Schenke , Oli e ; Osbe ghaus, Daniel
A icle — Published Ve sion
In e na ional T ade and he T ansmission o Tempe a u e
Shocks
En i onmen al and Resou ce Economics
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Schenke , Oli e ; Osbe ghaus, Daniel (2025) : In e na ional T ade and he
T ansmission o Tempe a u e Shocks, En i onmen al and Resou ce Economics, ISSN 1573-1502,
Sp inge Ne he lands, Do d ech , Vol. 88, Iss. 4, pp. 965-1007,
h ps://doi.o g/10.1007/s10640-025-00957-3
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/323358
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/
Accep ed: 6 Janua y 2025 / Published online: 30 Janua y 2025
© The Au ho (s) 2025
Oli e Schenke
oli e .schenke @zew.de
Daniel Osbe ghaus
osbe ghaus@a co .de
1 ZEW - Leibniz Cen e o Eu opean Economic Resea ch, L7,1, DE-68161 Mannheim,
Ge many
In e na ional T ade and he T ansmission o Tempe a u e
Shocks
Oli e Schenke 1· DanielOsbe ghaus1
En i onmen al and Resou ce Economics (2025) 88:965–1007
h ps://doi.o g/10.1007/s10640-025-00957-3
Abs ac
We examine how he ad e se impac s o wea he shocks a e dis ibu ed h ough he ade
ne wo k. Exploi ing a ich, heo e ically de i ed, ixed e ec s s uc u e, we ind signi i-
can nega i e sho - un e ec s o high empe a u e on expo s. A mon h wi h an a e age
empe a u e abo e 30
◦C
implies expo losses o a ound h ee pe cen . These e ec s a e
inc easing in he labou -in ensi y o expo s. Using ou s uc u al G a i y model, we assess
he gene al equilib ium incidence o hese empe a u e shocks. We ind ha equilib ium
adjus men s educe he economic cos s by a ound 20 pe cen , bu signi ican cos s a ise
also o coun ies no di ec ly exposed o high empe a u es.
Keywo ds In e na ional ade · Tempe a u e · Ex eme wea he · S uc u al G a i y
JEL Classi ica ion F14 · F18 · Q54 · Q56
1 In oduc ion
Clima e change is a global p oblem. The emission o g eenhouse gases impac s ecosys ems
and economies a ound he globe, independen ly o he loca ion o he emi e . In o de o
in e he sensi i i y o economic ac i i ies o clima e change a g owing li e a u e es ima es
how wea he in gene al, and empe a u e in pa icula , a ec s agg ega e economic ou comes.
Mos s udies in his ield analyse local ou comes o local wea he a ia ion (Dell e al.
2012; Bu ke e al. 2015). Bu economies a e no isola ed om each o he . As in e na ional
ade links he o une o economies, he economic impac s o local wea he e en s may
dissemina e h ough he global ade ne wo k, c ea ing a spa ial disen angling o he occu -
ence o wea he e en s and hei ull economic consequences. Fu he mo e, mos wea he
1 3
O. Schenke , D. Osbe ghaus
e en s occu only o a ew days o weeks, which may make i challenging o iden i y hei
impac in annual economic da a. Thus, in o de o comp ehensi ely assess he economic
cos s o wea he e en s, we need o unde s and he p opaga ion o he cos s o hese e en s
ac oss space using da a wi h high empo al esolu ion.
Ou s udy is—a leas o ou knowledge— he i s ha p o ides obus ex-pos empi i-
cal e idence o he in e na ional ansmission o he cos s o ex eme wea he e en s using
mon hly ade da a. Based on a e ined s uc u al G a i y model we show ha high em-
pe a u e e en s in one coun y cause economically and s a is ically signi ican cos s also in
coun ies no di ec ly exposed o he e en .
Recen ly, new economic geog aphy models s a ed o include spa ial he e ogenei y o cli-
ma e change impac s, showing ha he ma gins o spa ial economic adjus men s o clima e
impac s by ei he mig a ing o by shi ing specialisa ion pa e ns a e impo an mechanisms
o adap o clima e change (Con e e al. 2021; C uz and Rossi-Hansbe g 2024). Unde -
s anding he spa ial spillo e s o local wea he e en s has also impo an consequences o
cos -assessmen s o clima e change and hus he decision-making o policy make s. As he
equency and se e i y o ex eme wea he e en s is likely inc easing wi h u he global
wa ming he p opaga ion o he cos s o ex eme wea he e en s h ough he global ade
ne wo k is ele an o a comp ehensi e accoun o he social cos o ca bon (SCC), a key
igu e used in cos -bene i analysis ha pu s a mone a y alue on he impac s o clima e
change caused by one on o ca bon (Wagne e al. 2021). In he U.S., he o me T ump
adminis a ion e ised he o icial, o me ly global, SCC igu es, conside ing only domes ic
clima e impac s in con inen al Uni ed S a es. This has been e e sed by he Biden admin-
is a ion, mainly based on e hical and ai ness a gumen s. Bu i cos s o clima e change
a e p opaga ed h ough in e na ional ade, his p o ides an a gumen o ake in o accoun
global clima e impac s o he SCC calcula ion e en om a pu e sel ish pe spec i e.1 I is
he e o e impo an o measu e he ex en o clima e impac s on ade lows and hei p opa-
ga ion h ough he global ade sys em.
This pape he e o e aims a answe ing h ee esea ch ques ions: (i) Do ex eme wea he
e en s a ec expo s? (ii) I so, h ough which channels do hese e en s a ec bila e al ade
and which cha ac e is ics go e n he e ec s? (iii) And, inally, wha is he spa ial incidence
o he cos s o hese e en s and how much o hese cos s a ise in no di ec ly exposed
coun ies?
In o de o iden i y bo h pa ial and gene al equilib ium e ec s o wea he e en s on in e -
na ional ade, we build a s uc u al G a i y model, whe e wea he e en s a ec mon hly
ou pu and, consequen ly, impo e s ace supply losses om a ec ed expo e s. The model
explici ly desc ibes demand and p ice shi s, as impo e s espond wi h subs i u ion om
o he sou ces.
This model is hen es ima ed using i e decades o mon hly obse a ions o bila e al ade
and wea he da a, including mo e han 20,000 coun y pai s and abou 4.5 million obse a-
ions. Es ima ing a s uc u al G a i y model p o ides a well-sui ed amewo k o iden i y
he impac o wea he on agg ega e economic ou comes as he iden i ica ion builds on es ab-
lished ade heo y and a obus empi ical ela ionship. While we pe o m he empi ical
analyses o a ious ypes o ex eme wea he e en s, we ocus on episodes o high ambien
empe a u e in he main pa o he analysis. Consis en wi h ou de i ed gene al equilib ium
model, we a e able o exploi a ich and heo y-de i ed s uc u e o expo e (impo e )
×
1 See Ko chen (2018) o a heo e ical analysis o his a gumen .
1 3
966
In e na ional T ade and he T ansmission o Tempe a u e Shocks
yea , coun y pai
×
yea , and expo e (impo e )
×
calenda mon h ixed e ec s ha allows
o igh ly es ima e he empe a u e impac s on mon hly bila e al expo s.
We ind highly signi ican nega i e non-linea e ec s o high absolu e empe a u e and
ex eme empe a u e de ia ions in he expo ing coun y on he alue o con empo aneous
g oss expo s. In a mon h when he a e age empe a u e is a leas 30
◦C
, expo s dec ease
by 3.4 pe cen ela i e o a mon h wi h an a e age empe a u e below his h eshold. Using
an al e na i e speci ica ion o ex eme de ia ions om coun y-speci ic mean empe a u es,
we ind ha a op-pe cen ile empe a u e shock in he expo coun y educes he expo
alue by 2.1 pe cen .
We hen examine i speci ic expo e cha ac e is ics go e n he e ec size and ind ha
he impac ises wi h he labou -in ensi y o he expo s. This is consis en wi h he no ion
ha sho - un empe a u e impac s on expo s migh be go e ned by labou p oduc i i y o
labou supply e ec s in he expo ing coun y due o o physiological hea s ess, as sug-
ges ed by subs an ial mic o-empi ical e idence.
Equipped wi h hese es ima es, which in o m ou s uc u al G a i y model, we compu e
he coun e ac ual global ade equilib ia in absence o high empe a u e e en s. This allows
us o calcula e he cos -incidence o such an e en o he coun y di ec ly exposed o he
empe a u e shock as well o coun ies only indi ec ly exposed h ough ade links wi h he
a ec ed loca ion. Measu ing cos s as losses in ade ela i e o he coun e ac ual scena io
wi hou empe a u e shock, we ind ha he mean high empe a u e shock has s a is ically
signi ican global cos s o 360 million USD. Abou wo- hi ds o hese cos s appea in coun-
ies no di ec ly exposed o he empe a u e e en , sugges ing ha subs an ial pa s o he
cos s o hese shocks a e ansmi ed and p opaga ed h ough he ade ne wo k.
In a inal s ep, we analyse he magni ude o hese spillo e s unde clima e change p ojec-
ions based on wen y-yea mon hly empe a u e a e ages om global clima e models. We
ind ha unde a middle-o - he- oad clima e p ojec ion o he pe iod 2020–2039, annual
global ade is educed by abou 735 million USD due o addi ional high empe a u e e en s
ela i e o 2015.
1.1 Li e a u e Re iew
G owing mic o- and plan -le el e idence sugges s ha high ambien empe a u e has de -
imen al impac s on labou p oduc i i y and supply. Using su ey da a on ime alloca ion
o indi iduals, Zi in and Neidell (2014) ind e idence o a subs an ial educ ion o labou
supply in clima e-exposed indus ies such as ag icul u e, cons uc ion and manu ac u ing in
non-clima e-con olled acili ies on days wi h maximum empe a u e abo e 85
◦
F (29.4
◦
C).
Gi en he e idence on he empe a u e-p oduc i i y ela ionship o indi iduals, one
migh sugges ha such empe a u e e ec s p e ail also on plan -le el. Looking a he nea -
uni e se o Chinese manu ac u ing plan s om 1998 o 2007, Zhang e al. (2018) ind an
in e ed U-shape ela ionship be ween empe a u e and o al ac o p oduc i i y (TFP).
