Egge , Pe e ; E ha d , Ka ha ina
A icle
He e ogeneous e ec s o a i and non a i ade-policy
ba ie s in quan i a i e gene al equilib ium
Quan i a i e Economics
P o ided in Coope a ion wi h:
The Econome ic Socie y
Sugges ed Ci a ion: Egge , Pe e ; E ha d , Ka ha ina (2024) : He e ogeneous e ec s o a i and
non a i ade-policy ba ie s in quan i a i e gene al equilib ium, Quan i a i e Economics, ISSN
1759-7331, The Econome ic Socie y, New Ha en, CT, Vol. 15, Iss. 2, pp. 453-487,
h ps://doi.o g/10.3982/QE1994
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/320303
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-nc/4.0/
Quan i a i e Economics 15 (2024), 453–487 1759-7331/20240453
He e ogeneous e ec s o a i and non a i ade-policy
ba ie s in quan i a i e gene al equilib ium
Pe e H. Egge
Depa men o Managemen , Technology, and Economics, ETH Zü ich, CEPR, and CESi o
Ka ha ina E ha d
DICE, Hein ich Heine Uni e si y Düsseldo and CESi o
S uc u al quan i a i e wo k in in e na ional economics ypically models ade
cos s as a log-linea unc ion o exogenous ade-policy a iables. We p opose a
s uc u al app oach ha allows o a nonpa ame ic ela ionship and o ea -
ing a i and non a i ade-policy a iables as po en ially endogenous. The da a
ejec he assump ion o log-linea i y o ade cos s in bo h a i - and non a i -
policy a iables. We assess he e ec s o a unila e al inc ease o US a i s on Chi-
nese impo s by 10 pe cen age poin s and documen ha he es ima ed e ec s on
eal bila e al ade- low changes would be subs an ially unde es ima ed by s an-
da d app oaches.
Keywo ds. T ade policy, g a i y models, semipa ame ic me hods, nonpa ame -
ic me hods, gene alized p opensi y sco es.
JEL classi ica ion. C14, F13, F14.
1. In oduc ion
Vi ually e e y announcemen o a ade-policy in e en ion is ollowed by an a emp
o quan i y i s economic consequences. Economis s achie e his by employing pa ame-
e iza ions o s uc u al models, p edominan ly gene al-equilib ium models, which o -
malize he impac o ade-policy shocks. Coun less s udies ha e aimed a quan i y-
ing he e ec o ade libe aliza ions in his spi i .1One impo an insigh gained om
his wo k is ha he economic ou come esponses o a uni o m ade-policy shock di -
e ac oss coun ies due o hei dis inc undamen als. Howe e , despi e inco po a ing
ich mechanisms ha p oduce he e ogeneous esponses ac oss coun ies, con en ional
Pe e H. Egge : [email p o ec ed]
Ka ha ina E ha d : [email p o ec ed]
We g a e ully acknowledge nume ous help ul commen s on ea lie e sions o he manusc ip by h ee
anonymous e iewe s as well as by Chad Bown, Lo enzo Caliendo, A naud Cos ino , Jona han Ea on, Al-
onso Flo es-Lagunes, and Geo g Schau . We a e also hank ul o commen s om pa icipan s a se e al
semina s (Uni e si y o Salzbu g, Uni e si y o Wü zbu g, Uni e si y o Ox o d, Uni e si y o Mannheim)
and con e ences (Villa s Wo kshop on In e na ional Economics, PRONTO Wo kshop in Pa is, DEGIT in
No ingham, EEA in Gene a, ETSG in Pa is, S oos Sine gia Wo kshop, TRISTAN wo kshop in Bay eu h).
1See, o example, ecen wo k on B exi by B einlich e al. (2016), he T ansa lan ic T ade and In es men
Pa ne ship (TTIP) by Felbe may , Aichele, and Heiland (2016), o he 2018 ade wa be ween he US and
China by Fajgelbaum, Goldbe g, Kennedy, and Khandelwal (2019).
©2024 The Au ho s. Licensed unde he C ea i e Commons A ibu ion-NonComme cial License 4.0.
A ailable a h p://qeconomics.o g.h ps://doi.o g/10.3982/QE1994
454 Egge and E ha d Quan i a i e Economics 15 (2024)
quan i a i e gene al-equilib ium models o en impose a leas h ee impo an es ic-
i e assump ions.
Fi s , hey impose a homogeneous and log-linea ela ionship be ween ade-policy
a iables and ad alo em ade cos s and a log-linea di ec ela ionship be ween ad al-
o em ade cos s and ade lows.2He e ogeneous ade-cos e ec s eme ge only indi-
ec ly, and mainly h ough gene al-equilib ium epe cussions, ha is, h ough e ec son
consume and p oduce p ices (see, e.g., Ea on and Ko um (2002), Ande son and an
Wincoop (2003), A kolakis, Cos ino , and Rod iguez-Cla e (2012)). This pape demon-
s a es ha he da a appea o ejec a log-linea di ec ela ionship be ween ade policy
and ade lows and illus a es ha assuming i dep esses he he e ogenei y o equilib-
ium esponses o ade-policy changes.3
A second cus oma y es ic ion lies in he ocus on ade policy h ough a i s
alone.4Non a i ba ie s o ade a e associa ed wi h he applica ion o speci ic check-
ing ou ines a bo de s and he implemen a ion o s anda ds and p ocedu es wi h an
in en o p o ec domes ic supplie s (see Ande son (2016)). Non a i policy egula ions
ha e become ex emely impo an since he U uguay T ade Round (see Ho n, Ma oidis,
and Sapi (2010)). They ha e ga ne ed a en ion and become he ocus o in e es in he
con ex o he “new” p o ec ionism since he beginning o he Economic and Financial
C isis (Bown (2011), Baldwin and E ene (2012), Bown and C owley (2013)). The use o
non a i measu es is ele an i a i s and non a i ba ie s a e se join ly and no in-
dependen ly by policymake s, and i hei pa ial e ec s depend on each o he . We will
allow o he la e and documen ha esponses o a i changes end o depend on
non a i p o isions and ice e sa.5
A inal cus oma y bu es ic i e assump ion is ha ade-policy measu es a e
ea ed as andomly assigned o coun ies a he han chosen wi h an economic a io-
nale.6Economic heo y hypo hesizes ha coun ies choose a i s based on hei unda-
men als (see Bond (1990), Bond and Sy opoulos (1996), Bagwell and S aige (1999,2004),
Ossa (2011,2014), Felbe may , Jung, and La ch (2013), Caliendo, Feens a, Romalis, and
Taylo (2015)).
2Ea on and Ko um (2002) and Hende son and Millime (2008) conside nonpa ame ic di ec e ec s
o geog aphy on ade cos s and ade lows, bu he esul s do no poin o any s ong nonlog-linea i y
o he di ec e ec s a la ge geog aphical dis ance. The indings o Hillbe y and Hummels (2008) sugges ,
howe e , ha geog aphy induces nonlog-linea e ec s o e sho dis ances.
3In a ecen s udy, Adão, Cos ino , and Donaldson (2017) emphasize he impo ance o elaxing unc-
ional o m es ic ions in gene al equilib ium models o ade bu do no conside nonpa ame ic ade
cos s.
4See, o example, Romalis (2007)o Caliendo and Pa o (2015).
5Rela ed o his, Caliendo, Op omolla, Pa o, and S o za (2021) sugges ha , in quan i a i e models o
mig a ion, i is insu icien o con ol o di ec mig a ion cos s alone, bu o he policy domains need o be
be conside ed, oo.
6See, o example, Ea on and Ko um (2002)o Caliendo and Pa o (2015). I a all, endogenei y o ade
policy is mos ly conside ed h ough he (bina y) membe ship o coun ies in p e e en ial ade ag eemen s
(see 2007 (2007,2009)). Wi h he excep ion o Egge , La ch, S aub, and Winkelmann (2011) and Egge e al.
(2011), ela ed wo k on endogenous ade ag eemen s is conce ned wi h he di ec (pa ial) a he han he
o al (di ec plus indi ec ) e ec s on economic ou come.
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 455
The p esen pape pu s o wa d a quan i a i e amewo k and analysis o ade-
policy e ec s on ade cos s and ade lows in a uni ied amewo k o elax he a o e-
men ioned h ee assump ions. I builds on a mul icoun y, mul isec o quan i a i e
amewo k ha is consis en wi h a wide ange o ade models (compa e A kolakis,
Cos ino , and Rod iguez-Cla e (2012), Cos ino and Rod iguez-Cla e (2014)). To add ess
he a o emen ioned conce ns, he pape sugges s implemen ing he ollowing mul is ep
app oach.
