The Poli ical Cos o Re o ms
Alessand a Bon…glioliy
IAE-CSIC and CEPR
Gino Ganciaz
CREI, UPF and CEPR
FIRST DRAFT: Janua y 2010
This d a : Sep embe 2010
Abs ac
This pape o malizes in a ully- a ional model he popula idea ha poli icians
pe cei e an elec o al cos in adop ing cos ly e o ms wi h u u e bene… s and ec-
onciles i wi h he e idence ha e o mis go e nmen s a e no punished by o e s.
To do so, i p oposes a model o elec ions whe e poli ical abili y is ex-an e unknown
and in es men in e o ms is unobse able. On he one hand, elec ions imp o e ac-
coun abili y and allow o keep well-pe o ming incumben s. On he o he , poli icians
make oo li le e o ms in an a emp o signal high abili y and inc ease hei eap-
poin men p obabili y. Al hough in a a ional expec a ion equilib ium o e s canno
be ooled and hence eelec ion does no depend on e o ms, he s a egy o unde in-
es ing in e o ms is none heless sus ained by ou -o -equilib ium belie s. Con a y o
he con en ional wisdom, unce ain y makes e o ms mo e poli ically iable and may,
unde some condi ions, inc ease social wel a e. The model is hen used o s udy how
poli ical ewa ds can be se so as o maximize social wel a e and he desi abili y o im-
posing a one- e m limi o go e nmen s. The p edic ions o his heo y a e consis en
wi h a numbe o empi ical egula i ies on he de e minan s o e o ms and eelec ion.
They a e also consis en wi h a new s ylized ac documen ed in his pape : economic
unce ain y is associa ed o mo e e o ms in a panel o 20 OECD coun ies.
JEL Classi…ca ion: E6, H3
Keywo ds: Elec ions, Re o ms, Asymme ic In o ma ion, Unce ain y.
We hank Albe o Alesina, Allan D azen, Geo gy Ego o , Giacomo Ponze o, Jaume Ven u a, Fab-
izio Zilibo i and semina pa icipan s a he NBER Summe Ins i u e 2010, SED 2009 Annual Mee ing
(Is anbul), EDP Jambo ee 2010 (Ba celona), Uni e si y o Helsinki, IMT in Lucca, IAE and IEW-Zu ich
Uni e si y o commen s and Tomaz Cajne o ese ach assis ance. We acknowledge …nancial suppo
om he Fundación Ramon A eces.
yIns i u e o Economic Analysis-CSIC, Campus UAB, 08193, Bella e a, SPAIN. E-mail:
alessand a.bon…[email protected]
zCREI, Uni e si a Pompeu Fab a, Ramon T ias Fa gas, 25-27, 08005, Ba celona, SPAIN. E-mail:
ggancia@c ei.ca
1
1 In oduc ion
Among he undamen al ques ions in poli ical economy a e why go e nmen s o en ail o
adop e o ms ha a e widely belie ed o be wel a e-imp o ing and wha condi ions make
hei adop ion mo e likely. Fo ins ance, while mos obse e s end o ag ee ha p omo ing
p oduc ma ke compe i ion, p o iding ee access o ma ke s and educing public deb
a e o en essen ial o p ese e economic g ow h, he ex en o which such measu es a e
adop ed a ies eno mously ac oss coun ies. The exis ing li e a u e has iden i…ed se e al
explana ions o an an i- e o m bias, wi h mos o hem placing dis ibu ional con‡ic s
as he co ne s one. Fo ins ance, in e es g oups who may lose om a e o m may lobby
o block i (e.g., G ossman and Helpman, 2001). Al e na i ely, unce ain y abou he
dis ibu ion o cos s and bene… s may lead o a s a us quo bias (e.g., Fe nandez and Rod ik,
1991) o o a wa o a i ion be ween pa ies esul ing in an ine¢ cien delay (Alesina and
D azen, 1991). A popula iew among policy make s, ins ead, is ha go e nmen s a e
a aid o losing o es due o he sho - un cos s ha e o ms ypically en ail. As Jean-
Claude Juncke , P ime Minis e o Luxembou g and P esiden o he Eu og oup, once said,
“We all know wha o do, bu we don’ know how o ge eelec ed once we ha e done i .”
E e since he wo k o Pel zman (1992), howe e , economis s and poli ical scien is s did
no …nd e idence ha e o ms a¤ec ad e sely he chances o eelec ion.1I go e nmen s
do no seem o be sys ema ically punished a he ballo box o engaging in e o ms, he
ques ion o why hey a e so poli ically di¢ cul emains an un esol ed puzzle.
In his pape , we o malize in a ully- a ional model he iew ha poli icians pe cei e
an elec o al cos in adop ing e o m and we econcile i wi h he e idence ha hei
adop ion does no seem o ha e a nega i e e¤ec on eelec ion p ospec s. We do so
using a model o poli ical accoun abili y h ough elec ions whe e he key elemen s a e
in o ma ional asymme ies and unce ain y. Simila ly o he exis ing li e a u e, we …nd
ha incumben go e nmen s ha e oo weak an incen i e o in es in e o ms. Con a y
o exis ing wo ks, howe e , we …nd ha unce ain y is likely o make e o ms poli ically
mo e iable. Ou heo y builds on he p emise ha he cos o e o ms is immedia ely
obse able o o e s, while hei ac ual implemen a ion and payo¤s o en a e no . Fo
example, an inc ease in …scal p essu e is immedia ely e iden o axpaye s, while i may
ake ime o see how e¤ec i ely ax e enues a e spen . In such a si ua ion, he incumben
may no da e o emba k in e o ms, e en when hey a e unambiguously wel a e-imp o ing,
o he ea o losing o es and hence o¢ ce. This bias agains e o ms holds, e en when
ci izens a e a ional and awa e o poli ical incen i es, unde wo condi ions. The … s is
1We discuss his li e a u e mo e in de ails in he nex Sec ion.
2
ha abili y o poli icians is no di ec ly obse able o o e s, so ha i mus be in e ed
on he basis o pe o mance. The second is an in o ma ional asymme y be ween ci izens
and he incumben so ha he ac ual in es men in e o ms wi h u u e e u ns is no
obse ed by o e s. These assump ions gene a e a “poli ical cos ” o e o ms: he e is
an incen i e o in es less in e o ms in an e¤o o signal high abili y and hus inc ease
he eelec ion p obabili y. Despi e his, howe e , in a a ional-expec a ion equilib ium
o e s co ec ly o esee he s a egy o he incumben and also his expec ed abili y. As
a esul , poli icians canno manipula e hei eelec ion p obabili y. S ill, hei choice
o unde in es is sus ained by hidden in o ma ion ou o equilib ium: he ac ha he
poli ician can de ia e om his equilib ium s a egy in ways unknown o o e s.2
In e es ingly, he “poli ical cos ”o e o ms depends c ucially on he sou ces o unce -
ain y a¤ec ing he p ecision o he signal ha o e s can see. I obse able measu es o
pe o mance a e poo signals o abili y, o ins ance because he economy is going h ough
a pe iod o high u bulence, he pe cei ed p obabili y o being eelec ed becomes less
sensi i e o he choice o e o ms he eby educing hei poli ical cos . Thus, pe haps su -
p isingly, he ype o unce ain y highligh ed in his pape p omo es unambiguously he
adop ion o e¢ cien e o ms. Ye , he wel a e e¤ec s a e ambiguous because unce ain y
wo sens bo h elec o al accoun abili y, he eby inducing he incumben o pu less e¤o ,
and he abili y o selec good poli icians. By compa ing hese e¤ec s, we …nd a simple
condi ion o wel a e o inc ease o dec ease wi h a ious o ms o unce ain y.