Thei es ima es show ha o he a e age plan on a day wi h maximum empe a u e abo e
90
◦
F (32.24
◦
C), TFP dec eases by 0.56 pe cen ela i e o a day wi h 50–60
◦
F (10–15
◦
C), ansla ing in o an es ima ed ou pu loss o 0.45 pe cen o he a e age plan . Simila
indings a e documen ed using i m-le el da a om India. Somana han e al. (2021) p o ide
e idence ha annual plan ou pu alls by abou 2 pe cen i e e y day would wa m by 1
◦
1 3
967
O. Schenke , D. Osbe ghaus
C. This loss appea s o be d i en by a educ ion in he ou pu elas ici y o labou due o an
inc easing a e o absen eeism and a dec ease in labou p oduc i i y.
This esea ch p o ides he mic o-economic ounda ion o he mac o-le el impac o high
ambien empe a u es on economies. Measu ing agg ega e impac s o empe a u e changes
on economic g ow h a es has been he aim o a numbe o in luen ial s udies such as Dell
e al. (2012) and Bu ke e al. (2015). Assuming a log-linea ela ionship o empe a u e
and economic ac i i ies, Dell e al. (2012) ind a subs an ial nega i e e ec o empe a u e
changes on GDP g ow h, bu only in poo coun ies: a 1
◦C
ise in annual a e age empe a-
u e educes economic g ow h by abou 1.3 pe cen age poin s. Bu ke e al. (2015) a gue
ha he agg ega e impac o empe a u e on economic ou comes is non-linea , sugges ing a
conca e unc ion wi h economic p oduc i i y peaking a 13
◦C
. Based on sub-na ional da a,
Kalkuhl and Wenz (2020) suppo he e idence ha empe a u e a ia ion a ec s agg ega e
ou comes in a non-linea ashion.
Howe e , his li e a u e ocuses on local e ec s o local e en s. Gi en he economic
ele ance o in e na ional ade, ocusing on local empe a u e migh p o ide an incomple e
pic u e o he impac s o wea he e en s on he economy. As Jones and Olken (2010) say:
"in e na ional ade links he o unes o coun ies p o iding impo an condui s o geo-
g aphically limi ed clima ic impac s o ha e global economic e ec s." Using educed- o m
eg essions wi h p oduc -le el expo panel da a hey ind ha expo g ow h is educed by
2.0–5.7 pe cen age poin s in poo coun ies i annual empe a u e inc eases by 1
◦
C. P od-
uc -le el analyses show ha his is d i en in pa icula by ad e se e ec s on ag icul u e and
ligh manu ac u ing. This inding has been con i med by Dallmann (2019), who addi ionally
con ols o empe a u e (and p ecipi a ion) impac s a he impo e loca ion. Also using a
linea empe a u e speci ica ion in an annual ime-scale, she inds ha each 1
◦
C wa ming
in he expo e coun y educes bila e al expo s by 3.1 pe cen , bu does no ind signi ican
e ec s o he impo e ’s empe a u e.
Ou pape di e s in i e impo an aspec s om hese p e ious s udies. Fi s , ou es i-
ma ion is de i ed om a gene al equilib ium ade model. The de i ed es ima ed G a i y
equa ion has been p o ed o be empi ically obus in many applica ions in in e na ional
economics. Join ly wi h ou igh , heo e ically de i ed, high-dimensional ixed e ec s
s uc u e, his should minimize omi ed a iable bias, imp o e iden i ica ion and gene a e
obus es ima es o he wea he a ia ion e ec s on an agg ega e economic ou come such
as expo s.
Second, we use da a wi h a mon hly empo al esolu ion while mos o he p e ious
mac o-le el s udies ely on annual da a. Hea wa es ypically las only o a ew weeks. Bu
as agg ega e s a is ics such as GDP a e only a ailable on a qua e ly o e en annual basis,
i may be di icul o iden i y he e ec s o hose e en s in agg ega e da a wi h su icien
accu acy. The ew exis ing s udies using mon hly ade da a do ei he no analyze global
da a (such as Ka lsson (2021), ocusing on U.S. expo s), o do no s udy empe a u e e ec s
(Temba a and Takeuchi 2019; Felbe may e al. 2020). The mon hly empo al esolu ion
addi ionally enables an analysis o lagged impac s o wea he e en s. We ind ha expo s
a e mainly a ec ed du ing and di ec ly a e he e en bu do no ind e idence o a subs an-
ial compensa ion o he los expo s in pos -e en mon hs.
Thi d, p e ious s udies show ha he unc ional ela ionship be ween agg ega e eco-
nomic ou comes and empe a u e emains hea ily deba ed. Newell e al. (2021) es se e al
hund ed unc ional o ms and ind ha non-linea empe a u e speci ica ions domina e he
1 3
968
In e na ional T ade and he T ansmission o Tempe a u e Shocks
model se in e ms o p edic i e abili y. Based on ou la ge da a se , we es ima e he e ec s
o single
◦C
bins, he eby allowing o a high deg ee o lexibili y in he unc ional o m.
Fou h, while we, simila o Jones and Olken (2010), ind e idence o la ge e ec s o
high empe a u e e en s on expo s o manu ac u ed goods, we use a mo e di ec app oach
o iden i y labou -p oduc i i y e ec s as a key channel. Using inpu -ou pu da a we ind ha
high labou -in ensi y o expo s co ela es wi h a s onge nega i e impac o high empe a-
u e e en s on expo s.
Fi h, and o pa icula ele ance, by exploi ing ou es ima ed s uc u al model, we simu-
la e he equilib ium adjus men s caused by a high empe a u e episode. This enables us o
es ima e he cos -incidence o such e en s also o coun ies only indi ec ly a ec ed, aking
in o accoun hei oppo uni ies o adjus impo s in esponse o a shock ab oad.
Besides con ibu ing o li e a u e on wea he e ec s on in e na ional ade, ou pape
adds o a g owing li e a u e ha s udies he spa ial ansmission o na u al shocks such
as disas e s o wea he ex emes mo e explici ly. This is pa icula ly ele an as impac s
o clima e change a e and will be une enly sp ead ac oss coun ies and, as Cos ino e al.
(2016) poin ou : "[i]n a globalized wo ld, he impac o mic o-le el shocks depends no
only on hei a e age bu also on hei dispe sion o e space." Using a spa ially highly-
esol ed c op ield model, hey s udy gene al equilib ium adjus men s o c op p oduc i i y
shocks om clima e change. They show ha adjus ing plan ed c op ypes in esponse o
changes in compa a i e ad an age is an impo an o ce o educe cos s o clima e change
in ag icul u e. Rela i e o his, in e na ional ade plays only a mino ole in alle ia ing he
consequences o clima e change. Simila ly, Con e e al. (2021) de elop a calib a ed spa-
ial economic model o explo e changes in specializa ion as an adap a ion mechanism. In
con as , we do no add ess his ma gin explici ly as in ou model each coun y p oduces a
speci ic, non-homogeneous good. The eby, we implici ly abs ac om adjus men s in he
domes ic p oduc ion p ocesses.
Desme e al. (2021) s udy he changing spa ial dis ibu ion o he economic ac i i y due
o clima e change induced sea le el ise in a calib a ed model o he wo ld economy a a
1◦
by
1◦
esolu ion and show as well ha he dynamic spa ial equilib ium adjus men is an
impo an adap a ion mechanism. Lowe esol ed coun y-le el compu able gene al equi-
lib ium (CGE) models in o med by clima e impac p ojec ions, such as Schenke (2013)
and Kni el e al. (2020) also poin ou ha in e na ional ade is an impo an edis ibu ion
mechanism o clima e impac cos s such ha o some egions hese impo ed impac s can
be esponsible o a subs an ial pa o he o al cos o clima e change. Di e en o hese
s udies, which ely on ca e ully calib a ed models o u u e clima e condi ions, we exploi
pas wea he and ade da a o es ima e he dispe sion o hese e ec s ac oss space.
A he i m le el, Ba o and Sau agna (2016) ind ha when one o hei supplie s is
hi by a la ge na u al disas e , i ms expe ience an a e age d op o 2–3 pe cen age poin s in
sales g ow h. This is suppo ed by indings o Pank a z and Schille (2024) who show ha
i m-pe o mance is nega i ely a ec ed i la ge supplie s o hese i ms ha e been exposed
o ex eme wea he e en s bu also ha he downs eam i ms espond and adjus hei sup-
ply chains o less exposed supplie s. While hese pape s s udy i m-le el esponses o exog-
enous shocks, we ocus on agg ega e impac s on he mac o-le el.
The emainde o he pape de elops in Sec . 2 he analy ical gene al equilib ium G a -
i y model om which we de i e ou es ima ion equa ion. Sec ion 3 discusses he empi ical
app oach and in oduces he es ima ion amewo k. Sec ion 4 p esen s he da a and he
1 3
969
O. Schenke , D. Osbe ghaus
cons uc ion o he a iables. Sec ion 5 shows he es ima ed empe a u e e ec s on bila e al
ade. These es ima es lay he g oundwo k o he coun e ac ual simula ions, p esen ed in
Sec . 6, including ex-an e simula ions based on u u e clima e p ojec ions. Finally, Sec . 7
concludes he analysis.
2 Model
We build on a simple gene al equilib ium ade model whe e each coun y p oduces a spe-
ci ic a ie y which is aded wi h he es o he wo ld—i.e. goods a e di e en ia ed by
o igin as in A ming on (1969). Consume s ha e cons an elas ici y o subs i u ion (CES)
p e e ences o hese coun y-speci ic goods. The CES-A ming on gene al equilib ium
model, whose heo e ical unde pinning goes back o Ande son (1979), is he wo kho se
model in s uc u al G a i y esea ch (Head and Maye 2014). We ex end his model in wo
impo an dimensions: Fi s , we model how wea he shocks can a ec he p oduc ion o
ou pu . Second, we ake in o accoun in a-annual a ia ion in p oduc ion and consump ion.
2.1 Consump ion
Each poin in ime can be cha ac e ized by he se union o yea index
∈{1, .., T}
and
calenda mon h
m∈{1, .., 12}
. Each coun y
j∈{1, .., N}
is popula ed by a ep esen a-
i e agen wi h CES u ili y. As we explain below, he model con ols o impo an known
de e minan s o wea he shocks such as he geog aphy and in a-annual clima e a ia ion.
