G anzie a, Eleono a; Jalasjoki, Pi kka; Palo ii a, Ma i a
Wo king Pape
The bias and e iciency o he ECB in la ion p ojec ions: A
s a e dependen analysis
Wo king Pape , No. 1/2021
P o ided in Coope a ion wi h:
No ges Bank, Oslo
Sugges ed Ci a ion: G anzie a, Eleono a; Jalasjoki, Pi kka; Palo ii a, Ma i a (2021) : The bias and
e iciency o he ECB in la ion p ojec ions: A s a e dependen analysis, Wo king Pape , No. 1/2021,
ISBN 978-82-8379-189-1, No ges Bank, Oslo,
h ps://hdl.handle.ne /11250/2755864
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/246122
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by-nc-nd/4.0/deed.no
The Bias and E iciency o he
ECB In la ion P ojec ions: a
S a e Dependen Analysis
NORGES BANK
RESEARCH
1 | 2021
ELEONORA GRANZIERA
PIRKKA JALASJOKI AND
MARITTA PALOVIITA
WORKING PAPER
NORGES BANK
WORKING PAPER
XX | 2014
RAPPORTNAVN
2
Wo king pape s a No ges Bank, a 1992/1 il 2009/2 kan bes illes o e e-pos :
Facili ySe ices@no ges-bank.no
F a 1999 og sene e e publikasjonene ilgjengelige på www.no ges-bank.no
Wo king pape s inneholde o skningsa beide og u edninge som anlig is ikke ha å
sin endelige o m. Hensik en e blan anne a o a e en kan mo a kommen a e a
kollege og and e in e esse e. Synspunk e og konklusjone i a beidene s å o
o a e nes egning.
Wo king pape s om No ges Bank, om 1992/1 o 2009/2 can be o de ed by e-mail:
Facili ySe ices@no ges-bank.no
Wo king pape s om 1999 onwa ds a e a ailable on www.no ges-bank.no
No ges Bank’s wo king pape s p esen esea ch p ojec s and epo s (no usually in hei
inal o m) and a e in ended in e alia o enable he au ho o bene i om he commen s o
colleagues and o he in e es ed pa ies. Views and conclusions exp essed in wo king
pape s a e he esponsibili y o he au ho s alone.
ISSN 1502-8190 (online)
ISBN 978-82-8379-189-1 (online)
The Bias and Eciency o he ECB Ina ion P ojec ions: a S a e
Dependen Analysis
∗
Eleono a G anzie a
†
Pi kka Jalasjoki
‡
Ma i a Palo ii a

Ap il 28, 2021
Abs ac
We es o bias and eciency o he ECB ina ion o ecas s using a conden ial da ase
o ECB mac oeconomic qua e ly p ojec ions. We in es iga e whe he he p ope ies o he
o ecas s depend on he le el o ina ion, by dis inguishing whe he he ina ion obse ed by
he ECB a he ime o o ecas ing is abo e o below he a ge . The o ecas s a e unbiased and
ecien on a e age, howe e he e is e idence o s a e dependence. In pa icula , he ECB ends
o o e p edic (unde p edic ) ina ion a in e media e o ecas ho izons when ina ion is below
(abo e) a ge . The magni ude o he bias is la ge when ina ion is abo e he a ge . These
esul s hold e en a e accoun ing o e o s in he ex e nal assump ions. We also nd e idence
o ineciency, in he o m o unde eac ion o news, bu only when ina ion is abo e he a ge .
Ou ndings bea impo an implica ions o he ECB o ecas ing p ocess and ul ima ely o i s
communica ion s a egy.
Keywo ds
: Fo ecas E alua ion, Fo ecas Eciency, Ina ion Fo ecas s, Cen al Bank Commu-
nica ion
JEL classica ion
: C12, C22, C53, E31, E52
∗
This Wo king Pape should no be epo ed as ep esen ing he iews o No ges Bank, he Bank o Finland o
he Eu osys em. The iews exp essed a e hose o he au ho s and do no necessa ily eec hose o No ges Bank,
he Bank o Finland o he Eu osys em. We hank Ge gely Ganics, Juha Kilponen, Ja mo Kon ulainen, Ba ba a
Rossi, Ta e ik Sekhposyan, and con e ence pa icipan s a he 14 h In e na ional Con e ence on Compu a ional and
Financial Econome ics, 2nd Vienna Wo kshop on Economic Fo ecas ing, 28 h Annual Symposium o he Socie y
o Nonlinea Dynamics and Econome ics, 3 d Fo ecas ing a Cen al Banks Con e ence, he 2019 Con e ence on
Real-Time Da a Analysis, Me hods and Applica ions. We also hank an anonymous e e ee om he No ges Bank
Wo king Pape se ies.
†
Co esponding Au ho . No ges Bank; email: [email p o ec ed]
‡
Bank o Finland. Email: pi kka.jalasjoki@bo .

Bank o Finland. Email: ma i a.palo ii a@bo .
1 In oduc ion
Fo ecas ing is an essen ial pa o mone a y policy making and published o ecas s a e a he
co e o cen al bank communica ion s a egy, pa icula ly o cen al banks a ge ing ina ion. In
ac , cen al bank ina ion o ecas s aec p i a e sec o expec a ions (see e.g. Fujiwa a (2005),
Hube (2014,2015,2017) and Lyziak and Palo ii a (2018)). Ina ion p ojec ions published by
mone a y au ho i ies a e likely o gain u he p ominence in expec a ions managemen and se e
as an addi ional policy ool. In he cu en en i onmen o low in e es a es and low ina ion, as
he maneu e ing space o adi ional ools is limi ed, policies di ec ly impac ing agen s' ina ion
expec a ions can be used o s abilize economic condi ions (Coibion e al.,2020). In addi ion, as
cen al banks econside hei s a egies and assess make-up ules such as a e age ina ion a ge ing,
published o ecas s migh be used o c ea e he expec a ion ha ina ion will o e shoo i s a ge .
Howe e , published ina ion p ojec ions ha e gone unde sc u iny. The accu acy o cen al
banks' o ecas s dec eased du ing he nancial c isis (Po e (2011), S ock on (2012), Alessi e al.
