Soybean yield prediction in argentina using climate data

Basco, Emiliano; Elías, Diego; Agui e, Maximiliano Gómez; Pas o e, Luciana
Wo king Pape
Soybean yield p edic ion in a gen ina using clima e da a
Economic Resea ch Wo king Pape s, No. 117
P o ided in Coope a ion wi h:
Economic Resea ch Depa men (ie), Cen al Bank o A gen ina
Sugges ed Ci a ion: Basco, Emiliano; Elías, Diego; Agui e, Maximiliano Gómez; Pas o e, Luciana
(2025) : Soybean yield p edic ion in a gen ina using clima e da a, Economic Resea ch Wo king
Pape s, No. 117, Banco Cen al de la República A gen ina (BCRA), In es igaciones Económicas (ie),
Buenos Ai es
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/322395
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-sa/4.0/
Wo king Pape N 117 | 2025
Soybean Yield P edic ion in A gen ina
Using Clima e Da a
Economic Resea ch
Wo king Pape s | 2025 | N 117
Soybean Yield P edic ion in A gen ina Using Clima e Da a
Emiliano Basco
Banco Cen al de la República A gen ina
Diego Elías
Banco Cen al de la República A gen ina
Maximiliano Gómez Agui e
Banco Cen al de la República A gen ina
Luciana Pas o e
Banco Cen al de la República A gen ina
May 2025
2 | BCRA | Wo king Pape s 2017
Wo king Pape s, N 117
Soybean Yield P edic ion in A gen ina Using Clima e Da a
Emiliano Basco
Banco Cen al de la República A gen ina
Diego Elías
Banco Cen al de la República A gen ina
Maximiliano Gómez Agui e
Banco Cen al de la República A gen ina
Luciana Pas o e
Banco Cen al de la República A gen ina
May 2025
ISSN 1850-3977
Elec onic Edi ion
Reconquis a 266, C1003ABF
Ciudad Au ónoma de Buenos Ai es, A gen ina
Phone | 54 11 4348-3582
Email | in es ig@bc a.gob.a
Websi e | www.bc a.gob.a
The opinions exp essed in his pape a e he sole esponsibili y o i s au ho s
and do no necessa ily ep esen he posi ion o he Cen al Bank o A gen ina.
The Wo king Pape s se ies is comp ised o p elimina y ma e ial in ended o s imula e
academic deba e and ecei e commen s.
This pape may no be e e enced wi hou au ho iza ion om he au ho s.
1
Soybean yield p edic ion in A gen ina using clima e da a: A spa ial
panel and ime se ies app oach
*
Emiliano Basco†, Diego Elías†, Maximiliano Gómez Agui e†, Luciana Pas o e
†
May 2025
Abs ac
Ag icul u e, and especially soybean p oduc ion, has a c i ical ole in A gen ina’s economy, as a
majo con ibu o o GDP and expo e enue. This pape s udies he impac o clima e a iabili y
on soybean yields in A gen ina using a no el depa men -le el da ase spanning 1980–2023. We
es ima e a ixed e ec s spa ial e o model (SEM) o quan i y he e ec s o wea he shocks—
measu ed by ex eme hea , p ecipi a ion, and ENSO phases—while con olling o economic and
echnological ac o s such as seed echnology and ela i e p ices. Ou esul s show ha ex eme
hea signi ican ly educes yields, while mode a e ain all boos s hem up o a nonlinea h eshold.
El Niño phases inc ease yields, whe eas La Niña e en s a e de imen al. Technological adop ion
and a o able p ice signals also enhance p oduc i i y. These indings highligh he impo ance o
accoun ing o bo h clima e dynamics and spa ial dis ibu ions when es ima ing ag icul u al
ou comes. Time se ies models p o ide a s ong empi ical basis o o ecas ing soybean yields and
in o ming policy decisions unde inc easing clima e unce ain y.
These models can be employed as e ec i e ools o an icipa ing yield ou comes unde di e en
clima e scena ios and u ilized in s ess es exe cises. This wo k p o ides aluable insigh s o
policymaking decisions, con ibu ing o p epa e o po en ial economic impac s s emming om
clima e isks on A gen ina’s ag icul u al sec o .
JEL Code: Q10, Q12, C13, C32, C33
Key wo ds: Soybean Yields, A gen ina, Fo ecas ing, Model Selec ion
*
We a e g a e ul o Tamma Ca le on and To s en Ehle s o hei aluable commen s and sugges ions, which
ha e g ea ly imp o ed his wo k. We also hank Pablo Edua do Abba e, Resea ch Di ec o a he Ins i u o
Nacional de Tecnología Ag opecua ia, and C is ian Russo, Chie o GEA a he Bolsa de Come cio de
Rosa io, o hei insigh ul con ibu ions and sec o al expe ise. We app ecia e he eedback ecei ed om
pa icipan s a he 14 h BIS CCA Resea ch Con e ence, Bogo a, 3 Decembe and a he Resea ch
Depa men ’s in e nal semina a he Cen al Bank o A gen ina (BCRA). The iews exp essed in his pape
a e hose o he au ho s and do no necessa ily e lec he iews o he BCRA o i s au ho i ies.
†
Economic Resea ch, BCRA. E-mail: e[email p o ec ed]ob.a , delias@bc a.gob.a ,
mau icio.gomezagu[email p o ec ed] and luciana.[email p o ec ed]b.a

2
1. In oduc ion
Ag icul u e is a undamen al sec o wi hin he A gen ine economy. In 2021, p ima y ag icul u e
and ag i ood alue chains con ibu ed an es ima ed 16% o he na ion’s GDP (Wo ld Bank, 2024).
Ag icul u al expo s accoun o a ound 60% o he coun y’s o al expo s, emphasizing he
sec o 's impo ance in gene a ing o eign cu ency and suppo ing economic s abili y.
The soybean complex is he main ag icul u al p oduc ion chain in A gen ina, cha ac e ized by a
s ong expo p o ile. I in ol es he indus ializa ion o p ima y g ain p oduc ion and ep esen s
he coun y’s la ges expo chain, accoun ing o nea ly 28% o o al expo s in 2022, su passing
bo h he ce eal and au omo i e sec o s. Soybeans a e he second mos impo an ag icul u al c op,
ollowing co n, a e aging a p oduc ion o 46 million ons annually o e he las i e campaigns
(2018-2022). Ano he impo an aspec o he soybean sec o is i s ola ili y, since i expe iences
high p oduc ion a iabili y om yea o yea compa ed o o he c ops.