The high ac abili y o ou model allows us o use i o answe wo subs an i e no -
ma i e ques ions. Fi s , we s udy he e¤ec o poli ical ewa ds on social wel a e. By
inc easing he alue o s ay in powe , highe ewa ds exace ba e he unde in es men in
e o ms, bu also induce he incumben o exe mo e e¤o . We cha ac e ize he socially
op imal le el o compensa ion a ising om his ade-o¤ and …nd ha poli icians should
be ewa ded mo e when e¤o is ela i ely mo e impo an han e o ms and unce ain y
is high. Second, since he poli ical cos o e o ms a ises because incumben s ca e abou
eelec ion, we ask unde wha condi ions imposing a one-pe iod e m limi may be wel a e
imp o ing. A e m limi p omo es he adop ion o e o ms, bu educes poli ical accoun -
abili y and hence e¤o , and gi es up he bene… o keeping well-pe o ming poli icians
in o¢ ce. We …nd ha imposing such a limi is wel a e educing when unce ain y is high
and e¤o is impo an . Finally, we show ha he main esul s o ou model a e obus
o al e na i e assump ions. In e es ingly, when incumben s a e awa e o hei abili y and
he e is complemen a i y be ween abili y and he e u n om e o ms, he model can e en
2Glaese e al. (2005) show how a simila mechanisms may explain s a egic ex emism in a model
whe e policy s a emen s a e no di ec ly obse able and pa ies compe e o o e s.
3
gene a e a posi i e co ela ion be ween e o ms and eelec ion. We also show ha , when
e o ms a e modelled as a disc e e choice, hei poli ical cos is p opo ional o hei size.
The con‡ic o in e es be ween o e s and poli icians is an old heme in a as li -
e a u e.3Ou pape builds on agency models whe e he ole o elec ions is o selec he
mos compe en poli ician. This app oach has been used ex ensi ely (e.g. see No dhaus,
1975, Alesina, 1987, Rogo¤ and Sibe , 1988, Rogo¤, 1990, Pe sson and Tabellini, 1990,
Lohman, 1998 and D azen, 2000a) o explain poli ical business cycles, i.e., he incen i e
o incumben s o pe o m well jus be o e elec ions so as o appea alen ed o he o e s.
Despi e some simila i ies, hese pape s do no s udy he ole o unce ain y and economic
shocks on poli ical incen i es o unde ake e o ms, while we a e no in e es ed in s udying
how poli ical incen i es a y in elec ion and non-elec ion yea s.
In line wi h he li e a u e on elec o al accoun abili y, ini ia ed by Ba o (1973), we con-
side he disciplining ole o elec ions on e¤o choices. Ye , we a gue ha , when i comes
o in es ing in e o ms, elec o al incen i es a e mo e likely o induce myopic beha io .
Recen con ibu ions by Alesina and Tabellini (2007, 2008) compa e models o elec o al
accoun abili y and ca ee -conce ns o s udy how he op imal accoun abili y mechanism
depends on he cha ac e is ics o policy asks. The e¤ec o poli ical compensa ion on
he quali y o poli icians has ins ead been s udied, among o he s, by Caselli and Mo elli
(2004), Besley (2004), Besley and Sma (2007), and Ma ozzi and Me lo (2008). Once
again, none o hese pape s ocuses on unce ain y and on he choice o e o ms.
Impo an con ibu ions on he poli ical economy o e o ms in he p esence o un-
ce ain y a e Fe nandez and Rod ik (1991), Ciccone (2004), Cukie man e al. (1992),
Alesina and D azen (1991). The in‡uen ial pape by Fe nandez and Rod ik (1991) has
shown how unce ain y ega ding he dis ibu ion o gains and losses may lead o a s a us
quo bias. Alesina and D azen (1991) ha e ins ead shown ha e o ms may be pos poned
due o a wa o a i ion. D azen (2000b) discusses why e o ms may be mo e likely in
pe iods o c isis. Alesina and Cukie man (1990), ins ead, ha e shown ha unce ain y
allows he poli icians o ollow hei mos p e e ed policy, e en a he expenses o o e s.
Simila ly o hese pape , we …nd ha poli icians ha e a bias agains adop ing e o ms.
Ou esul , howe e , is based on a ional myopia a he han edis ibu ional con‡ic s
and can explain why e o ms a e so poli ically di¢ cul e en when hey do no seem o
be punished by o e s. Mo o e , he esul ha unce ain y lowe s he poli ical cos o
e o ms is o ou knowledge no el. Finally, a ecen li e a u e on ins i u ional change has
emphasized ha (ad e se) economic shocks may speed up he ansi ion owa ds mo e
democ a ic poli ical egimes (see o example Acemoglu and Robinson, 2001 and 2006,
3Pe sson and Tabellini (2000) p o ide an excellen in oduc ion o his …eld.
4
and he e idence in B ückne and Ciccone, 2009). Di¤e en ly om hese pape s, howe e ,
we es ic a en ion o economic e o ms in ep esen a i e democ acies only.
Finally, he gene al insigh ha unce ain y may imp o e he equilib ium along some
dimensions in agency models has been explo ed in Dewa ipon , Jewi and Ti ole (1999),
Holms öm (1999), and P a (2005), among o he s. Ye , he a gumen ha unce ain y
may imp o e wel a e by s imula ing he adop ion o e o ms wi h u u e payo¤s has no
been made in he li e a u e and appea s o … s o de ele ance o unde s and he poli ical
economy o e o ms.
The es o he pape is o ganized as ollows. Sec ion 2 discusses he empi ical obse a-
ions mo i a ing ou analysis. I e iews he exis ing e idence ha eelec ion p obabili y
is a¤ec ed by economic ou comes, bu no by he adop ion o e o ms, and i discusses he
empi ical de e minan s o e o ms. I also un eils a new pa e n in he da a: in a panel o
OECD coun ies, pe iods o high economic ola ili y a e associa ed o a highe p opensi y
o implemen e o ms. Sec ion 3 builds a wo-pe iod model o elec ions and e o ms ha
can a ionalize hese empi ical obse a ions. I con ains he main esul o he pape ,
ha e o ms en ail a poli ical cos , e en i hei equilib ium choice does no a¤ec he
eelec ion p obabili y o he incumben go e nmen . I also shows ha , by educing he
pe cei ed poli ical cos o e o m, unce ain y p omo es hei adop ion. Sec ion 4 exam-
ines he implica ions o he model o social wel a e. I p o ides condi ions o unce ain y
o be wel a e imp o ing, i shows how poli ical ewa ds can be se so as o maximize
he expec ed u ili y o ci izens and s udies he desi abili y o imposing a one- e m limi .
Sec ion 5 explo es he obus ness o he main esul s o al e na i e assump ions. Sec ion
6 concludes.
2 Mo i a ing E idence
In his Sec ion, we e iew he main s ylized ac s mo i a ing his pape . Fi s , he e is
g owing consensus ha , while economic pe o mance o en a¤ec s he p obabili y ha
poli icians s ay in powe , he incidence o …scal and s uc u al e o ms – despi e he
economic and poli ical cos s hey may en ail –does no . Second, e o ms a e mo e likely
o occu du ing imes o c isis. Since he li e a u e has no explo ed he link be ween
unce ain y and e o ms, we p o ide some o iginal e idence on i and …nd ha economic
ola ili y is associa ed o a mo e e o mis s ance on behal o go e nmen s.
5