Thus, he ealisa ion o a wea he shock is ex-an e unknown and economic agen s a e myo-
pic wi h espec o he occu ence o hese idiosync a ic wea he shocks. Hence, we assume
ha he ep esen a i e agen maximizes he u ili y a each poin in ime independen ly o
pas o u u e expec a ions o decisions.
Thus, u ili y o he ep esen a i e agen in coun y j in yea and mon h m is desc ibed by
U
j m =
(
N
∑
i=1
λ
1−σ
σ
iC
σ−1
σ
ij m
)σ
σ−1
,
(1)
whe e
Cij m
deno es coun y j’s consump ion o he speci ic a ie y impo ed om coun y
i a poin in ime
{ , m}
. This consump ion exp ession can be decomposed in wo compo-
nen s: Fi s , he e is agg ega e annual consump ion o good i in j in yea ,
Cij
. Second,
holiday seasons, accoun ing o exogenous in en o y managemen mo i a ions, as well as
o he ac o s shape he in a-annual demand a ia ion, cap u ed by he exogenous consump-
ion shi e
ϕjm
, which is no malized such ha
∑12
m=1
ϕ
jm =1
. Hence,
Cij m =Cij ϕjm
.
λi>0
desc ibes an exogenous p e e ence pa ame e o goods om coun y i and
σ>1
is he elas ici y o subs i u ion among a ie ies o di e en o igins.
While ade balances a e exogenously ixed ac oss yea s, we assume ha in a-annually
coun ies can un ade balance su pluses o de ici s. Thus, consume s in j maximize equa-
ion (1) subjec o he annual budge cons ain
∑i
τ
ij
p
i
C
ij =
E
j
, whe e
Ej
is o al
1 3
970
In e na ional T ade and he T ansmission o Tempe a u e Shocks
annual expendi u e o consump ion,
pi
deno es ac o y-ga e p ices o good i and
τij
a e
icebe g ade cos s. Impo an de e minan s o hese ade cos s a e ime-in a ian cha ac e -
is ics such as he geog aphical dis ance be ween expo e and impo e and common cul u al
a ibu es. In addi ion, ade cos s may also ha e a ime- a ying componen as new ade
ag eemen s come in o o ce o imp o emen s in in as uc u e educe anspo cos s. No e
ha we assume ha bila e al ade cos s change only annually a he han mon hly. While
in eali y, ade ag eemen s o in as uc u e imp o emen s come in o o ce in a pa icula
mon h, his assump ion simpli ies ou iden i ica ion s a egy.2 We also assume ha ac o y-
ga e p ices
pi
a e s icky and change only annually as menu cos s impede p ice adjus men s
and con ac ual ag eemen s ix p ices o ce ain pe iods.3
Sol ing he ep esen a i e agen s op imiza ion p oblem yields annual demand
C
ij =
(
λiτij pi
Pj )1−σ
Ej
,
(2)
whe e he associa ed consume p ice index in coun y j is gi en by
P
j =
(
N
∑
i=1
(λiτij pi )1−σ
)
1
1−σ
.
(3)
Fo gi en p ices, we can hus de ine he p opensi y o coun y j o spend on impo s o good
i a da e
{ , m}
wi h
θij m =pi τij Cij ϕjm
.
2.2 Ou pu and Wea he Shocks
Bu j’s impo spending p opensi y o good i canno be sa is ied in any case. Fi s , also
he p oduc ion in i is exposed o exogenous, in a-annual shi s and subjec o seasonal
a ia ion. This is mos ob ious in he case o ag icul u al p oduc ion which depends on
ha es ing cycles. O he d i e s o hese coun y-speci ic in a-annual cycles a e holiday
seasons o annual cyclical wea he impac s which also a ec p oduc ion and anspo in a-
s uc u e such as opical cyclones. This is cap u ed by he weigh pa ame e
φim
, which
desc ibes he exogenous mon hly coun y-speci ic a ia ion o ou pu , no malized such ha
∑12
m=1
φ
im =1
.
Second, he e a e po en ial wea he shocks a ec ing ou pu beyond hese cyclical pa -
e ns.4 Le us assume ha
Wi m =exp(ρ1i m(Di m))
desc ibes he po en ial wea he shock
a ec ing ou pu in coun y i in yea and calenda mon h m. I he indica o
1i m(Di m)
is
equal o one a wea he shock ma e ialises and p oduc ion o good i is exposed o he ex eme
2 Mos ade cos s changes a e long- e m and hei empo al implemen a ion p obably unco ela ed o mon hly
wea he a ia ions. This assump ion should he e o e no lead o biased es ima es o wea he impac s.
3 As seen below, p esumably empe a u e-induced labou -p oduc i i y e ec s in he manu ac u ing sec o a e
a key de e minan o he measu ed agg ega e impac on ade. These goods a e o en ela i ely speci ic and
a e no aded on global spo ma ke s bu ha e a he s icky p ices. Fo ins ance, Apel e al. (2005) ind ha
he median Swedish i m in hei sample adjus s p ices jus once a yea .
4 Mo i a ed by he abo e discussed mic o- (e.g., Somana han e al. (2021)) and mac o-le el (e.g., Bu ke e al.
(2015)) s udies, wea he shocks a ec ou pu in ou model. Howe e , one could also model wea he shocks
on p e e ences o ade cos s.
1 3
971
O. Schenke , D. Osbe ghaus
wea he e en
Di m
. In gene al, his could be a mon h o ex eme high o low empe a u e,
hea y o poo ain all, o high wind speed. O he wise,
1i m(Di m)=0
. The pa ame e
ρ
measu es hen he wea he shock’s impac on ou pu . Iden i ying
ρ
is one o he key aims o
ou empi ical exe cise.
Le us deno e
Xi m
as he o al ee on boa d ( .o.b) alue o expo s o i. Then,
X
i m
=∑jX
ij m, whe e
Xij m
is he alue o ac ual bila e al expo s ne o in a-annual
supply shi s and wea he shocks om coun y i o j.
2.3 In e na ional T ade
Hence, o gi en p ices, ac ual expo s o coun y i o j a ime
{ , m}
can be exp essed as
Xij m =θji m φim Wi m
. Sol ing o
θji m
and plugging his in o he demand equa ion
(2) leads o
X
ij m =
(
λiτij pi
Pj )1−σ
Ej ϕjm φim Wi m
.
(4)
Annual o al expo s o coun y i in yea a e hus
X
i =
N
∑
j=1
12
∑
m=1
φim Wi m
(
λiτij pi
Pj
)1
−σ
Ej ϕjm
.
A e ea anging his exp ession and di iding by
(
λ
i
p
i )1−σ
, ollowing Ande son and
Van Wincoop (2003), we de ine he e m on he igh hand side as
Π
1−σ
i =
N
∑
j=1
12
∑
m=1 (
τij
Pj
)1
−σ
Ej ϕjmφim Wi m
,
(5)
he so-called ou wa d mul ila e al esis ance e m. Since
(
λ
i
p
i )1−σ=
X
i
/
Π1−σ
i
, we plug
his in o equa ion (4) and ge he G a i y equa ion ha desc ibes he mon hly expo s o
coun y i o j:
X
ij m =Xi φim Wi m Ej ϕjm
(
τij
Πi Pj )1−σ
.
(6)
Equa ion (6) is he equa ion we a e going o es ima e. I is his equa ion ha ansmi s
he wea he shock om he expo e s loca ion o he impo e , a ec ing he a ailabili y o
goods om i. Assuming ha
ρ<1
, he occu ence o a wea he shock in i educes bila e al
impo o j by
exp(ρ)
a ha pa icula poin in ime. Ce e is pa ibus, households in j ace a
po en ial impo loss. Bu igno ing he gene al equilib ium esponse om subs i u ion and
p ice adjus men s may be misleading. Wi h a posi i e
σ
, consume s in j a e able o subs i u e
goods o di e en o igin, so a wea he -caused sho age o supply om one coun y can, a
leas pa ially, be compensa ed by impo s om o he loca ions.
1 3
972
In e na ional T ade and he T ansmission o Tempe a u e Shocks
abno mally high empe a u e shocks, iden i ied by obse a ions in he highes pe cen ile o
he coun y-speci ic empe a u e dis ibu ions. This speci ica ion, p esen ed in column (2)
in Table 1, con i ms a con empo aneous sho - un non-linea empe a u e e ec on expo s.
Compa ing columns (1) and (2) in Table 1, he ques ion a ises whe he he wo speci-
ica ions desc ibe he same economic p ocesses, o whe he he e ec s o absolu ely and
abno mally high empe a u es a e independen impac s. In he la e case, he unde lying
impac channels may di e , indica ed by signi ican ma ginal e ec s in an es ima ion includ-
ing bo h empe a u e speci ica ions. The e o e, in column (3), we include bo h absolu ely
and abno mally high empe a u e, and in column (4) we addi ionally s udy he in e ac ion
be ween hem. Fo expo e s, bo h empe a u e a iables (
D 30
i m
and
Dp99
i m
) emain indi idu-
ally signi ican . Mo eo e , he e ec s a e independen om each o he as he e is no s a is i-
cally signi ican in e ac ion e ec . Hence, bo h e ec s can be obse ed in he sho e m:
a subs an ial e ec on expo s om episodes o absolu e ho empe a u e, and a somewha
smalle e ec o mon hs o abno mally high empe a u es, gi en he coun y-speci ic em-
pe a u e dis ibu ion.
In he Appendix, we also es a empe a u e impac model wi h a linea , as well as a
quad a ic speci ica ion, as o en assumed in he li e a u e — see, o example, Missi ian and
Schlenke (2017) o Bu ke e al. (2015). The esul s, p esen ed in columns (1) and (2) in
Table 4 in he Appendix, con i m a non-linea e ec o empe a u e on expo s. Mo eo e ,
we assess he ques ion whe he he e ec s o absolu e and abno mally ho mon hs a ec he
ex ensi e o in ensi e ma gin o bo h. The e o e, we es ima e linea p obabili y models o
he bina y a iable whe he he e is ade be ween coun ies (columns 3 and 4), and es ic
he sample o posi i e expo lows (columns 5 and 6). The esul s (see Table 4 in he Appen-
dix) sugges ha he e ec s s em om changes a he in ensi e ma gin: Tempe a u e e en s
ha e no signi ican impac on he decision o whe he expo s occu o no , bu educe he
alue o expo s in he subsample o posi i e expo lows.
