ContentslistsavailableatScienceDirect
Journal
of
Mathematical
Economics
j o u r n al ho m e p age :w w w . e l s e v i e r . c o m / l o c a t e / j m a t e c o
Existence,
optimality
and
dynamics
of
equilibria
with
endogenous
time
preference
Selman
Erol
a,
Cuong
Le
Van
b,c,d,
Cagri
Saglam
a,∗aDepartmentofEconomics,BilkentUniversity,Turkey bCNRS,ParisSchoolofEconomics,France
cExeterUniversityDepartmentofEconomics,UnitedKingdom dDEPOCEN,VietNam
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:Received25July2009 Receivedinrevisedform 20December2010 Accepted26December2010 Available online 19 February 2011 JELclassification:
C61 D90 O41 Keywords:
Endogenoustimepreference Optimalgrowth
Competitiveequilibrium Multiplesteady-states
a
b
s
t
r
a
c
t
Thispaperstudiesthedynamicimplicationsoftheendogenousrateoftimepreferencedependingon thestockofcapital,inaone-sectorgrowthmodel.Theplanner’sproblemispresentedandtheoptimal pathsarecharacterized.Weprovethatthereexistsacriticalvalueofinitialstock,inthevicinityofwhich, smalldifferencesleadtopermanentdifferencesintheoptimalpath.Indeed,weshowthatadevelopment trapcanariseevenunderastrictlyconvextechnology.Incontrastwiththeearlycontributionsthat considerrecursivepreferences,thecriticalstockisnotanunstablesteadystatesothatifaneconomy startsatthisstock,anindeterminacywillemerge.Wealsoshowthatevenunderaconvex–concave technology,theoptimalpathcanexhibitglobalconvergencetoauniquestationarypoint.Themultipliers systemassociatedwithanoptimalpathisproventobethesupportingpricesystemofacompetitive equilibriumunderexternalityanddetailedresultsconcerningthepropertiesofoptimal(equilibrium) pathsareprovided.Weshowthatthemodelexhibitsgloballymonotonecapitalsequencesyieldinga richersetofpotentialdynamicsthantheclassicmodelwithexogenousdiscounting.
© 2011 Elsevier B.V. All rights reserved.
1. Introduction
Theoptimalcapitalaccumulationmodelshavebeenatthecore
of thetheoryof economicgrowth anddynamics. Based onthe
dynamicconsumptionandsavingdecisionsoftheeconomicagents
driven by intertemporal utilitytrade-offs between current and
futureconsumption,thekeycomponentsofthesemodelsturnout
tobetherateoftimepreferenceandthetechnology.The
classi-caloptimalgrowthmodelsandmuchofthesubsequentliterature
ongrowthfocusontheconvexstructuresofthetechnologyand
preferencesthatguarantee themonotonicalconvergenceof the
sequenceofoptimalstockstowardsauniquesteadystate.Sucha
structure,however,implythatthemodelcannotbeusedto
under-standthedevelopmentpatternsthat differconsiderablyamong
countriesinthelongrun(seeQuah,1996;Barro,1997;Barroand
Sala-i-Martin,1991).
∗ Correspondingauthor.Tel.:+903122901598;fax:+903122665140. E-mailaddresses:selman@bilkent.edu.tr(S.Erol),levan@univ-paris1.fr (C.LeVan),csaglam@bilkent.edu.tr(C.Saglam).
Toaccountforthesenon-convergentgrowthpaths,avarietyof
one-sectoroptimalgrowthmodelsthatincorporatesomedegree
of marketimperfectionsbased ontechnologicalexternaleffects
and increasingreturnshavebeenpresented.Withinamodel of
capital accumulation withconvex–concave technology, Dechert
andNishimura(1983),MitraandRay(1984)havecharacterized
optimal paths and prove the existence of threshold dynamics
that generate developmentor poverty traps(see Azariadis and
Stachurski,2005,forarecentsurvey).Inthesemodels,economies
withlowinitialcapitalstocksorincomesconvergetoasteadystate
withlowpercapitaincome,whileeconomieswithhighinitial
cap-italstocksconvergetoasteadystatewithhighpercapitaincome.
Indeed,theintroductionofincreasingreturnsalsomakesitpossible
forprivatereturnstotheaccumulationofcapitalstocktobe
com-plementarywiththeaggregatestockleadingtoindeterminaciesor
continuumofequilibria(e.g.,BenhabibandPerli,1994;Benhabib
andFarmer,1996;seeNishimuraandVenditti,2006forextensive bibliography).
Ageneraltendencyinthesestudieswithmultiplesteadystates,
indeterminacyorcontinuumofequilibriaisthattheyaremostly
devotedtotheanalysisofthetechnologycomponentleavingthe
timepreferenceessentiallyunalteredwithanexogenouslyfixed
0304-4068/$–seefrontmatter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jmateco.2010.12.006
geometricdiscounting.Inthispaper,yettheothermechanism,the
endogeneityoftimepreferencewillbeputforwardtoexplain
the-oreticallywhythedifferencesinpercapitaoutputlevelsamong
countriespersistinthelongrun.
Consideringthepreferencesoftheagentstoberecursive,the
earlycontributionsonthetheoryofendogenoustimepreference
postulate that anagent’s discountratedepends onthelevel of
present and future consumption (e.g.,Uzawa, 1968; Lucasand
Stokey,1984;Epstein,1987;Obstfeld,1990).Thesemodelsnearly
exclusivelyassumeanincreasingmarginalimpatiencesothatthe
agentsgetmoreimpatientastheygrowricher.Sucha
specifica-tionensuresstableoptimalcapitalsequencesthatconvergetoa
uniquesteadystateindependentoftheinitialconditions.However,
recentworkboththeoretically(e.g.,BeckerandMulligan,1997;
Stern,2006)andempirically(e.g.,Lawrance,1991;Samwick,1998; Fredericketal.,2002)proposethattheagentsgetmorepatient
astheygrowricherandimposethat thediscount ratedepends
closelyuponthestockofwealth(seeHamadaandTakeda,2009,
forarecentsurvey).Thisisinparalleltotheideathatthestockof
wealthisakeytoreachbetterhealthservicesandbetterinsurance
marketssothatthemorewealthiertheagentgets,themorepatient
itbecomes.Byconsideringdiscountfactorasafunctionof
con-sumption,Mantel(1998)studiestheimpactofdecreasingrateof
timepreferenceonoptimalgrowthpathofaneconomywitha
pri-maryfocusonthemonotonicitypropertiesofoptimalconsumption
andinvestment.Das(2003)derivesasetofsufficientconditions
forstabilityanduniquenessinthesamecontextinacontinuous
timeoptimalgrowthmodel.Stern(2006),alongthelinesofBecker
andMulligan(1997),letindividualsspendresourcestoincrease
theappreciationofthefuture.Heprovidesnumericalexamplesof
multiplesteadystatesandaconditionallysustainedgrowthpath.
