Aas ei , Knu A e; Jo e, Anne So ie; Ra azzolo, F ancesco
Wo king Pape
Fo ecas ing Recessions in Real Time
Wo king Pape , No. 2014/02
P o ided in Coope a ion wi h:
No ges Bank, Oslo
Sugges ed Ci a ion: Aas ei , Knu A e; Jo e, Anne So ie; Ra azzolo, F ancesco (2014) : Fo ecas ing
Recessions in Real Time, Wo king Pape , No. 2014/02, ISBN 978-82-7553-785-8, No ges Bank, Oslo,
h ps://hdl.handle.ne /11250/2495983
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/210051
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
2014 | 02
Fo ecas ing ecessions in eal ime
Wo king Pape
Mone a y Policy
Knu A e Aas ei , Anne So ie Jo e and F ancesco Ra azzolo
Wo king pape s a No ges Bank, a 1992/1 il 2009/2 kan bes illes o e e-pos :
se icesen e @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:
se icesen e @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 1502-8143 (online)
ISBN 978-82-7553-785-8 (online)
Fo ecas ing ecessions in eal ime∗
Knu A e Aas ei †Anne So ie Jo e‡F ancesco Ra azzolo§
Feb ua y 7, 2014
Abs ac
We e iew se e al me hods o de ine and o ecas classical business cycle u ning
poin s in No way. In he pape we compa e he B y - Boschan ule (BB) wi h a
Ma ko Swi ching model (MS), using al e na i e in ages o No wegian G oss Do-
mes ic P oduc (GDP) as he business cycle indica o . The iming o business cycles
depends on he in age and he me hod used. BB p o ides he mos easonable de -
ini ion o business cycles. The o ecas ing exe cise, whe e he models a e augmen ed
wi h su eys o inancial indica o s, espec i ely, leads o he conclusion ha he
BB ule applied o densi y o ecas s o GDP augmen ed wi h ei he he consume
con idence index o a inancial condi ions index p o ides he mos imely p edic-
ions o peaks. Fo oughs, augmen ing wi h su eys o inancial indica o s does
no inc ease o ecas abili y.
JEL-codes: C32, C52, C53, E37, E52
Keywo ds: Fo ecas densi ies; Tu ning Poin s; Real- ime da a
∗This Wo king Pape should no be epo ed as ep esen ing he iews o No ges Bank. The iews
exp essed a e hose o he au ho s and do no necessa ily e lec hose o No ges Bank. We hank Hilde
Bjø nland, Ma celle Chau e , Nicholas Fawce , Lau an Fe a a, Domenico Giannone, Ch is ian Schu-
mache and Shaun Vahey o help ul commen s. We also hank semina and con e ence pa icipan s a
he wo kshop on “New Me hods o Fo ecas ing Mac oeconomic Da a” a he Uni e si y o No ingham,
he 33nd Annual In e na ional Symposium on Fo ecas ing, he 7 h In e na ional Con e ence on Compu-
a ional and Financial Econome ics, he 9 h Annual CIRANO-CIREQ Wo kshop on “Da a Re ision in
Mac oeconomic Fo ecas ing and Policy”, Deu che Bundesbank and No ges Bank.
†No ges Bank, Knu -A e.Aas [email p o ec ed]
‡Co esponding au ho : No ges Bank, [email p o ec ed]
§No ges Bank and BI No wegian Business School, F ancesco.Ra [email p o ec ed]
1
1 In oduc ion
Sho - e m analysis in cen al banks and o he policy ins i u ions is in ended o p o ide
policy make s, and possibly a la ge audience, wi h assessmen s o he ecen pas and
cu en business cycle. Poin and densi y o ecas s o a iables o in e es a e o en
p o ided. Howe e , he analysis o cu en economic condi ions does no ely jus on
his in o ma ion, and he e is a long adi ion in business cycle analysis and ela ed
esea ch o sepa a ing pe iods o con ac ion om pe iods o expansion, see Schumpe e
(1954). Policy decisions a y depending on whe he he economy is in an expansiona y
o a ecessiona y pe iod. Mos o he esea ch has ocused on US da a, whe e he NBER
cycle is ega ded as he o icial e e ence cycle. Mos o he coun ies ha e no de ined
such an “o icial” da ing o classical business cycles.