Fi s , ade da a—he e, o 115 coun ies and 128 sec o s—a e decomposed o ex-
ac in o ma ion on exogenous p oduce -coun y-sec o undamen als as well as o al
ad alo em ade ic ions. Second, machine-lea ning algo i hms a e used o decom-
pose a i and non a i policy ba ie s in o hei de e minis ic (p edic ed) and esidual
( andom) pa s. In line wi h he li e a u e on op imal ade policy (Bond and Sy opoulos,
1996), Ossa (2011,2014), Felbe may , Jung, and La ch (2013), we impose ha ade policy
and ade lows depend on he same undamen als in hei educed o m. Thi d, we es i-
ma e he link unc ion o ad alo em ade cos s on a i and non a i ba ie s, using he
join densi y o hei andom componen s. This app oach is a mul i a ia e gene aliza-
ion o he dose- esponse- unc ion es ima ion in Flo es, Flo es-Lagunes, Gonzalez, and
Neumann (2012). We p esen e idence o a nonlog-linea mapping o he ade-policy
a iables wi h ad alo em ade cos s.
We ind ha he ma ginal e ec o an inc ease in a i s is e y s ong o e y low and
medium a i ba ie s, while i is much weake and e en close o ze o o e y high a -
i ba ie s, especially, when non a i ba ie s a e high. Non a i ba ie s inc ease ade
cos s, in pa icula , o e y low ini ial le els o non a i ba ie s. Fo medium le els o
non a i ba ie s, ma ginal e ec s on ade cos s can ac ually be ade-cos - educing,
which is owed o he bene icial e ec s o some echnical ba ie s o ade. These pa -
e ns a e consis en wi h a i and non a i a oidance a high le els o hese cos s (see
Fisman and Wei (2004), Ja o cik and Na ciso (2008), Sequei a (2016), and Demi and
Ja o cik (2018)), and wi h a lacking usage o g an ed p e e en ial a i s a low mos a-
o ed na ion a i le els (see He in (1986), F ancois, Hoekman, and Manchin (2006),
Es e adeo dal, F eund, and O nelas (2008), Fugazza and Nici a (2013), K ishna, Sala-
manca, Suzuki, and Ma incus (2021)).
To demons a e he impo ance o hese nonlinea i ies, we eed he es ima ed
ade-policy g adien s in o a quan i a i e mul icoun y, mul isec o gene al-equilib ium
model o ade and e alua e he e ec o a unila e al inc ease in US a i s on Chinese
impo s o 10 pe cen age poin s. The e ec s o his pa icula policy change would be
se e ely unde es ima ed by a cus oma y modeling o ade cos s as log-linea in a i s
compa ed o he lexible-g adien app oach p oposed in his pape . The a e age educ-
ion ac oss all ea ed sec o -le el US impo sha es om China is abou 7 pe cen age
poin s la ge wi h he lexible-g adien app oach and o al US impo s om China (e al-
ua ed a benchma k income le els) all by 6% as compa ed o only 3% unde an ad al-
o em speci ica ion.
456 Egge and E ha d Quan i a i e Economics 15 (2024)
2. The e ec o ade policy in g a i y models o in e na ional ade
Quan i a i e wo k in in e na ional economics on he e ec o ade-policy changes is
almos exclusi ely based on g a i y models o a leas h ee easons: (i) he empi i-
cal success o s uc u al g a i y es ima ion (Head and Maye ,2014), (ii) he wide ange
o heo e ical gene al-equilib ium ade models leading o a g a i y equa ion includ-
ing models wi h appealing mic o ounda ions (Cos ino and Rod iguez-Cla e,2014), and
(iii) he pa simonious na u e ha allows esea che s o quan i y he (wel a e) conse-
quences o changes in undamen als elying on only e y ew key pa ame e s—in pa ic-
ula , es ima es o he so-called ade elas ici y—while s ill being able o ake in o accoun
gene al-equilib ium e ec s.7
Conside he simples case o such a g a i y model—a single-sec o A ming on
model whe e di e en coun ies a e endowed wi h a ixed quan i y, Qi, o dis inc goods
and each coun y is popula ed by a ep esen a i e consume wi h cons an elas ici y o
subs i u ion (CES) p e e ences o e hese goods. Bila e al ade lows, Xij , be ween an
expo e iand an impo e jwill be gi en by he well-known CES demand o he ex-
po e ’s good, which is a unc ion o he good’s p ice (mill p ice, Pii, imes ad alo em
ade cos s, Dij ), he p ice index in j,Pj, and he impo e ’s o al expendi u e (which
equals income), Yj. The elas ici y αis a linea unc ion o he elas ici y o subs i u ion
speci ied in he CES p e e ences:
Xij =(PiiDij )αYj
Pj
(1)
No e ha bila e al sales consis o an expo e -speci ic pa , an impo e -speci ic com-
ponen , and bila e al ade cos s. The mill p ice o expo e i’s good, Pii, is he o al alue
o he good coun y iis endowed wi h, Yi, di ided by o al endowmen Qi,Pii =Yi/Qi.
While he quan i y is exogenously gi en, he p ice o he good is endogenously de e -
mined in gene al equilib ium. The impo e -speci ic pa consis s o o al expendi u e
o pa ne coun y j,Yj, di ided by he p ice index p e ailing in ha coun y, Pj.8The
p ice index i sel is a unc ion o mill p ices Pkk o all coun ies, k=1...J, and ade cos s
Dkj be ween all coun ies k=1...Jand j.
Using his model o ade-policy analysis, i is cus oma y o assume ha ad alo em
ade cos s, Dij , a e p opo ional o ad alo em a i s. Then he di ec e ec o a i s
on ade— he e ec be o e gene al-equilib ium adjus men s o p ices and income—is
log-linea and go e ned by α. No e ha his di ec e ec is uni o m o all coun ies and
i espec i e o hei undamen als. Hence, a educ ion in a i s would ha e he same
di ec e ec in a coun y i espec i e o i s p e ious a i le el, i s implemen a ion o
non- a i ba ie s o how ha d i is o ac ually apply p e e en ial ma ke access o mali-
ies. Only once gene al-equilib ium adjus men s a e aken in o accoun , he e ogeneous
e ec s o ade policy can eme ge h ough he endogenous adjus men o mill p ices.
7We p o ide de ails on he gene al equilib ium o mula ion o he speci ic pa simonious ade model
ou lined in his pape in a Supplemen al Appendix (Egge and E ha d (2024)) o his pape .
8No e ha we abs ac om a i e enues he e; his is elaxed la e .
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 457
While he A ming on model is admi edly s ylized, he gene al s uc u e o bila e al
ade lows ou lined abo e holds o a wide ange o models some o hem inco po a ing
sophis ica ed mic o- heo e ical mechanisms. In hese so-called g a i y models, ade
lows, Xs
ij , be ween an expo e iand an impo e jwi hin a sec o sdepend mul iplica-
i ely on h ee componen s: supply-po en ial ac o s ha a e expo e -sec o -speci ic,
As
i, demand-po en ial ac o s ha a e impo e -sec o -speci ic, Bs
j, and ic ion ac o s
ha a y a he sec o -coun y-pai le el, Ds
ij , and whose impac on ade lows is go -
e ned by he ade elas ici y, αs:
Xs
ij =As
iBs
jDs
ij αs.(2)
Theo e ical models esul ing in a g a i y- ype model o in e na ional ade lows di -
e mainly wi h espec o he s uc u al in e p e a ion o he expo e -sec o -speci ic
componen , As
i, and he impo e -sec o -speci ic componen , Bs
j, bu no wi h espec
o he bila e al componen , Ds
ij .9This bila e al componen is ypically simply e e ed o
as “icebe g- ype” ade cos s and assumed o encompass all po en ial ic ions o ade
in some unspeci ied way.10 Absen any heo e ical guidance, mos empi ical wo k as-
sumes a log-linea ade-cos unc ion. Jus as abo e, modeling changes in ad alo em
a i s will lead o he same es ic i e, uni o m di ec e ec s on ade lows i espec i e
o he unde lying complex mic o- heo e ical mechanisms.
How can we be e unde s and how bila e al ade cos s Ds
ij and ade policy a e e-
la ed?
To begin, we can ob ain an es ima e o hese cos s in logs,
ds
ij ,11 by decomposing
p oduc -le el bila e al expo s (in logs) in o hei p oduc -le el impo e and expo e -
speci ic componen s:
xs
ij =as
i+bs
j+αsds
ij ,(3)
es ima ing a linea ixed-e ec s eg ession o ob ain es ima es o as
iand bs
jand using
ade elas ici ies om he li e a u e (Kee, Nici a, and Ola eaga (2008)) o back ou an
es ima e o bila e al sec o -speci ic ade cos s:
ds
ij =1
αsxs
ij −
as
i−
bs
j.(4)
Ou objec i e is o comp ehend he na u e o ade cos s u he . To his end, we will
ex end he no ion o ade policy beyond i s ocus on a i s aking in o accoun non a i
ba ie s and allow o po en ially nonlinea e ec s o ade policy on ade cos s.