2.1 C isis, Re o ms and Reelec ion
The hypo hesis ha eelec ion p ospec s depend on cu en economic pe o mance ecei ed
ea ly suppo in he wo ks o Fai (1978, 2008), Kiewie and Ri e s (1985) and Alesina and
Rosen hal (1995). In pa icula , Kiewie and Ri e s (1985) …nd, using da a on U.S. and
Wes e n Eu opean elec ions, ha a 1-pe cen decline in eal income is associa ed wi h
a educ ion o he incumben pa y’s o e sha e o be ween 0.5 pe cen and 1 pe cen .
Mo e ecen ly, a se ies o pape s ha e p o ided e idence ha good economic pe o mance
inc eases he likelihood ha poli icians s ay in powe . B ende and D azen (2008) show on
a sample o 73 coun ies ha high g ow h du ing he e m in o¢ ce inc eases he eelec ion
p obabili y, pa icula ly in less de eloped coun ies. Using a sample o 21 OECD coun ies,
Bu i e al. (2010) …nd ha high le els and g ow h a es o he cyclical componen o GDP
ha e a posi i e impac on he chances o eelec ion o incumben go e nmen s. Finally,
Wol e s (2007) p o ides e idence om U.S. gube na o ial elec ions ha good economic
pe o mance inc eases he likelihood ha incumben pa ies s ay in o¢ ce.
On he con a y, many pape s ha e ailed o iden i y empi ically he o en-blamed
elec o al cos o e o ms. Bu i e al. (2010), using da a on p oduc and labo ma ke
e o ms ac oss OECD coun ies, show ha e o ms in gene al do no a¤ec he eelec ion
p obabili y o incumben go e nmen s. Alesina, Pe o i and Ta a es (1998) s udy he
poli ical consequences o …scal adjus men s in a c oss sec ion o OECD coun ies and …nd
ha …scal aus e i y has posi i e a he han nega i e poli ical e¤ec s. B ende and D azen
(2008) …nd ha loose …scal policies ha e a nega i e e¤ec on he p obabili y o eelec ion
in a panel o 74 coun ies o e he pe iod 1960-2003. Pel zman (1992), B ende (2003),
and D azen and Esla a (2010) examine he e¤ec o …scal pe o mance on eelec ion a
he s a e and local le el in a single coun y ( he Uni ed S a es, Is ael, and Colombia,
espec i ely) and …nd ha o e s punish— a he han ewa d— loose …scal policies.
2.2 C isis, Vola ili y and he Likelihood o Re o ms
The li e a u e on he de e minan s o e o ms such as ma coeconomic s abiliza ion, ade
libe aliza ion and de egula ion, is as . A e e iewing he expe iences o de eloping
coun ies wi h ma ke -o ien ed e o ms, Tommasi and Velasco (1996) conclude ha he
hypo hesis ha c ises lead o s abiliza ion is pa o he con en ional wisdom. Sys ema ic
empi ical wo ks on he de e minan s o e o ms (see, among o he s, Alesina and A dagna,
1998, D azen and Eas e ly, 2001, Hamann and P a i, 2002) con… m ha he adop ion o
s abiliza ion plans aimed a educing in‡a ion, go e nmen de…ci and he black ma ke
p emium, is mo e likely in pe iods o c isis, i.e., when in‡a ion, de…ci and black ma ke
6
p emium a e pa icula ly high. Recen ly, Alesina e al. (2006) p o ide e idence om
a la ge panel o coun ies ha …scal e o ms a e mo e likely o occu du ing imes o
c isis, when new go e nmen s ake o¢ ce and when go e nmen s a e “s ong.”Pe iods o
economic c isis a e also ound o a o he adop ion o s uc u al e o ms a ge ed o he
ma ke s o …nancial ins umen s (Abiad and Mody, 2003) goods and se ices, and labo
(Høj e al., 2006). Al hough c isis and ola ili y a e likely o be co ela ed, he e is o
ou knowledge no e idence on he ela ionship be ween e o ms and economic unce ain y.
This is un o una e because in pe iods o high ola ili y i migh be easie o policy make s
o blame economic shocks ins ead o aking esponsibili y o he cos s o hei ac ions,
and his may elax hei poli ical cons ain s. In o he wo ds, go e nmen s may pe cei e
unpopula e o ms o be poli ically mo e iable in imes o u moil. We now p o ide some
p elimina y e idence on his hypo hesis.
To do so, we s udy how economic ola ili y is empi ically ela ed o he likelihood
and he size o de…ci s abiliza ion in a panel o 20 OECD coun ies obse ed be ween
1975 and 2000.4To s a wi h, we ollow he exis ing li e a u e in cons uc ing a disc e e
indica o o e o m using da a on he annual a ia ion o cen al go e nmen de…ci as
a a io o GDP (DEFICIT), om he IMF Go e nmen Finance S a is ics (2001). We
de…ne as a e o m an episode o annual all in DEFICIT abo e he 80 h pe cen ile o he
empi ical dis ibu ion, co esponding o educ ions by mo e han 1.17 pe cen age poin s
in ou sample. Ou measu e o mac oeconomic ola ili y is he s anda d de ia ion o he
ou pu gap, i.e., he di¤e ence be ween he ac ual and he po en ial GDP o e po en ial
GDP, as compu ed by he OECD based on es ima ions o he p oduc ion unc ions. This
a iable is mean o cap u e unexpec ed a ia ions in economic pe o mance.
Figu e 1 plo s he o e all numbe o e o ms obse ed be ween 1975 and 2000 in he
sample agains he s anda d de ia ion o he ou pu gap in he same pe iod and p o ides
a syn he ic desc ip ion o he da a. The line in e pola ing he poin s exhibi s a posi i e
and signi…can slope, sugges ing he exis ence o posi i e co ela ion be ween ola ili y
and e o ms.
A … s s ep o co obo a e his g aphical e idence is o es whe he he c oss-sec ional
co ela ion holds a e con olling o o he a iables. The coe¢ cien s es ima ed wi h
o dina y leas squa es a e epo ed in Table 1. In column 1, we simply eplica e he linea
in e pola ion plo ed in Figu e 1, showing a posi i e and signi…can co ela ion be ween
ola ili y o e he pe iod and he numbe o e o ms. Nex , we con ol o ini ial eal
4These coun ies a e: Aus alia, Aus ia, Belgium, Canada, Denma k, Finland, F ance, Ge many,
G eece, I eland, I aly, Japan, Ne he lands, No way, Po ugal, Spain, Sweden, Swi ze land, Uni ed King-
dom, Uni ed S a es.
7
FRA
BEL
DEU
ITA
DNK
CHE
NLD
AUS
USAAUT
JPN CAN
GRC
GBR
SWE
NOR
ESP
IRL
PRT
FIN
0 2 4 6 8
Numbe o De ici S abiliza ions, 1975-2000
1 2 3 4 5
S anda d de ia ion o ou pu gap, 1975-2000
De ici S abiliza ions Fi ed alues: slope 0.95, -s a 2.6
De ici S abiliza ion = del a(de ici /GDP) < -1.17
OECD coun ies, 1975-2000
Economic Vola ili y and De ici S abiliza ion
Figu e 1: Economic Vola ili y and De…ci S abiliza ions.
GDP pe capi a and he a e age le el o DEFICIT, in columns 2 and 3. While hese
a iables a e no signi…can ly co ela ed wi h he numbe o e o ms, he coe¢ cien o
ola ili y emains posi i e and signi…can . Following he li e a u e on c ises and e o ms,
we also con ol o he numbe o …scal c ises, de…ned as episodes in which go e nmen
de…ci as a sha e o GDP is abo e he 20 h pe cen ile (i.e., o e 7.5 pe cen ). As columns
4 and 5 show, he numbe o …scal c ises is no signi…can ly co ela ed o he numbe
o e o ms. Finally, we also con ol o he ollowing poli ical a iables: he numbe o
le -wing go e nmen s (le ), he numbe o go e nmen s in he … s wo yea s o o¢ ce
(younggo ) and a dummy o pa liamen a y sys ems. These indica o s a e ob ained om
he 2006 elease o he Da abase o Poli ical Ins i u ions compiled by he Wo ld Bank.
Consis en ly wi h p e ious e idence (e.g., Alesina e al., 2006), a la ge numbe o le -wing
go e nmen s is associa ed o mo e …scal adjus men s, while he coe¢ cien s o he o he
poli ical a iables a e no signi…can . Th oughou all speci…ca ions, economic ola ili y
emains signi…can ly and posi i ely co ela ed o he numbe o e o ms.5
To be e exploi he in o ma ion in he da ase we now use he annual panel o
5The esul s a e also obus o con olling o a e age poli ical ac ionaliza ion and p opo ional (as
opposed o majo i a ian) sys em. The es ima ed coe¢ cien s o hese a iables a e non signi…can .
8
es ima e he ollowing logi o he likelihood o e o ms:
P [REFi = 1 jSDi 1; 5;Xi 1] = exp (b0+b1SDi 1; 5+b2Xi 1+ui+ei )
1 + exp (b0+b1SDi 1; 5+b2Xi 1+ui+ei )
(1)
whe e REFi is he dummy indica o o …scal adjus men in coun y iand yea ,SDi 1; 5
is he s anda d de ia ion o he ou pu gap o e he … e-yea pe iod be ween 1and
5,Xi 1is a ec o o con ol a iables, uiis he coun y …xed e¤ec and ei is he e o
e m. All eg esso s a e lagged one pe iod o accoun o he ac ha policies may be
decided he yea be o e hey a e enac ed, and o a oid simul anei y. Mo eo e , we always
include among he eg esso s he le el o de…ci o con ol o he ac ha e o ms may
be mo e likely when he de…ci is pa icula ly high.
The esul s a e epo ed in Table 2. The … s speci…ca ion sugges s ha he p obabili y
o e o ms inc eases wi h ola ili y, and, consis en ly wi h he empi ical li e a u e on he
de e minan s o e o ms, i is highe he la ge is he de…ci as a a io o GDP in he
p e ious pe iod. Since bo h a iables a e signi…can , we keep hem in he es ima ion
and add o he co a ia es in he ollowing eg essions. Fi s , we con ol o indica o s o
economic ac i i y such as he ou pu gap and eal GDP pe capi a, o le he likelihood o
e o ms a y o e he cycle. While he coe¢ cien o he ou pu gap, in column 2, is no
signi…can ly di¤e en om ze o, he one o eal ou pu , in column 3, is nega i e, poin ing
o he ac ha e o ms a e adop ed mo e o en in pe iods o bad economic pe o mance.
In column 4, we eplace eal GDP pe capi a wi h he indica o o …scal c isis (i.e., de…ci
le el abo e 7.5 pe cen o GDP) and …nd a signi…can and posi i e coe¢ cien , con… ming
he exis ing e idence ha a …scal adjus men is mo e likely a e a …scal c isis. When
we conside bo h economic pe o mance and de…ci c ises, in column 5, all co a ia es
emain signi…can , which poin s o he impo ance o all hese a iables as de e minan s o
e o ms. Finally, column 6 shows ha poli ical ac o s such as he p oximi y o an elec ion,
ideology and he enu e o he go e nmen do no signi…can ly a¤ec he likelihood o …scal
s abiliza ion. Independen ly o he co a ia es, he sign, magni ude and signi…cance o he
coe¢ cien o economic ola ili y emain unal e ed.
The e idence in Tables 1 and 2 e e s o a disc e e indica o o e o ms, whose de…ni ion
depends on a h eshold alue o he annual a ia ion in he de…ci o GDP a io, and
hence may be subjec o disc e ion. To ci cum en his p oblem and o s udy he in ensi e
ma gin o e o ms, we now eplica e he panel analysis o Table 2 using as dependen
a iable he annual change in de…ci , DEFICIT. We he e o e es ima e:
DEFICITi =DEF ICITi 1+1SDi 1; 5+2Xi 1+i+i ;(2)
9
Imposing = eand a=aein o (7) and hen using (9), he eelec ion p obabili y
u ns ou o be
p= P  +" =1
2;
which is jus he uncondi ional p obabili y ha he incumben be mo e able han he
popula ion a e age. Thus, in equilib ium he choice o e o m does no a¤ec he p obabili y
o eelec ion. Ye , wha d i es he “poli ical cos ” o e o ms (i.e., @p=@ < 0in 12) is
hidden in o ma ion ou o equilib ium: he ac ha poli icians can de ia e om hei
equilib ium s a egy in ways unknown o o e s. No e also ha he “poli ical cos ” o
e o ms would disappea i he e we e no unce ain y abou .
Finally, we can sol e o E +1, i.e., he ex-an e expec ed abili y o he poli ician in
o¢ ce in he second pe iod, gi en he equilib ium beha io o o e s and he incumben .
Wi h p obabili y (1 p), he poli ician will be a new d aw wi h expec ed abili y . Wi h
p obabili y p, i will ins ead be an incumben who, by i ue o he o ing s a egy, is
expec ed o be be e han he a e age. Hence:
E +1 = (1 p)+pE +1 jb
 (18)
=+
2;
whe e  ep esen s he “selec ion e¤ec ”, ha is, he di¤e ence be ween he ex-an e ex-
pec ed abili y o a eelec ed incumben and he a e age. This is equal o he a e age o he
pos e io belie unca ed om below a , minus he uncondi ional mean. Using s anda d
p ope ies o no mal dis ibu ions yields:
=22