In his pape we e iew some al e na i e me hods o de ine classical business cycle
u ning poin s o he No wegian economy. The No wegian business cycle is nei he ully
synch onized wi h he cycles o o he Scandina ian coun ies, no wi h he Eu opean o
he US cycles. One eason o his assyme y is No way’s posi ion as a small open econ-
omy wi h la ge expo s o ene gy (gas and oil) goods. We compa e he non-pa ame ic
B y - Boschan ule (BB) as in Ha ding and Pagan (2002) wi h he pa ame ic au o e-
g essi e Ma ko Swi ching model (MS) as in Hamil on (1989). Bo h me hods a e applied
o qua e ly mainland No wegian G oss Domes ic P oduc (GDP).1
Mac oeconomic da a, and in pa icula GDP, a e subjec o e isions o e ime.
Benchma k e isions, bu also e isions due o new in o ma ion may change GDP sub-
s an ially a e se e al yea s. Economic decisions a e no immune o such changes (see
O phanides and an No den (2002)). Hamil on (2011) also shows ha business cycle
da ing exhibi s impo an di e ences when al e na i e in ages o da a a e conside ed.
We es ima e business cycle u ning poin s using he me hods lis ed abo e, compa ing
esul s using ex-pos e ised da a and da a a ailable in eal ime.
Finally, da a a e eleased wi h subs an ial delays. Hence economic decisions mus
1Since he olume o expo s o oil and gas a e disconnec ed om he domes ic business cycle, we
exclude his om GDP.
2

ely on o ecas s o he missing ecen in o ma ion.
While he e is a la ge body o li e a u e on poin o ecas s (see Timme mann (2006)
o a ecen e iew), and a g owing li e a u e on densi y o ecas ing, esea ch on classical
u ning poin p edic ion, in pa icula o coun ies o he han he US, has been mo e
limi ed. See Anas e al. (2008) and Billio e al. (2012) o applica ions o he Eu o A ea.
Ma ko swi ching models can p oduce poin , densi y and u ning poin o ecas s o
GDP, whe eas o ecas s o ecen and u u e alues o GDP a e p oduced in o de o use
he BB ule o p edic u ning poin s. Bo h me hods can be augmen ed wi h addi ional
leading indica o s. A leading indica o is in e p e ed as a a iable ha summa izes he
common cyclical mo emen s o some coinciden mac oeconomic a iables in a imely
manne . We ocus ou analysis on wo classes o da a: inancial indices and su ey da a.
Ha ey (1989) was one o he i s au ho s o documen he ela ionship be ween
inancial a iables and mac oeconomic agg ega es, ocusing on he links be ween he
e m s uc u e and consump ion g ow h. Mo e ecen ly, Coch ane and Piazzesi (2005)
and Lud igson and Ng (2009) ha e ex ended such analysis. The e a e se e al s udies
documen ing ha he in o ma ion ob ained om su eys has high o ecas ing powe o
mac oeconomic a iables. See, o example, Thomas (1999), Meh a (2002), Fama and
Gibbons (1984), and Ang e al. (2007) o he use o quan i a i e su eys, and Hansson
e al. (2005), Abbe ge (2007), Cla e ia e al. (2007) and Lui e al. (2010a,b) o he
applica ion o quali a i e su eys. Speci ically, o No wegian da a Næs e al. (2011)
and Aas ei and T o ik (2012) documen he ole o inancial indica o s, and Ma insen
e al. (2014) apply su ey da a o o ecas No wegian economic agg ega es.
We ind ha he BB p o ides easonable es ima es o business cycles o e he las
h ee o ou decades, while business cycle down u ns es ima ed by he MS end o be
oo sho and migh no be de ec ed a all. When p edic ing business cycle peaks in eal
ime, inancial and su ey da a seem o add subs an ial p edic abili y. The BB applied o
densi y o ecas s augmen ed wi h ei he a inancial condi ion index, consume con idence
o No ges Bank’s egional ne wo k su ey p o ide supe io p edic ions compa ed o he
ones om Ma ko Swi ching models. T oughs a e p edic ed in a less imely manne .
3
In con as o o ecas ing peaks, augmen ing he models wi h inancial and su ey da a
does no imp o e ough p edic abili y.
The es o he pape is o ganized as ollows: he nex sec ion desc ibes he modeling
amewo k and discusses how business cycle u ning poin s a e de ined. The hi d sec ion
p esen s da a and da ing o business cycles in No way o e he pas ou decades. The
ou h sec ion ocuses on he p edic ion o u ning poin s in eal ime, desc ibes he
ecu si e o ecas ing exe cise and p esen s he esul s. Sec ion 5 concludes.