On he one hand, non a i policy egula ions ha e become ex emely impo an
since he U uguay T ade Round (see Ho n, Ma oidis, and Sapi (2010)) and hey ha e
ecen ly become he ocus o in e es in he con ex o he “new” p o ec ionism since
9In he simple A ming on model abo e, o ins ance, Ai=(Yi/Qi)αand Bj=Yj/Pj.
10No e ha we es ic ou analysis o ade models ea u ing a cons an ade elas ici y and ha his
assump ion has consequences, o example, o wel a e.
11Th oughou he pape , ha s will indica e es ima es and lowe case le e s, x, will e e o he log o a
a iable, X.
458 Egge and E ha d Quan i a i e Economics 15 (2024)
he beginning o he Economic and Financial C isis (Bown (2011), Baldwin and E ene
(2012), Bown and C owley (2013)). Non a i ba ie s o ade a e associa ed wi h he ap-
plica ion o speci ic checking ou ines a bo de s and he implemen a ion o s anda ds
and p ocedu es wi h an in en o p o ec domes ic supplie s (see Ande son (2016)). In
con as o a i s, ce ain non a i ba ie s migh also ha e ade-enhancing e ec s, o
example, by es ablishing us in p oduc s h ough s anda ds o dec easing ansac ion
cos s (WTO (2012)).
On he o he hand, se e al s udies ques ion he assump ion o log-linea ade cos s
based on a leas i e a gumen s. Fi s , he nexus be ween a i and non a i ba ie s
and ade cos s is a ec ed by misdecla a ions a cus oms (see, e.g., Demi and Ja o cik
(2018), Fisman and Wei (2004), Ja o cik and Na ciso (2008), Sequei a (2016)). Second, a
nonlog-linea ela ionship be ween ade-policy a iables and ade cos s eme ges once
some ade-policy measu es al e p ices nonp opo iona ely (as is he case wi h speci ic
a i s) a he han p opo iona ely (see, e.g., Hummels and Skiba (2004), I a azabal,
Moxnes, and Op omolla (2015)). Thi d, wo k on he size and ole o p e e ence ma gins
sugges s ha a ailable p e e en ial ma ke access is unde used (see, e.g., He in (1986),
F ancois, Hoekman, and Manchin (2006), Es e adeo dal, F eund, and O nelas (2008),
Fugazza and Nici a (2013), K ishna e al. (2021)). Fou h, he unce ain y abou expec ed
applied a i and non a i ba ie s may induce nonlog-linea e ec s o ade policy on
ade cos s (see, e.g., Handley and Limao (2017), Handley and Limao (2015), Pie ce and
Scho (2016), and C owley, Song, and Meng (2016)). Finally, nonlinea i ies in he ela-
ionship be ween ade-policy ins umen s and e ec i e ade cos s may ela e o he
incomple e bu ade-cos -dependen pene a ion o consume ma ke s (see, e.g., A ko-
lakis (2010)).12
In a i s s ep, we will e alua e how ade cos s,
ds
ij , a y along he dimensions o wo
measu es o ade policy: a i s, τs
ij =ln(1+ s
ij ), and non a i ba ie s, ηs
ij =ln(1+ns
ij ),
each exp essed in ad alo em e ms, in a po en ially nonlinea way using a highe -o de
polynomial app oxima ion while con olling o cus oma y exogenous ade ba ie s
and 4-digi sec o ixed e ec s.13 In his exe cise, we ea τs
ij and ηs
ij as exogenous de e -
minan s o
ds
ij .
We illus a e he esul s g aphically in Figu e 1. The uppe panel illus a es a so-
called dose- esponse unc ion: i s a es how log s a u o y applied a i - and non a i -
ba ie ad alo em a es map in o e ec i e log ad alo em ade cos s ds
ij .The wolowe
panels ep esen he wo-dimensional a e age g adien o he dose- esponse unc ion
wi h espec o he wo policy a iables a e aging o e he espec i e o he ade-policy
a iable. The g adien unc ion co esponds o he ma ginal e ec s o a i and non-
a i ade policy a di e en le els o ade policy. Wha hese igu es sugges is ha ,
o a i s, he g adien is la e a he lowe and he highe end o he suppo , and o
12Tha he ade-cos unc ion may be nonlinea in i s a gumen s inds also s ong suppo in he li e a-
u e assessing dis ance e ec s on ade (Hillbe y and Hummels,2008).
13Da a on 4-digi sec o al ade lows, xs
ij , obackou
ds
ij and applied a i s, s
ij , a he coun y-pai -sec o
le el o he yea 2011 a e aken om he T ade Analysis In o ma ion Sys em (TRAINS) Da abase con ained
in he Wo ld Bank’s WITS Da abase. Da a on ad alo em equi alen s o non a i ba ie s, ns
ij , a e a ailable
om Kee, Nici a, and Ola eaga (2016). Fo mo e de ails, see he espec i e da a desc ip ion in Sec ion 4.
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 459
Figu e 1. T ade cos s as a lexible polynomial unc ion o ade policy.
non a i ba ie s, i also la ens ou a he highe end. This is consis en wi h he e a-
sion/a oidance as well as he unde usage a gumen s om ea lie wo k.
Howe e , his e idence mus be aken wi h a g ain o sal as (i) i makes an ad hoc
(polynomial) unc ional o m assump ion and (ii) i ea s he ade-policy measu es τs
ij
and ηs
ij as exogenous. While he la e is ue o many empi ical s udies on he e ec s
o ade policy in quan i a i e ade models, ea lie wo k conside s hei endogenei y
in heo y as well as in empi ical analyses (see Bond and Sy opoulos (1996), Ossa (2011,
2014), Felbe may , Jung, and La ch (2013), Caliendo e al. (2015)). The nex sec ion p o-
poses an app oach owa ds modeling ade cos s as a lexible unc ion o ade-policy
a iables ha a e join ly de e mined wi h ade lows by he same exogenous unda-
men als as in he jus -men ioned ea lie wo k.
3. Econome ic me hodology
In modeling he dependence o ade cos s on a i and non a i ade-policy ba ie s,
we wan o pe mi a su icien ly lexible unc ional o m and o con ol o he unda-
men als, which join ly de e mine ade lows and ade policy (see Bond (1990), Bond
460 Egge and E ha d Quan i a i e Economics 15 (2024)
and Sy opoulos (1996), Ossa (2011,2014), Felbe may , Jung, and La ch (2013), Caliendo
e al. (2015)).14
In his con ex , i is impo an o no e ha s uc u al ade models in he quan i-
a i e li e a u e we wish o speak o do no only pe mi decomposing ade lows in o
hei supply, demand, and ic ion componen s as ou lined in equa ion (2). They also
iden i y measu es o he endogenous componen s in supply ac o s, and hence, hey
pe mi iden i ying composi e measu es o he exogenous undamen als o ade lows
and policy. Speci ically, in cus oma y quan i a i e ade models, he e is a undamen al
d i e o ade a he expo e -sec o le el, which we will deno e by Fs
i. This could be
an exogenous endowmen as in he simple A ming on model, a supply-side pa ame e
such as p oduc i i y (see Ea on and Ko um (2002)) o a demand-side pa ame e like
p e e ences o his pa icula coun y-sec o ’s good (see Ande son and an Wincoop
(2003)). This componen is exogenous o he model. Mo eo e , in g a i y- ype gene al
equilib ium models, endogenous ac o p ices ensu e ha ma ke s clea . We deno e his
endogenous componen by Ws
i. The sensi i i y o ade lows o hese coun y-sec o -
speci ic ac o p ices is go e ned by an elas ici y, which is ypically iden ical o (o a
leas , code e mined by) he ade elas ici y αs. No e ha all h ee componen s o he
expo -sec o - speci ic componen o ade lows in equa ion (2) a e log-addi i e in en-
dogenous (Ws
i) and exogenous de e minan s (Fs
i) o expo e po en ial:15
As
i=Fs
iWs
iαs.(5)
This s uc u al decomposi ion is impo an , as a la ge body o wo k on endogenous
ade policy conside s he undamen als behind he (op imal) choice o ade-policy pa-
ame e s o be he same as he ones behind he endogenous ac o and ou pu p ices,
which co-de e mine supply and demand, and hence, ade lows (see Bond and Sy-
opoulos (1996), Baie and Be gs and (2004), Bond, Riezman, and Sy opoulos (2004),
Ossa (2011,2014), Felbe may , Jung, and La ch (2013), Caliendo e al. (2015)). Acco d-
ing o his li e a u e, all ha ma e s o he sys ema ic de e mina ion o ade policy
a e he sec o -coun y-pai na u al ade cos s, he sec o -coun y undamen al d i e s
o supply po en ial, and he sec o -le el ade elas ici ies. Condi ional on hese ac o s,
ade policy is s ochas ically independen o o andom o ade lows. The eason is ha ,
upon a comple e decomposi ion o ade lows in a gene ic gene al-equilib ium se ing
14The endogenei y o ade-policy ba ie s can be add essed by ei he ins umen al- a iable (IV) es i-
ma ion o app oaches elying on an assump ion o condi ional mean independence (CMI). IV es ima ion
in he p esen con ex would equi e ha shi e s o ade policy could be ound, which a e independen
o any o he measu able o unmeasu able ade-cos ac o s. CMI, by con as , equi es us o model he
endogenous componen o ade-policy a iables obse ed in he da a. Using ou side ins umen s in a
mul icoun y quan i a i e gene al-equilib ium se ing appea s unna u al while he li e a u e sugges s ha
ade-policy a iables a e condi ionally—on he undamen als de e mining he endogenous model ou -
comes join ly wi h policy—mean independen .