q2
+2
"2
= 22
g: (19)
No e ha eelec ed poli icians end o be be e han he a e age and mo e so when abili y
is highly dispe sed ( he e is no bene… om selec ion i poli icians a e all alike) and when
noise is low (so ha i is less likely o eelec bad bu lucky poli icians).
4 Wel a e Analysis
We now explo e he implica ions o he model o social wel a e. To s a wi h, we com-
pa e he equilib ium de i ed abo e wi h a cons ained e¢ cien benchma k and show ha
poli icians choose a subop imally low le el o e o ms. Then, we examine he impac o
unce ain y on wel a e and de i e condi ions o he e¤ec o be posi i e. We also s udy
how he poli ical ewa d a¤ec s wel a e and show how i can be se so as o maximize he
16

expec ed u ili y o ci izens. Finally, we use he model o examine he ole o elec ions and
whe he o no i is socially desi able o impose a one- e m limi o he poli ician in o¢ ce.
Using (3), (5) and (18), expec ed ex-an e social wel a e is:
W= +a ++
2+ ( ):(20)
whe e and asol e (15) and (16), espec i ely. The abo e equilib ium is ine¢ cien . A
bene olen social planne subjec o he same in o ma ion se would choose e o ms F B
so as o equa e he social bene… o he social cos :
 0 F B= 1:(21)
Compa ing (21) o (15), i is immedia e o see ha he le el o e o ms chosen by he
poli ician is oo low. This ine¢ ciency a ises om he poli ical cos ha e o ms impose
on he incumben : he ac ha , by de ia ing om he equilib ium s a egy, he can
inc ease his chance o be eelec ed.13 In sum (p oo in he ex ):
P oposi ion 2 In he abo e en i onmen , in es men in e o ms, , is below he le el
ha would maximize social wel a e.
No e also ha , since Wis inc easing in a, a social planne would se e¤o o i s
maximum, amax.
4.1 Unce ain y and Wel a e
In he nex p oposi ion, we cha ac e ize how unce ain y a¤ec s ex-an e expec ed social
wel a e (20).
P oposi ion 3 The e¤ec on social wel a e o he a iance o noise (2
") and o abili y
(2
) is ambiguous:
@W
@2
"
>0()  g2
00 ( )> 2
+2(22)
@W
@2