A gen ina plays a c i ical ole in global ag icul u al ma ke s, being he hi d la ges p oduce and
expo e o soybeans wo ldwide and he op expo e o soybean oil and soybean meal. Oilseed
expo s a e a signi ican componen o A gen ina's ade, ep esen ing o e 31% o o al expo s
in ecen yea s (e.g., 2021 and 2022). This sec o 's signi ican con ibu ion e lec s i s impo ance
o bo h domes ic economic s abili y and global ood ma ke s.
Clima e e en s such as d ough s in he Pampas Region ha e se e e epe cussions, dec easing
yields o essen ial c ops such as soybeans, co n, and whea , and consequen ly a ec ing hei en i e
supply chains. The impac s o hese declines also in luence mul iple sec o s wi hin he economy.
Reduc ions in ag icul u al p oduc ion caused by d ough s ha e a spillo e e ec , wi h
implica ions o ups eam and downs eam indus ies.
The ag icul u al sec o 's ulne abili y o d ough s also has b oade economic consequences. A
down u n in c op yields educes iscal e enue, lowe s in e na ional ese es, and pu s p essu e
on he exchange a e, impac ing A gen ina's economic esilience.
Fo example, he ecen d ough in 2022-2023 had a signi ican impac on he yield and p oduc ion
o A gen ina's main c ops: whea , co n, and soybeans. The e ec s we e el di ec ly in ag icul u al
p oduc ion and, h ough he ag ibusiness chain, ex ended o o he sec o s. The subs an ial decline
in expo s also c ea ed a shock o go e nmen e enue and he balance o paymen s due o i s
in luence on he exchange a e and in e na ional ese es.
Two majo clima ic phenomena ha impac mois u e condi ions and can cause d ough s in he
p oduc i e egions a e La Niña and El Niño. La Niña esul s om cooling along he equa o ial
Paci ic Ocean, leading o lowe - han-a e age ain all and causing d ough s o a ying se e i y
ac oss Sou h Ame ican egions. In con as , El Niño is a clima e e en ma ked by he wa ming o
3
he equa o ial Paci ic Ocean, which ends o b ing abo e-a e age p ecipi a ion in A gen ina. In
his pape we conside he in luence o El Niño and La Niña indica o s, along wi h he es o he
clima e in o ma ion.
In his pape , we examine he impac o clima e a iabili y on soybean yields in A gen ina by
combining a unique depa men le el da ase co e ing 1980–2023 wi h daily wea he
obse a ions— empe a u es and p ecipi a ion—aligned o he c op’s phenological calenda . We
augmen hese clima ic con ols wi h economic and echnological a iables, including
in e na ional soybean and e ilize p ices, ansgenic seed adop ion a es, and land use change
indica o s. We employ spa ial panel econome ic echniques o cap u e bo h local ixed e ec s
and spa ially co ela ed unobse ables, imp o ing ou unde s anding o how clima e and egional
in e dependencies shape yield ou comes.
Ou empi ical s a egy un olds in wo s ages. Fi s , we es ima e a ixed e ec s spa ial e o model
(SEM) panel o quan i y he long un ela ionship be ween wea he shocks—measu ed by
g owing deg ee days, cumula i e hou s abo e 30 °C, p ecipi a ion, and El Niño and La Niña
indica o s—and soybean yields, while con olling o economic and echnological co a ia es. We
ind ha ex eme hea consis en ly dampens yields, mode a e p ecipi a ion inc eases hem up o a
nonlinea h eshold, and lagged El Niño anomalies (wa m phases) enhance yields while La Niña’s
d y condi ions a e de imen al. Technological p og ess, as p oxied by ansgenic seed use and
a o able p ice a ios, u he ele a es yields, and he highly signi ican spa ial e o e m con i ms
he impo ance o unobse ed, egionally co ela ed ac o s.
In he second s age, we ansla e hese insigh s in o o ecas ing exe cises by cons uc ing di e en
ime se ies models based on he a ailable wea he in o ma ion. We e alua e o ecas pe o mance
ia olling window exe cises and he Giacomini–Whi e es , benchma king agains he USDA’s
mon hly yield o ecas s wi h same da a a ailabili y. We ind ea ly in o ma ion models ha
sys ema ically ou pe o m USDA o ecas s, while hose models wi h in o ma ion close o he end
o he g owing season, ou o ecas s a e as eliable as USDA ones.
These indings ca y impo an policy implica ions. By demons a ing ha clima e d i en models
can deli e accu a e yield p ojec ions mon hs be o e ha es , we p o ide a ool o an icipa ing
o eign exchange p essu es and in o ming mone a y and iscal policy esponses. Ou amewo k
also lays he g oundwo k o cons uc ing clima e s ess scena ios—simula ing d ough s, hea
wa es, o ENSO e en s— o e alua e po en ial economic ulne abili ies in A gen ina’s
ag icul u al sec o . Finally, because he me hodology elies on spa ially disagg ega ed yield and
wea he da a, i can be eadily adap ed o o he c ops o egions, unde sco ing he b oade alue
o high esolu ion clima e da a in mode n ag icul u al isk managemen and policy planning.
4
This pape con ibu es o he unde s anding in he ela ionship be ween clima e a iabili y and
soybean yields in A gen ina. We in oduce a la ge da a se ha le e ages geo e e enced,
delega ion‑le el eco ds o unco e bo h local and egional clima e–yield ela ionships and o
accoun o spa ial in e dependencies among p oduc ion zones. This new da a se allows us o ge
a spa ial e o panel model ha e eals how ex eme hea , nonlinea p ecipi a ion, and ENSO
phases a ec soybean yields, while cap u ing echnological gains. Finally, we p o ide s a is ically
obus and accu a e o ecas s benchma ked agains USDA p ojec ions.
The pape is s uc u ed as ollows: i s , we p esen a li e a u e e iew ela ed o he in luence o
clima e on ag icul u e. In Sec ion 3, we p esen some concep s ha a e c ucial o unde s anding
soybean p oduc ion in A gen ina, along wi h ou da a sou ces, ea men o he da a, and he
calcula ions in ol ed in cons uc ing he da ase . Sec ion 4 con ains an o e iew o he
econome ic me hodology employed, and we p esen he esul s o ou analyses. The las pa
concludes and discusses he implica ions o ou indings and ecommenda ions o u u e
esea ch.