15Adão, Cos ino , and Donaldson (2017) es ablish a quan i a i e ade model suppo ing nonpa ame ic
e ec s o p e e ences and echnology—bo h o which a e ing edien s o Fs
i—on ade lows in s om i o j.
The in e es he e is on a nonpa ame ic link be ween endogenous ade policy and o e all ade cos s on
he one hand, and ade lows on he o he hand, an issue which is no add essed in Adão, Cos ino , and
Donaldson (2017). Hence, he wo app oaches appea complemen a y o each o he .
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 467
Table 2. Examples o es ima ed coun y-sec o sales undamen als e sus es ima ed coun y-
sec o ixed e ec s in he da a.
Sec o
S uc u al Me al Mo o Vehicles S uc u al Me al Mo o Vehicles
Coun y ˆ
s
iˆ
as
i
China −22.86 6.30 5.38 5.22
Ge many −20.74 11.21 3.99 7.60
Japan −23.21 10.90 0.39 7.66
Uni ed S a es −21.14 9.95 3.66 6.44
Mexico −27.19 6.89 −0.26 3.92
India −27.13 6.36 1.73 4.24
B azil −25.61 5.15 −0.31 2.49
helps us lea ning he subse o ele an (polynomial and in e ac ion) e ms o hese ac-
o s, which a e nume ous o e e y sec o and coun y pai . This pa o he analysis
is in e es ed in sepa a ing he condi ional mean o a i and non a i policy measu es
om he s ochas ic pa , νs
τ,ij and νs
η,ij :
τs
ij =gτqs
ij +νs
τ,ij ,ηs
ij =gηqs
ij +νs
η,ij . (10)
I is c ucial o allow o a high deg ee o lexibili y in es ima ing gm(·). We achie e his by
applying a powe ul app oach ha es ima es he ela ionship nonpa ame ically using
mul i a ia e adap i e eg ession splines (MARS) ollowing F iedman (1991). In o de o
accoun o any coun y-speci ic cha ac e is ics in policy o ma ion, we allow o ixed
e ec s ac oss impo ing coun ies jand expo ing coun ies i.27
The model selec ion o he MARS model along wi h c oss- alida ion s a is ics is p e-
sen ed in Figu e 2. The selec ed model is able o explain 50% o he a ia ion in τs
ij and
39% o he a ia ion in ηs
ij . Up o almos one-hal o he basic a iables in qs
ij as well as
he impo ing-coun y and expo ing-coun y ixed e ec s a e chosen as p edic o s and
en e he model in a o al o 223 and 188 e ms (such as in e ac ions o powe s), espec-
i ely.
A undamen al conce n wi h espec o machine-lea ning algo i hms is o e i ing.28
No e ha he employed algo i hm con ains a backwa d passage in which a subse o
p e iously selec ed eg esso s (e.g., polynomial e ms o in e ac ion e ms) is dele ed o
a oid o e i ing. We can assess he deg ee o o e i ing by means o c oss- alida ion.
Speci ically, in he p esen con ex we gene a e 5- old c oss- alida ed models o a i
and non a i ade ba ie s. Fo each old, he algo i hm builds a MARS model wi h he
27The MARS algo i hm applied in his con ex has wo s eps. Fo wa d passage: A e es ima ing an in-
e cep , piecewise-linea basis unc ions o he co a ia es a e added i e a i ely (allowing o in e ac ions).
Backwa d passage: In o de o a oid o e i ing, a subse o he p e iously selec ed e ms is dele ed, and he
inal model is selec ed based on he minimiza ion o he gene alized c oss- alida ion (GCV) sco e.
28In ac , unde i ing would be much mo e o a conce n economically han o e i ing he e, because
wha is key is ha he esidual is no sys ema ically in o med by he exogenous undamen als. Howe e ,
s a is ically, one may s ill ca e abou o e i ing.
468 Egge and E ha d Quan i a i e Economics 15 (2024)
Figu e 2. Model selec ion o gτ(·)and gη(·).
in- old da a (90% o he comple e da a) and uses his model o measu e he R2 om
p edic ions made on he ou -o - old da a (10% o he comple e da a). The mean o hese
ou -o - old R2s a is ics amoun s o 0.494 (wi h a s anda d de ia ion o 0.0.20) o he
a i model (gτ(qs
ij )) and 0.289 (wi h a s anda d de ia ion o 0.138) o he non a i
model (gη(qs
ij )). These s a is ics a e qui e close o he R2s a is ics o he model es ima ed
on he ull da a, in pa icula , o a i s. While we abs ain om an in-dep h analysis o
he educed- o m models o he wo ade-policy a iables τs
ij and ηs
ij he e, we discuss
some ela ionships in a Supplemen al Appendix o his pape .
The esiduals νs
ij =(νs
τ,ij ,νs
η,ij )o he eg essions in equa ion (10) se e as es ima es
o he wo condi ional (quasi- andomized) a i and non a i ade-policy- ea men
a iables whose join densi y has o be es ima ed. Fo mos o he subsequen analy-
sis, we es ima e he join densi y o he la e assuming a bi a ia e no mal dis ibu ion
and es ima e he pa ame e s o he dis ibu ion by maximum likelihood (see, e.g., Imai
and Van Dyk (2004), Hi ano and Imbens (2004), Klu e, Schneide , Uhlendo , and Zhao
(2012), o he assump ion o no mal densi ies in he con ex o uni a ia e-con inuous-
ea men -e ec s es ima ion). We conduc an al e na i e nonpa ame ic es ima ion ha
allows o a maximum deg ee o lexibili y ollowing Li and Racine (2006).
The es ima ed bi a ia e densi y by one o he a o emen ioned me hods se es as
an es ima e o he p opensi y o ge ing andomly assigned o a speci ic uple o a -
i and non- a i -ba ie le els o any coun y pai and sec o . The densi y as a com-
pac (p opensi y) sco e can be ob ained no only o obse ed bu e en o po en ial
(hypo he ical) ade-policy ea men le els. We will e e o his densi y as gene alized
p opensi y sco e. Howe e , his compac sco e is only meaning ul, i he co a ia es in qs
ij
a e simila o all uni s {ij s }wi h a simila le el o he es ima ed join densi y. Towa ds an
assessmen o he la e , we en o ce a common suppo and disca d obse a ions wi h
ex eme join -densi y-sco e alues. Speci ically, we ollow Flo es e al. (2012) in de ining
he common suppo and ex end hei me hodology o mul i a ia e ea men s.29
29De ails ega ding he de ini ion o he common suppo can be ound in he Supplemen al Appendix.
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 469
Besides ensu ing common suppo , we will es i o any obse a ion wi h he same
GPS he p obabili y o a speci ic le el o ade-policy ea men is independen o he
obse able de e minan s qs
ij ollowing Hi ano and Imbens (2004). Fo each co a ia e in
qs
ij ,wemayconduc a - es unde he null hypo hesis ha he mean o he co a ia e is
he same ac oss g oups ha co espond o di e en le els o ade policy. Speci ically,
we pe o m such a es uncondi ionally e sus condi ionally on he GPS. The espec i e
es s a is ics a e epo ed in he Supplemen al Appendix. We show ha condi ioning on
he GPS imp o es he sha e o balanced co a ia es (a he 5% le el) om 31% o 96%.
We use he s ochas ic componen o he wo ade-policy measu es in a con ol unc-
ion which is employed in a second s ep, whe e he unc ional o m o he causal ela-
ionship be ween he ade-policy measu es and o e all sec o -coun y-pai ade cos s
a e in he limeligh . This s ep de e mines wha we will call he dose- esponse unc ion.