>0() 2
+ 22
"2g
00 ( )> 2(23)
P oo . See Appendix
13 The same dis o ion leading o subop imal e o ms would a ise e en in he absence o elec ions i u u e
poli ical compensa ion was inc easing in cu en economic pe o mance, as empi ically show by Di Tella
and Fisman (2004) using da a on US gube na o ial sala ies.
17
Va (epsilon)
W
Figu e 2: Wel a e and Unce ain y (2
")
The a iance o noise (2
") has con as ing e¤ec s on wel a e. Fi s , P oposi ion 1 shows
ha noise p omo es in es men in e o ms. Gi en ha e o ms a e always subop imally
low, his e¤ec ends o inc ease social wel a e. Second, P oposi ion 1 also shows ha noise
educes e¤o and his ends o lowe social wel a e. Thi d, by making luck ela i ely mo e
impo an , a highe noise aises he p obabili y o ous a alen ed incumben o o con… m
a bad one. Thus, 2
" educes he selec ion p emium, ; and hence social wel a e. The
… s e¤ec domina es he o he wo, so ha noise u ns ou o be wel a e imp o ing, when
e o ms a e ela i ely mo e impo an han e¤o and selec ion. This is mo e likely o be
he case when he poli ical cos o e o ms, g, is high (so ha unde in es men is se e e),
e¤o is no e y aluable (low ) and abili y is e y concen a ed (low 2
). Gi en ha
g!0when 2
"! 1, condi ion (22) canno be sa is…ed when 2
"is high enough. Thus,
he nega i e wel a e e¤ec mus domina e i noise is su¢ cien ly high.
Wi hou addi ional es ic ions, wel a e can be a highly non-mono onic unc ion o 2
".
Some examples a e epo ed in Figu e 2 o he case ( ) = and a small . The dashed
line ep esen s he asymp o ic le el o wel a e as 2
"! 1, i.e., when e o ms a e op imal
bu he e is no bene… om selec ion. The uppe cu e displays he ela ionship be ween W
and 2
" o a high alue o 2
. When abili y is e y dispe sed, selec ion is so impo an ha
an inc ease in noise is always wel a e- educing, despi e i s posi i e e¤ec on e o ms. The
18
lowe cu e co esponds o he opposi e scena io in which he e ogenei y in abili y is e y
low, so ha selec ion is no e y use ul. In his case, wel a e inc eases wi h unce ain y
un il 2
"becomes e y la ge ( he poin a which he cu e becomes downwa d sloping is
no shown). Two in e media e examples make he non-mono onici y mo e e iden .
The a iance o abili y (2
) has con as ing wel a e e¤ec s oo. On he one hand, mo e
dispe sion in poli ical abili y inc eases e o ms (P oposi ion 1) and he selec ion p emium,
(as can be seen om 19). These e¤ec s end o inc ease social wel a e. On he o he
hand, P oposi ion 1 shows ha mo e he e ogenei y educes e¤o . The posi i e wel a e
e¤ec will domina e when e o ms and selec ion a e ela i ely mo e impo an han e¤o .
Tha is, when g is high (so ha unde in es men is se e e), e¤o is no e y aluable
(low ) and abili y is dispe sed (high 2
). F om (23), i is immedia e o see ha he
posi i e wel a e e¤ec o he e ogenei y mus domina e i 2
is su¢ cien ly high.
4.2 Op imal Poli ical Rewa d
Wha a e he wel a e e¤ec s o poli ical ewa ds, ? I en s om o¢ ce inc ease, he
poli ician will ca e mo e abou eelec ion and his will induce him o exe mo e e¤o ,
bu also o in es less in e o ms (see P oposi ion 1).14 The esul ing ade-o¤ sugges s
ha he e migh exis a socially op imal le el o poli ical ewa ds. This possibili y is
wo h explo ing because, al hough includes psychological en s and p i a e bene… s ha
may be di¢ cul o con ol, he pay o poli icians in powe can pa ly be chosen by he
socie y and a ies conside ably ac oss coun ies.15 Thus, we now u n o he analysis o
he op imum and i s de e minan .
Di¤e en ia ing expec ed social wel a e, (20), w. . . yields:
@W
@ =@a
@ + 0( )1@
@
The … s e m is he ma ginal alue o poli ical ewa ds, MB (): one addi ional uni o
inc eases e¤o by @a=@, wi h a alue p opo ional o . The second e m is ins ead
he ma ginal cos , MC (): an ex a uni o biases e o ms downwa d (@ =@ < 0)
and he cos o his is p opo ional o he se e i y o unde in es men in equilib ium,
[ 0( )1] >0. Using (15) and (16) and simpli ying e ms, he … s -o de condi ion o
14 Some e idence ha he wage paid o poli icians a¤ec s hei pe o mance is p o ided, by Besley (2004)
o he U.S., Fe az and Finan (2009) o B azil, and Gaglia ducci and Nannicini (2009) o I aly.
15 Fo example, Besley (2004) epo s ha he US p esiden is paid a ound $400,000, he B i ish p ime
minis e $270,000, while he F ench p esiden $70,000. See Die meie e al. (2005) o a pionee ing a emp
a quan i ying and decomposing he e u ns o a ca ee in he US Cong ess.
19
an in e io op imum is:
MB () = 2=@
@ =MC ():(24)
To cha ac e ize he solu ion o (24), in Figu e 3 we plo bo h MB and MC as a unc ion
o . The MB cu e is ep esen ed by he dashed, ‡a , line.16 The MC cu e (solid line)
s a s a ze o and may be non mono onic ( he Figu e shows wo possible cases). I s slope
is analyzed o mally in Lemma 1:
Lemma 1 The ma ginal cos o is inc easing i and only i
00 ( )2> 000 ( )g:
P oo . See Appendix
The condi ion in Lemma 1 is always sa is…ed when !0and also i 000 ( )<0.
Ye , when 000 ( )>0(e.g., o = wi h 0<  < 1), MC may be hump-shaped. In
wha ollows, we es ic a en ion o he mos in e es ing case in which MB and MC
in e sec once and only once o e he ele an ange 2[0;(amax= 1) =g].17 Unde
his es ic ion, he solu ion, , o (24) is unique and in e io , and he MC cu e mus be
posi i ely sloped a . We s udy he compa a i e s a ics o  o changes in pa ame e s
in he ollowing p oposi ion.18
P oposi ion 4 The socially op imal poli ical ewa d,
=2 00 ( )
g;
is inc easing in he a iance o bo h noise (2
") and abili y (2
), and in he alue o e¤o
():
@
@2
"
>0; @
@2