2 Business cycle da ing app oaches
Following Bu ns and Mi chell (1946), we de ine business cycles as luc ua ions in agg e-
ga e economic ac i i y. This is he classical business cycle cha ac e ized by peaks and
oughs, desc ibing de elopmen s in he le el o economic ac i i y ac oss many sec o s.
An al e na i e concep is he g ow h cycle. Economic luc ua ions a e cha ac e ized by
“high” o “low” g ow h, mos commonly ela i e o end g ow h. An a ac i e ea u e
o he classical business cycle is ha i is no necessa y o calcula e unobse ed end
g ow h. This is pa icula ly impo an when i comes o o ecas ing u ning poin s, since
he unce ain y in he measu emen o end g ow h is a i s highes a he end o he
ime se ies o commonly used wo-sided il e s.
Classical business cycles in he US a e de ined by he Business Cycle Da ing Com-
mi ee o he Na ional Bu eau o Economic Resea ch (NBER). The commi ee decides
when a u ning poin occu s, i.e. in which mon hs a ecession espec i ely s a s and
ends. Decisions a e made by delibe a ion based on a ailable da a, hence announcemen s
o u ning poin s a e no e y imely. The Decembe 2007 peak was announced Decem-
be 1, 2008 and he ollowing June 2009 ough was announced Sep embe 20, 2010. The
da ing o he u ning poin s is no mally no e ised.
A numbe o me hods ha e been sugges ed in o de o de elop mechanical algo i hms
o calcula ing he s a and end o ecessions, in pa icula o US da a whe e ecessions
de ined by he NBER se e as benchma ks. See Hamil on (2011) o a su ey. He e, we
4
concen a e on wo di e en me hods.
2.1 B y - B oschan
B y and Boschan (1971) desc ibe a me hod ha was able o (almos ) eplica e he
business cycles in he US as measu ed by he da ing commi ee o he NBER. Ha ding
and Pagan (2002) build on he wo k by B y and Boschan and de elop an algo i hm o
de ec ing u ning poin s in qua e ly da a. The p ocedu e picks po en ial u ning poin s
and subjec s hem o condi ions ha ensu e ha ele an c i e ia o business cycles a e
me .
In he i s s ep, he BB p ocedu e iden i ies a po en ial peak in a qua e i he alue
is a local maximum. Co espondingly, a po en ial ough is iden i ied i he alue is a
local minimum. Sea ching o maxima and minima o e a window o 5 qua e s seems
o p oduce easonable esul s. A e po en ial u ning poin s a e iden i ied, he choice o
inal u ning poin s depends on se e al ules o ensu e al e na ing peaks and oughs and
minimum du a ion o phases and cycles. Fo mally, de ini ions o peaks can be w i en
∧ = 1{(y −2, y −1)< y >(y +1, y +2)}(1)
Co espondingly o oughs:
∨ = 1{(y −2, y −1)> y <(y +1, y +2)}(2)
When o ecas ing peaks and oughs, he alues on he igh -hand side o he equa ions
a e eplaced by o ecas s. Fo mally:
∧ = 1{(y −2, y −1)< y , P ob(y +1, y +2)< y )>0.5}(3)
and
∨ = 1{(y −2, y −1)> y , P ob(y +1, y +2)> y )>0.5}(4)
The business cycle can be in e p e ed as a s a e S , which akes he alue 1 in
expansions and 0 in ecessions. Tu ning poin s occu when he s a e changes. The
5
ela ionship be ween he business cycle and he local peaks and oughs can be w i en
as
S =S −1(1 − ∧ −1) + (1 −S −1)∨ −1) (5)
I he economy is in an expansion, S −1= 1. I no peak occu ed in ( -1), hen ∧ −1= 0
and i ollows ha he s a e S = 1. On he o he hand, i he e is a peak in ( -1) hen
∧ −1= 1 and he s a e changes o S = 0. The s a e will emain a 0 un il a ough is
de ec ed.