5.2 Es ima ing pa ial (di ec ) e ec s o ade policy on ade cos s and ade lows
In his subsec ion, we es ima e he e ec o a i and non a i ade-policy a iables
on bila e al ade cos s, ds
ij , as ob ained om he p ocedu e in equa ion (3). The co -
esponding model o es ima e he so-called uni -le el dose- esponse o de i ed ade
cos s ˆ
ds
ij o he ade-policy ba ie s in ms
ij eads
ˆ
ds
ij =kms
ij , ms
ij +con olss
ij θ+ξs
ij . (11)
We addi ionally condi ion on a linea unc ion o all co a ia es in his s ep o he analysis
h ough he inclusion o con olss
ij in equa ion (11), as sugges ed by Imai and Van Dyk
(2004). No e ha his is no he same as assuming a linea ela ionship be ween na u al
ade ba ie s and ade cos s—in ac , we a e agnos ic abou hei impac on ade cos s
beyond hei (po en ially nonlinea ) ela ionship and in e ac ion wi h endogenous ade
policy. We hen es ima e he unc ional o m o k(·)in equa ion (11)byapolynomial
app oxima ion whose o de is chosen based on he Aikake in o ma ion c i e ion (AIC).
We allow o bo h policy a iables, hei in e ac ion, he GPS as well as any in e ac ion
wi h he GPS o en e he uni -le el dose- esponse unc ion up o a polynomial o o de
10.
The es ima ed coe icien s do no ha e any economic meaning. Howe e , he poly-
nomial model p o ides us wi h a unc ional o m o k(·)in equa ion (11) ha allows o
e alua ing he a e age causal e ec o changes in a i and non a i ade-policy a i-
ables on ade cos s a any po en ial le el o he policy a iables in he ou se . Hence, we
can de ine a g id o ade-policy le els o which we a e in e es ed in he le el o ade
cos s and es ima e he la e using he unc ional o m o equa ion (11). Figu e 3plo s
he bi a ia e dis ibu ion o he ade-policy da a on he such de ined g id o m=(τ,η)
and shows ha he majo i y o he da a is loca ed a ela i ely low le els o policy ade
ba ie s bu ha he e is a ia ion in he a i and as well as he non a i ba ie dimen-
sion.
470 Egge and E ha d Quan i a i e Economics 15 (2024)
Figu e 3. Dis ibu ion o he da a ac oss he 25 ×25 g id.
We e alua e he expec ed condi ional dose- esponse unc ion,k(m)=E[k(m,
(m,qs
ij ))], as an a e age om he size-nsample h ough
ˆ
k(m)=1
n
i∈J
j∈J
s∈S
ˆ
km,ˆ
m,qs
ij , (12)
whe e Jand Sdeno e he se s o coun ies and sec o s in he da a.
Figu e 4displays he a e age dose- esponse unc ion (12)aswellas he95%con i-
dence bounds ha a e based on 100 boo s ap samples o he es ima ion based on a
no mally dis ibu ed GPS in he le panel and he es ima ion based on a nonpa ame i-
Figu e 4. A e age dose- esponse unc ion o log ade cos s and log ade-policy a iables wi h
95% con idence bounds.
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 471
Figu e 5. G adien s w. . τand w. . . ηwi h 95% con idence bounds o di e en le els o ade
policy (no mal densi y).
cally dis ibu ed GPS in he igh panel.30 The wo a ian s o he GPS es ima ion a e o
mino impo ance o he o e all shape o he dose- esponse unc ion.
As expec ed, ade cos s a e gene ally inc easing in he wo ade-policy a iables.
The e ec is, howe e , s ongly nonlinea and depends on he le el o ade policy in he
ou se . In o de o in es iga e he nonlinea i ies in mo e de ail, we p esen he g adien
o he a e age dose- esponse unc ion wi h espec o τand ηin Figu e 5. Fo a be e
illus a ion, we p esen h ee slices o he g adien in each dimension. Each slice co e-
sponds o ei he low (g id poin s 1–5), medium (g id poin s 11–15), o high (g id poin s
21–25) le els o a i o non a i ba ie s, espec i ely.
No e ha he g adien is e e ing o he change in log ade cos s, ˆ
ds
ij , in esponse
o a change in policy ba ie s by one g id-poin di e ence in he dose- esponse unc-
ion, which amoun s o 0.02. To pu he g adien in o con ex , ecall ha he cus oma y
assump ion in mos gene al equilib ium models o ade is ha ade policy en e ade
cos s log-linea ly in an ad alo em ashion. Hence, ad alo em a i s inc ease ade cos s
one o one. T ansla ed o he g id de ined in his exe cise, a one-g id-poin change in
a i s should inc ease log ade cos s by 0.02 a any poin on he g id. This is a below
he maximum g adien o 0.3 ob ained in his exe cise poin ing o a ole o a i s beyond
he pu e ad alo em e ec . A he same ime, he g adien is e ec i ely ze o o some
30The boo s ap was conduc ed as ollows. 100 boo s ap samples o he o al sample (see Table 1)we e
d awn ensu ing ha all impo e -sec o combina ions exis in each boo s ap sample. This co esponds
o impo e -sec o block-boo s apping, and all s eps o he p ocedu e a e es ima ed o each boo s ap
sample. This leads o di e en ly-sized boo s ap samples ha a e hen used in he subsequen s eps. This
p ocedu e accoun s o he imp ecision in he measu emen o es ima ed (de i ed) a iables ha a e used
in la e s ages, such as log sales undamen als, ˆ
s
i,o adecos s, ˆ
ds
ij .
472 Egge and E ha d Quan i a i e Economics 15 (2024)
Figu e 6. G adien s w. . τand w. . . ηwi h 95% con idence bounds o di e en le els o ade
policy— echnical e sus non echnical ade ba ie s (no mal densi y).
pa s o he g id and a subs an ial sha e o he obse a ions is placed in exac ly hose
pa s o he g id. The esul s sugges ha ade policy is a he ine ec i e in hese cases.
Fo e e ence, we also plo he g adien o he “nai e” eg ession unde lying Figu e 1 o
di e en slices and adjus ed o he scaling such ha he g adien e e s o he change in
log ade cos s w. . . a one-g id-poin change. We e e o his as “nai e g adien ” in Fig-
u e 5. The e e ence illus a es ha his app oach en ails a poo app oxima ion o he
ma ginal e ec o changes in ade policy, in pa icula , in a i s, bo h in e ms o e ec
size and shape as a unc ion o a i le els.
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 473
The di e en panels in Figu e 5illus a e nicely ha a gi en change in ade pol-
icy has a e y di e en e ec depending on whe e in he ou se ade policy lies a he
momen o e alua ion and ha hese di e ences a e pa icula ly s a k when conside ing
la ge-scale ade-policychanges. A change in a i policy has a e y s ong posi i e e ec
on ade cos s o e y low le els o a i ba ie s when non a i ba ie s a e medium o
high. This ma ginal e ec is subs an ially smalle o e y low le els o a i and non a -
i ba ie s, which migh be explained by unused p e e ence ma gins (see He in (1986),
F ancois, Hoekman, and Manchin (2006), Es e adeo dal, F eund, and O nelas (2008),
Fugazza and Nici a (2013)). The e o o comply wi h any equi emen s in o de o ob-
ain a p e e en ial a i ea men migh be simply oo bu densome, especially, when
he gains om compliance a e a he low, leading o an unused p e e ence ma gin and
o a less p onounced ma ginal impac o a i s on ade cos s and ade lows. In con-
as , he e is i ually no ma ginal e ec o a change in a i s when a i s a e a ound
10%. The ma ginal e ec o a change in a i s emains ze o also beyond a a i le el o
15% o high le els o non a i ba ie s. Howe e , he ma ginal e ec becomes s ongly
posi i e o low and medium le els o non- a i ba ie s as we mo e beyond 15% and
ades slowly ou as we app oach a le el o 40%. The esul ha he ma ginal e ec o a
change in a i s ades ou o high a i le els and, in pa icula , in case o high non a -
i ba ie s, is well in line wi h he li e a u e on a oidance s a egies o high ba ie s o
ade (see Fisman and Wei (2004), Ja o cik and Na ciso (2008), Sequei a (2016), Demi
and Ja o cik (2018)).