In this paperweadapt theclassicoptimalgrowthmodel to
includeanendogenousrateoftimepreferencedependingonthe
stockofcapitalandanalyzetheimplicationsontheequilibrium
dynamics.Todoso,theplanner’sproblemisfirstpresentedandthe
optimalpathsarecharacterized.Weshowthatadevelopmenttrap
canariseevenunderastrictlyconvextechnology.Inotherwords,
weprovethatthereexistsacriticalvalueofinitialstock,inthe
vicinityofwhich,smalldifferencesleadtopermanentdifferences
intheoptimalpath.IncontrastwiththeearlycontributionsbyKurz
(1968),whoassumescapitaldependentpreferences,byBealsand Koopmans(1969),Iwai(1972),whoassumerecursivepreferences,
andmorerecently,byStern(2006),whoassumesendogenoustime
preferencedependingonfutureorientedresources,wenotedthat
thecriticalstockisdefinitelynotanunstablesteadystatesothat
ifaneconomystartsatthisstock,anindeterminacywillemerge.
Ontheotherhand,underasimilarconditiontothatofDechertand
Nishimura(1983),wealsoshowthatevenunderaconvex–concave
technology,theoptimalpathcanexhibitglobalconvergencetoa
uniquestationarypoint.Later,themultiplierssystemassociated
withanoptimalpathis proventobethesupportingprice
sys-temofacompetitiveequilibriumunderexternalityanddetailed
resultsconcerningthepropertiesofoptimal(equilibrium)paths
areprovided.Weshowthatthemodelexhibitsgloballymonotone
capitalsequencesyieldingarichersetofpotentialdynamicsthan
theclassicmodelwithexogenousdiscounting.
Thekeyreasonofourresultsisthatweconsiderthattherateof
timepreferencedecreaseswiththestockofwealthincontrastwith
exogenouslyfixeddiscountfactorassumedinstandardmodels.In
particular,thediscountfactorattributedtotheutilityof
consump-tionatperiodtincreaseswiththelevelofcapitalstockavailable
forproductionatperiodt.Accordingly,thelowerthestockof
cap-ital,thehigherthesacrificeofpostponingpresentconsumptionin
exchangeforfutureconsumption.InlinewithBeckerandMulligan
(1997),Stern(2006),animmediateimplicationofthisassumption
isthatastheagentgrowswealthier,thelevelofinvestmentin
capi-talthatwillbeappreciatedinthefutureincreasesandthereforethe
agentwillbecomemorepatient.Indeed,recentempiricalstudies
usingexperimentaldataalsosupportthevalidityofthishypothesis
ofwealthcausespatience(e.g.,Harrisonetal.,2002;Ikedaetal.,
2005).
Incorporatingsuchanhypothesis,animportantaspectofour
modelisthatwhileallowingvariationintherateoftime
prefer-ence,atthesametimeitmaintainstimeconsistency.Aremarkable
featureofouranalysisisthatourresultsdonotrelyon
particu-larparameterizationoftheexogenousfunctionsinvolvedinthe
model,rather,itprovidesamoreflexibleframeworkinregardsto
thediscountingoftime,keepsthemodelanalyticallytractableand
usesonlygeneralandplausiblequalitativeproperties.
Therestofthepaperisorganizedasfollows.Section2describes
themodelandprovidesthedynamicpropertiesofoptimalpaths.
Section3presentstheexistenceofacompetitiveequilibriumwith
externalityandstudiestheequilibriumdynamics.Finally,Section
4concludes.
2. Model
Themodeldiffersfromtheclassicoptimalgrowthmodelby
theassumptions ondiscounting. Ratherthanassumingthatthe
levelofdiscountonfutureutilityisanexogenousparameter,we
assumethatitisendogenousdependingonthepathofcapitalstock.
Formally,themodelisstatedasfollows:
max{ct,xt+1}∞t=0 ∞
t=0 t s=1 ˇ(xs) u(ct), subjectto∀
t,ct+xt+1≤f(xt),∀
t,ct≥0,xt≥0, x0≥0,given,Wemakethefollowingassumptionsregardingthepropertiesof
thediscount,utilityandtheproductionfunctions.
Assumption1. ˇ:R+→R++iscontinuous,differentiable,
stric-tly increasing and satisfies supx>0ˇ(x)=ˇm<1, supx>0ˇ
(x)<+∞.
Assumption2. u:R+→R+ is continuous,twicecontinuously
differentiableandsatisfiesu(0)=0.Moreover,itisstrictly
increas-ing,strictlyconcaveandu(0)=+∞(Inadacondition).
Assumption3. f :R+→R+ is continuous, twicecontinuously
differentiableandsatisfiesf(0)=0.Moreover,itisstrictlyincreasing
andlimx→+∞f(x)<1.
Foranyinitialconditionx0≥0,whenx=(x0,x1,x2, ...)issuch
that0≤xt+1≤f(xt)forallt,wesayitisfeasiblefromx0andtheclass
ofallfeasibleaccumulationpathsisdenotedby(x0).Itmaybe
easilyverifiedthatifx0<x0,then(x0)⊂(x0).Aconsumption
sequencec= (c0,c1,...) isfeasiblefromx0≥0whenthereexists
x∈(x0)with0≤ct≤f(xt)−xt+1.
Notingthattheconstraintswillbebindingattheoptimumas
theutilityandthediscountfunctionsarestrictlyincreasing,we
introducethefunctionUdefinedonthesetoffeasiblesequencesas
U(x)= ∞
t=0 t s=1 ˇ(xs) u(f(xt)−xt+1). (1)ThepreliminaryresultsaresummarizedinthefollowingLemma
whichhasastandardproofusingTychonovtheorem(seeLeVan
Lemma1. Let ¯x bethegreatestpointx≥0suchthatf(x)=x.Then,
(a)Forany x∈(x0), we have xt≤A(x0) for all t, where A(x0)=
max
x0, ¯x .(b)(x0)iscompactintheproducttopologydefinedonthespaceof
sequencesx.