(2014), Fawce e al. (2015)). Also, ollowing he c isis many mone a y au ho i ies ha e consis en ly
o e es ima ed he a e o ina ion (I e sen e al. (2016), Kon ogeo gos and Lamb ias (2019)), which
has emained pe sis en ly below a ge . Repea ed la ge and sys ema ic p ojec ion e o s pose a chal-
lenge o cen al banks as hese e o s may inc ease he isk o deancho ing o ina ion expec a ions
and de e io a e he c edibili y o mone a y au ho i ies.
Agains his backg ound we analyze he p ope ies o he Eu osys em/Eu opean Cen al Bank
s a (he ea e ECB) p ojec ions o ina ion. Fi s , we check whe he he o ecas s a e biased by
es ing whe he he o ecas e o s ha e mean ze o (Holden and Peel,1990). Second, we in es iga e
whe he p ojec ions a e ecien by es ing whe he he ex-pos o ecas e o is unco ela ed wi h
he p e ious o ecas e isions, ollowing No dhaus (1987) and Coibion and Go odnichenko (2015).
As a no el con ibu ion we ela e bias and eciency o he economic condi ions a he ime
o o ecas ing. Specically, we ask whe he sys ema ic o ecas e o s a e ela ed o he le el o
ina ion, by dis inguishing whe he ina ion is abo e o below a ge when he o ecas s a e made.
We conjec u e ha , because o he ECB p ice s abili y manda e, he le el o ina ion obse ed a
he ime o o ecas ing migh inuence he way in which new in o ma ion is inco po a ed in he
o ecas s.
2

We analyze he ECB p ojec ions o he Eu o A ea HICP ina ion a e. We ocus only on
ina ion because he manda e o he ECB is dened in e ms o p ice s abili y. Ou da a include
eal ime es ima es o cu en -qua e alues and eal ime p ojec ions un il eigh qua e s ahead.
No e ha in e e y qua e he ECB publishes annual p ojec ions o ull calenda yea s, while ou
da ase includes o iginal conden ial qua e ly p ojec ions, which a e una ailable o he public.
1
On a e age, we nd no sign o bias no ineciency in ina ion o ecas s. Howe e , we documen
a signican die ence in he p ope ies o he o ecas s depending on he le el o ina ion a he
ime o o ecas ing. In pa icula , we de ec a sys ema ic bias, bo h s a is ically and economically
signican , o medium e m o ecas s when condi ioning on he le el o ina ion. This sugges s
ha when ina ion is lowe (highe ) han he a ge , he ECB ends o o e p edic (unde p edic )
ina ion. The e o e, we conclude ha he e is a sys ema ic bias owa ds he a ge . We also
documen ha he magni ude o he bias is conside ably la ge in absolu e alue when ina ion is
abo e he a ge .
The ECB p ojec ions a e condi ional o ecas s, i.e. hey a e based on assump ions ega ding
a pa h o u u e alues o ele an mac oeconomic and nancial a iables. One migh a gue ha
sys ema ic e o s in hese ex e nal assump ions migh esul in o sys ema ic e o s in ina ion
o ecas s. Howe e , we nd ha , while he p edic ion e o s in ina ion a e co ela ed wi h e o s
in he ex e nal assump ions, he ina ion o ecas s exhibi a signican sys ema ic bias e en a e
con olling o e o s in hese assump ions.
Biased o ecas s a e no necessa ily i a ional. Capis an (2008) asc ibes he bias o an asym-
me ic loss unc ion o he cen al bank, and shows ha o e p edic ion occu es when he mone a y
au ho i y is mo e conce ned abou ina ion abo e han below a ge . Cha emza and Ladley (2016)
show ha a bias owa ds he a ge can esul om he o ing sys em and dynamics wi hin he
mone a y policy commi ee. He be (2020) shows ha i is op imal o he mone a y au ho i y o
sys ema ically o e p edic (unde p edic ) agg ega e condi ions in ecessions (expansions) in o de
o bias agen s' belie s, i agen s ha e he e ogeneous p io s abou he s a e o he economy. In a
labo a o y expe imen , Duy and Heinemann (2021) nd ha hei es subjec cen al banke s ac
s a egically and make announcemen s ha de ia e om he " ue" o ecas s in o de o manage
agen s' ina ion expec a ions. While we do no iden i y he sou ce o he s a e dependen bias, we
1
The ECB s a ed o publish qua e ly p ojec ions in 2017Q2.
3
no e ha i is consis en wi h a s a egic beha iou o a cen al bank aiming a s ee ing expec a ions
owa ds he a ge .
Rega ding eciency, we nd ha o ecas e isions do p edic o ecas e o s, bu only when he
de ia ions o ina ion om he a ge a e posi i e. In pa icula , when ina ion is abo e he a ge ,
he es ima ed coecien associa ed wi h he e isions is posi i e. This means ha , o example,
when he e ision is posi i e, i.e. he new o ecas is highe han he o ecas made in he p e ious
pe iod, he o ecas e o is posi i e, i.e. he ECB unde p edic s, so he o ecas was no e ised
"enough". These ndings poin o an "unde - eac ion" o he ECB o new in o ma ion, and can be
due o in o ma ion igidi ies (Coibion and Go odnichenko,2015) o smoo hing (Tillman,2011).
2
A la ge numbe o pape s has analyzed he bias and eciency o he o ecas s p oduced by he
Fede al Rese e: Clemen s e al. (2007), Capis an (2008), Sinclai e al. (2010), Messina e al.
(2015) and El-Shagi e al. (2016) o G eenbook o ecas s, and Rome and Rome (2008) and A ai
(2016) o FOMC. We depa om hese s udies in wo dimensions:  s , we analyze bias and
eciency o he ECB p ojec ions, which ha e no been s udied p e iously, wi h he excep ion o
Kon ogeo gos and Lamb ias (2019).