The pa e n is a he di e en when conside ing he ma ginal e ec wi h espec o
non a i ba ie s. We obse e a e y s ong e ec o low le els o non a i ba ie s, in
pa icula , when a i s a e high. By con as , we obse e basically no o e en nega i e
ma ginal e ec s o in e media e le els o ηac oss all le els o a i ba ie s and a s ong
ma ginal inc ease in ade cos s o high le els o non a i ade ba ie s when a i s a e
low. As men ioned in he In oduc ion, in con as o a i s, non a i ba ie s migh en-
ail a dec ease in ade cos s, in pa icula , in he case o non echnical measu es. Since,
he da a on he ad alo em equi alen s o non a i ba ie s allow o di e en ia ing be-
ween echnical and non echnical ba ie s o ade, we analyze hei di e en ial e ec s
in Figu e 6. While he g adien o a i ba ie s is basically unchanged in he wo sub-
analyses, we see ha he nega i e g adien in he non a i -ba ie dimension is almos
en i ely d i en by echnical ba ie s o ade (compa e WTO (2012)). Figu e 6sugges s
ha echnical ba ie s o ade ha e a s ong ma ginal impac on ade cos s o low pol-
icy ba ie s, bu a nega i e e ec o in e media e le els o echnical ba ie s in place.
We ake his esul as suppo o he no ion ha , while non a i measu es a e cos ly on
a e age, he e exis s an in e media e le el o , pa icula ly, non echnical ba ie s o ade
whe e a highe le el may ac ually be ade enhancing.
5.3 Pu ing he es ima ed ade-policy g adien s in con ex
The nonlinea ela ionship o ade cos s and ade policy migh s em om a ious
sou ces (c . Sec ion 2). We canno include measu es o hese sou ces in he eg essions
explici ly, because hey a e pa ly unc ions o ade policy, and hence, endogenous jus
474 Egge and E ha d Quan i a i e Economics 15 (2024)
like ade policy i sel .31 Howe e , pu ing hese measu es in con ex wi h he g adien
may shed ligh on he oo s o he a ia ion in ade-cos esponses o ade policy. We
do so in Figu e 7.
We s a wi h he ole o ade-policy unce ain y. Ea lie wo k has demons a ed ha
such unce ain y is an impo an ac o in de e mining he ac ual e ec o ade policy.
Fo ins ance, he indings in Handley and Limao (2015)andPie ce and Scho (2016)
sugges ha ade-policy changes con ain a signal abou u u e ade-policy unce ain y.
We hypo hesize ha ade-policy unce ain y a ec s he shape o he ade-cos unc ion
and i a ies ac oss a i le els. In o de o p oxy o unce ain y, we use he unexplained
a ia ion om a i s -o de au o eg essi e eg ession, whe e we eg ess he annual ap-
plied a i le el o any coun y pai and sec o on i s lagged alue in all yea s be ween
2001 and 2011 o each a i cell using TRAINS da a. The le panel o Figu e 7sugges s
ha a lowe le el o a i p edic abili y, ha is, a highe a i unce ain y, is associa ed
wi h highe applied a i le els. In pa icula , he measu e o a i unce ain y ises sub-
s an ially mo e s ongly o a i le els beyond 10–20%. The la e is exac ly whe e he
g adien o ade cos s wi h espec o a i s τis s ongly posi i e.
Fu he po en ial a ionales o a he e ogeneous impac o a i s on ade cos s a e
a oidance s a egies a high a i s on he one hand, and he nonuse o a ailable a -
i p e e ences a low mos - a o ed na ion a i s on he o he hand (see He in (1986),
F ancois, Hoekman, and Manchin (2006), Es e adeo dal, F eund, and O nelas (2008),
Fugazza and Nici a (2013)). We would assume a i a oidance s a egies o be mo e
p e alen in coun ies wi h high le els o co up ion (see Fisman and Wei (2004), Ja-
o cik and Na ciso (2008), Sequei a (2016), Demi and Ja o cik (2018)). We p oxy o he
lack o co up ion by aking a measu e o anspa ency a he coun y le el o 2006 om
T anspa ency In e na ional as an in e se measu e o co up ion. We map he la e o
a i s by he espec i e coun ies’ densi ies a di e en p oduc s and le els o a i s. In
o de o p oxy o he nonuse o a ailable a i p e e ences, we use he p e e ence ma -
gin ( he di e ence be ween he mos a o ed na ion a i and he p incipally-a ailable
minimum a i in a ade ag eemen ). The esul s in Figu e 7sugges he ollowing.
Fi s , he g adien is posi i ely co ela ed wi h a g ea e anspa ency in he low- and
high- a i ange, while a medium a i s, anspa ency is nega i ely ela ed o a highe
g adien o ade cos s w. . . a i s. This is consis en wi h a oidance being nega i ely
co ela ed wi h anspa ency. Second, he a i g adien is ela i ely independen o he
p e e ence ma gin a low a i le els. The la e is consis en wi h an unde exploi a ion
o a i p e e ences a low a i le els.
A u he impo an de e minan o he e ec i eness o ade policy is he p esence
o ules o o igin, which impose mo e es ic ions and cos s on he access o p e e en ial
ea men in ade ag eemen s, in pa icula , ega ding non a i ba ie s. K ishna e al.
(2021) show ha he ixed cos s o mee ing ules o o igin dec ease wi h he expe ience
o he i m in ob aining p e e en ial a i s. Gene ally, we would expec ha he impac
o changes in ade policy is s onge , whene e ewe ules o o igins a e in place. We
31Fo example, he “ a i wa e ,” he gap be ween bound and applied a i s, o he co ela ion be ween
lagged and con empo aneous a i s as a measu e o a i a ia ion in ime o a i unce ain y a e bo h
unc ions o applied a i s.
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 475
Figu e 7. G adien s and po en ial co a ia es.
assess he ole o ules o o igin o he ade-cos g adien wi h espec o ade policy
using he p oduc -le el incidence o ules o o igin in he No h Ame ican F ee T ade
A ea (NAFTA) based on da a compiled by Conconi, Ga cía-San ana, Puccio, and Ven-
u ini (2018). We summa ize he associa ed indings in he igh panel o Figu e 7.The
igu e sugges s ha he g adien o ade cos s w. . . non a i ba ie s is nega i ely as-
socia ed wi h he p e alence o ules-o -o igin p o isions a di e en non a i ba ie
le els.
O e all, he ela ionship be ween ade policy and ade cos s iden i ied in his pape
is well aligned wi h e idence in he li e a u e on ade-policy se ing and unce ain y as
well as co up ion, a i a oidance, and ade-policy s ingency.
5.4 Quan i ica ion o o al (gene al-equilib ium) e ec s o ade policy on ade lows
A key insigh o quan i a i e ade models is ha e en homogeneous pa ial e ec s o
a i and non a i ba ie s on ade cos s ma e ialize in he e ogeneous esponses o
economic ou comes h ough gene al equilib ium esponses. Howe e , he e ogeneous
pa ial ea men e ec s o ade-policy a iables as po ayed in Figu e 5will add o
and ampli y he he e ogenei y o o al ade-cos ea men e ec s in gene al equilib-
ium. This subsec ion is conce ned wi h a quan i ica ion o his ampli ica ion o he o al
ade-policy ea men e ec s.32
To illus a e he ele ance o ou es ima es, we will analyze he e ec o a unila e al
inc ease o US a i s on all Chinese impo s by 10 pe cen age poin s—a policy change
ha was discussed and pa ly implemen ed in a simila ein o e he las yea s. In pa ic-
ula , we a e in e es ed in compa ing he es ima ed ou comes based on a homogeneous
g adien o ade cos s wi h espec o a i s (assuming a one- o -one esponse o log
32Fo his analysis, i is necessa y o ocus on a somewha mo e agg ega ed sample o he da a used
abo e. The eason is ha a gene al-equilib ium analysis needs o ely on a ull da a se o all coun y pai s
and sec o s co e ed, while his was no necessa y wi h he analysis o pa ial ade-policy ea men e ec s
on he ea ed. The e o e, we ocus on a subse o 41 indi idual coun ies and one es o he wo ld (ROW)
o 97 sec o s in he espec i e analysis.
476 Egge and E ha d Quan i a i e Economics 15 (2024)
ade cos s, ds
ij , oachangeinτs
ij ) and he one co esponding o he lexible g adien
as es ima ed abo e. Th oughou he analysis, we keep he de ici sha e in e ms o o al
spending cons an o any coun y.33 De ails on he compu a ion o he gene al equilib-
ium a e delega ed o he Supplemen al Appendix.
In he con ex o he expe imen —an inc ease o US a i s on Chinese impo s— he
ac ual change in ade cos s esul ing om he inc ease in a i s is on a e age 19%, and
hus 9 pe cen age poin s highe han he pu e ad alo em e ec o a i s alone (which
is 10%). This implies ha he majo i y o a ec ed ade lows be ween China and he US
lies in an a ea o he g adien whe e he esponse o ade cos s o a change in a i s is
pa icula ly s ong. Indeed, he a e age a i s le ied on hese lows in he benchma k
economy lie a 5%, which is exac ly he domain ea u ing s ong ma ginal esponses in
Figu e 5.