2. Rela ed Li e a u e
A numbe o s udies examine he e ec s o clima e a ia ions on c op yields. Fo example,
Schlenke and Robe s (2009) s udied he impac o a ia ions o empe a u e on co n, soybeans,
and co on yields in he Uni ed S a es, using a panel da a app oach wi h nonlinea empe a u e
e ec s. They ound ha ex eme hea has signi ican nega i e impac s on yields, and de ined hea
h esholds o each c op, abo e which p oduc i i y is educed.
Deschênes and G eens one (2007) explo ed he e ec s o annual a ia ions in empe a u e and
p ecipi a ions o es ima e hei in luence on ag icul u al p o i s, using coun y le el panel da a in
he Uni ed S a es. They es ima ed long- un e ec s and ound he e ogeneous and o e all small
ou comes ac oss he US.
Chen e al. (2013) conduc ed an analysis o co n and soybean yields in China, using spa ial panel
econome ic echniques. They modeled he beha io o p o i -maximizing a me s, and es ima ed
changes on yields using clima e a iables, socioeconomic a iables and a iables ep esen ing
a me s’ adap a ion beha io s. They also ound nonlinea and asymme ic ela ionships be ween
yields and clima e a iables.
Miao, Kanna and Huang (2016) examined he e ec s o clima e a iables and c op p ices on co n
and soybean yields and ac eage in he Uni ed S a es, using a la ge spa ial panel da ase . They
ound ha p ices ha e s a is ically signi ican e ec s on co n yields, and ha he po en ial e ec
5
o clima e change on p oduc ion is nega i e bu highly he e ogeneous, depending on clima e
scena ios and models.
Fo A gen ina, Co nejo and Ahumada (2021) analyzed he long- e m ela ionships be ween
clima ic, echnological, and economic ac o s and c op yields. They ound ha soybean yields
adjus o echnological imp o emen s and ha high empe a u es ha e a nega i e e ec . They also
ound e idence o CO2 e iliza ion. Co nejo (2021), using in o ma ion on clima e a iables
published in ad ance and wi h a equency highe han soybean yields, ound ha he e a e
o ecas ing gains when conside ing he maximum empe a u e du ing he plan 's g ow h cycle.
Addi ionally, when using p ecipi a ion da a, he model based on annual da a ou pe o ms he
o he s.
These s udies p esen di e en me hodologies o assess he ela ionship be ween clima e and
ag icul u al p oduc i i y. Al hough echnological ad ancemen s ha e imp o ed o e all yields,
clima e e en s, pa icula ly d ough s, con inue o ha e a nega i e in luence on ag icul u al
p oduc ion.
3. Desc ip i e Analysis and Da a
3.1. Fa me s p oduc ion unc ion
Ou s udy employs a p oduc ion unc ion as a amewo k o es ima e he e ec s o clima e
condi ions on yields. This me hod can be used o isola e he impac o wea he on speci ic c ops,
while i is also necessa y o cap u e he beha io o a me s and ake in o accoun o he ange o
compensa o y esponses hey pe o m in eac ion o clima e a iabili y.
Soybean p oduc ion, in pa icula , is de e mined by wo key ac o s: he a ea plan ed and he
yields achie ed. The decision ega ding he a ea o be cul i a ed is made in ad ance, based on
expec a ions o p o i abili y and isk. P oduce s ake in o accoun c op p ices, mac oeconomic
condi ions, and po en ial unding cons ain s.
On he o he hand, yield a iabili y depends on mul iple ac o s, including he applica ion o
inpu s ( e ilize s, pes icides), he adop ion o ad anced ag icul u al echnologies, and he clima ic
condi ions du ing he g owing season. P ice a iables can also in luence a me s' decisions on
inpu use and p oduc ion in ensi y.
Ou s udy ocuses on yield a iabili y as i is he a iable which is mo e a ec ed by clima e
shocks. While changing wea he pa e ns, empe a u e and p ecipi a ion play a undamen al ole
in de e mining p oduc i i y, a me s decide he plan ed a ea be o e knowing he ull ex en o
wea he condi ions.
12
Figu e 6: a e age p ecipi a ion de ia ion om his o ical mean s soybean yields
3.5. La Niña / El Niño
Unde s anding he e ec s o El Niño and La Niña is c ucial when analyzing A gen ina’s clima e
condi ions and hei impac on soybean p oduc ion. These phenomena signi ican ly in luence
p ecipi a ion pa e ns in a ious egions wo ldwide, which, in u n, a ec c op yields. Al hough
hese shi s can a y sligh ly be ween El Niño e en s, he mos in ense changes ypically emain
consis en ac oss speci ic egions and seasons.
The El Niño–Sou he n Oscilla ion (ENSO) is moni o ed h ough sea su ace empe a u e (SST)
anomalies in he equa o ial Paci ic Ocean. The p ima y measu e used is he Oceanic Niño Index
(ONI), which a e ages SST anomalies in he Niño 3.4 egion (170°W–120°W).
El Niño condi ions occu when SST anomalies exceed 0.5°C o i e consecu i e mon hs,
indica ing a wa ming phase. La Niña condi ions a e cha ac e ized by SST anomalies below –
0.5°C, signaling a cooling phase. These clima e anomalies dis up global ain all pa e ns,
a ec ing a ious ag icul u al egions wo ldwide.
Unlike o he soybean-p oducing egions, A gen ina expe iences educed ain all du ing La Niña
yea s, leading o d ough condi ions ha signi ican ly impac c op yields. His o ical obse a ions
con i m a nega i e co ela ion be ween La Niña e en s and soybean p oduc ion in A gen ina, as
shown in Figu e 7. This unde sco es he clima ic ulne abili y o A gen ine ag icul u e o ENSO
luc ua ions.

13
Figu e 7: Rela ionship be ween changes in soybean yield and El Niño/La Niña episodes
We also explo e whe he ea ly signals o La Niña e en s— ypically a ailable be o e he sowing
season—ha e any measu able impac on soybean plan ing decisions, pa icula ly in e ms o o al
plan ed a ea. The e idence sugges s ha , in he A gen ine con ex , an icipa ed La Niña condi ions
do no signi ican ly a ec he a ea sown wi h soybeans. As a esul , he p ima y channel h ough
which La Niña a ec s soybean p oduc ion in A gen ina is h ough i s ad e se e ec s on yields.