2.2 Ma ko Swi ching
The e is a long adi ion o using nonlinea models o cap u e he asymme y and he
u ning poin s in business cycle dynamics. Among such classes o models, Ma ko
Swi ching (MS) models (see o example Gold eld and Quand (1973), Hamil on (1989),
Clemen s and K olzig (1998), Kim and Mu ay (2002), Kim and Pige (2002), and
K olzig (2000) o u he ex ensions) a e dominan . We conside an au o eg essi e MS
model o GDP g ow h simila o Hamil on (1989), whe e only he in e cep is allowed
o swi ch be ween egimes:
y =νs +φ1y −1+. . . +φpy −p+u , u
i.i.d.
∼ N(0, σ2) (6)
= 1, . . . , T, whe e νs is he MS in e cep , φl, wi h l= 1, . . . , p, a e he au o eg essi e
coe icien s; and {s } is he egime-swi ching p ocess, ha is an m-s a es e godic and
ape iodic Ma ko -chain p ocess. This p ocess is unobse able (la en ) and s ep esen s
he cu en phase, a ime , o he business cycle (e.g. con ac ion o expansion). The e-
o e, he MS model does no equi e knowledge o y +1 and y +2, as he BB ule does, o
de ine he cycle a ime . The la en p ocess akes in ege alues, say s ∈ {1, . . . , m},
and has ansi ion p obabili ies P(s =j|s −1=i) = pij, wi h i, j ∈ {1, . . . , m}. The
ansi ion ma ix Po he chain is
P=




p11 . . . p1m
.
.
..
.
.
pm1. . . pmm





6
Figu e 2. Business cycle da ing using in age published Feb ua y 2011
(a) BB (b) MS
The loga i hmic ans o ma ion o GDP published in Feb ua y 2011. The shaded a eas indica e ecessions
da ed by he B y - Boschan (BB) ule ( he le hand side) and he Ma ko Swi ching (MS) model ( he
igh hand side).
Figu e 3. Business cycle da ing using Vin age published Feb ua y 2010
(a) BB (b) MS
The loga i hmic ans o ma ion o GDP published in Feb ua y 2010. The shaded a eas indica e ecessions
da ed by he B y - Boschan (BB) ule ( he le hand side) and he Ma ko Swi ching (MS) model ( he
igh hand side).
13

Figu e 4. Unemploymen a e and business cycle da ing (BB) based on GDP
Unemploymen a e (labo o ce su ey). The shaded a eas indica e ecessions da ed by he BB ule.
ecession is de ined by bo h me hods. The MS me hod de ines a e y sho ecession, 3
qua e s, wi h a peak in 2008Q2. The BB me hod de ines he peak in he same qua e ,
bu he ough does no occu un il he second qua e o 2009 - a ecession las ing 5
qua e s.
Wi h he wo olde in ages, he MS me hod de ines a one-qua e long ecession
in 2002Q4, unde ec ed in he in age published in Feb ua y 2012. Fu he mo e, he
ecession in 2008-2009 de ined by he MS me hod is one qua e longe when using he
wo oldes in ages, while i is wo qua e s sho e de ined by he BBQ me hod when
using he oldes in age.
An al e na i e way o assessing how easonable he u ning poin s a e, is o compa e
hei iming wi h de elopmen s in he unemploymen a e, which has he ad an age
o no being e ised, see igu e 4. Excep o he mild ecession in he ea ly 1980,
de elopmen s in he unemploymen a e suppo s he iming o he h ee peaks de ined
by he BB me hod. We would no expec he same cong uence o he iming o he
oughs, since p oduc ion no mally picks up be o e unemploymen imp o es. The ough
in unemploymen occu s in he hi d qua e o 1987, suppo ing he iming o a peak in
1987Q2 as de ined by he BB me hod. In he nex ecession, he unemploymen a e has a
14
ough in he second qua e o 2001, again suppo ing he iming o he peak in 2001Q1.
In con as o he wo down u ns discussed abo e, he ough in unemploymen in he
second qua e o 2008 occu s in he same qua e as he peak de ined o his ecession.
The labo ma ke no mally con ains mo e ic ions han he goods- and se ice ma ke s,
leading o a delayed eac ion o shocks in he labo ma ke . The eason o he mo e
speedy eac ion in he labo ma ke in 2008 could be a ea u e o he da a equency.
Wi h mon hly a iables, labo ma ke eac ions could s ill be delayed compa ed wi h
goods- and se ice ma ke eac ions. Bu i is also possible ha his can be explained
by he na u e o he shocks ha occu ed du ing he hi d qua e o 2008.5
4 Fo ecas ing No wegian u ning poin s in eal ime
I we in e p e he u ning poin s de ined by he qua e ly e sion o he B y-Boschan
me hod as desc ibing u ning poin s in he “ ue” No wegian classical business cycles,
one in e es ing ques ion is hen: Is i possible o p edic he u ning poin s in eal ime?