3
No e ha se e al ins i u ional die ences dis inguish he
ECB om he Fede al Rese e. The ECB's manda e is dened in e ms o p ice s abili y, while
he Fed has a dual manda e. The Fed has an explici a ge o 2% o ina ion, while he ECB
aims a keeping ina ion below, bu close o 2%. Finally, he Fed p oduces wo se s o o ecas s:
he Tealbook (ea lie G eenbook) o ecas s, which ep esen s a o ecas s and a e kep conden ial
o  e yea s, and he FOMC p ojec ions, which summa ize he o ecas s o he FOMC membe s
and a e a ailable o he public in eal ime. In con as , he ECB p ojec ions a e s a o ecas s,
simila ly o he Tealbook, bu hey a e eleased o he public in he same qua e in which hey a e
p oduced, like he FOMC o ecas s. These die ences migh aec he p ope ies o he o ecas s
and he e o e he ndings documen ed o he Fed's o ecas s migh no hold in he ECB con ex .
Second, only ew pape s in es iga e whe he he p ope ies o he o ecas s a e s a e dependen ,
and hey dene he s a es acco ding o he phases o he business cycle, i.e. ecessions o expansions,
2
Some ecen s udies analyze o ecas eciency using su ey-based expec a ions o households and p o essional
o ecas e s and epo conic ing esul s ega ding unde - o o e - eac ion o news (Fuh e (2018), Angele os e al.
(2020), Bo dalo e al. (2020), Kohlhas and Wal he (2021)). These s udies igno e s a egic mo i es in mone a y
policy communica ion, which migh be p esen in o ecas s made by an ina ion- a ge ing cen al bank.
3
Kon ogeo gos and Lamb ias (2019) ocus on o ecas ing accu acy, eciency, a ionali y and op imali y and
conclude ha he ECB p ojec ions o ina ion a e unbiased and ecien on a e age.
4
a he ime when he ou come is ealized. We die because we in es iga e whe he he p ope ies
o he o ecas s a e ela ed o he
le el o ina ion
wi h espec o he a ge and we dene he s a e
based on he le el o ina ion
a he ime o o ecas ing
. The e o e, we ela e bias and eciency o
he ECB manda e and i s o ecas ing p ocess.
The es o he pape is o ganized as ollows: sec ion 2 p esen s he da a, sec ion 3 summa izes
he baseline econome ic amewo k and empi ical esul s, sec ion 4 desc ibes he s a e dependen
analysis and esul s, sec ion 5 concludes.
2 Da a
We analyze he ECB p ojec ions o he yea -o e -yea Eu o A ea Ha monized Index o Con-
sume P ices (HICP) ina ion a e a he qua e ly equency. Ou da a include eal ime es ima es
o cu en -qua e alues (nowcas es ima es) and eal ime p ojec ions un il eigh qua e s ahead.
No e ha in e e y qua e he ECB publishes
annual
p ojec ions o ull calenda yea s, while
ou da ase includes o iginal conden ial
qua e ly
p ojec ions, which a e una ailable o he public.
The sample uns om 1999Q4 o 2019Q4, esul ing in 81 obse a ions o o ecas e alua ion o he
nowcas and 73 o he eigh s ep-ahead o ecas s.
4
The mac oeconomic p ojec ions a e p oduced by he ECB s a in Ma ch and Sep embe , and
by bo h ECB s a and expe s in na ional cen al banks in June and Decembe . Howe e , in each
qua e mon hly ina ion p ojec ions a e p o ided by na ional cen al bank expe s o o ecas ing
ho izon up o 11 mon hs, h ough he Na ow Ina ion P ojec ion Exe cise (NIPE). Finally, he
p ojec ions a e based on a se ies o assump ions and a e p oduced combining models as well as expe
knowledge and judgemen .
5
The mac oeconomic p ojec ions play a key ole in mone a y policy
decision-making, as he o ecas s a e p esen ed o he Go e ning Council ahead o i s mone a y
policy delibe a ions.
Panel A o Figu e (1) shows he yea -o e -yea Eu o A ea HICP ina ion om 1999Q4 o
2019Q4. Ou sample pe iod includes he ela i ely s able p e-c isis yea s, ollowed by he nancial
c isis, he so e eign deb c isis and mo e ecen ly a pe iod o pe sis en ly low ina ion.
4
We e alua e he o ecas s agains he in age published by Eu os a in Ap il 2020. We do no expe imen wi h
o he in ages since bo h means and medians o da a e isions o eal- ime obse ed ina ion a e close o ze o.
5
Alessi e al. (2014) and Kon ogeo gos and Lamb ias (2019) p o ide u he de ails on he ECB o ecas ing
p ocess.
5
Panel B o Figu e (1) summa izes he p ojec ions o die en o ecas ho izons o e he whole
sample. Fo each o ecas o igin he gu e shows selec ed quan iles o he p ojec ed pa h om
nowcas ing o wo yea s ahead. The solid lines show he median p ojec ion o each ho izon con-
di ional on he a e o ina ion in he qua e when he o ecas s a e p oduced. In pa icula , he
blue ( ed) line shows he median ina ion o ecas s i ina ion is abo e (below) 1.8% a he ime o
o ecas ing. The ed (blue) shaded a eas deno e he 10 h and 25 h (75 h and 90 h) pe cen iles o
he dis ibu ion condi ional on ina ion being below (abo e) a ge . Rega dless o hei ini ial le el,
he o ecas s e e owa ds he a ge a he quickly, wi hin he  s 4 qua e s, and a en ou a
longe ho izons. Also, when ina ion is abo e a ge a he ime o o ecas ing, he e is e idence o
unde shoo ing ( o ecas ing ina ion o all below a ge ) om 4 o 10 qua e s ahead.
3 Tes ing o Bias and Eciency
In his sec ion, we analyze he p ope ies o he ECB ina ion o ecas s. Fi s , we es o
bias
by es ing whe he he o ecas e o s ha e ze o mean (Holden and Peel (1990), Capis an (2008),
Kon ogeo gos and Lamb ias (2019)):
 ,h =y −y | −h=α0,h +u ,h h= 0, ..8
(1)
whe e
 ,h
is he
h
-s ep ahead o ecas e o ,
y
is he ealized alue and
y | −h
is he o ecas o
y
made a ime
−h.