This ein o ces he ele ance o ocusing on yield esponses— a he han ac eage—as he key
ma gin o adjus men when modeling clima e impac s and building p edic i e models.
1
A key economic mechanism ha could mi iga e he impac o lowe yields is p ice adjus men ,
which may dampen he b oade mac oeconomic e ec s. I La Niña–induced d ough s we e o
signi ican ly educe global soybean supply, in e na ional p ices could ise, pa ially o se ing
a me s’ income losses. This, in u n, could help s abilize expo e enues, consump ion, and
in es men . Howe e , his adjus men mechanism is less s aigh o wa d in he case o A gen ina.
The main eason is ha La Niña does no a ec all majo soybean-p oducing egions equally.
While A gen ina ends o su e om d ough s and yield losses, B azil and pa s o he Uni ed
1
In mee ings wi h specialis s om he ag icul u al sec o , hey we e speci ically asked whe he he isk o
El Niño/La Niña e en s was aken in o accoun in plan ing plans (such as he choice o c op mix, he
mo e o less in ensi e use o inpu s, e c.), as his could in oduce some o m o endogenei y. Thei
esponse was nega i e.
14
S a es o en expe ience a o able g owing condi ions, leading o inc eased ou pu in hose egions
(see Figu e 8). As a esul , global supply may no all signi ican ly, and p ice inc eases a e limi ed.
This is consis en wi h he e idence ha he ENSO index does no sys ema ically an icipa e
in e na ional soybean p ices.
Figu e 8: El Niño and La Niña Teleconnec ions Map (Lenssen, Godda d & Mason, 2020)
Figu e 8.a. El Niño
Figu e 8.b. La Niña
15
3.6. Technology and o he a iables
In addi ion o clima e e ec s, i is necessa y o conside he con ex o soybean cul i a ion in
A gen ina. This sec ion desc ibes some ac o s ha allowed i s expansion in olume and e i o y,
which ha e led o he inco po a ion o speci ic a iables in he models.
Comme cial soybean p oduc ion began in A gen ina in he 1970s. Acco ding o Cadenazzi (2009),
soybean was ini ially in oduced as a second c op a e whea , as i s ease o managemen and
adap abili y ep esen ed a mo e p o i able o a ion, eplacing he adi ional ag icul u e/li es ock
o a ion.
In he 1980s, soil e osion e ec s we e obse ed in he Pampas egion due o in ensi e ag icul u al
ac i i ies. Howe e , by he mid-1990s, soybean p oduc ion expanded wi h echnological
ad ancemen s ha also allowed o he use o lowe -quali y land bo h in he co e zone and ou side
o i , expanding he ag icul u al on ie and inc easing yields.
In e na ionally, a sus ained inc ease in soybean demand was obse ed, d i en by popula ion
g ow h and he need o ood, as well as he g owing demand o o he uses. Be e p ices in he
global ma ke also s imula ed supply. As men ioned ea lie , be o e plan ing begins, p oduce s will
de ine he a ea o be cul i a ed, aking in o accoun he ma ke con ex and expec a ions. Fo his
eason, he in e na ional p ice o soybeans and i s ecen ola ili y we e inco po a ed as a iables.
The in oduc ion o ansgenic soybean in 1996, speci ically RR soybean, ma ked a u ning poin
in ag icul u al p oduc ion in A gen ina. This a ie y o soybeans is esis an o he b oad-spec um
he bicide glyphosa e, which elimina es all weeds. This ansla es in o cos educ ions as i is easy
o apply.
The p ac ice o no- ill a ming also sp ead du ing his pe iod. I in ol es elimina ing plowing
while he esidue om he p e ious ha es conse es mois u e and se es as e ilize . Sowing is
ca ied ou wi h specially designed machines, wi h minimal soil dis u bance. This me hod educes
ieldwo k ime and soil e osion.
No- ill a ming, along wi h he use o RR soybean and glyphosa e, complemen each o he since
he o me leads o an inc ease in weed quan i y. These echniques, along wi h he de elopmen
o new ag ochemicals such as he bicides, pes icides, and e ilize s, as well as he de elopmen
o speci ic machine y, con ibu ed o he expansion o soybean p oduc ion, as hey inc ease yields
and educe p oduc ion cos s. This makes i easible o p oduce in a eas ha we e p e iously no
iable, esul ing in inc eases in bo h yields and cul i a ed a eas.
We inco po a ed he p ice and ola ili y o Diammonium Phospha e (DAP), a e ilize commonly
used o soybean p oduc ion, as a signal o he a iable cos s o ag icul u al p oduce s.
16
Addi ionally, a a iable ep esen ing he p opo ion o he plan ed a ea whe e ansgenic a ie ies
we e es ablished was included as an indica o o he echnological ad ancemen in he p oduc ion
sec o , as his ep esen s a shi in he soybean p oduc ion model. In o ma ion on echnological
ad ancemen in each delega ion was no a ailable.
Wi h espec o land use, ollowing Chen e al. (2013), we cons uc ed an indica o o land use
change, compu ing he yea -on-yea educ ion in he plan ed a ea o c ops o he han soybean
inside each delega ion. This indica es he po ion o cul i able land ha can be made a ailable o
soybean p oduc ion each yea , and e en ually ha e an e ec on yields.
4. Me hodological App oach and Resul s
4.1. S a egy o Model Selec ion
In o de o de elop a ool wi h obus p edic i e capaci y o p edic soybean yields in A gen ina,
we explo e di e en speci ica ions using delega ion-le el da a om ac oss he en i e coun y.
These speci ica ions di e based on he ype o a iable used, he equency o he a iables
inco po a ed, he combina ion o equencies (mon hly, annual, and qua e ly), and he
cons uc ion o indica o s om he same a iable. This gene a es a mul iplici y o da a s uc u es
ha can be used o p edic yield, each o which equi es di e en es ima ion me hods.