Equipped wi h his se ies o es ima ion esul s o con empo aneous e ec s o empe a-
u e, we ob ain one obus inding: Mon hs wi h high empe a u es, ei he measu ed in abso-
lu e o ela i e e ms, lead o a s a is ically and economically signi ican educ ion o expo s
(1) (2) (3) (4)
Abs. Temp. Abn. Temp. Bo h In e ac ion
D 30
i m
−0.0336*** −0.0253** −0.0166
(0.0100) (0.0124) (0.0140)
D 30
j m
−0.0124** −0.0092 −0.0095
(0.0056) (0.0058) (0.0059)
Dp99
i m
−0.0210*** −0.0206*** −0.0203***
(0.0076) (0.0076) (0.0077)
Dp99
j m
−0.0149** −0.0148** −0.0149**
(0.0064) (0.0064) (0.0063)
D
30
i m ×
D
p99
i m
−0.0236
(0.0190)
D
30
j m ×
D
p99
j m
0.0019
(0.0089)
Obse a ions 3821155 3821155 3821155 3821155
Table 1 Con empo aneous em-
pe a u e e ec s
ρ
on expo s
The dependen a iable is
bila e al expo s in cu en
USD. Index i (j) indica es he
expo ing (impo ing) coun y.
S anda d e o s in pa en heses
1 3
979
O. Schenke , D. Osbe ghaus
ela i e o mon hs acing lowe (o less ex eme) empe a u e le els. This e ec is well cap-
u ed by he pa simonious models ocusing on
D 30
i m
o
Dp99
i m
(columns (1) and (2) in Table
1). Re iewing he p io li e a u e on empe a u e e ec s on mac o-economic ou comes (see
li e a u e e iew in Sec . 1), e eals a clea ocus on absolu e empe a u e speci ica ions, as
he e is a sound and obus mic o-economic and physiological empi ical ounda ion o hese
e ec s, while he e is less heo e ical and mic o-econome ic suppo o economic impac s
om abno mally high empe a u es. Fu he analysis e eals ha he obse ed e ec is p i-
ma ily d i en by episodes o ex emely high absolu e empe a u es in wa me coun ies,
a he han by episodes o ela i ely high empe a u es in colde coun ies, which a e mod-
e a e in absolu e e ms (see Fig. 13 in he Appendix). The e o e, in he emainde o he
analysis, we a e going o ocus on impac s o high absolu e empe a u e (
D 30
i m
). Howe e ,
we eplica e all analyses wi h speci ica ions based on abno mally high empe a u es, epo
he esul s in he Appendix, and highligh po en ial quali a i e di e ences.
5.2 Lagged Tempe a u e E ec s
We ha e shown ha mon hs o high empe a u es ha e a de imen al e ec on expo s in
he mon h o hei occu ence. Howe e , i is impo an o unde s and he du a ion o hese
impac s. A e hey only sho -li ed o do hey ha e longe e m consequences? Fo assess-
ing his ques ion, we es ima e a ini e dis ibu ed lags model wi h ou mon hs be o e and
wel e a e he empe a u e e en . To educe compu a ional complexi ies, we ely on he
pa simonious model wi h a dummy o mon hs wi h a e age empe a u e g ea e o equal
30
◦C
. Figu e 3 depic s he es ima ed coe icien s, ela i e o a mon h whe e empe a u e
has been below 30
◦C
.
The es ima ion con i ms he con empo aneous e ec o a empe a u e shock on expo s
in he mon h o he e en (−6.7 pe cen ,
p=0.031
). We also ind a lagged nega i e e ec
h ee mon hs a e he e en (−5.5 pe cen ,
p<0.001
) and a posi i e e ec a e se en
mon hs (+4.2 pe cen ,
p=0.009
). The es ima ed e ec s p io o he empe a u e e en ,
se ing as placebos, a e s a is ically non-signi ican . The cumula i e e ec o e he pe iod
o one yea a e he e en emains nega i e, albei a non-signi ican le els (see Fig. 15
in he Appendix). We conclude ha empe a u e shocks on expo s mani es mainly in he
sho e m du ing and di ec ly a e he e en . The e ec is nei he subs an ially agg a a ing
o e ime no is he e any e idence o a subs an ial ca ching up o compensa ion o los
expo s in pos -e en mon hs, such ha he cumula i e e ec a e one yea is in he same
o de o magni ude as a e a ew mon hs. Fo abno mally high empe a u es, he analysis
o lagged e ec s yields simila esul s: The e ec p o es o be sho -li ed and only exis en
o expo e s (see Figs. 16 and 17 in he Appendix). Howe e , he ini ially nega i e e ec
o abno mally ho mon hs is ollowed by some (s a is ically non-signi ican ) posi i e lagged
e ec s, such ha he cumula i e e ec a e one yea o he e en is no s a is ically di e en
om ze o.
In summa y, i can be s a ed ha ac oss all model speci ica ions he empe a u e shock
in he expo ing coun y is mo e ele an han in he impo ing coun y, suppo ing simila
indings based on educed- o m, pa simonious es ima es wi h annual ade da a (Dallmann
2019). These iden i ied empe a u e impac s a e ela i ely sho -li ed. Consequen ially, we
ocus in he emainde o he analysis on he con empo aneous empe a u e e ec s in he
1 3
980
In e na ional T ade and he T ansmission o Tempe a u e Shocks
expo ing coun y. Howe e , we keep he empe a u e in he impo ing coun y as a co a i-
a e in all es ima ions.
5.3 He e ogeneous E ec s o Tempe a u e
So a , we es ima ed a e age e ec s o high empe a u e e en s on expo s ac oss he ull
sample. Bu he e ec magni ude migh be condi ional on coun y o ade low cha ac e is-
ics. Unde s anding hese di e ences allows us o in e mo e p ecisely abou impac chan-
nels and economic mechanisms ansla ing he empe a u e shock in economic ou comes.
One impo an impac channel iden i ied by p io mic o- and plan -le el e idence sug-
ges s ha high absolu e ambien empe a u e educes p oduc i i y and supply o labou
(Somana han e al. 2021; Zhang e al. 2018; Zi in and Neidell 2014). Hence, we inco -
po a e he labou in ensi y o yea - and impo e -speci ic expo s (
labou in ij
). I labou
p oduc i i y (o supply) is a majo de e minan o he magni ude o he ad e se empe a u e
e ec on expo s as sugges ed, he es ima ed e ec should a y wi h he labou in ensi y o
bila e al expo s.9
We augmen he basic es ima ion equa ion (10) wi h in e ac ion e ms o he he e ogene-
i y a iable wi h
D 30
i m
. Di ec e ec s o labou in ensi y and o he he e ogenei y a iables
a e no es ima ed as hey a e pe ec ly collinea wi h ixed e ec s a coun y pai -yea le el.
9 While we ocus he e on labou in ensi y gi en he mic o-economic e idence, we simila ly es o o he
po en ial sou ces o e ec he e ogenei y in he Appendix. Inspi ed by Dell e al. (2012), we in e ac he em-
pe a u e e ec wi h he annual income in he expo ing coun y (
gdpi
) and he annual sha e o ag icul u al
p oduc ion in he expo e ’s GDP (including o es y, ishe ies and hun ing,
ag ii
). We u he hypo hesize
ha he expo e ’s esilience owa ds clima e change (measu ed by he annual ND-GAIN index,
ndgaini
) may go e n he esponse o a empe a u e shock. Finally, e ec s may a y wi h he p oduc composi ion
o o al expo s (Jones and Olken 2010). The e o e, we in e ac he empe a u e shock wi h p oduc -speci ic
sha es o o al expo s in he p eceding ou yea s (e.g.,
p F oodij
o ood and li e animal p oduc s).
Fig. 3 Lagged impac on expo s o an a e age mon hly empe a u e o a leas 30
◦C
. Es ima ed e -
ec s
ρl
o a ho mon h on expo s, including 95-pe cen con idence in e als. The e ec s a e ela i e
o a mon h wi h a empe a u e below 30
◦C
. Lagged impac s o ho mon hs a he impo e loca ion a e
depic ed in Fig. 14 in he Appendix
1 3
981
O. Schenke , D. Osbe ghaus
We es ima e ma ginal e ec s o
D 30
i m
on expo s o e di e en le els o he he e ogene-
i y a iable. Figu e 4 depic s he esul s o labou in ensi y, and Fig. 19 in he Appendix
summa izes simila plo s o he o he po en ial he e ogenei y a iables.
The esul s sugges ha he con empo aneous e ec s o absolu ely ho mon hs is indeed
go e ned by he labou in ensi y o expo s. The in e ac ion e ec shown in Figu e 4 is
highly signi ican (
p=0.003
). A 10 pe cen age poin s inc ease in labou in ensi y o expo s
is associa ed wi h an inc ease o he ad e se impac by app oxima ely 5 pe cen age poin s.
On he con a y, he e is no signi ican in e ac ion e ec wi h ela i e empe a u e ex emes
(
Dp99
i m
, see Figu e 18 in he Appendix). This is b oadly in line wi h p io mic o-economic
and physiological li e a u e (see e.g., Dunne e al. (2013)), which pos ula es ha absolu e
empe a u e le els a e c ucial o po en ial declines o wo k capaci y.10 Hence, an unusually
wa m summe in a cold en i onmen would be ea ed as an ex eme empe a u e e en in
Dp99
i m
, bu has no expec ed ad e se impac on labou p oduc i i y. In a simila ein, Jones
and Olken (2010) de ec a nega i e e ec o high empe a u es on expo s o ligh manu ac-
u ing goods and specula e ha p oduc i i y e ec s o wo ke s migh be esponsible. Ou
esul s p o ide u he e idence ha labou p oduc i i y o supply is key o unde s anding
he mechanism behind he iden i ied e ec s o empe a u e shocks on agg ega ed expo s.
No e ha ou da a se con ains only ade in goods bu no in se ices. Since se ices a e
ypically mo e labo -in ensi e han goods, ou es ima e o he ull impac o empe a u e
shocks on expo s he e o e likely ep esen s a conse a i e assessmen and a lowe bound
o he ull e ec .
10 While o he ac o s such as wind speed and humidi y a e impo an , subs an ial losses o wo k capaci y a e
gene ally only obse ed a absolu e empe a u e le els o highe han 25
◦C
.