Then, he null o unbiasedness is
α0,h = 0
.
Second, we in es iga e whe he p ojec ions a e
inecien
, i.e. whe he he e was addi ional
in o ma ion eadily a ailable o he o ecas e s ha could ha e been used o imp o e he accu acy
o he p ojec ions. Specically, we es whe he he o ecas e o is unco ela ed wi h he p e ious
o ecas e ision: i he o ecas e ision be ween
−h−1
and
−h
can p edic he o ecas e o ,
hen he in o ma ion ha became a ailable a
−h
was no p ope ly inco po a ed in he e ised
p ojec ions, i.e. he o ecas s we e o e ly o insucien ly adjus ed.
6
This app oach was o iginally
sugges ed by No dhaus (1987) and has been ecen ly adop ed by Lahi i and Sheng (2008), Coibion
and Go odnichenko (2015), Fuh e (2018) and Bo dalo e al. (2020). Fo ecas eciency can be
6
The alidi y o ou esul s is con med using al e na i e es s o bias and eciency, ollowing Mince and
Za nowi z (1969) and Loungani e al. (2013). Resul s o hese exe cises a e a ailable om he au ho s upon eques .
6
his nding.
4.3 The Role o Ex e nal Assump ions
The ECB p ojec ions a e condi ional on a numbe o ex e nal assump ions, i.e. echnical assump-
ions ega ding he u u e de elopmen s o he in e na ional economic en i onmen .
13
Then, one
could a gue ha
sys ema ic
e o s ound in he ina ion p ojec ions migh be d i en by
sys ema ic
e o s in ex e nal assump ions, which esul in bias and ineciency.
Gi en ou access o he assump ions o he ex e nal a iables, we a e in a unique posi ion o es
his conjec u e. We ocus on h ee a iables which a e conside ed he main d i e s o ina ion: a
sho e m in e es a e (3M EURIBOR), he EURUSD exchange a e and he B en C ude oil p ices.
The assump ions made by he ECB o he oil p ices and in e es a es a e ob ained om u u es
p ices while he exchange a e is assumed o be cons an h ough he o ecas ho izon, consis en
wi h he p edic ion om a andom walk model. No e ha hese condi ioning assump ions a e
comple ely exogenous, he e o e no aec ed by he ECB p ojec ions published in he same qua e .
In o de o assess he ole o ex e nal assump ions in ina ion p ojec ions, we add he o ecas
e o s in ex e nal assump ions as addi ional p edic o s in he bias eg ession models (1) and (3).
Dene he e o in he ex e nal assump ion
i
a ime
o ho izon
h
as:
ζi
,h =xi
−xi
| −hi= 1, .., 3; h= 0, ..8
(5)
whe e
xi
is he ealiza ion and
xi
| −h
is he alue assumed a
−h
o da e
. The a iable
xi
is he le el o in e es a e o exchange a e, o he g ow h a e o oil p ices. Fi s , we es whe he
he e o s in he ex e nal assump ions a e biased in bo h he uncondi ional and s a e dependen
cases:
ζi
,h =αi
0+ei
,h i= 1, .., 3; h= 0, ..8
(6)
ζi
,h =αi
0+αi
1,hd −h+ei
,h i= 1, .., 3; h= 0, ..8
(7)
13
Kon ogeo gos and Lamb ias (2019) nd ha e o s in hese ex e nal assump ions dec ease he accu acy o he
ECB p ojec ions o ina ion.
13

whe e
d −h
is he dummy cons uc ed as in 4.1. I we we e o nd some sys ema ic bias in he
condi ional model, one could a gue ha i explains he sys ema ic bias obse ed in he ina ion
o ecas e o s. Then, we un he ollowing eg ession, which includes he e o
ζi
,h
as an addi ional
p edic o :
 ,h =βi
0,h +βi
1,hd −h+βi
2,hζi
,h +ui
,h i= 1, .., 3; h= 0, ..8
(8)
whe e,
 ,h
is he o ecas e o o p edic ing ina ion a o ecas o igin
o ho izon
h
. I he
es ima ed pa ame e s
βi
0,h
and
βi
1,h
we e no signican in (8), while
βi
2,h
we e signican , he bias
in he ina ion o ecas s could be a ibu ed o he bias in he ex e nal assump ions.
Ideally, o ca y ou ou in es iga ion, we would  s cons uc a coun e ac ual se ies o
adjus ed
e o s o ina ion by using in e nal ECB models condi ioned on he ue ealiza ion o he ex e nal
a iables, and hen check whe he he adjus ed e o s a e biased. Howe e , we a e unable o
cons uc such coun e ac ual se ies based on he o ecas ing models used in eal ime o o include
he ECB expe judgemen . We hink ha ou simple app oach is sensible because i sys ema ic
e o s in he ex e nal assump ions we e d i ing he o ecas e o s, hen he dummy should no ha e
addi ional explana o y powe .
Resul s o eg essions (6) and (7) a e displayed in Table (4) o he Appendix. Panel A shows
ha in e cep e m in he eg essions o he sho e m a e is nega i e, meaning ha on a e age
he sho e m a e is assumed o be la ge han he ealized alue. This holds a e e y ho izon,
al hough he coecien s a e signican only a longe ho izons. In he s a e dependen eg essions
he cons an is s ill nega i e and signican , while he dummy is posi i e bu no s a is ically
signican ly die en om ze o, sugges ing ha he e is sys ema ic bias only when ina ion is
abo e a ge . The bias is la ge in absolu e e ms, up o
70
basis poin s, when ina ion is abo e
a ge .
The esul s o he exchange a e show no e idence o bias on a e age (Panel B). Howe e ,
when ina ion is abo e (below) he a ge he exchange a e e sus he US dolla is assumed o be
weake (s onge ) han ealized. The coecien s associa ed wi h he cons an and he dummy a e
signican o all ho izons highe han wo and he bias inc eases wi h he ho izon.