(c)Uiswelldefinedanditiscontinuousover(x0)withrespectto
theproducttopology.
Itisnowclearthattheinitialoptimalgrowthmodelisequivalent
to:
max{U(x)|x∈(x0)}.
2.1. Existenceofoptimalpaths
Theexistence ofan optimalpath followsfromthefact that
(x0)iscompactfortheproducttopologydefinedonthespace
ofsequencesxandUiscontinuousforthisproducttopology.The
positivityofoptimalconsumptionandcapitalstockpathsfollow
fromtheInadacondition.
Proposition1.
(i)Thereexistsanoptimalpathx.Theassociatedoptimal
consump-tionpath,cisgivenby
ct=f(xt)−xt+1,
∀
t.(ii) Ifx0>0,everysolution(x,c)totheoptimalgrowthmodelsatisfies
ct>0,xt>0,
∀
t. (2)Proof. Itiseasy.
2.2. Valuefunction,Bellmanequation,optimalpolicy
Inordertocharacterizethebehavioroftheoptimalpaths,we
willproceedbydefiningthevaluefunctionandanalyzingthe
prop-ertiesoftheoptimalpolicycorrespondence.ThevaluefunctionV
isdefinedby:
∀
x0≥0,V(x0)=max{xt+1}∞t=0 ∞ t=0 t s=1 ˇ(xs) u(f(xt)−xt+1)|∀
t,0≤xt+1≤f(xt),x0≥0,given . (3)Theboundsondiscountingtogetherwiththeexistenceof
maxi-mumsustainablecapitalstockguaranteeafinitevaluefunction.
UndertheAssumption(1)andtheAssumption(2),onecan
imme-diately show that the value function is non-negative, strictly
increasingandcontinuous(seeStokeyetal.,1989).Giventhese,
Bellmanequationfollows.
Proposition2.
(i) V(0)=0,andV(x0)>0ifx0>0.Ifxisanoptimalpath,then
V(x0)= ∞
t=0 t s=1 ˇ(xs) u(f(xt)−xt+1).(ii)Visstrictlyincreasing.
(iii)Viscontinuous.
Proposition3.
(i)VsatisfiesthefollowingBellmanequation:
∀
x0≥0,V(x0)=max u(f(x0)−x)+ˇ(x)V(x)|0≤x≤f(x0) . (4)(ii) Asequencex∈(x0)isanoptimalsolutionifandonlyifitsatisfies:
∀
t,V(xt)=u(f(xt)−xt+1)+ˇ(xt+1)V (xt+1). (5)Proof. SeeLeVanandDana(2003).
Theoptimalpolicycorrespondence,:R+→R+isdefinedas
follows: (x0)=argmax
u(f(x0)−y)+ˇ(y)V(y)|y∈[0,f(x0)] .It is important to note that although the utility function is
strictlyconcave,thesolutionmaynotbeuniqueasthe
multiplica-tionofdiscountfunctiondestroystheconcavestructureneededfor
uniqueness.Thenon-emptinessandtheclosednessoftheoptimal
correspondenceanditsequivalencewiththeoptimalpathfollow
easilyfromthecontinuityofthevaluefunctionbyastandard
appli-cationofthetheoremofthemaximum.
Proposition4.
(i)(0)={0}.
(ii)Ifx0>0andx1∈(x0),then0<x1<f(x0).
(iii)isclosedandhenceuppersemi-continuous.
(iv)Asequencex∈(x0)isoptimalifandonlyifxt+1∈(xt),
∀
t.(v)The optimalcorrespondence isincreasingsothat if x0<x0,
x1∈(x0)andx1∈(x0)thenx1<x1.
Proof. (ii)Followseasilyfrom(2).(iii)FollowsfromtheTheorem
ofMaximum.(iv)Followsfrom(5).(v)SeeDechertandNishimura
(1983),orAmiretal.(1991).
Theincreasingnessofiscrucialfortheconvergenceofoptimal
paths,hencefortheanalysisofthelong-rundynamics.In Stern
(2006),assuminganendogenous timepreference dependingon
thefuture-orientedresources,theincreasingnessofhasbeen
provenbyusingthestrictconcavityofˇ.Sucharestrictiononthe
curvatureofthediscountfunctionisnotnecessaryinoursetup.
Moreover,wehavealsoproventhattheoptimalcorrespondence,
isnotonlyclosedbutalsouppersemi-continuous.
Withthepositivityoftheoptimalconsumptionandthestockof
capital,Eulerequationeasilyfollows.
Proposition5. Whenx0>0,anysolutionx∈(x0)satisfiestheEuler
equation:
∀
t,u(f(xt)−xt+1)=ˇ(xt+1)u(f(xt+1)−xt+2)f(xt+1)
+ˇ(x
t+1)V (xt+1) (6)
Proof. Recallthat0<xt+1<f(xt),
∀
t.Takeanyn,andconsiderthepathofcapitalxydefinedasfollows:xy
t =xt,
∀
t /=n+1andxyn+1=y∈Y,whereY=
f−1(xn+2),f(xn).Notethatitisawelldefined
openinterval includingxn+1 becausext+1<f(xt)and xt+2<f(xt+1).
Bychoosingy∈Y,weguaranteethaty<f(xt)andxt+2<f(y)sothat
xy∈(x
0).FromtheoptimalityofxwehaveU(x)≥U(xy),
∀
y∈Y.Let(y)=U(xy).Then(y)ismaximizedatzeroinY,which
sug-geststhat(0)=0.Withsomealgebra,thisendstheproof.
Inastandardoptimalgrowthmodelwithgeometricdiscounting
andtheusualconcavityassumptionsonpreferencesand
theproperties oftheoptimalpath iseasilyfoundby usingthe
firstorderconditionstogetherwithenvelopetheoremby
differ-entiatingthevaluefunction.However,inourmodel,theobjective
functionincludesmultiplicationofadiscountfunction.This
gen-erallydestroystheusualconcavityargumentwhichisusedinthe
proofofthedifferentiabilityofvaluefunctionandtheuniquenessof
theoptimalpaths(seeBenvenisteandScheinkman,1979;Araujo,
1991).