Fig. 4 Es ima ed e ec o an a e age mon hly empe a u e o a leas 30
◦C
on expo s o di e en le els
o labou in ensi y. Es ima ed con empo aneous e ec s
ρ
o
D 30
i m
on expo s o gi en le els o labou
in ensi y, including 95-pe cen con idence in e als. The e ec s a e ela i e o a mon h wi h empe a u e
below 30
◦C
. Labelled alues a he x-axis a e he
5 h
,
50 h
, and
95 h
pe cen ile o
labou in ij
1 3
982
In e na ional T ade and he T ansmission o Tempe a u e Shocks
Mos o he o he analyzed a iables show no signi ican in e ac ion wi h he empe a u e
e ec (Fig. 19 in he Appendix). In ou s uc u al G a i y model, he empe a u e e ec s on
expo s do nei he signi ican ly a y wi h economic de elopmen , he sha e o ag icul u al
goods in he expo ing coun y’s p oduc ion, o he expo e ’s assessed esilience o clima e
change, no o e ime (Fig. 21 in he Appendix). Simila ly, mos o he p oduc ca ego y
sha es do no in e ac wi h empe a u e impac s — wi h he excep ions o expo s cha ac e -
ized by high sha es o C ude Ma e ials (mo e ad e sely a ec ed) and Mine al Fuels (less
a ec ed). These esul s, howe e , a e compa ible wi h he in e p e a ion ha labou p oduc-
i i y is he unde lying channel o empe a u e e ec s — as labou in ensi y is ela i ely
high o he o me and low o he la e .
5.4 Ex ensions: Impac s o P ecipi a ion and S o ms
While we ocus on he e ec s o high empe a u e, he da a and he employed empi ical
me hodology gene ally allow o an equi alen analysis o he impac s o o he wea he
phenomena. We a e pa icula ly in e es ed in wea he e en s ha may be a ec ed by clima e
change, and he e o e addi ionally assess impac s o ex eme p ecipi a ion and s o ms. The
employed da a and ob ained esul s a e summa ized in Appendix 8.4..
Rega ding p ecipi a ion, we do no ind any con empo aneous e ec s on expo s o
impo s (see Table 6). Conside ing he po en ial impac channels o hyd ological e en s on
p oduc ion p ocesses, his non-e ec may be plausible: p ecipi a ion is mos ele an o he
p oduc ion o ag icul u al goods. Bu hese goods ha e ce ain g owing pe iods, and hei
p ocessing also needs ime, such ha a con empo aneous e ec o a lack o p ecipi a ion on
he p oduc ion and hence on expo s is unlikely. Bu pe haps i is he combina ion o high
empe a u e wi h low p ecipi a ion ha leads o d ough s and nega i e impac s on expo s.
I high empe a u e e en s a e highly co ela ed wi h episodes o low p ecipi a ion, ou
es ima ed empe a u e e ec s migh mask d ough e ec s. We ind a posi i e bu low co -
ela ion o he appea ance o low p ecipi a ion e en s wi h high empe a u e e en s (Spea -
man’s
ρ=0.07
). This aligns wi h he indings o an es ima ion model ha inco po a es
low-p ecipi a ion e en s as con ols, yielding empe a u e e ec s nea ly iden ical o hose o
he baseline model (see Table 6 in he Appendix).
Ex eme high p ecipi a ion, in con as , may ha e ad e se e ec s on ade, e.g. by looded
anspo in as uc u e o p oduc ion acili ies. The ac ha we do no ind such an e ec
may be due o an inaccu a e measu e o lood in ensi y. As a mon hly mean alue, ou p e-
cipi a ion measu e may no p ope ly indica e sho - e m ex eme e en s which las only o
one o wo days. Fu he mo e, o being ha m ul o he economy, he occu ence o in ense
p ecipi a ion e en s mus coincide wi h he loca ion o ulne able asse s o in as uc u e —
issues ha a e no su icien ly de ec able by coun y-mon h a e ages.
Simila ly, we ind only limi ed e idence o he exis ence o s o m impac s on expo s.
Linea and quad a ic speci ica ions o mon hly maximum wind speed alues yield insig-
ni ican e ec s. Ex eme wind speed e en s, de ined as being in he coun y-speci ic op
pe cen ile o wind speeds, a e wi hou e ec as well. Howe e , we ind non-linea e ec s
o e y in ense s o ms in absolu e e ms (o 140 kno s maximum wind speed and highe )
when analyzing a lexible model using wind speed bins. In hese mon hs, expo s dec ease
subs an ially (up o 7 pe cen ), which may hin o impai ed anspo in as uc u e o p o-
duc ion acili ies.
1 3
983
O. Schenke , D. Osbe ghaus
6 Coun e ac ual Simula ions
Ou p e ious analysis e ealed ha expo s a e nega i ely a ec ed by ex eme empe a u e
e en s. A high empe a u e episode in he expo ing coun y educes ade in he mon h o
he e en . Bu he global cos s o such an e en emain unclea since impo e s a e able o
ei he sou ce goods om somewhe e else o compensa e o los impo s by pu chasing
mo e in la e pe iods om he same sou ce. We s udied possible compensa ion ac oss ime
using dis ibu ed lag models bu did no ind obus e idence o his (see Sec . 5.2).
Bu impo e s could also seek o compensa ion ac oss space. Al hough in ou model
impo s o di e en o igin a e impe ec subs i u es only, buye s can adjus he sou ce o
hei impo s and, a leas pa ially, ecoup losses on one ade link by addi ional impo s
om o he sou ces. This demand adjus men has, o cou se, epe cussions on ela i e p ices
and a ailable income, changing he wo ld ade equilib ium.
The e o e, we compu e he new global ade equilib ium esul ing om a empe a u e
shock, he global cos s, and hei dis ibu ion on di ec ly and indi ec ly a ec ed coun ies.
Finally, we app oach he ques ion how hese global cos s will e ol e unde clima e change.
6.1 Me hodology
As discussed abo e, and as shown by Fally (2015), we a e able o e ie e om ou PPML
ixed e ec s es ima ion o (10) he unde lying s uc u al G a i y model, desc ibed in equa-
ions (5)– (9). See Appendix 8.1. o he de i a ion o he model om he ixed e ec
es ima ion.
Howe e , he e a e wo pa ame e s which we do no obse e. One is he p e e ence
pa ame e
λi
, which in o ms he p icing equa ion (8). Bu we can ew i e equa ion (8) such
ha i p o ides he p ice change ela i ely o he es ima ed baseline which is independen o
his s uc u al pa ame e .
∆
pi =
(
Xi
ˆ
X
ˆ
Xi X )
1
1−σ
ˆ
Πi
Πi
,
whe e he ha deno es es ima ed pa ame e s, i.e. he p edic ed expo s and he es ima ed
ou wa d mul ila e al esis ance e ms.
The second no e ie able key pa ame e is he elas ici y o subs i u ion be ween a ie ies
o di e en o igin,
σ
. We ake his om he li e a u e and in o m ou model by he elas ici y
es ima e o Simono ska and Waugh (2014), who es ima e i using disagg ega e p ice and
ade- low da a. Thei es ima e yields an elas ici y o oughly ou . Howe e , we a e going
o conduc ex ensi e sensi i i y analyses and es he obus ness o ou indings wi h espec
o
σ
. Ob iously, he ease o subs i u ion om one good o ano he subs an ially in luences
he magni ude o he e ec s.
Equipped wi h ou es ima ed model and he addi ional pa ame e , we un coun e ac ual
simula ions, compu ing he hypo he ical ade equilib ium in absence o a high absolu e
empe a u e e en . This allows o assess he ull in e na ional ade cos s o a empe a-
u e shock, accoun ing o equilib ium adjus men s in he ade ne wo k. Bu i needs o be
poin ed ou ha we only obse e in e na ional ade lows, omi ing domes ic impac s on
p oduc ion and income as well as domes ic adjus men e ec s. Fo ins ance, a empe a u e
1 3
984
In e na ional T ade and he T ansmission o Tempe a u e Shocks
shock migh induce a ealloca ion o esou ces wi hin he coun y and subsequen ly also
a ec impo s and expo s. Gi en he a he sho e m na u e o he empe a u e shock, sub-
s an ial equilib ium adjus men s o domes ic esou ces seems a he imp obable. Howe e ,
wi hou obse ing he ull domes ic impac in he a ec ed coun y, we a e also no able o
de i e comp ehensi e wel a e e ec s o empe a u e shocks om ou simula ions. Ra he ,
ou calcula ions p o ide he cos s o empe a u e shocks in e ms o los ade.11
To adminis e he coun e ac ual simula ions and compu e he new equilib ia we need o
educe he model dimensionali y in e ms o spa ial and empo al co e age. Fi s , we educe
he numbe o coun ies inco po a ed in he analysis. This mo e pa simonious model co -
e s he bila e al mon hly ade o 43 coun ies (lis ed in Table 8 in he Appendix). The ade
wi hin his sub-sample makes up oughly 90 pe cen o he global ade olume. Wi h his
educed da a se we boo s ap 100 imes he es ima ion o he s uc u al G a i y equa ion
(11) wi h speci ic expo e
×
yea , impo e
×
yea , and coun y-pai
×
yea ixed e ec s. The
mean coe icien o he wea he shock
D 30
i m
on expo s es ima ed wi h his smalle da a se
spanning 43 coun ies only is almos iden ical o he poin es ima o based on he ull da a
se in ou p e e ed speci ica ion epo ed in column (1) o Table 1. In he nex s ep, we
selec he yea 2015 — he yea whe e we ha e he bes da a co e age o mon hly bila e al
ade — o ou assessmen .12
The boo s apped ixed e ec s o 2015 plus he es ima ed empe a u e shock coe icien
calib a e ou CES-A ming on mon hly- ade model desc ibed in sec ion 2. As hese es i-
ma ed coe icien s, join ly wi h ou assumed ade elas ici y, ully and consis en ly speci y
ou ade model, we a e eady o conduc coun e ac ual analyses. Fo each se o boo -
s apped es ima es, we andomly d aw an obse ed high empe a u e e en in 2015, and
compu e he coun e ac ual global ade equilib ium assuming his e en would no ha e
happened. We epo he mean loss o impo s as well as he 95 pe cen -con idence in e al
o he coun y expe iencing he empe a u e shock (di ec ly a ec ed) and all o he coun ies
(indi ec ly a ec ed).