>0; @
@ >0:
P oo . See Appendix
An inc ease in unce ain y (due o ei he abili y dispe sion o noise) a¤ec s he ma -
ginal cos o poli ical compensa ion while lea ing i s ma ginal bene… una¤ec ed. Highe
16 The MB cu e is ‡a because e¤o is linea in , which in u n depends on he assump ion o a
quad a ic cos o a. Ou esul s gene alize o su¢ cien ly con ex cos unc ions.
17 Recall ha he e exis s an uppe bound o e¤o , amax . The e o e, i is ne e op imal o o¤e a
compensa ion highe han he minimum le el inducing he maximum e¤o .
18 Thus, he uppe bound on a ules ou he a he ex eme and no e y ealis ic case in which social
wel a e is maximized o ! 1 and a! 1.
20
gamma
MC, MB
Figu e 3: Opimal : Solid = MC, Dashed = MB
unce ain y means ha chance plays a bigge ole in eelec ion, implying ha becomes
less eac i e o . F om (24) we see ha his educes he ma ginal cos o . As a
esul , op imal poli ical compensa ion inc eases. A highe alue o e¤o , , aises he
ma ginal bene… o while lea ing he ma ginal cos una¤ec ed. The e o e, he op imal
compensa ion inc eases wi h he alue o e¤o .
4.3 Te m Limi
A key eason why e o ms a e oo low is ha incumben s ca e no only abou social
wel a e, bu also hei eelec ion. Thus, a way o align he incen i es o poli icians o
do e o ms and hose o he socie y would be o ule ou he possibili y o eelec ion by
imposing a one-pe iod e m limi . This would se bo h pand @p=@ o ze o and es o e he
… s -bes in es men in e o ms. Ye , wi hou elec o al incen i es, incumben s will pu less
e¤o . Mo eo e , by excluding eelec ion, ci izens o ego he oppo uni y o e aining well
pe o ming candida es.19 The e o e, despi e i s nega i e e¤ec on e o ms, he p ospec
o eelec ion may be in he in e es o he socie y, al hough a one- e m limi migh be
19 See Besley and Case (1995) o e idence ha e m limi s do appe o a¤ec policy choices. Ye , om
hei esul s i is di¢ cul o so ou he e¤ec s on e¤o and on e o ms.
21

op imal unde some condi ions. We now explo e his possibili y o mally.
The equilib ium unde a one- e m limi is cha ac e ized by = FB,a=and = 0.
Ruling ou eelec ion is socially op imal i i g an s an ex-an e expec ed social wel a e,
WTL, highe han W, i.e., when:
WTL W=  F B FB ( ( ) )
2+2g>0:
The … s e m in squa e b acke s, which is always posi i e, is he gain om a e m limi
due o he highe in es men in e o ms. The second e m, ins ead, is he loss in social
wel a e o gi ing up selec ion and lowe ing e¤o . Rea anging and using (19) yields ha
a one- e m limi is socially op imal i and only i :
  F B FB ( ( ) )> g 2
+2(25)
This condi ion is mo e likely o hold when selec ion is no e y use ul o e¤ec i e, abili y
has low alue, and e o ms a e highly needed. Mo e p ecisely, an inc ease in  aises he
RHS o (25) he eby making he op imali y o a e m-limi less likely.
The e¤ec o unce ain y is ins ead mo e complex. Fi s , no e ha 2
and 2
"do
no a¤ec WT L, bu ha e ambiguous e¤ec s on W, as discussed in P oposi ion 3. Mo e
unce ain y lowe s he LHS o (25), because i educes he unde in es men in e o ms
( ecall, con e ges mono onically o FB as ei he 2
! 1 o 2
"! 1). This ends
o make a e m limi less a ac i e. Ye , he e¤ec on he RHS o (25) may depend on
he sou ce o unce ain y. When he e is mo e poli ical he e ogenei y (highe 2
) he e is
mo e o gain om ex-pos selec ion, bu he e is also less e¤o in an equilib ium wi h
eelec ions. The … s e¤ec domina es, so ha he RHS o (25) inc eases i :
@g2
+2
@2