Tothisend,thenextpropositionshowsthatthevaluefunction
isdifferentiablealmosteverywhereandthereexistsaunique
opti-malpathfromalmosteverywherewithoutanyassumptiononthe
curvatureofˇ.
Proposition6.
(i)LeftderivativeofVexistsateveryx0>0andV−(x0)=u(f(x0)−
(x0))f(x0),where(x0)=min(x0).
(ii)RightderivativeofVexistsateveryx0>0andV+(x0)=u(f(x0)−
(x0))f(x0),where(x0)=max(x0).
Proof. SeeAskriandLeVan(1998);LeVanandDana(2003)forthe
proofsbasedontheClarkegeneralizedgradientsandthestandard
optimalgrowthtools,respectively.
Nowwe are abletoseethe relationbetween
differentiabil-ityofthevaluefunctionandtheuniquenessoftheoptimalpath.
Noticethat,givenx0,ifoptimalx1isuniquesothat(x0)is
single-valuedand(x0)=(x0)=x1,fromtheaboveproposition,wesee
thatV
−(x0)=V+(x0),henceVisdifferentiableatx0,andviceversa.
Thisanalysiscanbegeneralizedtoanyperiodtotherthanzero.
Theseallowustoclaimthattheoptimalcorrespondenceissingle
valuedanddifferentiablealmosteverywhere.
Proposition7.
(i)Ifxisanoptimalpathfromx0,thenVisdifferentiableatanyxt,
t≥1.Ifxisanoptimalpathfromx0,thereexistsauniqueoptimal
pathfromxtforanyt≥1.
(ii)Visdifferentiableatx0>0ifandonlyifthereexistsaunique
optimalpathfromx0.
(iii) Visdifferentiablealmosteverywhere,i.e.,the optimal pathis
uniqueforalmosteveryx0>0.
(iv)isdifferentiablealmosteverywhere.
Proof. SeeLeVanandDana(2003).
Wehaveprovedthattheoptimalcorrespondenceissingle
valuedanddifferentiablealmosteverywhere.Inadditiontothis,
wehavealsoshownthatthereexistsauniqueoptimalpathfrom
almostanyinitialcapitalstock.Theseresultswillprovetobecrucial
inanalyzingthedynamicpropertiesoftheoptimalpaths.
2.3. Dynamicpropertiesoftheoptimalpaths
Apointxsisanoptimalsteadystateifxs= (xs),sothatthe
stationary sequence xs= (xs,xs,...,xs,...) solves the problem:
max{U (xs) :xs∈ (xs)}.Ifxsisdifferentfromzero,thenthe
asso-ciatedoptimalsteadystateconsumptionmustbestrictlypositive
fromInadacondition.Hence,fromEulerequation(6),thissteady
statewillsolve:
u(f(x)−x)=ˇ(x)u(f(x)−x)f(x)+ˇ(x)V (x). (7)
ByProposition3,weknowthatthestationaryplaneveryperiod
equaltoxsisoptimalfromxsifandonlyifitsatisfies
V
xs=u(f(xs)−xs)+ˇ(xs)V
xs. (8)
Followingfrom(7)and(8),thissteadystatewillsatisfy:
u(f(x)−x)=[1−ˇ(x)][1ˇ −ˇ(x)f(x)]
(x) u
(f(x)−x). (9)
Wewillnowprovethattheendogenousrateoftimepreference
preservesthemonotonicityoftheoptimalpathsandprovidethe
conditionunderwhichtheconvergencetoanoptimalsteadystate
isguaranteed.
Proposition8. Theoptimalpathxfromx0ismonotonic.
Proof. Sinceisincreasing,ifx0>x1,wehavex1>x2.By
induc-tion,xt>xt+1,
∀
t.Ifx1>x0,usingthesameargumentyieldsxt+1>xt,∀
t.Nowifx1=x0,thenx0∈(x0).Recallthattheoptimalpathisuniqueaftert=1.Sincex0∈(x0),xt=x0,
∀
t.Itisimportanttonotethatthemonotonicityoftheoptimalpaths
hasbeenprovedwithoutanyassumptiononthecurvatureof
nei-thertheproductionnorthediscountfunction.Itisalready well
knownthatinmulti-sectornonclassicaloptimalgrowthmodels,
onecaneasilyrefertolatticeprogrammingandTopkistheoremin
ordertoprovethattheoptimalpathsaremonotoniciftheplanner’s
criterionfunctionissupermodular(seeAmiretal.,1991).However,
followingthisapproachinamodelwithtimepreferencedepending
onthefuture-orientedresources,Stern(2006)assumesastrictly
concavediscountfunction.
Asamonotonerealvaluedsequencewilleitherdivergeto
infin-ityorconvergetosomerealnumber,thefactthatoptimalcapital
sequencesare monotone proves tobe crucial in analyzing the
dynamicpropertiesandthelong-runbehaviorofourmodel.
Proposition9. Thereexistsanε>0suchthatifsupx>0f(x)<((1−
ε)/ˇm),anyoptimalpathconvergestozero.
Proof. Lettheoptimalpathxconvergetox>0.Then,by
consid-eringtheEulerequationast→∞,weobtain
u(f(x)−x) u(f(x)−x) = [1−ˇ(x)][1−ˇ(x)f(x)] ˇ(x) ≥ 1− ˇm ˇ sup
1−ˇmsup x>0f (x) , where ˇsup≡supx>0ˇ(x). Note that (u/u)=(((u)2−
u.u)/(u)2)>0,i.e.,u/uisincreasing.LetB=argmax
x>0(f(x)−x).
Inaccordancewiththese,wehave:
u(f(B)−B) u(f(B)−B) ≥ u(f(x)−x) u(f(x)−x) ≥ 1−ˇm ˇ sup
1−ˇmsup x>0 f(x) .Defining ε=(ˇsup /(1−ˇm))((u(f(B)−B))/(u(f(B)−B))),
supx>0f(x)≥(1−ε/ˇm) when the optimal path converges to
x>0.Thus,supx>0f(x)<(1−ε/ˇm)impliesthattheoptimalpath
convergestozero.
Inaclassicoptimalgrowthmodel,iff(0)≤1/ˇ,thenthe
opti-malpathconvergestozero.Asweconsidertheendogenousnature
oftimepreference andallowfornonconvexities inthe
produc-tiontechnology, we have slightlydeviated from this condition.