6.2 Global T ade-loss Incidence o Local Tempe a u e Shocks
Equipped wi h ou es ima ed s uc u al G a i y model, we aim a answe ing how a single
high empe a u e e en changes he wo ld ade equilib ium and impac s di ec ly and indi-
ec ly a ec ed coun ies. As a i s app oxima ion o compu ing he global in e na ional
ade losses o he a e age empe a u e shock, an analys could jus sum up he educ ion
o impo s om he di ec ly a ec ed expo e ac oss all impo e s, igno ing subs i u ion
and income e ec s. This is shown in he le ba in he le panel o Figu e 5. This "naï e"
app oach leads o agg ega ed in e na ional ade losses o 2015-USD 454 million o he
a e age high empe a u e e en . Howe e , as he igh ba in he same panel o Figu e 5
shows, when equilib ium adjus men s a e aken in o accoun , impo e s a e able o pa ially
11 No e ha in wel a e e ms — igno ing any ic ions and igidi ies — he maximum cos s a s ake o he
impo ing coun y a e he o al wel a e gains om ade wi h he expo ing coun y exposed o he wea he
e en .
12 The global clima e in 2015 has been cha ac e ized by an El Niño si ua ion — a phase o he El Niño-
Sou he n Oscilla ion (ENSO) clima e phenomena ha in luences sea empe a u es and wea he in la ge pa s
o he globe (Blunden and A nd 2016). As a consequence, 2015 saw hea wa es in F ance, high empe a u es
and d ough in Sou h Ame ica, in pa icula in A gen ina, B azil, and Colombia, as well as se e e d ough
in Sou h A ica.
1 3
985
O. Schenke , D. Osbe ghaus
subs i u e hei pu chases om he di ec ly exposed coun y o o he sou ces, and global
ade losses dec ease o 360 million USD, a educ ion o abou 20 pe cen ela i e o an
assessmen ha igno es equilib ium e ec s. This is a clea indica ion o in e na ional ade’s
po en ial o dis ibu e and educe cos s o such shocks h ough equilib ium adjus men s and
subs i u ion.
This o al amoun o impo losses appea s in di ec ly and indi ec ly a ec ed coun ies.
Indi ec ly a ec ed impo e s can only pa ially subs i u e hei impo losses ia al e na i e
sou ces. In his sense, in e na ional ade ansmi s a sha e o he cos s o he empe a u e
e en . A cos assessmen o hese e en s no including hese c oss-bo de e ec s is he e o e
incomp ehensi e. The igh panel o Figu e 5 shows his. O he 360 million USD o al
in e na ional ade losses appea 136 million USD in he di ec ly a ec ed expo ing coun y
due o losses in pu chasing powe caused by he empe a u e-induced expo educ ion. This
means ha he g oup o coun ies ha ha e no been di ec ly expe ienced he empe a u e
e en bea impo losses o 224 million USD om he a e age single empe a u e shock.
This a e a e age cos s o abou 5.3 million USD pe empe a u e shock o an indi ec ly
a ec ed coun y ia hese ade spillo e s. While his e ec is no huge, ou boo s apped
con idence in e al sugges s ha hese cos s a e s a is ically signi ican di e en om ze o
gi en a i e pe cen signi icance le el.
Bu hese indi ec cos s a e dis ibu ed une enly ac oss coun ies. As G a i y heo y ells
us, absolu e indi ec cos s o empe a u e shocks a e go e ned by he o al alue o impo s
and he e o e depend on he size o he impo ing coun y, as well as he ade cos s o ship-
ping a good o he impo ing coun y. Thus, la ge impo ing coun ies and coun ies wi h
lowe cos s o ade wi h he di ec ly a ec ed coun y ha e o bu den la ge absolu e cos s.
Figu e 6 anks he a e age indi ec cos s in a dec easing o de . While he coun y a he
25 h
pe cen ile aces cos s o abou 8.6 million USD, he coun y a he
75 h
pe cen ile aces
cos s o 2.3 million USD.
6.3 Sensi i i y Analysis
The magni ude o he gene al equilib ium e ec s is ully speci ied by he es ima es o he
unde lying G a i y model wi h one deg ee o eedom: The es ima ion o he G a i y sys em
Fig. 5 Gene al equilib ium ade losses o a e age high empe a u e e en . The le panel shows he ag-
g ega ed global ade loss (summing up di ec ly and indi ec ly a ec ed coun ies) o he a e age high
empe a u e e en (mon hly mean empe a u e a leas 30
◦C
). The le ba depic s he mean es ima e o
los impo s be o e equilib ium adjus men s (naï e). The igh ba shows es ima ed los impo s a e he
equilib ium-adjus men (equil.). The igh panel spli s he agg ega ed losses in o a e age losses o he
di ec ly a ec ed coun y and a e age agg ega ed losses o he indi ec ly a ec ed coun ies. Boo s apped
95 pe cen con idence in e al
1 3
986
In e na ional T ade and he T ansmission o Tempe a u e Shocks
does no iden i y he elas ici y o subs i u ion
σ
which go e ns he ease o adjus demand
and sou ce goods om somewhe e else. As discussed abo e, in ou cen al es ima es we
se
σ=4
, ollowing Simono ska and Waugh (2014). Howe e , i is key o unde s and how
sensi i e ou esul s a e depending on he choice o
σ
.
Figu e 7 plo s he e ec o he mean high empe a u e e en o alues o
σ
anging om
2.5 o 5.5. The igh panel shows he agg ega e impo losses o only indi ec ly a ec ed
coun ies. Assuming an elas ici y o subs i u ion o 2.5 leads on a e age o an impo loss o
abou 520 million USD. These losses dec ease wi h a subs i u ion elas ici y o 5.5 o abou
124 million USD o all impo ing coun ies. I is also impo an o no e ha he poin es i-
ma o o he mean empe a u e e en on he a e age indi ec ly a ec ed coun y, al hough
being small (2.9 million USD a
σ=5.5
), is s a is ically di e en om ze o wi h 95 pe cen
con idence o e he whole ange o he es ed elas ici ies. As heo y sugges s, he cos s a e
Fig. 7 Sensi i i y analysis–elas ici y o subs i u ion
σ
. Sensi i i y analysis o gene al equilib ium e ec s
condi ional on elas ici y o subs i u ion
σ
anging om 2.5 o 5.5. The le panel shows he ade losses
o he mean
>30
◦C
e en on di ec ly a ec ed coun ies. The igh panel shows he losses on indi ec ly
a ec ed coun ies. Con idence in e als a e boo s apped
Fig. 6 Dis ibu ion o indi ec cos s. Dis ibu ion o a e age cos s o he op-15 indi ec ly a ec ed coun-
ies in dec easing o de . Boo s apped 95 pe cen con idence in e al
1 3
987
O. Schenke , D. Osbe ghaus
lowe o highe elas ici ies as cos s can be mo e easily mi iga ed by adjus ing he sou cing
o goods.
The le panel o Fig. 7 shows he e ec s o di ec ly a ec ed coun ies. I cos -sha ing
is limi ed unde a low elas ici y o subs i u ion o 2.5, he a e age di ec ly a ec ed coun y
aces mean losses 220 million USD. Assuming an elas ici y o subs i u ion o 5.5, hese
losses dec ease o abou 100 million USD.
6.4 Cos s Unde Fu u e Clima e P ojec ions
So a , ou es ima ed s uc u al G a i y model has been applied o compu e ex-pos gene al
equilib ium e ec s o high empe a u e e en s. Howe e , he model allows also o compu e
he impac s on in e na ional ade — o bo h, di ec ly and indi ec ly a ec ed coun ies—
unde u u e clima e p ojec ions. We hus compu e he coun e ac ual impac a p ojec ion
o he u u e clima e would ha e on ou es ima ed wo ld economy o 2015 and compa e i
wi h he his o ic clima e means. The e is he e o e an impo an ca ea : Ou simula ions
assume ha he u u e clima e happens in he wo ld economy o 2015, igno ing any socio-
economic adjus men s and changes ha will happen. Bu ade cos s, o al expendi u es bu
also he ulne abili y o high empe a u e migh change subs an ially in u u e.13 We he e-
o e belie e ha his is a sensible app oach, and a e ully awa e o i s limi s.
We use da a p o ided by Wo ld Me eo ological O ganiza ion in he KNMI Clima e
Change A las. This da abase p o ides modelled mean empe a u es a he coun y-mon h
le el, bo h o he pas and o p ojec ions up o he yea 2100. We ely on he median ou pu
o he mul i-model ensemble CMIP5 (Coupled Model In e -compa ison P ojec Phase 5
used in he
5 h
IPCC Assessmen Repo ), and ocus on p ojec ions based on he Rep esen-
a i e Concen a ion Pa hway (RCP) 4.5. RCP4.5 is a a he "op imis ic" emission scena io
which assumes ha global emissions peek be o e 2050, and adia i e o cing s abilizes by
2080–2100. Howe e , due o he ine ia o he clima e sys em, p ojec ions o he nea u u e
do no a y subs an ially ac oss RCPs.
In ou analysis, we examine he consequences o he a e age clima e o 2020–2039
due o he easons discussed abo e. We compa e his pe iod o he la es a ailable his o ic
20-yea pe iod in he da a, which is 1980–1999. Thus, he di e ences o he his o ic clima e
a e no subs an ial. Fo bo h 20-yea pe iods, we compu e o each coun y and calenda -
mon h he p obabili y ha he mon hly mean empe a u e exceeds
30◦C
.
We hen compu e he di e ence in annual impo expendi u es, compa ing a scena io
wi h he his o ical dis ibu ion o high empe a u e e en s o a scena io wi h he expec ed
dis ibu ion o hea e en s gi en he clima e p ojec ion o 2020–2039. We exp ess hese
changes in pe cen o he his o ical baseline.
While he di e ences depic ed in Fig. 8 a e small, we obse e addi ional ade losses in
all coun ies ha a e pa o he simula ed wo ld ade equilib ium. While hese addi ional
losses a e small in Eu ope, hey become subs an ially highe in Asia (in pa icula India) and
Oceania, as well as in Sou h A ica.
I we sum his up ac oss coun ies, we ind ha annual global ade is educed by 735
million USD due o addi ional mon hs wi h high empe a u es in a p ojec ion o he a e age
13 An impo an ac o is u u e access o ai condi ioning. Da is and Ge le (2015) show ha wi h ising
income, demand o cooling will inc ease subs an ially in many o oday’s middle income and de eloping
coun ies.