Finally, he oil p ice assump ions a e no biased on a e age. In con as , he s a e dependen
analysis shows ha a e y sho ho izons he e is e idence o unde p edic ion (o e p edic ion) when
14
ina ion is highe (lowe ) han he a ge . This is consis en wi h he sys ema ic bias obse ed o
ina ion. Howe e , he size o he bias is small, amoun ing o 0.06p.p. a mos .
Resul s o eg essions (8) a e epo ed in Panel C1 h ough C3 o Table (1). Rega dless o he
ex e nal assump ion conside ed, adding he o ecas e o s o he bias eg ession does no al e he
sign, he magni ude no he signicance o he coecien s obse ed in Panel B o Table (1). The
in e es a e o ecas e o s a e signican only a ho izons one and wo and en e posi i ely, so ha
a la ge o ecas e o o in e es a es p edic s a la ge o ecas e o o ina ion. Simila ly, in
he bias eg ession augmen ed by he exchange a e o ecas e o s, hese e o s a e no signican .
Finally, he oil p ice o ecas e o s a e highly signican a sho e m ho izons and exhibi a posi i e
coecien , implying ha a highe o ecas e o o oil p ices ansla es in o a highe o ecas e o
o ina ion. While he magni ude o he coecien is again unchanged, he cons an and he
dummy a e signican only o h ee qua e s ahead o ecas s.
In sum, we nd e idence o sys ema ic o e (unde )p edic ion o medium e m p ojec ions when
ina ion is lowe (highe ) han he a ge e en when con olling o he e o s in ex e nal assump-
ions.
5 Conclusion
Using a conden ial da ase o ECB mac oeconomic qua e ly p ojec ions we documen h ee
no el ndings ha ela e he bias and eciency o he ECB ina ion o ecas s o he le el o ina ion
a he ime o o ecas ing: (i) a sys ema ic bias owa ds he a ge a medium ho izons, which implies
o e (unde ) p edic ion when ina ion is low (high); (ii) a la ge bias when ina ion is abo e he
a ge ; (iii) ineciency, esul ing in unde u iliza ion o new in o ma ion a medium ho izons when
ina ion is abo e a ge .
Ou esul s sugges ha looking a s a e dependence is c ucial. We a gue ha heo e ical
models a e unable o eplica e he obse ed s a e dependence in bias and eciency and hence, hey
migh need o be e o mula ed in o de o ma ch hese ea u es o he o ecas s. Gi en he lack
o heo e ical models ha can simul aneously accoun o s a e dependen bias and ineciency in
cen al banks' o ecas s, his pape has he po en ial o s a a line o heo e ical esea ch ha
a ionalizes hese ndings.
15
Ou in es iga ion o he p ope ies o he published p ojec ions is ele an also om a policy
pe spec i e, in pa icula o he ongoing ECB s a egy e iew: he o ecas s o m he basis o he
mone a y policy decisions o he ECB Go e ning Council and ep esen an impo an communica ion
ool. While he easons behind ou ndings a e beyond he scope o his pape , we conjec u e
ha he e idence o s a e dependen bias and ineciency migh be ela ed o s a egic mo i es
in mone a y policy communica ion. These a e pa icula ly s ong a he in e media e ho izons,
eec ing he medium- e m o ien a ion o mone a y policy making.
16
Figu e 1: HICP Ina ion and Ina ion P ojec ions
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
*
(a) Panel A: HICP Ina ion and Dis ibu ion o S a es Ac oss Time
01234567891011
o ecas ing ho izon
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
o ecas s
(b) Panel B: ECB P ojec ions o HICP Ina ion: Pa hs
Panel A. The gu e shows he yea -o e -yea HICP ina ion o he Eu o A ea and he pe iods in which
ina ion is below he a ge , i.e. 1.8% (shaded a eas). Panel B. The gu e shows o each o ecas ing ho izon
he maximum and minimum ina ion p ojec ions, he
10 h
and
25 h
(
75 h
and
90 h
) pe cen iles and he medians
condi ional on whe he ina ion was abo e o below he a ge (1.8%, g een dashed line) du ing each p ojec ion
exe cise.
17
Table 1: Bias in ECB Ina ion Fo ecas s
Fo ecas Ho izon
h= 0 1 2 3 4 5 6 7 8
Panel A: O e all Sample
c
-0.02 0.03 0.07 0.11 0.13 0.14 0.11 0.07 0.02
(0.01) (0.04) (0.08) (0.11) (0.14) (0.15) (0.15) (0.16) (0.16)
Panel B: Asymme y
c
-0.03* 0.09* 0.19** 0.32*** 0.37** 0.35* 0.31 0.22 0.14
(0.02) (0.05) (0.09) (0.13) (0.16) (0.18) (0.19) (0.20) (0.20)
d
0.03 -0.11 -0.25* -0.38* -0.44* -0.38 -0.34 -0.25 -0.16
(0.03) (0.08) (0.15) (0.20) (0.27) (0.29) (0.30) (0.32) (0.33)
Panel C1: Sho Te m In e es Ra e
c
-0.03* 0.11** 0.23** 0.37*** 0.44** 0.46** 0.42** 0.33 0.26
(0.02) (0.05) (0.09) (0.13) (0.17) (0.19) (0.20) (0.21) (0.22)
ζ
-0.12 0.36** 0.37** 0.27 0.24 0.25 0.21 0.17 0.17
(0.20) (0.13) (0.14) (0.13) (0.12) (0.09) (0.08) (0.07) (0.07)
d
0.03 -0.14* -0.28** -0.41** -0.48* -0.44 -0.39 -0.28 -0.18
(0.03) (0.08) (0.14) (0.20) (0.26) (0.28) (0.30) (0.31) (0.32)
Panel C2: Exchange Ra e
c
-0.03 0.10** 0.19** 0.32** 0.34** 0.26 0.15 0.02 -0.10
(0.02) (0.05) (0.09) (0.13) (0.17) (0.19) (0.19) (0.19) (0.18)
ζ
-0.72 -0.94 0.16 0.11 0.65 1.62 2.49 2.93 3.18
(1.00) (0.73) (0.88) (1.04) (1.17) (1.21) (1.16) (1.03) (0.93)
d
0.03 -0.13* -0.24 -0.37* -0.38 -0.21 -0.03 0.14 0.28
(0.03) (0.08) (0.15) (0.22) (0.28) (0.31) (0.31) (0.30) (0.30)
Panel C3: Oil P ices
c
-0.02 0.02 0.12 0.25** 0.29* 0.30* 0.27 0.19 0.10
(0.02) (0.04) (0.09) (0.13) (0.17) (0.17) (0.18) (0.18) (0.18)
ζ
0.78*** 1.66*** 1.24*** 1.18* 1.39 1.19 1.01 1.41 1.51
(0.30) (0.26) (0.45) (0.64) (0.85) (0.84) (0.86) (0.91) (0.93)
d
0.01 -0.05 -0.18 -0.35* -0.41 -0.40 -0.37 -0.25 -0.15
(0.03) (0.07) (0.14) (0.20) (0.26) (0.26) (0.28) (0.29) (0.29)
No e: Es ima ed coecien s om eg essions (1), (3) and (8). Newey-Wes s anda d e o s a e in pa en he-
sis. S a s deno e he 10% (*), 5% (**) and 1% (***) signicance le el. Fo eg essions in ol ing he sho
e m in e es a es, no e ha be o e 2006 an assump ion o cons an sho e m in e es a es was used.