The sha e o US spending on Chinese goods d ops by oughly one pe cen age poin
due o his policy change. Holding o al US spending a he benchma k le el, his
amoun s o a dec ease o Chinese impo s by mo e han 6% (95% con idence in e al
anging om 7.2% o 4.8%). An e ec ha is subs an ially la ge han he one implied by
he cus oma y ad alo em speci ica ion in which he d op in impo s is less han 3%.34
The sha e o US income gene a ed h ough a i s inc eases by 50% in he nonpa ame -
ic exe cise whe eas he inc ease is 54% in he ad alo em speci ica ion, which illus a es
he addi ional impac o a i s in inc easing ade cos s beyond he pu e ad alo em e -
ec . These agg ega e numbe s hide, howe e , subs an ial he e ogenei y ac oss impo
sha es om di e en Chinese sec o s as illus a ed in he le panel o Figu e 8.Ona e -
age, o ea ed sec o s—sec o s ha we e subjec o nonze o a i s in he benchma k—
he change in impo lows (a US benchma k income le els) is 7 pe cen age poin s lowe
(signi ican a he 5% le el) in he nonpa ame ic speci ica ion compa ed o he ad al-
o em speci ica ion. In some sec o s, he impo sha es d op by mo e han 50% in he
nonpa ame ic case illus a ing he wide ange o esponses unde he nonpa ame ic
speci ica ion as compa ed o he ad alo em exe cise, which exhibi s subs an ially less
a ia ion. No e ha o 92% o all bila e al ade sha e changes he di e ences be ween
he nonpa ame ic speci ica ion and he ad alo em speci ica ion is signi ican a he
5% le el.
The expe imen o a unila e al inc ease in a i s by he US owa d Chinese impo s
e eals ha cus oma y (ad alo em) app oaches o e alua ing he policy change would
se e ely unde es ima e he implied e ec s o his pa icula policy change. How a he
wo di e en speci ica ions di e ge depends on he espec i e s a us quo o ade pol-
icy, which di e s ac oss coun ies and sec o s. To illus a e how la ge he di e gence
is on a e age, le us conside an al e na i e expe imen ha implies a change in ade
cos s o all a i - ea ed ade lows in he da a and no only Chinese impo s o he
33No e ha ade imbalances do no a ec he key insigh s ega ding he esponse unc ion o ade cos s
and ade-policy a iables. The eason is ha ade imbalances a e indexed by (p oduc and) impo e bu
no by (p oduc and) coun y pai .
34No e ha since he es ima ion s a egy is o be in e p e ed as a ea men e ec on he ea ed, he
coun e ac ual ade-policy change applies only o hose ade lows ha a e subjec o a i s in he bench-
ma k.
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 483
A kolakis, Cos as, A naud Cos ino , and And es Rod iguez-Cla e (2012), “New ade
models, same old gains?” Ame ican Economic Re iew, 102 (1), 94–130. [454,455]
Bagwell, Kyle and Robe W. S aige (1999), “An economic heo y o GATT.” Ame ican
Economic Re iew, 89 (1), 215–248. [454]
Bagwell, Kyle and Robe W. S aige (2004), TheEconomicso heWo ldT adingSys em.
MIT P ess Books, Vol. 1. The MIT P ess. [454]
Baie , Sco L. and Je ey H. Be gs and (2004), “Economic de e minan s o ee ade
ag eemen s.” Jou nal o in e na ional Economics, 64 (1), 29–63. [460]
Baie , Sco L. and Je ey H. Be gs and (2007), “Do ee ade ag eemen s ac ually in-
c ease membe s’ in e na ional ade?” Jou nal o in e na ional Economics, 71 (1), 72–95.
[454]
Baie , Sco L. and Je ey H. Be gs and (2009), “Es ima ing he e ec s o ee ade
ag eemen s on in e na ional ade lows using ma ching econome ics.” Jou nal o In-
e na ional Economics, 77 (1), 63–76. [454]
Baldwin, Richa d E. and Simon J. E ene (2012), “Begga - hy-neighbou policies du ing
he c isis e a: Causes, cons ain s and lessons o main aining open bo de s.” Ox o d
Re iew o Economic Policy, 28 (2), 211–234. [454,458]
Bond, E ic W. (1990), “The op imal a i s uc u e in highe dimensions.” In e na ional
Economic Re iew, 31 (1), 103–116. [454,459]
Bond, E ic W. Raymond G. Riezman, and Cons an inos Sy opoulos (2004), “A s a egic
and wel a e heo e ic analysis o ee ade a eas.” Jou nal o In e na ional Economics,
64 (1), 1–27. [460]
Bond, E ic W. and Cons an inos Sy opoulos (1996), “The size o ading blocs ma ke
powe and wo ld wel a e e ec s.” Jou nal o In e na ional Economics, 40 (3–4), 411–437.
[454,455,459,460]
Bown, Chad P. (2011), “Taking s ock o an idumping, sa egua ds and coun e ailing du-
ies, 1990–2009.” The Wo ld Economy, 34, 1955–1998. [454,458]
Bown, Chad P. and Me edi h A. C owley (2013), “Impo p o ec ion, business cycles, and
exchange a es: E idence om he g ea ecession.” Jou nal o In e na ional Economics,
90 (1), 50–64. [454,458]
B einlich, Holge , Swa i Dhing a, Saul Es in, Hanwei Huang, Gianma co O a iano,
Thomas Sampson, John Van Reenen, and Jona han Wadswo h (2016), “BREXIT 2016—
policy analysis om he cen e o economic pe o mance.” Technical Repo , Cen e o
Economic Pe o mance. [453]
Caliendo, Lo enzo Robe Feens a, John Romalis, and Alan Taylo (2015), “Ta i educ-
ions, en y, and wel a e: Theo y and e idence o he las wo decades.” CEPR Discus-
sion Pape s 10962, C.E.P.R. Discussion Pape s 2015. [454,459,460,481]
484 Egge and E ha d Quan i a i e Economics 15 (2024)
Caliendo, Lo enzo Luca Da id Op omolla, Fe nando Pa o, and Alessand o S o za
(2021), “Goods and ac o ma ke in eg a ion: A quan i a i e assessmen o he EU en-
la gemen .” Jou nal o Poli ical Economy, 129 (12), 3491–3545. [454]
Caliendo, Lo enzo and Fe nando Pa o (2015), “Es ima es o he ade and wel a e e ec s
o NAFTA.” Re iew o Economic S udies, 82 (1), 1–44. [454]
Conconi, Paola, Manuel Ga cía-San ana, Lau a Puccio, and Robe o Ven u ini (2018),
“F om inal goods o inpu s: The p o ec ionis e ec o ules o o igin.” Ame ican Eco-
nomic Re iew, 108 (8), 2335–2365. [475]
Con e, Maddalena, Pie e Co e laz, and Thie y Maye (2022), “The CEPII g a i y
da abase.” Wo king Pape s 2022-05, CEPII esea ch cen e . [463]
Cos ino , A naud and And es Rod iguez-Cla e (2014), “T ade heo y wi h numbe s:
Quan i ying he consequences o globaliza ion.” In Handbook o In e na ional Eco-
nomics,Vol.4Else ie .[455,456,481]
C owley, Me edi h, Huasheng Song, and Ning Meng (2016), “Ta i sca es: T ade policy
unce ain y and o eign ma ke en y by Chinese i ms.” Camb idge Wo king Pape s in
Economics 1676, Facul y o Economics, Uni e si y o Camb idge. [458]
Demi , Banu and Bea a S. Ja o cik (2018), “Fo ensics, elas ici ies and Ben o d’s law: De-
ec ing ax aud in in e na ional ade.” CEPR Discussion Pape s 12798. [455,458,473,
474]
Ea on, Jona han and Samuel Ko um (2002), “Technology, geog aphy, and ade.” Econo-
me ica, 70 (5), 1741–1779. [454,460,481]
Egge , Pe e H., and Ka ha ina E ha d (2024), “Supplemen o ‘He e ogeneous e ec s o
a i and non a i ade-policy ba ie s in quan i a i e gene al equilib ium’.” Quan i a-
i e Economics Supplemen al Ma e ial, 15, h ps://doi.o g/10.3982/QE1994.[456]
Egge , Pe e H. and Ma io La ch (2011), “An assessmen o he Eu ope ag eemen s’ e -
ec s on bila e al ade, GDP, and wel a e.” Eu opean Economic Re iew, 55 (2), 263–279.