1 3
988
In e na ional T ade and he T ansmission o Tempe a u e Shocks
ha he e ec is on he in ensi e ma gin, as he e is no signi ican e ec on he bina y a i-
able indica ing posi i e ade (Models 3 and 4), bu a subs an ial e ec on he ade alue in
he sub-sample o posi i e expo s (Models 5 and 6). No e ha he numbe o obse a ions
in he Models 3 o 6 is lowe han in he Models 1 and 2, since we can only sa ely iden i y
ze o ade lows since he yea 2000.
Fo assessing he e ec s o high empe a u es in di e en ly exposed coun ies, we di ide
he sample in o ’cold’ and ’wa m’ expo e s, depending on hei maximum empe a u e
h oughou he analysis pe iod. ’Wa m’ coun ies exhibi maximum empe a u es abo e he
median o he maximum empe a u e dis ibu ion. Figu e 13 p esen s he esul s in he o m
o 5-deg ee empe a u e bins. Simila o he p oceeding analyses, hese esul s sugges ha
Fig. 12 Con empo aneous e ec s o empe a u e on expo s in 5
◦C
bins. Es ima ed coe icien s o em-
pe a u e bins in he expo ing coun y (le panel) and he impo ing coun y ( igh panel) and hei 95
pe cen con idence in e als, using 15–20
◦C
as he baseline ca ego y. Based on a PPML eg ession wi h
ixed e ec s a he coun y-pai -yea -le el, expo e -calenda mon h-le el, and impo e -calenda mon h-
le el (N=3,821,155)
Fig. 11 Con empo aneous e ec s o empe a u e on expo s in
1◦C
s eps, un es ic ed plo . Es ima ed
coe icien s o each
◦C
in he expo ing coun y (le panel) and he impo ing coun y ( igh panel) and
hei 95 pe cen con idence in e als, using 20
◦C
as he baseline ca ego y. Based on a PPML eg ession
wi h ixed e ec s a he coun y-pai -yea -le el, expo e -calenda mon h-le el, and impo e -calenda
mon h-le el (N=3,821,155)
1 3
995
O. Schenke , D. Osbe ghaus
e y high empe a u es a e he main d i e o he es ima ed nega i e e ec . Mode a ely high
empe a u es in absolu e e ms, e en i abno mally high o he exposed loca ions, do no
show a simila e ec on expo s - nei he in ’wa m’ no in ’cold’ coun ies.
Figu e 14 illus a es he lagged impac s o a empe a u e shock a he impo e ’s loca ion,
es ima ed as co a ia es in he con ex o assessing lagged impac s on expo s (p esen ed in
Fig. 3). The es ima ed coe icien s a e no s a is ically di e en om ze o h oughou he
i s yea a e he empe a u e shock.
Table 4 Linea and quad a ic empe a u e e ec s on expo s, es ima es o he ex ensi e and in ensi e ma gin
(1) (2) (3) (4) (5) (6)
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
empi m
0.0013 0.0022*
(0.0010) (0.0011)
emp2
i m
−0.0001**
(0.0000)
D 30
i m
0.0017 −0.0398***
(0.0035) (0.0125)
Dp99
i m
0.0009 −0.0235***
(0.0013) (0.0090)
empj m
−0.0005 −0.0002
(0.0008) (0.0009)
emp2
j m
−0.0000
(0.0000)
D 30
j m
0.0024 −0.0239***
(0.0016) (0.0073)
Dp99
j m
−0.0020** −0.0194**
(0.0009) (0.0086)
Obse a ions 3821155 3821155 2698697 2698697 1725813 1725813
Model 1 and 2: PPML es ima ions o bila e al expo s in cu en USD, including ze o expo s. Model 3 and
4: Linea p obabili y es ima ions o a bina y a iable indica ing he exis ence o posi i e expo s (ex ensi e
ma gin). Model 5 and 6: PPML es ima ions o bila e al expo s in cu en USD, excluding ze o expo s
(in ensi e ma gin). All eg essions a e wi h ixed e ec s a he coun y-pai -yea -le el, expo e -calenda
mon h-le el, and impo e -calenda mon h-le el. *
p<.1
**
p<.05
***
p<.01
(1) (2) (3)
Clus e ing 1 Clus e ing 2 Clus e ing 3
D 30
i m
−0.0336*** −0.0337*** −0.0337***
(0.0100) (0.0129) (0.0100)
D 30
j m
−0.0124** −0.0125** −0.0125**
(0.0056) (0.0062) (0.0055)
Obse a ions 3821155 3776991 3776991
Column 1: Clus e ed by impo e , expo e , and ime s ep ( eplica ion
o baseline speci ica ion). Column 2: Clus e ed by spa ial dis ance
be ween impo e and expo e . Column 3: Clus e ed by all ou
a iables. *
p<.1
**
p<.05
***
p<.01
. The numbe o
obse a ions is sligh ly lowe in column 2 and 3 due o missing da a
o dis ance
Table 3 Con empo aneous e -
ec s o empe a u e wi h di e -
en ly clus e ed e o s uc u es
1 3
996
In e na ional T ade and he T ansmission o Tempe a u e Shocks
In Fig. 15, we depic he cumula i e e ec s o
D 30
i m
du ing wel e mon hs a e he em-
pe a u e shock. The es ima ion con i ms he nega i e e ec on expo s in he i s mon hs
a e he empe a u e shock, and shows ha hese expo alues a e no eco e ed in subse-
quen pe iods. The e is no signi ican impac on impo s.
Rega ding ela i ely wa m empe a u e e en s (
Dp99
i m
), we depic he lagged and cumu-
la i e e ec s on expo s in Figs. 16 and 17, espec i ely. As o he case o absolu e ho
empe a u es, he e ec is concen a ed on expo e s, and sho -li ed. Howe e , in con as
o he
D 30
i m
-speci ica ion, he e ec is no agg a a ing in he i s ew mon hs bu he e is a
sligh endency owa ds ca ching-up, such ha he cumula i e e ec wi hin one yea a e
he e en is s a is ically equal o ze o.
In Fig. 18, we eplica e he in e ac ion analysis o he po en ial labou in ensi y channel
o empe a u e e en s in he op pe cen ile (
Dp99
i m
). In his speci ica ion, he e ec is no sig-
Fig. 14 Lagged impac on
impo s o an a e age mon hly
empe a u e o a leas 30
◦C
.
Es ima ed e ec s on impo s o a
ho mon h on expo s, including
95-pe cen con idence in e als.
The e ec s a e ela i e o a
mon h wi h a empe a u e below
30
◦C
. The model is es ima ed
wi h PPML and include em-
pe a u e a he expo e loca ion,
coun y-pai -yea , expo e -cal-
enda mon h, and impo e -calen-
da mon h ixed e ec s. S anda d
e o s a e mul i-way-clus e ed
a expo e , impo e and ime
s ep-le el
Fig. 13 Con empo aneous e ec s o empe a u e on expo s in 5
◦C
bins in ’cold’ and ’wa m’ coun ies.
Es ima ed coe icien s o empe a u e bins in ’cold’ expo ing coun ies (le panel) and ’wa m’ expo ing
coun ies ( igh panel) and hei 95 pe cen con idence in e als, using 15-20
◦C
as he baseline ca ego y.
Based on a PPML es ima ion wi h ixed e ec s a he coun y-pai -yea -le el, expo e -calenda mon h-
le el, and impo e -calenda mon h-le el (N=1,956,682 o ’cold’ subsample and 1,864,473 o ’wa m’
subsample
1 3
997
O. Schenke , D. Osbe ghaus
ni ican which may be a ionalized by he insigh s o p io s udies ha absolu e empe a u e
le els a e mo e impo an o e ec s on labou capaci y.
Figu es 19 and 20 summa ize he es ima es o a ious in e ac ion e ec s. Mos o he
analyzed po en ial he e ogenei y a iables show no signi ican impac on he es ima ed
e ec s o
D 30
i m
o
Dp99
i m
.
Figu e 21 analyses he e ogeneous e ec s o e ime. We in e ac
D 30
i m
wi h ime bins o
i e yea leng h in o de o check i he e ec o empe a u e shocks changes o e ime. We
ind s a is ically signi ican e ec s o high empe a u e shocks on expo s in ea ly yea s o
he sample bu ha e o in e p e hem wi h a g ain o sal as he sample is small and biased
owa ds a small numbe o de eloped coun ies. In la e pe iods, when he sample becomes
Fig. 16 Lagged e ec s o an a e age mon hly empe a u e in he op pe cen ile. Es ima ed e ec s o a ho
mon h on expo s (le panel) and impo s ( igh panel), including 95-pe cen con idence in e als. The
e ec s a e ela i e o a mon h wi h non-ex eme empe a u e. The model is es ima ed using PPML and
include coun y-pai -yea , expo e -calenda mon h, and impo e -calenda mon h ixed e ec s. S anda d
e o s a e mul i-way-clus e ed a expo e , impo e and ime s ep-le el
Fig. 15 Cumula i e e ec s o an a e age mon hly empe a u e o a leas 30
◦C
. Es ima ed cumula i e
e ec s o a ho mon h on expo s (le panel) and impo s ( igh panel), including 95-pe cen con idence
in e als. The e ec s a e ela i e o a mon h wi h a empe a u e below 30
◦C
. The model is es ima ed wi h
PPML and include coun y-pai -yea , expo e -calenda mon h, and impo e -calenda mon h ixed e ec s
1 3
998
In e na ional T ade and he T ansmission o Tempe a u e Shocks
mo e comp ehensi e, we ind mos ly nega i e e ec s o empe a u e shocks on expo s bu
no all o hem a e s a is ically signi ican .
E ec s o P ecipi a ion and S o ms
As o he case o empe a u e, ou main sou ce o his o ical p ecipi a ion da a is CCKP
(Wo ld Bank 2022a). As o he case o empe a u e, he alues a e mon hly means agg e-
ga ed a he coun y le el. In Table 5, we p esen he desc ip i e s a is ics o he addi ional
wea he a iables used in his and he subsequen sec ion.