18

Table 2: Eciency in ECB Ina ion Fo ecas s
Fo ecas Ho izon
h =
0 1 2 3 4 5 6 7
Panel A: O e all Sample
c
-0.02 0.02 0.05 0.09 0.12 0.12 0.09 0.04
(0.01) (0.04) (0.08) (0.12) (0.16) (0.16) (0.17) (0.17)
,h
0.03 0.11 0.20 0.36 -0.09 -0.59 -0.26 -0.32
(0.05) (0.10) (0.21) (0.30) (0.78) (0.95) (0.87) (0.91)
Panel B: Asymme y
c
-0.04* 0.06 0.16* 0.28** 0.37** 0.35** 0.30 0.20
(0.02) (0.05) (0.09) (0.12) (0.16) (0.18) (0.19) (0.20)
,h
0.04 0.17 0.22 0.56* 0.11 -0.50 -0.18 -0.70
(0.06) (0.13) (0.24) (0.31) (0.78) (1.14) (1.09) (1.28)
d
0.03 -0.09 -0.22 -0.34* -0.48* -0.40 -0.38 -0.27
(0.03) (0.08) (0.14) (0.19) (0.27) (0.30) (0.31) (0.33)
,h ×d
0.00 -0.18 -0.06 -0.26 0.10 0.65 0.26 0.29
(0.09) (0.19) (0.36) (0.49) (1.37) (1.69) (1.55) (1.69)
No e: Es ima ed coecien s om eg essions (2) and (4). Newey-Wes s anda d e o s a e in
pa en hesis. S a s deno e he 10% (*), 5% (**) and 1% (***) signicance le el.
19
Re e ences
Alessi, L., E. Ghysel, L. Ono an e, R. Peach, and S. Po e (2014). Cen al Bank Mac oeconomic
Fo ecas ing Du ing he Global Financial C isis: The Eu opean Cen al Bank and Fede al Rese e
Bank o New Yo k Expe iences.
Jou nal o Business and Economic S a is ics 32
(4), 483500.
Angele os, G.-M., Z. Huo, and K. A. Sas y (2020). Impe ec Mac oeconomic Expec a ions: E i-
dence and Theo y. NBER Wo king Pape s 27308, Na ional Bu eau o Economic Resea ch, Inc.
A ai, N. (2016). E alua ing he eciency o he omc's new economic p ojec ions.
Jou nal o Money,
C edi and Banking 48
(5), 10191039.
Bo dalo, P., N. Gennaioli, Y. Ma, and A. Shlei e (2020). O e eac ion in mac oeconomic expec a-
ions.
Ame ican Economic Re iew 110
(9), 27482782.
Capis an, C. (2008). Bias in Fede al Rese e Ina ion Fo ecas s: Is he Fede al Rese e I a ional
o Jus Cau ious?
Jou nal o Mone a y Economics 55
, 14151427.
Cha emza, W. and D. Ladley (2016). Cen al banks' o ecas s and hei bias: E idence, eec s and
explana ion.
In e na ional Jou nal o Fo ecas ing 32
, 804817.
Clemen s, M. P., F. Jou z, and H. O. S ekle (2007). An e alua ion o he o ecas s o he ede al
ese e: a pooled app oach.
Jou nal o Applied Econome ics 22
, 121136.
Coibion, O. and Y. Go odnichenko (2015). In o ma ion igidi y and he expec a ions o ma ion
p ocess: A simple amewo k and new ac s.
Ame ican Economic Re iew 105
(8), 26442678.
Coibion, O., Y. Go odnichenko, S. Kuma , and M. Pedemon e (2020). Ina ion expec a ions as a
policy ool?
Jou nal o In e na ional Economics 124
, 226.
Duy, J. and F. Heinemann (2021). Cen al bank epu a ion, cheap alk and anspa ency as
subs i u es o commi men : Expe imen al e idence.
Jou nal o Mone a y Economics 117
, 887
903.
El-Shagi, M., S. Giesen, and A. Jung (2016). Re isi ing he ela i e o ecas pe o mances o ed s a
and p i a e o ecas e s: A dynamic app oach.
In e na ional Jou nal o Fo ecas ing 32
, 313323.
20
Ellison, M. and T. J. Sa gen (2012). A de ence o he FOMC.
In e na ional Economic Re-
iew 53
(34), 10471065.
Fawce , N., L. Koe be , R. Masolo, and M. Wald on (2015). E alua ing UK Poin and Densi y
Fo ecas s om an Es ima ed DSGE Model: he Role o O-Model In o ma ion O e he Financial
C isis. Wo king Pape s 538, Bank o England.
Fuh e , J. (2018). In insic Expec a ions Pe sis ence: E idence om P o essional and Household
Su ey Expec a ions. Wo king Pape 2018-09, Fede al Rese e Bank o Bos on.