Ou s a ing poin is a panel da a model designed o unde s and he unde lying d i e s o yield
a ia ion. While i is c ucial o unde s and how clima ic a iables di ec ly a ec soybean yields
in A gen ina, hese ac o s ope a e wi hin an economic and inancial con ex ha can mi iga e o
exace ba e clima ic e ec s. By de ec ing he mos ele an a iables in bo h he clima e and he
economic and inancial aspec s—and by examining hei speci ic dis ibu ions—we a e able o
in o m he con igu a ion o al e na i e ime se ies model speci ica ions.
4.2. Panel Da a Model: Speci ica ion and Resul s
In his sec ion, we p esen a de ailed analysis o ou panel eg ession me hods and esul s. Fi s ,
we es he panel uni oo hypo hesis, and hen we p oceed o es ima e he p oposed model using
he ixed e ec s es ima o o a s ochas ic p oduc ion unc ion. All panel uni oo es s
2
—ac oss
mul iple speci ica ions—consis en ly ejec he null hypo hesis o a uni oo , indica ing ha he
se ies a e s a iona y in le els and do no equi e di e encing p io o es ima ion.
Second, we es ima e he p oposed model using he ixed e ec s es ima o o a s ochas ic
p oduc ion unc ion. This app oach elies in he p oduc ion unc ion men ioned in Sec ion 3.1,
along wi h p oduc ion incen i e a iables, o achie e impac es ima es by a ying one o mo e
2
Le in, Lin & Chu (LLC) es , Im es , Pesa an & Shin (IPS) and Dickey Fulle
17
inpu a iables, such as empe a u e, p ecipi a ion, he sea su ace empe a u e indica o o he
Paci ic Ocean, and he use o incen i e a iables like soybean p ices and inpu cos s along wi h
hei ola ili y.
We u ilize he ixed e ec s model o ou panel da a o wo easons. The main eason is ha he
ixed e ec s model allows us o es ima e a uni -speci ic e ec o each delega ion in he model.
Addi ionally, he ixed e ec s model does no equi e he es ic i e assump ion ha he speci ic
e ec o he delega ion is independen o he included co a ia es, as is he case wi h he andom
e ec s model (See Appendix 2 o he Hausman es esul s). Also, we a e using he o al
popula ion (all he soybean p oducing delega ions), which p e en s us om needing a andom
e ec s es ima ion o gene alize sample da a o a b oade popula ion.
The dependen a iable o his model is he loga i hm o soybean yields o each delega ion (as
desc ibed abo e). We es ima e he ollowing eg ession:
log𝑌
𝑖𝑡 = 𝑍𝑖𝑡 𝛽 + 𝑐𝑖+ 𝜀𝑖𝑡 (1)
𝜀𝑖𝑡 = 𝜌 ∑𝑊𝑖,𝑗𝜀𝑗𝑡 + 𝜙𝑖𝑡𝑗 (2)
Whe e log 𝑌
𝑖𝑡 indica es log c op yield in delega ion i and yea . The e m 𝑍𝑖𝑡𝛽 includes wea he
a iables, a ime end and quad a ic ime end, and o he echnological, land use change and
economic a iables. The e m 𝑐𝑖 is included o accoun o he delega ions’ ixed e ec s, and 𝜀𝑖𝑡
is he e o e m.
Following Chen e al. (2013) and Schlenke e al. (2006), we allow o spa ial co ela ion in he
e o e m. In equa ion 2, 𝜌 is he spa ial co ela ion ac o , 𝑊𝑖,𝑗 is a spa ial weigh ing ma ix ha
iden i ies neighbo s o each delega ion, and 𝜙𝑖𝑡a e he e o e ms ha a e independen ly
no mally dis ibu ed wi h E=0 and a iance= 𝜎2. Panel models ha include an in e ac ion e ec
in he e o e m-called SEM (spa ial e o models)-, indica e ha uni s migh ha e a simila
beha io because o sha ed unobse ed cha ac e is ics (Elho s , 2017). In his case, yield migh
be in luenced by ac o s ha a e no included in he model, such as egional policies, soil quali y
o seed a ie ies used, ha a e spa ially co ela ed o e delega ions.
We es o he signi icance o he spa ial e o e m and ound i o be s a is ically signi ican (see
Appendix 3).
The model speci ica ion equi es comple e da a o all delega ions; consequen ly, we exclude
se en delega ions wi h missing o ze o alues om he es ima ion. The excluded delega ions add
up o 2% o he o al p oduc ion on a e age.

18
Table 2: SEM wi h spa ial ixed e ec s.
Coe icien
S d. E o
z
P>|z|
T end
-2,323
0,907
-2,560
0,010
Quad a ic end
0,001
0,000
2,570
0,010
El Niño empe a u e anomaly (-1)
0,030
0,006
4,590
0,000
T ansgenic Soybean
0,328
0,087
3,760
0,000
Ra io o soybean p ice/ e ilize p ice (-1)
0,119
0,081
1,470
0,140
LUC indica o
0,087
0,051
1,700
0,088
GDD
-0,000
0,000
-2,540
0,011
GDD o e 30ºC
-0,000
0,000
-2,280
0,023
P ecipi a ion
0,002
0,000
7,530
0,000
P ecipi a ion squa ed
-0,000
0,000
-5,470
0,000
Spa ial co ela ion
0,608
0,028
21,640
0,000
Numbe o g oups = 32
Panel leng h = 35
The SEM panel es ima ion esul s a e shown in Table 2, and he summa y o he e ec s ound is:
• GDD and GDD o e 30ºC: These a iables ep esen he g owing deg ee days and
cumula ed hou s wi h empe a u es abo e 30°C du ing he g owing season espec i ely.
Coe icien s a e bo h nega i e, small (due o scaling o he a iables), and s a is ically
signi ican . A nega i e coe icien associa ed wi h he a iable GDD o e 30ºC sugges s
ha empe a u es abo e 30 deg ees nega i ely impac yields. The coe icien o he
a iable GDD is also nega i e, sugges ing ha he e ec o high empe a u es is dominan
o e he summe soybean g owing season, ha ing a nega i e e ec on yields o e all.
• P ecipi a ion and P ecipi a ion squa ed: These a iables e lec he cumula i e
p ecipi a ion and he squa ed cumula i e p ecipi a ion be ween July and Ma ch along he
same campaign. The coe icien s associa ed wi h hese a iables (posi i e and nega i e,
espec i ely) show a nonlinea ela ionship be ween p ecipi a ion and yields: mo e
p ecipi a ion inc eases yields, bu beyond a ce ain h eshold, he impac s a s o inc ease
a a dec easing a e.