[454]
Egge , Pe e H., Ma io La ch, Ke in E. S aub, and Raine Winkelmann (2011), “The ade
e ec s o endogenous p e e en ial ade ag eemen s.” Ame ican Economic Jou nal: Eco-
nomic Policy, 3 (3), 113–143. [454]
Es e adeo dal, An oni, Ca oline F eund, and Emanuel O nelas (2008), “Does egion-
alism a ec ade libe aliza ion owa d nonmembe s?” The Qua e ly Jou nal o Eco-
nomics, 123 (4), 1531–1575. [455,458,473,474]
Fajgelbaum, Pablo D., Pinelopi K. Goldbe g, Pa ick J. Kennedy, and Ami K. Khandelwal
(2019), “The e u n o p o ec ionism.” NBER Wo king Pape s 25638, Na ional Bu eau o
Economic Resea ch, Inc. [453]
Felbe may , Gab iel, Rahel Aichele, and Inga Heiland (2016), “Going deep: The ade
and wel a e e ec s o TTIP e ised.” i o Wo king Pape Se ies 219, i o Ins i u e—Leibniz
Ins i u e o Economic Resea ch a he Uni e si y o Munich. [453]
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 485
Felbe may , Gab iel, Benjamin Jung, and Ma io La ch (2013), “Op imal a i s, e alia-
ion, and he wel a e loss om a i wa s in he Meli z model.” Jou nal o In e na ional
Economics, 89 (1), 13–25. [454,455,459,460]
Fisman, Raymond and Shang-Jin Wei (2004), “Tax a es and ax e asion: E idence om
missing impo s in China.” Jou nal o Poli ical Economy, 112 (2), 471–500. [455,458,473,
474]
Flo es, Ca los A., Al onso Flo es-Lagunes, A u o Gonzalez, and Todd C. Neumann
(2012), “Es ima ing he e ec s o leng h o exposu e o ins uc ion in a aining p o-
g am: The case o job co ps.” The Re iew o Economics and S a is ics, 94 (1), 153–171.
[455,461,464,468]
Fon agné, Lionel, Houssein Guimba d, and Gianluca O e ice (2022), “Ta i -based
p oduc -le el ade elas ici ies.” Jou nal o In e na ional Economics, 137. (C). [477,480]
F ancois, Joseph, Be na d Hoekman, and Mi iam Manchin (2006), “P e e ence e osion
and mul ila e al ade libe aliza ion.” Wo ld Bank Economic Re iew, 20 (2), 197–216.
[455,458,473,474]
F iedman, Je ome H. (1991), “Mul i a ia e adap i e eg ession splines.” Ann. S a is .,19
(1), 1–67. [467]
Fugazza, Ma co and Alessand o Nici a (2013), “The di ec and ela i e e ec s o p e e -
en ial ma ke access.” Jou nal o In e na ional Economics, 89 (2), 357–368. [455,458,473,
474]
Handley, Kyle and Nuno Limao (2015), “T ade and in es men unde policy unce ain y:
Theo y and i m e idence.” Ame ican Economic Jou nal: Economic Policy, 7 (4), 189–222.
[458,474]
Handley, Kyle and Nuno Limao (2017), “Policy unce ain y, ade, and wel a e: Theo y
and e idence o China and he Uni ed S a es.” Ame ican Economic Re iew, 107 (9),
2731–2783. [458]
Head, Kei h and Thie y Maye (2014), “G a i y equa ions: Wo kho se, oolki , and cook-
book.” In Handbook o In e na ional Economics,Vol. 4 (Gi a Gopina h, Elhanan Help-
man, and Kenne h Rogo , eds.), 131–195, Else ie . [456]
Hende son, Daniel J. and Daniel L. Millime (2008), “Is g a i y linea ?” Jou nal o Applied
Econome ics, 23 (2), 137–172. [454]
He in, Jan (1986), “Rules o o igin and di e ences be ween a i le els in EFTA and in
he EC.” Vol. 13, EFTA Occasional Pape . [455,458,473,474]
Hillbe y, Russell and Da id Hummels (2008), “T ade esponses o geog aphic ic ions:
A decomposi ion using mic o-da a.” Eu opean Economic Re iew, 52 (3), 527–550. [454,
458]
Hi ano, Keisuke and Guido W. Imbens (2004), “The p opensi y sco e wi h con inuous
ea men s.” Applied Bayesian modeling and causal in e ence om incomple e-da a pe -
spec i es, 226164, 73–84. [461,462,468,469]
486 Egge and E ha d Quan i a i e Economics 15 (2024)
Ho n, Hen ik, Pe os C. Ma oidis, and And é Sapi (2010), “Beyond he WTO? An
ana omy o EU and US p e e en ial ade ag eemen s.” The Wo ld Economy, 33 (11),
1565–1588. [454,457]
Hummels, Da id (2007), “T anspo a ion cos s and in e na ional ade in he second e a
o globaliza ion.” Jou nal o Economic Pe spec i es, Summe , 21 (3), 131–154. [480]
Hummels, Da id and Alexand e Skiba (2004), “Shipping he good apples ou ? An empi -
ical con i ma ion o he Alchian-Allen conjec u e.” Jou nal o Poli ical Economy, 112 (6),
1384–1402. [458]
Imai, Kosuke and Da id A.Van Dyk (2004), “Causal in e ence wi h gene al ea men
egimes.” Jou nal o he Ame ican S a is ical Associa ion, 99 (467). [461,462,468,469]
I a azabal, Al onso, And eas Moxnes, and Luca Da id Op omolla (2015), “The ip o he
icebe g: A quan i a i e amewo k o es ima ing ade cos s.” The Re iew o Economics
and S a is ics, 97 (4), 777–792. [458]
Ja o cik, Bea a S. and Gaia Na ciso (2008), “Di e en ia ed p oduc s and e asion o im-
po a i s.” Jou nal o In e na ional Economics, 76 (2), 208–222. [455,458,473,474]
Kee, Hiau Looi, Alessand o Nici a, and Ma celo Ola eaga (2008), “Impo demand elas-
ici ies and ade dis o ions.” The Re iew o Economics and S a is ics, 90 (4), 666–682.
[457,463,481]
Kee, Hiau Looi, Alessand o Nici a, and Ma celo Ola eaga (2016), “T ade auds, ade
eas ici ies and non- a i measu es.” Technical Repo , Wo ld Bank. [458,463]
Klu e, Jochen, Hilma Schneide , A ne Uhlendo , and Zhong Zhao (2012), “E alua ing
con inuous aining p og ammes by using he gene alized p opensi y sco e.” Jou nal o
he Royal S a is ical Socie y: Se ies A (S a is ics in Socie y), 175 (2), 587–617. [468]
K ishna, Kala, Ca los Salamanca, Yu a Suzuki, and Ch is ian Volpe Ma incus (2021),
“Lea ning o use ade ag eemen s.” NBER Wo king Pape s 29319, Na ional Bu eau o
Economic Resea ch, Inc. [455,458,474]
Li, Qi and Je ey Sco Racine (2006), Nonpa ame ic Econome ics: Theo y and P ac ice.
Economics Books, Vol. 1. P ince on Uni e si y P ess. [468]
Ossa, Ralph (2011), “A “new ade” heo y o GATT/WTO nego ia ions.” Jou nal o Poli i-
cal Economy, 119 (1), 122–152. [454,455,459,460]
Ossa, Ralph (2014), “T ade wa s and ade alks wi h da a.” Ame ican Economic Re iew,
104 (12), 4104–4146. [454,455,459,460]
Pie ce, Jus in R. and Pe e K. Scho (2016), “The su p isingly Swi decline o US manu-
ac u ing employmen .” Ame ican Economic Re iew, 106 (7), 1632–1662. [458,474]
Romalis, John (2007), “NAFTA’s and CUSFTA’s impac on in e na ional ade.” The Re iew
o Economics and S a is ics, 89 (3), 416–435. [454]
Sequei a, Sand a (2016), “Co up ion, ade cos s, and gains om a i libe aliza ion:
E idence om sou he n A ica.” Ame ican Economic Re iew, 106 (10), 3029–3063. [455,
458,473,474]
Quan i a i e Economics 15 (2024) He e ogeneous e ec s o ade policy 487
WTO (2012), “Wo ld ade epo 2012: T ade and public policies: A close look a non-
a i measu es in he 21s cen u y.” Technical Repo , Wo ld T ade O ganiza ion. [458,
473]
Co-edi o Mo en Ra n handled his manusc ip .
Manusc ip ecei ed 30 Sep embe , 2021; inal e sion accep ed 23 Janua y, 2024; a ailable on-
line 23 Janua y, 2024.
The eplica ion package o his pape is a ailable a h ps://doi.o g/10.5281/zenodo.10489950.
The Jou nal checked he da a and codes included in he package o hei abili y o ep oduce
he esul s in he pape and app o ed online appendices.