4.1 Fo ecas ing exe cise
In his exe cise we will concen a e on he la es ecession. P io o No embe 2011,
e isions o seasonally adjus ed na ional accoun s al e ed da a as a back as in he ea ly
1980s. Hence, da a in ages om he pe iod p io o No embe 2011 a e inc easingly
di e en om he la es published in age as we mo e backwa ds in ime. We would
no expec , using eal- ime da a, o be able o p edic u ning poin s a back in ime
de ined by he la es a ailable in age. This is also o some ex en ue o he la es
ecession, since da a a e inal only a e h ee yea s o e isions. Howe e , hese a e
e isions based on new in o ma ion and we canno a oid aking his in o accoun .
5Labo ma ke a iables a e commonly used o de ec ing u ning poin s. F om igu e 4, we would
expec he unemploymen a e o be use ul o de ec ing peaks, bu no o de ec ing oughs. We
augmen ed he Ma ko Swi ching model wi h se e al labo ma ke measu es; hou s wo ked, employmen
and wo al e na i e unemploymen a es. These did no imp o e he imeliness o u ning poin de ec ion
compa ed o inancial ma ke a iables.
15
We will apply he wo me hods desc ibed in sec ion 2using eal- ime da a and com-
pa e hei abili y o o ecas u ning poin s. The Ma ko swi ching echniques al eady
compu e p edic ed p obabili ies o being in one egime o he o he (i.e. in ecession
o expansion). The qua e ly B y-Boschan equi es p edic ions o GDP o be able o
o ecas a u ning poin in eal ime. We p oduce p edic i e densi ies o GDP om an
au o eg essi e model AR(p), whe e we ix p= 4 as in he Ma ko swi ching example.
Bo h he BB and he MS can be augmen ed wi h leading indica o s. A leading
indica o is in e p e ed as a a iable ha summa izes he common cyclical mo emen s
o some coinciden mac oeconomic a iables in a imely manne .
Fo augmen ing he BB, we ollow Aas ei e al. (2011) and p oduce p edic i e
densi ies om di e en bi a ia e ec o au o eg essi e models using GDP and one single
leading indica o :
Y =α+
p
X
i=1
βiY −i+ ,  ∼N(0,Σ),(9)
whe e Y = (y1, , y2, ) and y1, and y2, deno es GDP g ow h and g ow h in he leading
indica o , espec i ely.
Se e al pape s, such as E ans (2005), Giannone e al. (2008) and Aas ei e al.
(2014), show ha accoun ing o he imeliness o da a is c ucial o nowcas ing and
sho - e m o ecas ing. The exploi a ion o imely in o ma ion leads o imp o emen in
he o ecas accu acy. I is he e o e essen ial o ake in o accoun ha leading indica o s
a e a ailable p io o he GDP elease.6The models a e pu in o a s a e space o m
and Kalman il e echniques can easily be applied o deal wi h missing da a, i.e. he
unbalanced da a p oblem.
The o ecas unce ain ies a e ob ained h ough simula ions, whe e he inal densi ies
a e de i ed using ke nel smoo hing echniques. By applying he Kalman il e we can
6Some o he leading indica o s included he e a e o mon hly equency, and we need o b idge he
mon hly indica o s wi h qua e ly GDP. This is done by cons uc ing qua e ly a e ages o he mon hly
se ies. I a mon hly se ies only con ains one o wo mon hs o a qua e , we simply cons uc he a e age
o he one o wo obse a ions om he qua e o in e es , see Ba igi e al. (2004) and Angelini e al.
(2011) o a mo e de ailed discussion o al e na i e b idge equa ions.