However,itmustbenotedthattheintervalforsupx>0f(x)that
guaranteestheconvergencetozerowillalwaysbeasubsetof(0,
1/ˇm)andthiscannotbeimprovedinanessentialway.
Wewillnow presenttheconditionunderwhich the
conver-gencetoanoptimalsteadystateis guaranteedand analyzethe
behaviorof theoptimalpathswhen we would have uniqueor
multipleoptimalsteadystates.
Proposition10. Assumex0>0.Letinfx>0ˇ(x)=ˇ.Iff(0)>1/ˇ,
thentheoptimalpathconvergestoanoptimalsteadystatexs>0.
Proof. Sinceˇ(·)isincreasing,ˇ=ˇ(0).Notethat[fˇ](·)isa
pathanditconvergestozero.Since[fˇ](·)iscontinuous,there
existsNsuchthatn>Nimpliesf(xn)ˇ(xn)>1.
IfxisanoptimalpaththenitsatisfiestheEulerequation.Hence,
foreveryt,inparticular,fort≥N,wehave:
u(f(xt)−x t+1) =u(f(xt+1)−xt+2)ˇ(xt+1)f(xt+1) +ˇ(x t+1)V (xt+1) >u(f(xt+1)−xt+2)ˇ(xt+1)f(xt+1) >u(f(x t+1)−xt+2), implyingthat f(xt)−xt+1<f(xt+1)−xt+2.
Then,weobtainthat:
0<f(xN)−xN+1<lim
y→0[f(y)−y] =0,
leadingtoacontradiction.
Sincethesequencexismonotonicandbounded,itconvergesto
somexs>0.Nowx
t→xs,xt+1∈(xt),xt+1→xs.Hence,bytheupper
semi-continuityofthecorrespondence,wegetxs∈(xs).
Withconvextechnologyand exogenouslyfixedtime
prefer-encerateinastandardoptimalgrowthmodel,iff(0)>1/ˇthen
thereexistsaunique(non-zero)steadystatetowhichalloptimal
pathsconvergeindependentlyoftheirinitialstate.Inourmodel,
duetotheexistenceofthemaximumsustainablelevelofcapital
stock,whenf(0)>1/ˇ,theoptimalpathconverges
monotoni-callytowardsanoptimalsteadystate.Thisconditioncanberecast
asF(0)>r+ıbydefiningtherateofinterestrby1/(1+r)=ˇand
thefunctionfbyf(k)=F(k)+(1−ı)kwhereFistheproduction
func-tionand ıistherateof depreciationofcapital. Accordingly,in
ourmodel,whenthecostofinvestmentislow,theoptimalpath
convergestoanoptimalsteadystate.
Wewillnowconsidertheoptimalpathdynamicsinthelong
runandshowthatourmodelcansupportuniqueoptimalsteady
statewithglobalconvergenceandmultiplicityofoptimalsteady
stateswithlocalconvergence.Indeed,wewillshowthatourmodel
exhibitsglobalconvergenceevenunderconvex–concave
technol-ogyandmultiplicityofoptimalsteadystatesevenunderconvex
technology.
Case1(Globalconvergence). Assumex0>0.Letinfx>0ˇ(x)=ˇand
f(0)>1/ˇ.Considerthecasewherethereexistsauniquesolution
xsto(9).ByProposition10,weknowthattheoptimalpaths
can-notconvergeto0.Then,anyoptimalpathconvergestotheunique
optimalsteadystatexs,irrespectiveoftheirinitialstate.
Whenthereexistsauniquesolutionto(9),itisclearfromabove
thatitwillbeanoptimalsteadystate.However,whenthereexist
multiplesolutionstothesteady-stateequation(9),thestationary
sequencesassociatedwiththesesteadystatesmayormaynotbe
optimal.Anaturalquestionisthenwhetherthesesolutionswill
constitutea long-runequilibrium intheoptimalgrowthmodel
andifso,willthatenableustoshowwhetheraneconomywith
wealthdependentendogenousdiscountingcanexhibitthreshold
dynamicsevenunderconvextechnologyornot.Devotedtothis
end,first,wewillconsiderthecasewherewehaveexactlytwo
optimalsteadystates,forthesakeofsimplicity.Weshallprovethat
thiswouldimplytheexistenceofacriticalstockandlocal
conver-gence.Second,wewillshowthatourmodelcanindeedsupport
twooptimalsteadystatesinanexample.
Case2(Localconvergence). Assumex0>0.Letinfx>0ˇ(x)=ˇand
f(0)>1/ˇ.Considerthecasewherewehaveexactlytwooptimal
steadystates,xl<xh,forthesakeofsimplicity.Supposethatxhis
unstablefromtheright.Giventheexistenceofamaximum
sustain-ablecapitalstock,asanyoptimalpathfromx0>xhhastoconverge
toanoptimalsteadystate,therewillbeanothersteadystatelarger
thanxh,acontradiction.Hence,xhisstablefromtheright.Itisalso
impossibletohavexlunstablefromtheleft,sincetheoptimalpaths
cannotconvergeto0.Thesealreadyimplytheexistenceofacritical
stockandtheemergenceofthresholddynamics.
Whenever both of theoptimal steady states turn out tobe
saddle-pointstable,itisclearthattherewillexistacriticalstock
ofcapitalbelowwhichtheoptimalpathwillconvergetoxland
abovewhichtheoptimalpathwillconvergetoxh.However,xhcan
bestablefromtherightbutunstablefromtheleft,andsimilarlyxl
canbestablefromtheleftbutunstablefromtheright.Thesecases
canoccuronlywhentheoptimalpolicyistangenttoy=xlineatxh,
andxl,respectively.Itisimportanttonotethatevenunderthese
casesthresholddynamicsemerge.Thefollowingproposition
pro-videstheformalanalysisofthelocalconvergencewhenwehave
twooptimalsteadystates.
Proposition11. Assumex0>0.Letinfx>0ˇ(x)=ˇandf(0)>1/ˇ.
Supposethereareexactlytwooptimalsteadystates.Letthehighand
thelowoptimalsteadystatesbe,respectivelyxhandxl.Thenthere
existsxc∈[xl,xh]suchthatanyoptimalpathxstartingfromx0,
con-vergestoxlifx0<xc,andconvergestoxhifx0>xc.