Fujiwa a, I. (2005). Is he cen al bank's publica ion o economic o ecas s inuen ial?
Economics
Le e s 89
, 255261.
Gomez-Ba e o, S. and J. A. Pa a-Polania (2014). Cen al bank s a egic o ecas ing.
Con empo a y
Economic Policy 32
(4), 802810.
Ha mann, P. and F. Sme s (2018). The Fi s Twen y Yea s o he Eu opean Cen al Bank: Mon-
e a y Policy. Wo king Pape s 2219, Eu opean Cen al Bank.
He be , S. (2020). S a e-Dependen Cen al Bank Communica ion wi h He e ogeneous Belie s.
mimeo.
Holden, K. and D. A. Peel (1990). On es ing o unbiasedness and eciency o o ecas s.
The
Manches e School 58
(2), 120127.
Hube , P. (2014). Fomc o ecas s as a ocal poin o p i a e expec a ions.
Jou nal o Money,
C edi and Banking 46
(7), 13811420.
Hube , P. (2015). Do cen al bank o ecas s inuence p i a e agen s? o ecas ing pe o mance
e sus signals.
Jou nal o Money, C edi and Banking 47
(4), 771789.
Hube , P. (2017). Quali a i e and quan i a i e cen al bank communica ion and ina ion expec-
a ions.
B.E. Jou nal o Mac oeconomics 17
(1), 141.
I e sen, J., S. Laséen, S. L. H., and U. Söde s öm (2016). Real-Time Fo ecas ing o Mone a y
Policy Analysis: he Case o S e iges Riksbank. Wo king Pape s 318, S e iges Riksbank.
21
Kohlhas, A. N. and A. Wal he (2021). Asymme ic a en ion.
Ame ican Economic Re iew o h-
coming
.
Kon ogeo gos, G. and K. Lamb ias (2019). An Analysis o he Eu osys em/ECB P ojec ions. Wo k-
ing Pape s 2291, Eu opean Cen al Bank.
Lahi i, K. and X. Sheng (2008). E olu ion o o ecas disag eemen in a bayesian lea ning model.
Jou nal o Econome ics 144
, 325340.
Loungani, P., H. S ekle , and N. Tami isa (2013). In o ma ion igidi y in g ow h o ecas s: Some
c oss-coun y e idence.
In e na ional Jou nal o Fo ecas ing 29
, 605621.
Lyziak, T. and M. Palo ii a (2018). On he o ma ion o ina ion expec a ions in u bulen imes:
The case o he eu o a ea.
Economic Modelling 72
, 132139.
Messina, J., T. M. Sinclai , and H. S ekle (2015). Wha can we lea n om e isions o he g eenbook
o ecas s?
Jou nal o Mac oeconomics 45
, 5462.
Mince , J. and V. Za nowi z (1969, June).
The E alua ion o Economic Fo ecas s
, pp. 346. Eco-
nomic Fo ecas s and Expec a ions: Analysis o Fo ecas ing Beha io and Pe o mance. NBER
Books.
No dhaus, W. (1987). Fo ecas ing eciency, concep s, and applica ions.
Re iew o Economic and
S a is ics 69
, 667674.
Palo ii a, M., M. Haa io, P. Jalasjoki, and J. Kilponen (2021). Wha does "below, bu close o, wo
pe cen " mean? assessing he ecb's eac ion unc ion wi h eal ime da a.
In e na ional Jou nal
o Cen al Banking
, o hcoming.
Po e , S. (2011). The Failu e o Fo ecas he G ea Recession. Technical epo ,
h ps://libe ys ee economics.newyo k ed.o g/.
Rome , C. D. and D. H. Rome (2008). The omc e sus he s a: Whe e can mone a y policymake s
add alue?
Ame ican Economic Re iew 98
(2).
22
Table 7: Robus ness o Bias: Technical Recessions
Fo ecas Ho izon
h= 0 1 2 3 4 5 6 7 8
Panel A: O e all Sample
c
-0.02 0.03 0.07 0.11 0.13 0.14 0.11 0.07 0.02
(0.01) (0.04) (0.08) (0.11) (0.14) (0.15) (0.15) (0.16) (0.16)
Panel B: Asymme y
c
-0.02 0.07 0.14 0.25** 0.29* 0.32* 0.27 0.21 0.14
(0.02) (0.05) (0.09) (0.12) (0.16) (0.17) (0.18) (0.18) (0.19)
d
0.01 -0.11 -0.18 -0.28 -0.34 -0.45 -0.35 -0.30 -0.21
(0.03) (0.09) (0.18) (0.24) (0.31) (0.32) (0.34) (0.35) (0.36)
Coecien es ima es om eg ession (3). We dene he s a e acco ding o he phases o he business
cycle. Then, he dummy akes he alue o one du ing echnical ecessions, when ou pu g ow h is
below he 25 h quan ile o i s dis ibu ion.
Table 8: Bias in Published P ojec ions
Fo ecas Ho izon
cu en yea nex yea yea a e nex
Panel A: O e all Sample
c
0.02 0.14 0.12
(0.02) (0.12) (0.15)
Panel B: Asymme y
c
0.04 0.28** 0.18
(0.03) (0.14) (0.19)
d
-0.05 -0.35 -0.17
(0.05) (0.23) (0.30)
Coecien es ima es om eg ession (3).