16
ob ain condi ional o ecas s when he da a se is unbalanced. Mo e p ecisely, we use
he smoo hed co a iance ma ix o he p edic o s, which will esemble he mean squa ed
e o (MSE) ma ix o he sys em, and d aw om he no mal dis ibu ion o ob ain
simula ed o ecas s o each ho izon.7The median alue o he p edic i e densi ies is
hen used o ex end he eal- ime GDP le el wi h he o ecas s.8The median o ecas s
can be di ec ly compa ed wi h de ining ecessions using he MS when egime 1 has 50%
(o highe ) p obabili y. Since qua e ly GDP is eleased wi h a lag o app oxima ely 7
weeks, his means ha i we add o ecas s o 2 qua e s (i.e. a nowcas and a o ecas )
o he la es a ailable in age, we may a he ea lies p edic a u ning poin 7 weeks
a e i occu ed. I we add a h ee-s eps ahead o ecas , i would in heo y be possible
o o ecas a u ning poin 5 weeks be o e i occu s. As unce ain y inc eases wi h he
ho izon we will, howe e , con ine ou sel es o p edic i e densi ies one and wo s eps
ahead.
We employ leading indica o s ans o med in o di e ences as exogenous a iables in
he augmen ed MS au o eg essi e models. As discussed in he sec ion 2.2, he MS does
no equi e p edic ions o y +h, h = 1,2 o p oduce an indica o a ime .
We ocus ou analysis on wo classes o da a: inancial indices and su ey da a. Indi-
ca o models based on inancial da a and su ey da a a e likely candida es o de ec ing
u ning poin s ea ly. Publica ion is imely compa ed o GDP, and he na u e o he
s a is ics ensu es ha a wide ange o in o ma ion and conside a ions a e aken in o
accoun by inancial ma ke pa icipan s, see Næs e al. (2011), and by he esponden s
in he su eys, see Ma insen e al. (2014). Fo inancial da a we use mon hly a e ages
o daily obse a ions. All he su eys a e qua e ly. The e a e no mon hly su eys in
No way ha ha e been published long enough o be use ul o model-based o ecas ing.
They a e, howe e , eleased ea lie han GDP, hence indica o s a e in gene al a ailable
o qua e , while GDP is only a ailable o qua e −1. We ha e chosen wo inancial
7See o example L¨u kepohl (2005).
8Gene ally, comple e p obabili y dis ibu ions o e ou comes p o ide in o ma ion help ul o making
economic decisions; see, o example, he discussions in G ange and Pesa an (2000), Timme mann
(2006) and Gnei ing (2011).
17
a iables and h ee al e na i e su eys. The inancial indices a e:
•Financial condi ions index (FCI). The index is cons uc ed using a mon hly dy-
namic ac o model wi h inancial a iables, including in e es a es, money and
c edi and in e es a e sp eads.
•Rela i e sp ead (RS). The RS is a measu e o liquidi y and is calcula ed as he
quo ed sp ead as a ac ion o he midpoin p ice (mon hly a e age) and measu es
he implici cos o ading (a small numbe o ) sha es a he Oslo s ock exchange.
The su eys a e:
•The o e all business con idence indica o om he business endency su ey o
manu ac o ing, mining and qua ying (BTS). The su ey is conduc ed by S a is ics
No way in he las h ee weeks o he qua e and is published a he end o he
i s mon h in he nex qua e .
•The o e all consume con idence index (CC). The su ey is conduc ed by TNS
Gallup in he i h week o he qua e and is published a ound 4 weeks be o e he
end o he qua e .
•Expec ed g ow h in 6 mon hs (all indus ies), om No ges Bank’s egional ne wo k
su ey (RN). The su ey is conduc ed in he i s hal o he qua e and published
a ound h ee weeks be o e he end o he qua e .
4.2 Resul s
Resul s a e p esen ed in ables 1 o 3and summa ized in igu e 5. F om able 1, we
obse e ha all he augmen ed BB models p edic 2008Q3 as he peak qua e in eal
ime. The i s model o p edic he down u n is he model wi h he inancial condi ions
index (FCI) as he augmen ing a iable. As soon as da a o he second mon h o he
ou h qua e is known, his model pinpoin s a peak in he hi d qua e . All h ee
su ey models p edic he peak as soon as he su ey o he ou h qua e o 2008 is
18

Table 1. Fo ecas ing u ning poin s in eal ime - peaks
Model da e o de ec ion Peak qua e
BB wi h AR(4) 2009 19 May 2008Q3
BB wi h Consume con idence (CC) 2008 2 Decembe 2008Q3
BB wi h Business Tendency Su ey (BTS) 2009 28 Janua y 2008Q3
BB wi h Regional Ne wo k Su ey (RN) 2008 17 Decembe 2008Q3
BB wi h Financial condi ions index (FCI) 2008 1 Decembe 2008Q3
BB wi h RS 2009 1 Ma ch 2008Q3
Ma ko swi ching wi h AR(4) 2009 19 May 2008Q2
Ma ko Swi ch wi h change in CC 2009 1 June 2008Q1
Ma ko Swi ch wi h change in BTS 2009 27 Ap il 2008Q3
Ma ko Swi ch wi h change in RN 2008 17 Decembe 2007Q4
Ma ko Swi ch wi h change in FCI 2009 1 Ap il 2008Q1
Ma ko Swi ch wi h change in RS 2009 1 July 2008Q2
The able epo s he eal- ime p edic ed peak qua e and he exac da e o de ec ion using se e al
al e na i e me hods de ined in sec ion 2.