Proof. Take anyinitialcapital levelsy<z, and let y,z beany
two correspondingoptimalpaths.Since theoptimal
correspon-dence is increasing,weiteratively obtainyt<zt for allt. Hence
limt→∞yt≤limt→∞zt.Thus,ifzconvergestoxl,sodoesy.Ify
con-verges toxh, sodoesz. Let X1={x|there existsan optimalpath
fromxthat converges toxl}, x∗1=supX1, and X2={x|there exists
anoptimalpathfromxthatconvergestoxh},x∗2=infX2.Fromthe
above argument, x∗
1≤x2∗. Suppose otherwise, then there exists
small
1
,2
≥0suchthatx∗1−
1
∈X1,x2∗+2
∈X2,andx1∗−1
>x∗
2+
2
.However,thiscontradictswiththeabove,bysettingx∗2+2
=y<z=x∗1−
1
.Nowwewanttoshowthatx∗1=x2∗.Supposenot,thenthereexistsx∗
3suchthatx1∗<x3∗<x∗2.Butthenan
opti-mal path from x∗
3 can neither converge toxl norto xh by the
definitionsofx∗
1 andx∗2.Therefore,x∗1=x2∗.Definexc=x1∗=x∗2.If
x0<xc=x1∗,takesomesmall
1≥0withx0<x∗1−1∈X1.Bytheaboveargument,anyoptimalpathfromx0 convergestoxl.
Simi-larly,ifx0>xc=x∗2,takesomesmall
2 ≥0withx0>x2∗+2∈X2,which wouldshowthatanyoptimalpathfromx0 convergesto
xh.Noticethatxl∈X1andxh∈X2.Therefore,xl≤supX1=x∗1=xc=
x∗
2=infX2≤xh,i.e.,xc∈[xl,xh].
Remark1. Asithasbecomeclearbynow,ifisnottangentto
y=xlineateitherxlorxh,botharelocallystableandxc∈(xl,xh).
Notethatifisnottangenttoy=xlineateitherxlorxh,itmustbe
thecasethatjumpsovery=xlineatxc.Inthiscase,xcisnotan
unstableoptimalsteadystatebutagenuinecriticalpointleading
tothethresholddynamics.
Ifistangenttoy=xlineatxl,itcanonlybefromabovethe
y=xlinebecausexlisstablefromtheleft.Inthiscasewegetxc=xl.
Similarly,ifistangenttoy=xlineatxh,itcanonlybefrombelow
they=xlineandwegetxc=xh.
Inordertoprovideabetterexpositionofouranalysis,following
Stern(2006),wewillspecifyfunctionalformsinanexampleand
showthatourmodelcanindeedsupporttwooptimalsteadystates
Fig.1.Optimalpolicyafter300iterationsontheinitialzerovaluefunction.
Fig.2. Lowsteadystate=0.0012isoptimal.
Example1. Supposethat
u(c) =1c1− − , f(x) =Ax˛+(1−ı)x, ˇ(x) =−e−(x+)ε , where 0<{A,},0<{˛, ,ε}<1,0<<eε ,ande−ε<<
1.Checkthatf,u,andˇsatisfytheassumptionsets.Weemploy
thefollowingsetoffairlystandardcoefficients:
A=0.5, ˛=0.3, ı=0.15, =0.5, =0.97, =2.5,
=1, ε=0.9.
ItturnsoutthatthemaximumsustainablecapitalstockisA (x0)=
max
x0, ¯x,where ¯xis5.58431,andthereexistthreesolutionsto
(9).Theprecisevaluesare,xl=0.0012,xm=0.5865,andxh=3.8105.
Inordertodeterminewhichoftheseareactuallytheoptimalsteady
states,weanalyzetheoptimalpolicybymakinguseofBellman’s
operator.Fig.1showstheoptimalpolicyforiterationsofthe
Bell-manoperatoronthezerofunctionandindicatesthatxlandxhare
stableoptimalsteady states.Figs. 2–4present thedetailed
pic-turesoftheoptimalpolicyintheneighborhoodofthexl,xmand
xh,respectively.
Atfirstsight,onemightthinkthatthemiddlesteadystatexmis
anunstableoptimalsteadystateasitissurroundedbytwostable
optimalsteadystates.IncontrastwithStern(2006),eventhough
xmisasolutionofthestationaryEulerequation(9),Fig.1strongly
indicatesthatitisnotanoptimalsteadystate.Indeed,ifitwere,
Fig.3.Middlesteadystate(xm=0.586505)isnotoptimal!
Fig.4.Highsteadystate(xh=3.81057)isoptimal.
wouldhavetocrossy=xlineatxm.Moreover,weseethatthereis
genuinecriticalpointatxc=0.6548.Itisclearlynotanoptimalor
nonoptimalsteadystatebecausethenitwouldhavetosatisfy(9).
Asxcisnotanoptimalsteadystate,cannotcrossy=xlineat
xc,i.e.,xc∈/(xc).Ontheotherhand,Fig.1showsthatthegraphof
jumpsovery=xlineatxc.Asisuppersemicontinuous,itmustbe
thecasethatthereexistx
1∈(0,xc)∩(xc)andx1∈(xc,∞)∩(xc).
Therefore,xcisacriticalpointwhichisnotanunstablesteadystate.
Foranyinitialcapitalstocklevellowerthanxc,thesystemwill
faceadevelopmenttrap,enforcingconvergencetoaverylow
cap-itallevelxl,evenunderastrictlyconcaveproductionfunction.On
theotherhand,foranyinitialcapitallevelhigherthanxc,the
opti-malpathwillconvergetoxh.However,ifaneconomystartsatxc,
anindeterminacywillemerge:thesystemcanoptimallyfollowx
1,
andatthesametimeitcanoptimallyfallintoadevelopmenttrap
followingx
1.
Theexistenceofcriticalvalueisactuallyrecognizedsincethe
papersbyClark(1971);MajumdarandMitra(1982);Dechertand
Nishimura(1983)indiscretetimeandSkiba(1978)andAskenazy and Le Van (1999) in continuous time horizon. These studies
are mostly devoted to the analysis of the technology
compo-nent leaving thetime preference essentially unaltered withan
exogenouslyfixedgeometricdiscounting.Theyassumeaspecific
convex–concave technology under which the low steady state
turnsouttobeunstableandhighsteadystateturnsouttobe
sta-blesothatanoptimalpathconvergeseithertozeroortothehigh
steadystate.However,weshowthatevenunderstrictlyconcave
productionfunction,theeconomycanexhibita“trap”sothata
smalldifferenceswillleadtopermanentdifferencesintheoptimal path.