• El Niño empe a u e anomaly: This a iable e lec s an expanded yea ly Oceanic Niño
Index (ONI), ha shows he p esence o he El Niño and La Niña e en s along wi h hei
in ensi y. The a iable is lagged one yea , o show he condi ions p io and du ing he
ea ly s ages o each campaign. The coe icien is posi i e and s a is ically signi ican ,
sugges ing ha wa m condi ions (associa ed wi h he El Niño e en s) a e a o able o
yield. Con e sely, nega i e alues o ONI indexes indica e La Niña e en s ha p omo e
19
d y condi ions o e he soybean p oducing a eas o A gen ina, and would ha e a nega i e
e ec on yields.
Rega ding end, ma ke and echnology a iables, we inco po a ed he ollowing:
• T end and Quad a ic end: The linea and quad a ic ends ep esen long- e m e ec s.
The nega i e coe icien on he end indica es a sligh dec ease in yield o e ime; while
he posi i e quad a ic end sugges s ha he a e o change in yield may s abilize o e en
sligh ly inc ease in ecen yea s due o ac o s no included in he es ima ion.
• Ra io o soybean p ice/ e ilize p ice: I is he a io be ween he soybean p ice a iable
and he p ice o DAP (Diammonium Phospha e e ilize ) on he yea p io o sowing.
The posi i e coe icien obse ed in his a io indica es po en ial inc eases in a me
bene i s, which in u n lead o highe yields. This is likely due o a me s ying o inc ease
p oduc ion pe a ea in esponse o hei pe cep ion o highe u u e bene i s.
• LUC indica o : The land use change indica o shows he yea -on-yea educ ion in he
plan ed a ea o c ops o he han soybean inside each delega ion. The esul ing coe icien
is, al hough small, indica i e o he ac ha educ ions in a eas dedica ed o o he c ops
a e being shi ed o soybean plan ing.
• T ansgenic Soybean: The impac o he pe cen age o ansgenic soybean use yields a
posi i e and signi ican coe icien , indica ing ha as he use o ansgenic soybean in he
p oduc i e a ea has inc eased, i has con ibu ed o imp o ed yields ega dless o clima ic
impac s.
4.3. Time se ies: model speci ica ion and da a s uc u e
Using insigh s om he panel app oach, we cons uc ime‑se ies models o e alua e di e en
s a egies o p edic ing A gen ina’s agg ega e soybean yield.
The independen a iable emains he loga i hm o soybean yields. Fo he explana o y a iables,
we gene a e a ime se ies da ase based on he panel da a, conside ing di e en agg ega ion
me hods. Fi s , we a e age he delega ions’ clima e a iables o e he spa ially in e pola ed da a
used in he panel model o he whole soybean a eas. Second, we gene a e ano he da ase by
a e aging di ec ly all soybean-p oducing zone’s me eo ological s a ions eco ds. Las ly, we c ea e
ano he da ase wi h clima ic a iables measu ed in he co e zone only, in o de o gene a e a
leading indica o . The idea is o explo e a uni o m egion in e ms o clima e, p oduc ion, and
managemen .
We hen es di e en combina ions o a iables. Ou i s se o ime se ies models (G oup G1)
uses he same a iables as he panel da a es ima ion, whe e he wea he a iables we e a e aged
20
o e he whole soybean-p oducing zone. In he second s ep, we iden i y he op imal model—
based on en opy—using a iables om he panel‐model o key mon h o he g owing‐season
calenda . This yields di e en ime‐se ies models, each co esponding o he wea he in o ma ion
a ailable a successi e s ages (Decembe , Janua y, Feb ua y, and Ma ch).
To p o ide obus ness o he compa ison, we also inco po a e a second g oup o models designed
o cap u e di e en ea u es o he phenomenon a hand. This second se o models (G oup G2)
con ains combina ions o a iables di e en om hose on g oup G1, and also akes in o accoun
he a ailabili y o da a o di e en mon hs. The explana o y a iables a e he o al panel a e ages
o all he soybean p oducing egion.
The hi d g oup (G oup G3) akes in o accoun di e en o mula ions using a iables wi h annual,
mon hly, and qua e ly equencies, as is he case wi h ENSO. This mix o equencies equi es
he use o di e en es ima ion me hods such as MIDAS (Mixed Da a Sampling
3
). Models in g oup
G3 use as explana o y a iables he wea he a e ages using indica o s measu ed wi hin he co e
zone only. The independen a iable is he same as he p e ious g oups, he loga i hm o he o al
soybean yield in A gen ina.
4.4. Fo ecas compa ison
Ou model selec ion s a egy is d i en by he goal o an icipa ing o ou pe o ming he o ecas
esul s published by he US Depa men o Ag icul u e (USDA), a globally ecognized au ho i y
in ag icul u al p oduc ion o ecas ing. The e o e, we conside wo dimensions: he abili y o
an icipa e and he abili y o p edic .
In his con ex we use he e m "an icipa e" e e ing o he compa ison be ween wo models wi h
in o ma ion a ailable a di e en poin s in ime. A model wi h inpu s ha a e mo e ecen o close
in ime o he e en being p edic ed is expec ed o ha e g ea e p edic i e capaci y han a model
wi h inpu s ha a e mo e dis an in ime. When a model wi h in o ma ion ha is mo e dis an in
ime p edic s a con empo a y a iable be e han a model wi h mo e ecen in o ma ion, we say
ha he la e an icipa es he o me ; e en i hese wo models a e no able o di e en ia e hei
p edic i e capaci y ( hey p edic he same hing).
On he o he hand, i wo models’ inpu s a e a ailable a he same ime, he model ha be e
p edic s he a ge a iable will be conside ed o ha e be e o ecas ing powe and he e o e ha e
be e p edic i e capaci y.
To gene a e he o ecas s, once he panel and ime se ies models we e de ined, we di ided he
da ase in o aining and es ing segmen s, a p ocess o en used in ime-se ies analysis as se ing
3
See Ghysels, San a-Cla a and Valkano (2006)
21
an "es ima ion window." Wi hin his window, each model is ained o make p edic ions,
p ojec ing ‘h’ s eps ahead.