published. The consume con idence indica o is he i s su ey o be published on 2
Decembe , while he business endency su ey is no published un il 17 Janua y 2009.
Using he RS as he augmen ing ac o does no p oduce a e y imely o ecas o he
peak, since in his case he u ning poin in 2008Q3 is no o ecas ed un il 1 Ma ch 2009.
Wi h one excep ion, he Ma ko Swi ching models do no p edic a peak un il well
in o 2009. The MS model augmen ed wi h he No ges Bank’s egional ne wo k su ey
p edic s he peak when he su ey is published 17 Decembe . The p edic ed peak qua e
is as ea ly as 2007Q4, which seems oo ea ly o No way.
Ou esul s di e om Chau e and Pige (2008) who ind ha he Ma ko Swi ching
dynamic- ac o model p o ides mo e imely in o ma ion han he BB ule in US eal- ime
da ing. The di e en da a se , he al e na i e way o augmen ing he MS model and he
ac ha we augmen he BB ule wi h o ecas densi ies, meaning we can p edic up o
ime + 2, a e among he possible explana ions o he di e en esul s.
Table 2summa izes he peaks and oughs iden i ied in he p e ious sec ion using
19
Table 2. De ining u ning poin s ex pos
Me hod Peak qua e Leng h o down u n T ough qua e
BB 2008Q2 5 qua e s 2009Q3
Ma ko Swi ching 2008Q2 3 qua e s 2009Q1
The able epo s peak and ough qua e and leng h o he down u n using he in age published in
Feb ua y 2012 wi h wo al e na i e me hods.
ex-pos da a, mo e p ecisely he in age published in Feb ua y 2012. In eal ime he
peaks o ecas ed by BB occu 1 qua e la e han he peaks de ined ex pos using he
la es da a in age. Subs an ial da a e isions a e he main eason o he di e ence.
In igu e 6 we compa e he log le els o GDP in ages published in No embe 2008
and Feb ua y 2012, espec i ely. In he 2008 in age g ow h con inues h ough 2008,
al hough he g ow h a e is sligh ly nega i e in he i s qua e o 2008. In he Feb ua y
2012 in age, on he o he hand, g ow h is on a e age nega i e he i s h ee qua e s
o 2008.
Table 3con ains an o e iew o he models’ abili y o o ecas oughs in eal ime.
All he BB models o ecas a ough in 2009Q1 and mos o hem o ecas he ough
du ing he ollowing qua e . The i s models o p edic a ough a e he BB augmen ed
wi h AR(4) o ecas s and he BB augmen ed wi h ela i e sp ead. These models p edic
he u ning poin as soon as na ional accoun s o 2009Q1 we e published. The MS
models p o ide mo e mixed e idence, o en wi h subs an ial delays, wi h oughs anging
om 2008Q2 o 2009Q2.
Figu e 5 p o ides a g aphical summa y o all esul s. Since we do no ha e any
benchma k o compa e he esul s wi h, we discuss esul s in he ligh o wha seems
easonable gi en wha we know o economic de elopmen s. The aim is o o ecas u ning
poin s in eal ime as accu a ely as possible compa ed o an ex pos benchma k, and as
imely as possible.