In this respect,ouranalysisis in linewiththeearly
contri-butionsthat emphasizetheimportanceof thesubjective utility
functionandreleasetheoptimalgrowththeoryfromthealmost
takenforgrantedRamsey’sadditiveutilityfunction(e.g.,Bealsand
Koopmans,1969;Iwai,1972).Baseduponthestationaryordinal
utility function that implies a broad class of preference
struc-turesthatarerecursiveandcontainthetime-additivepreference
asa specialcase, theyshowtheexistenceofa criticalstock so
thattheinitialconditionscanaffectthelong-run optimalpath.
Adoptingastrictlyconcavediscountfunctiondependingonthe
future-orientedresources,Stern(2006)analyzesaseriesof
numer-icalexamplesthatalsoexhibitmultiplicityofsteadystateswith
merelocalconvergence.Besidesthese,Kurz(1968)introduces
cap-italasanargumentintheRamseytypeutilityfunctionandshows
alsothattheremayexistmultipleturnpikes.
Inallof thesestudiesmentioned above,thecritical stockof
capitalbelow which theoptimalprogram leadsonlytoforever
diminishingcapitalandgraduallyconvergingtodevelopmenttrap
isanunstableoptimalsteadystate.Ifitsohappensthatthe
ini-tiallevelofcapitalstockequalsthecriticalstock,theoptimalpath
willremainonthatgeneralizedturnpikeforever.However,inour
modelthecriticalstockisdefinitelynotanunstablesteadystateso
thatifaneconomystartsatthisstock,anindeterminacyemerges.
Comparedtotheoptimalgrowthmodelswithexogenoustime
pref-erence,this introducesafundamental differenceintheoptimal
pathdynamics.
3. Competitiveequilibriumwithexternality
Fromnowon,wewillassumethattheproductionfunctionfis
strictlyconcave.Wefirstdefinetheconceptsofequilibriumwith
externalityand competitiveequilibrium. Suppose weare given
asequenceofcapital ˜x =(x0, ˜x1,..., ˜xt,...)∈ [0,A(x0)]∞andthe
associatedsequenceofdiscountfactors ˜ˇ=( ˜ˇ1,..., ˜ˇt,...)where
˜
ˇt=ˇ (˜xt) ,
∀
t≥1. Giventhis fixedsequence ˜ˇ∈ˇ,ˇm∞,
con-siderthefollowingproblem:
maxx ∞
t=0 t s=1 ˜ ˇs u (f(xt)−xt+1) subjectto∀
t,0≤xt+1≤f(xt), x0>0,given.Let the solutionto theabove problem bex=(x0, x1, ...,xt,
...).Itdependson ˜ˇ,hence ˜x.Wewritex= ( ˜ˇ)= (ˇ(˜x)),and
hencex=(˜x).Anequilibriumwithexternalityassociatedwith
x0isasequenceofcapitalstockx∗=(x0,x1∗,...,x∗t,...)suchthat
x∗=˚(x∗).Alistofsequences(x∗,c∗,p∗,q∗)isacompetitive
equilib-riumwithexternalityofthiseconomyifthefollowingaresatisfied:
(a) c∗∈ ∞
+,x∗∈ ∞+,p∗∈( 1+\{0}),q∗∈R++,
(b)c∗solvestheconsumer’sproblem:
maxc ∞
t=0 t s=1 ˇ(x∗ s) u(ct) subjectto ∞ t=0 ∗ p tct≤q ∗x 0+∗where∗isthemaximumprofitofthefirm,
(c)x∗solvesthefirm’sproblem:
∗=maxx ∞
t=0 p∗ t(f(xt)−xt+1)−q∗x0 subjectto∀
t≥0,0≤xt+1≤f(xt), x0>0,given,(d)themarketsclearateveryperiod:
∀
t≥0,c∗t +x∗t+1=f(xt∗),x∗0=x0.
Thepurposeistoprovetheexistenceofacompetitive
equi-libriumofthisintertemporaleconomy.Inordertodoso,wewill
firstprovethatthereexistsafixedpointof˚,namelyan
equi-librium.Later,underanadditionalassumption,wewillshow
thatthisequilibriumisindeedacompetitiveequilibriumofthe
intertemporaleconomywithendogenoustimepreference.
Thevaluefunctionassociatedwiththeproblem(PE)takesthe
followingform: V(x0,ˇ1,ˇ2,ˇ3,...)=maxx ∞
t=0 t s=1 ˇs u(f(xt)−xt+1) subjectto 0≤xt+1≤f(xt), x0>0,given.VsatisfiestheBellmanequation:
V(x,ˇt+1,ˇt+2,..)=maxy ∈[0, f (x)]
u(f(x)−y)+ˇt+1V(y,ˇt+2,ˇt+3,..)
.Proposition12. ThefunctionViscontinuouswiththetopologyin
Rforx0andtheproducttopologyforˇ∈[ˇ,ˇm]∞.Moreover,itis
strictlyconcaveinx0. Proof. Let U(x0,ˇ,x)= ∞
t=0 t s=1 ˇs u (f(xt)−xt+1)wherex∈(x0).OnecanshowthatUiscontinuousin(x0,ˇ,x)for
thetopologyinRforx0andtheproducttopologyfor(ˇ,x).Itiswell
knownthat(·)isacontinuouscorrespondencefromR+intothe
spaceofsequencesendowedwiththeproducttopology(see,for
instance,LeVanandDana,2003).Since
V(x0,ˇ)=max
U(x0,ˇ,x)|x∈(x0)
,fromthemaximumtheorem,Viscontinuous.Thestrictconcavity
ofVwithrespecttox0followsfromthestrictconcavityoffandu.
Lemma2. Themapˇ iscontinuouswithrespecttotheproduct topology.
Proof. Itisstandard.
Proposition13. ˚iscontinuouswithrespecttotheproduct topol-ogy.
Proof. Fromthemaximumtheorem,thesolutionx= (ˇ)isupper
semi-continuouswithrespecttoˇ.Asthesolutionto(PE)isunique,
itisacontinuousfunctionwithrespecttoˇ.Since ˜ˇ=ˇ(˜x),themap
Toattainanequilibrium,theinitialsequenceofdiscountinghas
tobeconsistentwiththelevelthatis assumedwhentheagent
makeshersingledecisions.Thissuggeststhatthefixedpointsof˚
arecandidatesforcompetitiveequilibria.