Fo example, we es ima e he model using da a om 1989 o 2016 and we ob ain a p edic ion o
he soybean yields one yea ahead. Then we expand he sample o 1989 o 2017, and we o ecas
2018. I e a ing his p ocess esul s in a o RMSE o each model, o be compa ed wi h he USDA
p edic ions’ RMSE.
To compa e he models, we use he Giacomini and Whi e (2006) es , which allows us o e alua e,
compa e, and p io i ize di e en models based on hei p edic i e pe o mance. This ool only
indica es which model is he bes wi hin a se , bu i does no p o ide in o ma ion on how good
he model is in absolu e e ms. To o e come his limi a ion, we use he USDA o ecas s as a
benchma k o compa ison.
Ou benchma k is de i ed om a subse o he USDA’s se ies o p edic ions in hei mon hly
epo s. While USDA issues mon hly Oilseeds epo s, h oughou he yea , we ake in o accoun
he epo s published in Janua y o Ap il ( he ha es s a s in Ap il and las s un il June). The
epo s a e made a ailable in he i s days o each mon h, he e o e we conside hem o
inco po a e in o ma ion up o he end o he p e ious mon h. Fo example, o he o ecas
included in he Janua y epo , we assume i inco po a es in o ma ion up un il he las day o
Decembe o he p e ious yea .
Table 3 shows some o he esul s o he Giacomini-Wi h es s. The es esul s allow us o compa e
all he o ecas s coming om ou panel Spa ial E o Models (SEM) and Time se ies models om
g oups 1, 2 and 3 wi h published USDA o ecas s. While we an pai wise compa isons o all he
a ailable models, in his able we show he ones ha ou pe o m USDA o ecas s
4
.
Table 3: Model compa ison (GW esul s) on models ha pe o m be e han USDA o ecas s
Model 1
Model 2
Da a
A ailabili y
Coe .
S d.
E o
-S a is ic
P ob.
SEM_DEC
USDA_Jan_ epo
DECEMBER
-106,87
56,91
-1,87
0,1095
TS_G2_01_DEC
USDA_Jan_ epo
DECEMBER
-115,53
65,51
-1,76
0,1283
TS_G3_03_JAN
USDA_Feb_ epo
JANUARY
-150,24
80,26
1,87
0,1104
TS_G2_06_JAN
USDA_Feb_ epo
JANUARY
-137,35
72,93
-1,88
0,1087
SEM_FEB
USDA_Ma _ epo
FEBRUARY
-77,918
43,01
-1,81
0,1201
The i s wo columns show he models being compa ed. "Da a A ailabili y" e e s o he momen
when he da a becomes a ailable o inpu in o he model. Fo example, in he i s ow, he
SEM_DEC, - he spa ial e o model- and he USDA_Jan_ epo o ecas ake in o accoun da a
4
Resul s can be made a ailable upon eques o he au ho s.
28
Appendices
Appendix 1: Causali y es s
When analyzing wo lags, we ind ha he null hypo hesis ha p ices do no cause p oduc ion
canno be accep ed; howe e , A gen ine p oduc ion has no e ec on p ices.
Pai wise G ange Causali y Tes s
Sample: 1980 2024
Lags: 2
Null Hypo hesis:
Obs
F-S a is ic
P ob.
DLPROD does no G ange Cause DLPRECIO
31
0.57428
0.5701
DLPRECIO does no G ange Cause DLPROD
3.57279
0.0426
The null hypo hesis o no G ange causali y om plan ed a ea o yields canno be accep ed;
whe eas, he hypo hesis ha yields do no G ange -cause plan ed a ea canno be ejec ed.
On he o he hand, he null hypo hesis o no G ange causali y be ween plan ed a ea and ENSO
canno be ejec ed in ei he di ec ion, indica ing ha he e ec o ENSO on yields is isola ed om
any po en ial changes in plan ed a ea induced by ENSO.
Pai wise G ange Causali y Tes s
Sample: 1989 2022
Lags: 2
Null Hypo hesis:
Obs
F-S a is ic
P ob.
L_SUP does no G ange Cause ENSO
32
0.3726
0.6924
ENSO does no G ange Cause L_SUP
1.0931
0.3495
LOG_RINDE does no G ange Cause ENSO
32
2.6924
0.0859
ENSO does no G ange Cause LOG_RINDE
9.4367
0.0008
LOG_RINDE does no G ange Cause L_SUP
32
0.2634
0.7704
L_SUP does no G ange Cause LOG_RINDE
5.4377
0.0104

29
The e is no e idence o bidi ec ional G ange causali y be ween he in e na ional p ice and he
ENSO index, indica ing ha hey a e s a is ically independen in his con ex .
Pai wise G ange Causali y Tes s
Sample: 1989 2022
Lags: 2
Null Hypo hesis:
Obs
F-S a is ic
P ob.
ENSO does no G ange Cause DLPRECIO
32
0.3471
0.7100
DLPRECIO does no G ange Cause ENSO
0.5429
0.5875
30
Appendix 2: Hausman es o Fixed E ec s and Random E ec s panels
Since we ha e da a om all ele an delega ions (i.e., he sample co e s he en i e popula ion o
egions), panel ixed e ec s es ima ions a e p e e ed because hey cap u e he indi idual
he e ogenei y o each egion wi hou assuming ha hese di e ences a e andom. A andom
e ec s es ima ion would be mo e app op ia e i he da a we e a andom sample om a la ge se
o egions.
To es he c oss-sec ion andom e ec s, we conduc ed he Hausman Tes , wi h he ollowing
esul s:
Tes Summa y
Chi-Sq. S a is ic
Chi-Sq. d. .
P ob.
C oss-sec ion andom
45.666
10
1.648e-06
The esul s imply ha ixed e ec s should be used. This sugges s ha he unobse ed di e ences
be ween egions a e co ela ed wi h he explana o y a iables, in alida ing he key assump ion o
andom e ec s.
31
Appendix 3: Mo an and LM es s o spa ial e o co ela ion signi icance
The spa ial weigh ma ix used in he eg ession and o his es is a spa ial con igui y ma ix.
This spa ial con igui y ma ix includes pai wise compa isons o all delega ions, whe e he (i, j)
elemen o W is uni y i delega ions i and j sha e a common bounda y, and 0 o he wise. The ma ix
is no malized so ha he sum o he elemen s in each ow is equal o one.