The BB me hod used on he GDP le el appended wi h o ecas s gi en by su ey da a,
he RS o an AR(4) gi es consis en esul s, da ing he peak o 2008Q3 and he ough
20
Figu e 5. Tu ning poin o ecas s
12341234123412341234
B yBo schan wi h AR4
Ex pos
RT in
RT ou
B yBo schan wi h Consume Con idence (CC)
Ex pos
RT in
RT ou
B yBo schan wi h Business endency su ey (BTS)
Ex pos
RT in
RT ou
B yBo schan wi h Regional ne wo k su ey (RN)
Ex pos
RT in
RT ou
B yBo schan wi h Financial condi ions index (FCI)
Ex pos
RT in
RT ou
B yBo schan wi h sp ead (RS)
Ex pos
RT in
RT ou
Ma ko Swi cing wi h AR4
Ex pos
RT in
RT ou
Ma ko Swi cing wi h wi h Δ Consume Con idence (CC)
Ex pos
RT in
RT ou
Ma ko Swi cing wi h Δ Business endency su ey (BTS)
Ex pos
RT in
RT ou
Ma ko Swi cing wi h Δ Regional ne wo k su ey (RN)
Ex pos
RT in
RT ou
Ma ko Swi cing wi h Δ Financial condi ions index (FCI)
Ex pos
RT in
RT ou
Ma ko Swi cing wi h Δ sp ead (RS)
Ex pos
RT in
RT ou
2007
2008
2009
2010
2011
Fo each class o models, he igu e illus a es he leng h o he ecession in lines “Ex-pos ” using ex-pos
da a. Red deno es qua e s whe e he economy is in ecession. The lines deno ed “RT in” show which
qua e is he i s in he ecession in eal ime, i.e he peak qua e is he qua e be o e he qua e
ma ked in ed. In he lines deno ed “RT ou ”, he ed qua e is he ough qua e in eal ime. Black
e ical lines wi h eal- ime da a illus a e on which da e he u ning poin was de ec ed.
21
Table 3. Fo ecas ing u ning poin s in eal ime - oughs
Model da e o de ec ion T ough qua e
BB wi h AR(4) 2009 19 May 2009Q1
BB wi h Consume con idence (CC) 2009 1 June 2009Q1
BB wi h Business Tendency Su ey (BTS) 2009 28 July 2009Q1
BB wi h Regional Ne wo k Su ey (RN) 2009 10 June 2009Q1
BB wi h Financial condi ions index (FCI) 2009 1 July 2009Q1
BB wi h RS 2009 19 May 2009Q1
Ma ko swi ching wi h AR(4) 2009 19 May 2009Q1
Ma ko Swi ch wi h change in CC 2009 7 Decembe 2009Q2
Ma ko Swi ch wi h change in BTS 2010 27 Janua y 2009Q4
Ma ko Swi ch wi h change in RN 2008 17 Decembe 2008Q2
Ma ko Swi ch wi h change in FCI 2009 1 July 2009Q2
Ma ko Swi ch wi h change in RS 2010 1 Janua y 2009Q2
The able epo s he eal- ime p edic ed hough qua e and he exac da e o de ec ion using se e al
me hods.
o 2009Q1. Compa ed o using AR(4) o ecas s, he peak is p edic ed mo e imely, in
pa icula i using ei he Consume Con idence (CC) o a Financial Condi ions Index
(FCI) as an augmen ing a iable. The ough is in gene al o ecas ed wi h a much longe
delay han is he case when o ecas ing he peak. In con as o o ecas ing he peak,
AR(4) and RS a e now he mos imely al e na i e, i.e. o ecas abili y o he ough
does no inc ease by augmen ing he BB me hod by ei he su eys o inancial a iables.
The pu e Ma ko swi ching model pinpoin s 2008Q2 as he peak qua e , bo h in
eal ime and ex pos . Bu he eal- ime o ecas is no p oduced un il 19 May 2009.
By hen na ional accoun s o 2009Q1 a e a ailable and he accoun s o 2008 ha e been
e ised se e al imes. On he o he hand, he MS me hod o ecas s he ough in 2009Q1
a he same ime, which is as imely as he mos imely BB model. Augmen ing he
MS wi h any o he su eys, he inancial condi ion index o he RS, espec i ely, does
no seem e y help ul. These models a e no able o o ecas he u ning poin s in a
imely manne in eal ime. Peaks a e no o ecas ed un il well in o 2009. Fo ecas ed
22
Timme mann, A. (2006). Fo ecas combina ions. In G. Ellio , C. W. J. G ange , and
A. Timme mann (Eds.), Handbook o Economic Fo ecas ing, Volume 1, pp. 136–96.
Ams e dam: Else ie .
Ve maak, J., C. And ieu, A. Douce , and S. J. Godsil (2004). Re e sible jump Ma ko
chain Mon e Ca lo s a egies o Bayesian model selec ion in au o eg essi e p ocesses.
Jou nal o Time Se ies Analysis 25, 785–809.
29