>0() 2
+ 22
"> 2
Thus, i 2
is su¢ cien ly high, a e m limi is ne e op imal.
An inc ease in 2
", ins ead, wo sens selec ion and lowe s he RHS o (25). Since i also
lowe s he LHS, i is unclea whe he o no i makes a e m-limi mo e a ac i e. Despi e
his ambigui y, i can be shown ha , i 2
"is high enough, imposing he e m limi canno
be op imal. The eason is ha , as 2
"! 1,Wcon e ges o WTL and @W=@2
"<0( he
la e ollows om P oposi ion 3), implying ha Wmus con e ge om abo e. Thus, we
mus ha e W > WTL when he noise is su¢ cien ly high. We summa ize his discussion
in he ollowing P oposi ion (p oo in he ex ):
P oposi ion 5 The e exis s a h eshold le el o he e ogenei y in poli ical abili y, b2
, such
22
Va (epsilon)
W>W(TL)
W<W(TL)
Va ( he a)
Figu e 4: Unce ain y and Te m Limi
ha o 2
>b2
holding elec ions wi hou he e m limi is socially op imal. The e exis s
a h eshold le el o economic ola ili y, b2
", such ha o 2
">b2
"holding elec ions wi hou
he e m limi is socially op imal.
Mo e gene ally, he desi abili y o a e m limi is depic ed in Figu e 4 in he space
(2
"; 2
). In he egion below he solid line, wel a e is highe when a e m limi is in place.
Clea ly, i 2
is high enough so ha selec ion is su¢ cien ly impo an , a e m limi is
ne e op imal. Simila ly, i 2
"is high enough, imposing he e m limi canno be op imal
ei he . Mo eo e , as he …gu e shows, depending on pa ame e s, a e m limi may be
mo e likely o be op imal o in e media e alues o noise. Fo example, his is he case
when he deg ee o he e ogenei y in poli ical abili y co esponds o he dashed line. This
happens because, o low 2
"selec ion is e y e¤ec i e and e¤o is high, his compensa ing
he unde in es men in e o ms.
5 Robus ness
We now s udy he obus ness o he main esul s de i ed in Sec ion 3 o al e na i e
assump ions. In pa icula , we show ha when poli icians a e awa e o hei own abili y,
hei equilib ium choices ( and a) will depend on alen , . Ye , he poli ical cos o
23
e o m is s ill p esen in his e sion o he model and is s ill dec easing in he deg ee o
unce ain y hey ace. Mo eo e , when good poli icians a e also ela i ely be e a doing
e o ms, he model can gene a e a posi i e co ela ion be ween e o ms and eelec ion,
as ound in some o he empi ical li e a u e (e.g., Alesina e al. 1998 and B ende and
D azen, 2008). Las , we s udy he ex ensi e ma gin o e o ms in a e sion o he model
whe e he choice o is disc e e and …nd he new esul ha he poli ical cos is inc easing
in he size o e o ms.
5.1 Poli ical Abili y and Re o ms
I he incumben knows his abili y ibe o e choosing e o ms, he o ing s a egy does
no change since ci izens can only condi ion hei choice on y and hey canno disen angle
i om " . This means ha hey s ick o he ule o eelec ing any incumben deli e ing
an ou come abo e he expec ed mean, y y=+ae e, whe e  eand aea e he
equilib ium le el o e o ms and e¤o chosen by he a e age poli ician (wi h i=). The
p oblem acing incumben iis now di¤e en :
max
;agi i+aia2
i
2++pi(i+) + (1 pi)+ ( i)(26)
subjec o:
pi= P (y y) = P (i+ai i+" y)
= 1 H(yi+ iai);(27)
whe e His he c.d. . o "N0; 2
", wi h densi y h. The pe cei ed eelec ion p obabili y,
pi, is s ill dec easing in e o ms and inc easing in e¤o :
@pi
@ i
=h(yi+ iai)<0;(28)
@pi
@ai
=h (yi+ iai)>0:(29)
The … s -o de condi ions o he incumben ’s p oblem a e:
 0( i)=1@pi
@ ii+;(30)
ai=+@pi
@aii+:(31)
The main di¤e ence wi h espec o equa ions (12) and (13) is ha in es men in e o ms
and e¤o now a ies wi h abili y. In pa icula , compa ed o (21), incumben s wi h
24
e y low abili y (i<  ) o e in es in e o ms o olun a ily educe hei eelec ion
p obabili y, because he cos o ha ing a bad poli ician ou weighs hei p i a e bene… om
emaining in o¢ ce. All he o he poli icians (wi h i>  ) adop oo li le e o ms,
as in he p e ious sec ion. Simila ly, good poli icians pu mo e e¤o o inc ease hei
eelec ion p obabili y pi, because he alue o s aying in powe is p opo ional o i.
Al hough pi, iand aia e now abili y speci…c, i is use ul o cha ac e ize he equilib ium
o he a e age poli ician (wi h i=), because i coincides wi h he ex-an e expec ed
equilib ium. Imposing i=and a ional expec a ions ( i=  =  eand ai= a= ae) in o
(28)-(31), he … s -o de condi ions o he a e age-abili y incumben become:
 0( ) = 1 + h
a=1 + h
wi h h=22
"1=2. Compa ing hese condi ions o (15) and (16), we see ha he a e age
poli ician now makes less e o ms and pu s mo e e¤o han in Sec ion 3. This is because
he incumben aces a lowe deg ee o poli ical unce ain y (he knows his own ype, i)
and, as al eady discussed, lowe unce ain y implies less e o ms and mo e e¤o . Fo he
same eason,  and ado no depend on 2
anymo e, since he only sou ce o unce ain y
o he incumben is now he noise shock, ". Ye , a highe ola ili y o "inc eases ( educes)
he a e age le el o e o ms (e¤o ), p ecisely as in P oposi ion 1.
I is also in e es ing o no e ha , o poli icians o e y high o e y low abili y,
unce ain y may now inc ease he poli ical cos o e o ms. Fo mally, om (28), @pi=@ i
becomes a posi i e unc ion o 2
" o su¢ cien ly high o low i, because he densi y h()
in he ails ge s a e when he a iance inc eases. Thus, unce ain y lowe s in es men
in e o ms o e y good o e y bad poli icians, bu poli icians in he ails a e a mino i y
only. In sum, while he e¤ec s a e now mo e nuanced, unce ain y s ill unambiguously
educes he a e age poli ical cos o e o ms.
The esul ha good poli icians make less e o m may appea unappealing. Ye , his
esul can be o e u ned by assuming no only ha poli icians know hei own ype, bu
also ha poli ical abili y ma e s o e o ms. In pa icula , assume ha ( i; i), wi h
he () unc ion being supe modula (i.e., @2 ( i;i)
@ i@i>0). In wo ds, good poli icians a e
also ela i ely be e a doing e o ms. Then, he new … s -o de condi ion o ibecomes:
@ ( i; i)
@ i
= 1 @pi
@ ii+
P o ided ha @2 ( i;i)
@ i@iis su¢ cien ly high, be e poli ician will now make mo e e o ms.
Mo eo e , since be e poli icians ha e a highe chance o being eelec ed, his e sion o
25
[41] Hamann, A. Ja ie and Alessand o P a i (2002). “Why Do many S abiliza ions ail?
The impo ance o Luck, Timing and Poli ical Ins i u ion,” IMF Wo king Pape n.
02/228.
[42] Holms öm, Beng (1999). “Manage ial Incen i e P oblems: A Dynamic Pe spec i e,”
Re iew o Economic S udies 66(1), 169-182
[43] Høj, Jens, Thai Tanh Dang, Vincenzo Galasso and Giuseppe Nicole i (2006). “The
Poli ical Economy o S uc u al Re o m: Empi ical E idence om OECD Coun ies,”
OECD Economics Depa men Wo king Pape s, No. 501.
[44] Kiewie , D. Rode ick and Ri e s, Douglas (1985). “A Re ospec i e on Re ospec i e
Vo ing,”in Heinz Eulau and Michael S. Lewis- Beck, eds., Economic Condi ions and
Elec o al Ou comes: The Uni ed S a es and Wes e n Eu ope, New Yo k: Aga hon,
1985, pp. 207-31.
[45] Ki ie , Jan F. (1995). “On Bias, Inconsis ency and E¢ ciency o Va ious Es ima o s
in Dynamic Panel Da a Models,”Jou nal o Econome ics 68, 53-78.
[46] Lohmann, Susanne (1998). “Ra ionalizing he Poli ical Business Cycle: a Wo kho se
Model,”Economics and Poli ics 10(1), 1-17.
[47] Ma ozzi, And ea and An onio Me lo (2008). “Poli ical Ca ee s o Ca ee Poli i-
cians?,”Jou nal o Public Economics 92, 597-608.
[48] No dhaus, William (1975). “The poli ical business cycle,”Re iew o Economic S udies
42, 169–190.
[49] Pel zman, Sam (1992). “Vo e s as Fiscal Conse a i es,”Qua e ly Jou nal o Eco-
nomics, 107(2): 327–61.
[50] Pe sson, To s en and Guido Tabellini (1990). Mac oeconomic Policy, C edibili y and
Poli ics, Ha wood Academic Publishe s, London.
[51] Pe sson, To s en and Guido Tabellini (2000). “Poli ical Economics: Explaining Eco-
nomics Policy,”MIT P ess, Camb idge MA.
[52] P a , And ea (2005). “The W ong Kind o T anspa ency,”Ame ican Economic Re-
iew 95(3), 862-877.
[53] Rogo¤, Kenne h (1990). “Equilib ium Poli ical Budge Cycles,”Ame ican Economic
Re iew 80, 21–36.
32

[54] Rogo¤, Kenne h and Anne Sibe (1988). “Elec ions and Mac oeconomic Policy Cy-
cles,”Re iew o Economic S udies 55(1), 1-16.
[55] Tommasi, Ma iano and And es Velasco (1996). “Whe e A e We in he Poli ical Econ-
omy o Re o ms?”Jou nal o Policy Re o ms, Vol. 1, pp. 187–238.
[56] Windmeije , F ank (2005). “A Fini e Sample Co ec ion o he Va iance o Linea
E¢ cien Two-S ep GMM Es ima o s,”Jou nal o Econome ics 126, 25-51.
[57] Wol e s, Jus in (2007). “A e Vo e s Ra ional? E idence om Gube na o ial Elec-
ions,”Wha on School, Uni e si y o Pennsyl ania, mimeo.
7 Appendix
7.1 P oo o P oposi ion 1
Implici di¤e en ia ion o (15) wi h espec o 2
",2
and , and using (17) yields:
@
@2
"
=
00 ( )
@g
@2
"
=
00 ( )
g
2(2
+2
")>0(35)
@
@2

=
00 ( )
@g
@2

=
00 ( )
g
2(2
+2
")>0(36)
@
@ =g
00 ( )<0;(37)
since ma ginal e u ns o e o ms a e assumed o be dec easing ( 00 ( )<0).
Di¤e en ia ion o (16) wi h espec o 2
",2
and , and using (17) yields:
@a
@2
"
=@a
@2