29

Table 9: Eciency o ECB Ina ion Fo ecas s: Al e na i e Values o Ta ge
Fo ecas Ho izon
h =
0 1 2 3 4 5 6 7
Panel A:
π∗= 1.7
c
-0.04* 0.07 0.16* 0.28*** 0.36** 0.35** 0.30* 0.21
(0.02) (0.05) (0.09) (0.12) (0.16) (0.17) (0.18) (0.20)
,h
0.04 0.15 0.23 0.56** 0.21 -0.48 -0.33 -0.84
(0.06) (0.11) (0.21) (0.29) (0.71) (0.89) (0.84) (0.95)
d
0.03 -0.10 -0.22 -0.34* -0.47* -0.38 -0.42 -0.29
(0.03) (0.08) (0.14) (0.19) (0.27) (0.29) (0.30) (0.32)
,h ×d
0.01 -0.17 -0.05 -0.24 0.16 2.54 2.04 1.43
(0.09) (0.20) (0.38) (0.51) (1.59) (2.17) (1.85) (1.79)
Panel B:
π∗= 1.9
c
-0.06* 0.01 0.06 0.16 0.19 0.19 0.21 0.15
(0.02) (0.06) (0.11) (0.14) (0.20) (0.21) (0.22) (0.22)
,h
0.10 0.23 0.33 0.72** 0.18 -0.45 -0.16 -1.03
(0.07) (0.14) (0.27) (0.37) (0.90) (1.26) (1.23) (1.41)
d
0.06 0.03 0.02 -0.03 -0.01 0.04 -0.10 -0.08
(0.03) (0.08) (0.15) (0.20) (0.27) (0.30) (0.31) (0.33)
,h ×d
-0.06 -0.18 -0.10 -0.36 0.32 0.85 0.21 0.83
(0.10) (0.20) (0.37) (0.51) (1.39) (1.77) (1.62) (1.77)
No e: Es ima ed coecien s om eg essions (4). Newey-Wes s anda d e o s a e in pa en hesis.
S a s deno e he 10% (*), 5% (**) and 1% (***) signicance le el.
30
Table 10: Eciency o ECB Ina ion Fo ecas s: In o ma ion Se
Fo ecas Ho izon
h =
0 1 2 3 4 5 6 7
Panel A:
πI
−h=π −h| −h
c
-0.04** 0.06 0.16* 0.27** 0.36** 0.38** 0.32* 0.22
(0.02) (0.05) (0.09) (0.12) (0.16) (0.18) (0.19) (0.20)
,h
0.05 0.16 0.22 0.60** 0.13 -0.36 -0.04 -0.66
(0.06) (0.13) (0.24) (0.32) (0.80) (1.15) (1.08) (1.27)
d
0.05 -0.06 -0.22 -0.36* -0.51* -0.53* -0.46 -0.35
(0.03) (0.08) (0.15) (0.20) (0.27) (0.30) (0.31) (0.34)
,h ×d
0.03 -0.17 -0.13 -0.47 -0.45 -0.06 -0.25 0.02
(0.10) (0.20) (0.37) (0.50) (1.38) (1.70) (1.55) (1.70)
Panel B:
πI
−h=π −h
c
-0.05** 0.06 0.16 0.26** 0.36** 0.34* 0.26 0.16
(0.02) (0.05) (0.10) (0.13) (0.17) (0.19) (0.20) (0.21)
,h
0.06 0.16 0.25 0.65** 0.16 -0.42 -0.19 -0.82
(0.07) (0.14) (0.25) (0.34) (0.83) (1.19) (1.12) (1.30)
d
0.05 -0.08 -0.20 -0.32 -0.43 -0.35 -0.26 -0.14
(0.03) (0.08) (0.15) (0.20) (0.27) (0.30) (0.31) (0.33)
,h ×d
-0.01 -0.16 -0.16 -0.52 -0.16 0.42 0.23 0.51
(0.10) (0.20) (0.37) (0.50) (1.36) (1.73) (1.58) (1.72)
No e: Es ima ed coecien s om eg essions (4). Newey-Wes s anda d e o s a e in pa en hesis.
S a s deno e he 10% (*), 5% (**) and 1% (***) signicance le el.
31
Table 11: Eciency o ECB Ina ion Fo ecas s: Technical Recession
Fo ecas Ho izon
h =
0 1 2 3 4 5 6 7
Panel A: O e all Sample
c
-0.02 0.02 0.05 0.09 0.12 0.12 0.09 0.04
(0.01) (0.04) (0.08) (0.12) (0.16) (0.16) (0.17) (0.17)
,h
0.03 0.11 0.20 0.36 -0.09 -0.59 -0.26 -0.32
(0.05) (0.10) (0.21) (0.30) (0.78) (0.95) (0.87) (0.91)
h =
0 1 2 3 4 5 6 7
Panel B: Asymme y
c
-0.03 0.06 0.13 0.22** 0.28* 0.31* 0.26 0.18
(0.02) (0.04) (0.08) (0.11) (0.15) (0.17) (0.18) (0.19)
,h
0.01 0.00 0.07 0.37 -0.23 -0.04 0.39 -0.29
(0.05) (0.12) (0.23) (0.32) (0.84) (1.27) (1.07) (1.12)
d
0.01 -0.09 -0.13 -0.21 -0.32 -0.44 -0.36 -0.31
(0.03) (0.09) (0.16) (0.22) (0.30) (0.32) (0.34) (0.36)
,h ×d
0.09 0.31 0.48 0.35 1.33 0.04 -0.95 -0.47
(0.10) (0.20) (0.37) (0.51) (1.37) (1.71) (1.56) (1.67)
No e: Es ima ed coecien s om eg ession (4). We dene he s a e acco ding o he phases o
he business cycle. Then, he dummy akes he alue o one du ing echnical ecessions, when
ou pu g ow h is below he 25 h quan ile o i s dis ibu ion. Newey-Wes s anda d e o s a e in
pa en hesis. S a s deno e he 10% (*), 5% (**) and 1% (***) signicance le el.
32
Table 12: Eciency o Published
P ojec ions
Fo ecas Ho izon
cu en yea nex yea
Panel A: O e all Sample
c
0.02 0.12
(0.02) (0.09)
,h
0.00 1.39***
(0.03) (0.24)
Panel B: Asymme y
c
0.10*** 0.46***
(0.05) (0.09)
,h
-0.14** 0.30
(0.07) (0.28)
d
-0.10 -0.57***
(0.06) (0.16)
,h ×d
0.17* 1.04***
(0.09) (0.45)
No e: Es ima ed coecien s om eg es-
sion (4). Newey-Wes s anda d e o s a e in
pa en hesis. S a s deno e he 10% (*), 5%
(**) and 1% (***) signicance le el.
33