Fo e ec s o s o ms, we use da a on he maximum wind speed a he coun y-mon h
le el om he i o GAME da a se (Felbe may and G öschl 2014), which is based on wo
p ima y da a sou ces: Fi s , i uses he In e na ional Bes T ack A chi e o Clima e S ew-
Fig. 18 Es ima ed e ec o an
a e age mon hly empe a u e in
he op pe cen ile on expo s o
di e en le els o labou in en-
si y. Es ima ed con empo aneous
e ec s o
Dp99
i m
on expo s o
gi en le els o labou in ensi y,
including 95-pe cen con idence
in e als. The e ec s a e ela i e
o a mon h wi h non-ex eme
empe a u e. The model is es i-
ma ed wi h PPML and includes
coun y-pai -yea , expo e - and
impo e -calenda mon h ixed
e ec s. Labelled alues a he
x-axis a e he 5 h, 50 h, and 95 h
pe cen ile o
labou in ij
Fig. 17 Cumula i e e ec s o an a e age mon hly empe a u e in he op pe cen ile. Es ima ed cumula i e
e ec s o a ho mon h on expo s (le panel) and impo s ( igh panel), including 95-pe cen con idence
in e als. The e ec s a e ela i e o a mon h wi h non-ex eme empe a u e. The model is es ima ed wi h
PPML and include coun y-pai -yea , expo e -calenda mon h, and impo e -calenda mon h ixed e -
ec s. S anda d e o s a e mul i-way-clus e ed a expo e , impo e and ime s ep-le el
1 3
999
O. Schenke , D. Osbe ghaus
a dship (IBT ACS) which is p o ided by he Na ional Clima ic Da a Cen e o he Na ional
Oceanic and A mosphe ic Adminis a ion (NOAA) and con ains da a o indi idual hu i-
cane e en s. Second, in o de o cap u e o nadoes, summe and win e s o ms no included
in IBT ACS, he hu icane da a is ma ched o daily da a o he Global Su ace Summa y
o Day (GSOD) da a ( e sion 7) on maximum wind speed and wind gus om o e 9,000
wea he s a ions wo ldwide. Wind da a om i o GAME is a ailable a he coun y-mon h
le el o he pe iod 1979-2010.
In Table 6 we s udy he e ec o di e en models es ima ing he p ecipi a ion impac s on
expo s. In mo e de ail, we examine he linea and quad a ic e ec s o absolu e p ecipi a ion
(Models 1 and 2, espec i ely), and e ec s o ex emely d y and we mon hs (Models 3 and
4). We do no ind signi ican e ec s o p ecipi a ion in any o he speci ica ions. In Model 5,
we add ex emely d y mon hs as a con ol a iable o ou baseline es ima es. The es ima ed
e ec o high empe a u e on expo s emains almos iden ical bu episodes o low p ecipi-
a ion ha e no signi ican e ec . Model 6 shows ha also he in e ac ion o ex emely d y
mon hs wi h e y ho mon hs does no ha e a signi ican e ec on expo s.
Table 7 shows he esul s o simila es ima ions o wind speed impac s on expo s. The e
a e nei he e ec s in he linea and quad a ic speci ica ions, no do mon hs wi h coun y-
speci ic high wind speeds show an e ec on expo s.
Fig. 19 Es ima ed e ec s o an a e age mon hly empe a u e o a leas 30
◦C
o a ious he e ogenei y
a iables. Es ima ed con empo aneous e ec s o
D 30
i m
on expo s o gi en le els o a ious he e ogene-
i y a iables, including 95-pe cen con idence in e als. The e ec s a e ela i e o a mon h wi h a em-
pe a u e below 30
◦C
. The model is es ima ed wi h PPML and includes coun y-pai -yea , expo e - and
impo e -calenda mon h ixed e ec s. S anda d e o s a e mul i-way-clus e ed a expo e , impo e and
ime s ep-le el. Labelled alues a he x-axes a e he
5 h
,
50 h
, and
95 h
pe cen iles o he he e ogenei y
a iables. The epo ed p- alues e e o he signi icance o he in e ac ion e m (n.s.: p
≥
0.1)
1 3
1000
In e na ional T ade and he T ansmission o Tempe a u e Shocks
Howe e , using coun y-speci ic dis ibu ions may no be he app op ia e s a egy o
iden i ying non-linea e ec s o s o ms, since he occu ence o ha m ul wind speeds is no
equally dis ibu ed on coun ies (as he a iable combiP99), bu is mainly an issue in opi-
cal cyclone-exposed loca ions. The e o e, we use a simila s a egy as o empe a u e bins
(Fig. 12), and es ima e e ec s o absolu e wind speed bins in Fig. 22, using he bin wi h he
la ges numbe o obse a ions (40-45 kno s) as he baseline ca ego y. The esul s sugges
ha o wind speeds abo e 140 kno s, expo s dec ease subs an ially in he mon h o he
wea he e en , while impo s emain la gely una ec ed. These po en ial sho - e m and non-
linea e ec s o in ense s o ms on expo s a e beyond he scope o his analysis, and may be
a p omising a enue o u he esea ch.
Addi ional In o ma ion on Coun e ac ual Simula ions
Table 8 p esen s he lis o coun ies included in he simula ion analysis.
Fig. 20 Es ima ed e ec s o an a e age mon hly empe a u e in he op pe cen ile o a ious he e oge-
nei y a iables. Es ima ed con empo aneous e ec s o
Dp99
i m
on expo s o gi en le els o a ious he -
e ogenei y a iables, including 95-pe cen con idence in e als. The e ec s a e ela i e o a mon h wi h
non-ex eme empe a u e. The model is es ima ed wi h PPML and includes coun y-pai -yea , expo e -
and impo e -calenda mon h ixed e ec s. S anda d e o s a e h ee-way-clus e ed a expo e , impo e
and ime s ep-le el. Labelled alues a he x-axes a e he
5 h
,
50 h
, and
95 h
pe cen iles o he he e o-
genei y a iables. The epo ed p- alues e e o he signi icance o he in e ac ion e m (n.s.: p
≥
0.1)
1 3
1001
O. Schenke , D. Osbe ghaus
Table 5 Desc ip i e s a is ics o addi ional wea he a iables
Va iable Desc ip ion Mean S d. de . Minimum Maximum Obs.
p eci m
Mean p ecipi a ion in mm, a e age o
coun y a ea
94.67 93.57 0.00 1,063.69 49,597
p ecP 1i m
Mean p ecipi a ion in lowes coun y-
speci ic pe cen ile
0.015 0.121 0.00 1.00 49,597
p ecP 99i m
Mean p ecipi a ion in coun y-speci ic
op pe cen ile
0.010 0.099 0.00 1.00 49,597
windi m
Maximum wind speed in kno s 48.58 19.12 0.00 165.00 26,559
windP 99i m
Maximum wind speed in coun y-
speci ic op pe cen ile
0.007 0.082 0.00 1.00 26,559
Desc ip i e s a is ics a e calcula ed a he coun y-mon h-le el. Fo easons o b e i y, coun y-speci ic
s a is ics a e only epo ed o expo e s
Fig. 21 Es ima ed e ec s o a e age mon hly empe a u e o a leas 30
◦C
o e ime. Es ima ed e ec s
on impo s o a ho mon h on expo s o gi en ime pe iods, including 95-pe cen con idence in e als.
Based on an in e ac ion o he empe a u e a iable wi h 5-yea -dummy a iables. The e ec s a e ela i e
o a mon h wi h a empe a u e below 30
◦C
. The model is es ima ed wi h PPML and include empe a u e
a he impo e loca ion, coun y-pai -yea , expo e -calenda mon h, and impo e -calenda mon h ixed
e ec s. S anda d e o s a e h ee-way-clus e ed a expo e , impo e and ime s ep-le el
1 3
1002
In e na ional T ade and he T ansmission o Tempe a u e Shocks
Table 6 Con empo aneous p ecipi a ion e ec s on expo s
(1) (2) (3) (4) (5) (6)
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
D 30
i m
−0.0335*** −0.0331***
(0.0100) (0.0102)
D 30
i m ×p ecP 1i m
−0.0719
(0.0678)
p eci m
−0.0000 −0.0000
(0.0000) (0.0001)
p ec2
i m
−0.0000
(0.0000)
p ecP 1i m
0.0026 0.0026 0.0026
(0.0190) (0.0190) (0.0190)
p ecP 99i m
0.0019
(0.0080)
Obse a ions 3821155 3821155 3821155 3821155 3821155 3821155
All models a e es ima ed using PPML and include coun y-pai -yea , expo e -calenda mon h, and
impo e -calenda mon h ixed e ec s. In all es ima es we con ol o wea he e ec s in he impo ing
coun y
(1) (2) (3)
Model 1 Model 2 Model 3
windi
−0.0001 0.0003
(0.0001) (0.0005)
wind2
i
-0.0000
(0.0000)
windP 99i
0.0003
(0.0085)
Obse a ions 2155697 2155697 2155697
Table 7 Con empo aneous wind
speed e ec s on expo s
All models a e es ima ed using
PPML and include coun y-pai -
yea , expo e -calenda mon h,
and impo e -calenda mon h
ixed e ec s. In all es ima es we
con ol o wea he e ec s in he
impo ing coun y
1 3
1003
O. Schenke , D. Osbe ghaus
A gen ina Aus alia Aus ia Belgium Bangladesh B azil
Canada Swi ze -
land
China Colombia Czechia Ge -
many
Denma k Alge ia Spain F ance G ea
B i ain
Hun-
ga y
Indonesia India I eland I aq I aly Japan
Sou h
Ko ea
Mexico Malaysia Ne he -
lands
No way New
Zea-
land
Oman Philip-
pines
Poland Russia Saudi
A abia
Sin-
ga-
po e
Sweden Thailand Tu key Uni ed
S a es
Venezuela Vie
Nam
Sou h
A ica
Table 8 Lis o coun ies in used
in simula ions
Lis o coun ies used in
coun e ac ual simula ions.
T ade be ween hese coun ies
co e s 90 pe cen o he ade
olume o e he sample pe iod
Fig. 22 Con empo aneous e ec s o maximum wind speed on expo s in 5 kno s bins. Es ima ed coe -
icien s o wind speed bins in he expo ing coun y (le panel) and he impo ing coun y ( igh panel)
and hei 95 pe cen con idence in e als, using 40-45 kno s as he baseline ca ego y. Based on a PPML
eg ession wi h ixed e ec s a he coun y-pai -yea -le el, expo e -calenda mon h-le el, and impo -
e -calenda mon h-le el (N=2,153,305). S anda d e o s a e clus e ed a expo e , impo e and ime
s ep-le el
1 3
1004
