1263–97
Pe ick, S., K. Rehdanz, and U. J. Wagne (2011) “Ene gy use pa e ns in Ge man indus y:
E idence om plan -le el da a,” Jah büche ü Na ionalökonomie und S a is ik 231(3),
379–414
Rich e , P.M., and A. Schie sch (2017) “CO2emission in ensi y and expo ing: E idence om
fi m-le el da a,” Eu opean Economic Re iew 98, 373–391, ISSN 0014-2921
Schmid , K.M. (1997) “Manage ial incen i es and p oduc ma ke compe i ion,” The Re iew o
Economic S udies 64(2), 191–213, ISSN 00346527, 1467937X
Shu, P., and C. S einwende (2019) “The impac o ade libe aliza ion on fi m p oduc i i y and
inno a ion,” Inno a ion Policy and he Economy 19, 39–68
Umwel bundesam (2018) “En wicklung de spezifischen kohlendioxid-emissionen des deu schen
s ommix in den jah en 1990–2017,” ISSN 1862-4359