=g
2(2
+2
")<0(38)
@a
@ =g > 0:(39)
7.2 P oo o P oposi ion 3
Di¤e encing social wel a e (20) w. . . 2
"yields:
@W
@2
"
=@a
@2
"
+ 0( )1@
@2
"
+
2
@
@2
"
:
A e eplacing he e m in he b acke wi h (15), he de i a i es @a=@2
"and @ =@2
" om
(38) and (35), and @=@2
"=2
g= 2
+2
", ob ains:
@W
@2
"
=g
22
+2
"g2
00 ( )2
2:
33
This is posi i e i and only i he e m in b acke s is posi i e, i.e.,
@W
@2
"
>0()  g2
00 ( )> 2
+2:
Di¤e encing social wel a e (20) w. . . 2
yields:
@W
@2

=@a
@2

+ 0( )1@
@2

+
2
@
@2

A e eplacing he e m in he b acke wi h (15), he de i a i es @a=@2
and @ =@2
 om
(38) and (36), and @=@2
=g22
(2
+2
")1, ob ains:
@W
@2

=g
2(2
+2
")22
00 ( )g+2
+ 22
"
This is posi i e i and only i he e m in b acke s is posi i e, i.e.,
@W
@2

>0() 2
+ 22
"2g
00 ( )> 2:
7.3 P oo o Lemma 1
Recall:
MC () = @
@ =g
00 ( )
whe e we ha e subs i u ed (37). Di¤e encing MC ()w. . . yields:
@MC ()
@ =g
[ 00 ( )]2 000 ( )@
@ g
00 ( )
=g
00 ( )1 000 ( )
00 ( )2g:
This exp ession is posi i e i and only i he e m in b acke s is posi i e, i.e.
@MC ()
@ >0() 00 ( )2> 000 ( )g:
7.4 P oo o P oposi ion 4
F om:
=2 00 ( )
g
34
i is immedia e o see ha is inc easing in . To …nd he e¤ec o unce ain y, no e
ha : @
@2
"
=@
@2

=@
@g
@g
@2
x
=@
@g
g
2(2
+2
")
o x="; . Nex , de i e w. . . g:
@
@g =2 00 ( )2 000 ( )g
00( )
g2<0;
unde he condi ion in Lemma 1, which is assumed o be sa is…ed a . Thus, @=@2
"=
@=@2
>0:
35
SD 0.946** 1.023** 0.930*** 0.929** 0.993*** 0.814**
[0.371] [0.455] [0.332] [0.384] [0.334] [0.332]
log(GDP_75) 0.453
[2.386]
DEFICIT -0.022 0.158
[0.163] [0.140]
#_CRISIS_DEF 0.022 0.084 0.091
[0.071] [0.098] [0.071]
#_le 0.139**
[0.063]
#_younggo -0.136
[0.063]
pa liamen a y -0.162
[0.694]
R-squa ed 0.148 0.150 0.149 0.153 0.216 0.489
Obse a ions 20 20 20 20 20 20
No e. Dependen a iable: # o De ici Re o m = numbe o yea s wi h
he annual change in DEFICIT (= go e nmen de ici as a sha e o
GDP) smalle han o equal o -1.17% o e he sample pe iod.
Reg esso s a e: SD = s anda d de ia ion o he ou pu gap o e he
sample pe iod; log(GDP_75) = log o eal GDP pe capi a in 1975;
DEFICIT in pe iod a e age; #_CRISIS_DEF= numbe o yea s wi h
DEFICIT less han o equal o -7.5; #_le = numbe o yea s wi h le -
wing go e nmen s; #_younggo = numbe o yea s wi h go e nmen s
in he i s wo yea s o o ice; pa liamen a y= dummy o
pa liamen a y (as opposed o p esiden ial) sys ems. Reg essions a e
pe o med wi h o dina y leas squa es. Robus s anda d e o s a e
epo ed in b acke s. ***, ** and * deno e signi icance a 1, 5 and 10
pe cen le el.
OECD Coun ies, 1975-2000 - C oss Sec ion
Table 1. Economic Vola ili y and he Numbe o Re o ms
SD_1 0.290* 0.320** 0.346** 0.260*** 0.348*** 0.320**
[0.159] [0.163] [0.164] [0.070] [0.161] [0.164]
DEFICIT_1 0.153*** 0.114** 0.231*** 0.300* 0.317*** 0.321***
[0.047] [0.052] [0.058] [0.157] [0.076] [0.078]
OUTPUTGAP_1 0.093
[0.060]
log(GDP_1) -4.288*** -3.934*** -3.925***
[1.471] [1.480] [1.510]
CRISIS_DEF_1 1.502** 1.502** 1.270*
[0.675] [0.675] [0.693]
elec ion_1 -0.238
[0.334]
le _1 -0.330
[0.364]
younggo _1 -0.511
[0.372]
Obse a ions 342 342 342 342 342 342
Coun ies 19 19 19 19 19 19
Coun y-FE Yes Yes Yes Yes Yes Yes
Table 2. Economic Vola ili y and he Likelihood o Re o ms
OECD Coun ies, 1975-2000 - Panel
No e. Dependen a iable is De ici Re o m = dummy aking alue 1 i he annual
change in DEFICIT is less han o equal o -1.17%, ze o o he wise. Co a ia es a e
lagged alues o : DEFICIT = go e nmen de ici as a sha e o GDP; SD = s anda d
de ia ion o he ou pu gap o e he p e ious i e yea s; OUTPUTGAP; log(GDP) =
log o eal GDP pe capi a; CRISIS_DEF= dummy aking alue 1 i DEFICIT is less
han o equal o -7.5; elec ion = dummy o legisla i e and/o execu i e elec ions; le
= dummy o le -wing go e nmen s; younggo = dummy o go e nmen s in he i s
wo yea s o o ice. Reg essions a e pe o med wi h panel logi wi h coun y ixed
e ec s. S anda d e o s a e epo ed in b acke s. ***, ** and * deno e signi icance a
1, 5 and 10 pe cen le el.

SD_1 -0.332** -0.344** -0.352** -0.351** -0.369*** -0.334**
[0.148] [0.152] [0.146] [0.148] [0.145] [0.149]
DEFICIT_1 0.843*** 0.852*** 0.775*** 0.725*** 0.682*** 0.687***
[0.047] [0.047] [0.047] [0.047] [0.047] [0.047]
OUTPUTGAP_1 -0.024
[0.048]
log(GDP_1) 0.0001*** 0.0001*** 0.0001***
[0.00005] [0.00005] [0.00005]
CRISIS_DEF_1 -1.317*** -1.151*** -1.099***
[0.434] [0.428] [0.431]
elec ion_1 0.269
[0.263]
le _1 0.353
[0.286]
younggo _1 0.518*
[0.290]
R-squa ed 0.743 0.745 0.732 0.743 0.739 0.742
Obse a ions 346 346 346 346 346 346
Coun ies 20 20 20 20 20 20
Table 3. Economic Vola ili y and Re o ms
No e. Dependen a iable is DEFICIT = go e nmen de ici as a sha e o GDP.
Reg esso s a e lagged alues o : DEFICIT; SD = s anda d de ia ion o he ou pu
gap o e he p e ious i e yea s; ou pu gap; log(GDP) = log o eal GDP pe
capi a; CRISIS_DEF = dummy aking alue 1 i DEFICIT is less han o equal o -
7.5; elec ion = dummy o legisla i e and/o execu i e elec ions; le = dummy o
le -wing go e nmen s; younggo = dummy o go e nmen s in he i s wo yea s
o o ice. Reg essions a e pe o med wi h leas squa es (LSDV) wi h coun y ixed
e ec s. S anda d e o s, in b acke s, a e co ec ed o he e oskedas ici y and
consis ency wi h Ki ie (1995) p ocedu e. ***, ** and * deno e signi icance a 1, 5
and 10 pe cen le el.
OECD Coun ies, 1975-2000 - Panel LSDV