The Mo an es is conduc ed o e he c oss-sec ion o he delega ions’ soybean yields o each
yea . The esul s a e:
Yea
Mo an I
p- alue
Yea
Mo an I
p- alue
Yea
Mo an I
p- alue
1989
4,197
2,70E-05
2002
2,639
8,32E-03
2015
3,629
2,85E-04
1990
1,504
1,33E-01
2003
3,100
1,94E-03
2016
4,603
4,15E-06
1991
2,024
4,30E-02
2004
3,549
3,87E-04
2017
0,917
3,59E-01
1992
4,149
3,34E-05
2005
4,519
6,23E-06
2018
3,197
1,39E-03
1993
0,325
7,45E-01
2006
2,271
2,31E-02
2019
2,932
3,37E-03
1994
2,857
4,28E-03
2007
0,898
3,69E-01
2020
3,031
2,44E-03
1995
2,002
4,53E-02
2008
0,925
3,55E-01
2021
2,602
9,26E-03
1996
3,326
8,80E-04
2009
3,061
2,20E-03
2022
2,843
4,47E-03
1997
1,739
8,20E-02
2010
0,262
7,93E-01
1998
3,894
9,87E-05
2011
3,148
1,64E-03
1999
2,950
3,18E-03
2012
6,473
9,62E-11
2000
3,612
3,03E-04
2013
4,096
4,21E-05
2001
0,232
8,17E-01
2014
0,659
5,10E-01
32
Appendix 4: SEM Panel models o o ecas ing
We de elop an al e na i e a iable o all SEM o ecas ing models (SEM_DEC, SEM_JAN,
SEM_FEB and SEM_MAR) ha aims o cap u e wi hin-yea empe a u e pa e ns mo e
e ec i ely, which may p o ide aluable insigh s o yield p edic ion.
To show he p ocedu e applied, he ollowing able desc ibes he SEM_MAR model. The main
di e ence be ween he Spa ial E o Model (SEM) in Table 2 and he SEM_ MAR model lies in
how hey inco po a e he equency o he El Niño empe a u e anomaly.
Speci ically, when we e e o he El Niño empe a u e anomaly (-1) in he SEM Table 2 model,
we a e e e ing o he a e age empe a u e du ing he calenda yea ha s a s jus be o e he
campaign. In his con ex , he anomaly (-1) includes empe a u e in o ma ion bo h p io o and
du ing he campaign. While his annual a e age anomaly p o ides use ul in o ma ion abou he
o e all impac o empe a u e on yield, i may also obscu e impo an in a-annual pa e ns ha
a e ele an o o ecas ing.
Coe icien
S d. E o
z
P>|z|
T end
-2,748
0,951
-2,890
0,004
Quad a ic end
0,001
0,000
2,900
0,004
El Niño empe a u e anomaly (Dec(-1))
0,178
0,044
4,070
-
El Niño empe a u e anomaly (Ma (-1))
-0,180
0,060
-2,990
0,003
El Niño empe a u e anomaly (June(-1))
0,057
0,038
1,500
0,134
T ansgenic Soybean
0,312
0,095
3,290
0,001
Ra io o soybean p ice/ e ilize p ice (-1)
0,173
0,084
2,060
0,040
LUC indica o
0,084
0,051
1,660
0,097
GDD
-0,000
0,000
-2,620
0,009
GDD o e 30ºC
-0,000
0,000
-2,260
0,024
P ecipi a ion
0,002
0,000
7,560
-
P ecipi a ion squa ed
-0,000
0,000
-5,570
-
Spa ial co ela ion
0,595
0,029
20,580
0,000
Numbe o g oups = 32
Panel leng h = 35
In he able abo e we show he qua e ly pa e n, whe e we inco po a e he qua e ly a iables
ha end in Decembe , Ma ch and June, p e ious o he s a o he campaign. The di e ence in
signs be ween empe a u e in di e en pe iods indica es ha di e en combina ions o qua e ly
empe a u e could gi e di e en esul s o he soybean yields e en i he a e age is he same.
33
Appendix 5: Es ima ion esul s: SEM wi h spa ial ixed e ec s
SEM
Decembe
SEM Janua y
SEM
Feb ua y
SEM Ma ch
T end
-2,060
-2,192*
-3,084***
-2,748***
(1,257)
(1,152)
(1,017)
(0,951)
Quad a ic end
0,001*
0,001*
0,001***
0,001***
(0,000)
(0,000)
(0,000)
(0,000)
El Niño empe a u e
anomaly (-3Q)
0,219***
0,220***
0,202***
0,178***
(0,059)
(0,054)
(0,047)
(0,044)
El Niño empe a u e
anomaly (-2Q)
-0,261***
-0,252***
-0,213***
-0,180***
(0,080)
(0,073)
(0,065)
(0,060)
El Niño empe a u e
anomaly (-1Q)
0,139***
0,113**
0,068
0,057
(0,051)
(0,047)
(0,042)
(0,038)
T ansgenic Soybean
0,347***
0,339***
0,372***
0,312***
(0,130)
(0,118)
(0,102)
(0,095)
Ra io o soybean
p ice/ e ilize p ice (-1)
0,257**
0,194*
0,196**
0,173**
(0,112)
(0,104)
(0,090)
(0,084)
LUC indica o
0,077
0,076
0,079
0,084*
(0,052)
(0,052)
(0,051)
(0,051)
G owing deg ee days
-0,0004*
-0,0003**
-0,0003***
-0,0002***
(0,000)
(0,000)
(0,000)
(0,000)
GDD o e 30ºC
-0,000058
-0,000007
-0,000139
-0,000260**
(0,000)
(0,000)
(0,000)
(0,000)
P ecipi a ion
0,001***
0,002***
0,002***
0,002***
(0,000)
(0,000)
(0,000)
(0,000)
P ecipi a ion squa ed
-
0,000001***
-0,000001***
-0,000001***
-0,000001***
(0,000)
(0,000)
(0,000)
(0,000)
Spa ial co ela ion
0,694***
0,668***
0,624***
0,595***
(0,023)
(0,025)
(0,028)
(0,029)
Numbe o g oups = 32
Panel leng h = 35
S anda d e o s a e epo ed in pa en heses.
*, **, *** indica es signi icance a he 90%, 95%, and 99% le el, espec i ely