Proposition14. ˚hasafixedpoint.
Proof. ˚isacontinuousmappingfrom[0,A(x0)]∞into[0,A(x0)]∞.
Sincethedomainiscompactfortheproducttopologyandalsoa
convexset,Schaudertheoremconcludesthatthereexistsx∗∈[0,
A(x0)]∞whichsatisfiesx∗=(x∗).
Proposition15. If(x0,x∗1,x∗2...)isanequilibriumwithexternality
associatedwithx0,then(x1∗,x2∗,x3∗...)isanequilibriumwith
exter-nalityassociatedwithx∗
1.
Proof. LetV(x,ˇt+1,ˇt+2,..)bethevaluefunctionatperiodt+1.
Foranytwehave
V(x∗
t,ˇt+1,ˇt+2,..) =maxy∈[0,f(x)]
u(f(x)−y)+ˇt+1V(y,ˇt+2,ˇt+3,..)
=u(f(x∗t)−x∗t+1)+ˇt+1V(x∗t+1,ˇt+2,ˇt+3,..)
Particularlyfort≥2,weseethat(x∗1,x∗
2,...)isanequilibrium
withexternalityassociatedwithx∗
1.
Proposition16. Anyequilibriumwithexternalityx∗associatedwith
someinitialcapitalstocksatisfiestheEulerequation:
u(f(x∗
t)−x∗t+1)=u(f(x∗t+1)−x∗t+2)ˇ(x∗t+1)f(xt+1∗ ). (10)
Proof. Itiseasy.
Fortherest,weassumethatthefunction[ˇf](·)isdecreasing
and[ˇf](0)>1.Notethatlimx→∞[ˇf](x)<1.
Proposition17. Theequilibriumwithexternalityassociatedwithx0
isunique.
Proof. Taketwoequilibriawithexternality(x0,x1,x2,...)and
(x0,x1,x2,...)associated withx0.Supposethatx1>x1.Let the
associatedconsumptionpathsbe(c0,c
1,...)and(c0,c1,...).
First,wewillprovebyinductionthatx
t>xt,forallt≥1and
c
t<ct for all t. Trivially, c0 =f(x0)−x1<f(x0)−x1=c0. Then
by (10), u(c
1)[ˇf](x1)=u(c0)>u(c0)=u(c1)[ˇf](x1) which
impliesu(c
1)>u(c1)as[ˇf]isadecreasingfunction.Hence,c1<
c
1.
Now suppose x
t>xt and ct<ct for some t. f(xt)−
x
t+1=ct<ct =f(xt)−xt+1, so xt+1 >xt+1. Again, by (10),
u(c
t+1)[ˇf](xt+1)=u(ct)>u(ct)=u(ct+1)[ˇf](xt+1) which
impliesu(c
t+1)>u(ct+1 )as[ˇf]isadecreasingfunction.Hence,
x
t+1>xt+1andct+1 <ct+1.
Nowconsiderthemaximizationproblemgiventhediscounting
sequenceˇ(x).Thepathx
isfeasiblefromx0,andyieldsstrictly
higherutilitythanxgiventhediscountingsequenceˇ(x)because
c
t<ctforallt.However,thisisacontradictionwithxbeingafixed
pointwhich wouldimplythat x itselfistheuniquemaximizer
giventhediscountingsequenceˇ(x).Thereforex
1=x1.Letx1=
x
1=x1.
Notice that Proposition 15 implies that (x1,x2,x3,...) and
(x1,x2,x3,...)aretwoequilibriawithexternalityassociatedwith
x1.Thenapplyingthesamearguments,wegetx2=x2.Finally,by
inductivelyapplyingProposition15andtheabovearguments,we
getx
t=xt forallt.
Corollary1. If[ˇf](x0)=1forsomex0,thentheuniquefixedpoint
associatedwithx0isthestationarysequence(x0,x0,x0,...).
Proof. Letˇ=ˇ(x0)andconsiderthebasicneoclassicalmodelwith
fixeddiscountfactorˇ.Astheinitialcapitalisalreadyx0whichis
steadystatecapitallevelintheneoclassicalgrowthmodel,(x0,x0,
x0,...)willsolvethesocialplannerproblem.Thismeansthat(x0,
x0,x0, ...)istheuniquemaximizergiventhediscountsequence
(ˇ,ˇ, ...).Thus,itisafixedpointassociatedwithx0,andbythe
abovepropositionitistheuniquefixedpointassociatedwithx0.
Proposition18. Any equilibriumwithexternality monotonically
convergestoapointxssatisfying[ˇf](xs)=1.Moreover,ifthereare
multiplesuchpoints,itmonotonically convergestotheonethat is
closesttotheinitiallevelofcapital.
Proof. Onecanshowthatthesetofsolutionsto[ˇf](x)=1
con-stituesacompactinterval,sayX=[xsmin,xs
max].Notethatitiswell
possibletohavexs
min=x s
maxorxmins <x s
max.Lettheinitialcapital
levelbex0andtheassociatedequilibriumwithexternalitybex∗.
ByCorollary1,ifx0∈Xwehavex∗=(x0,x0,...)whichconverges
totheclosestsolutionx0to[ˇf](x)=1.
Considerx0<xsmin.Supposex∗doesnot–monotonically
con-verge–toxs
min.Thenx∗asasequenceeitherpassesfromtheregion
(0,xs
min]totheregion(x
s
min,∞),orstrictlydecreasesintheregion
[0,xs
min],atleastonceatsomepointintime.Formally,thereexists
someteitherwithx∗
t ≤xsmin<x∗t+1orwithx∗t+1<x∗t ≤xsmin.
Supposethatforsomet,xt≤xmins <xt+1.As1=[ˇf](xmins )≥
[ˇf](xt+1∗ ),(10)impliesu(c∗
t)≤u(ct+1∗ ).Thenbyconcavityofuwe
getf(x∗t)−x∗
t+1≥f(x∗t+1)−x∗t+2,whichyieldsx∗t+1<xt+2∗ asx∗t <
x∗
t+1.Hence, x∗t+1<x∗t+2 and xsmin<x∗t+2.Inductively we obtain
xs
min<x∗t+1<xt+2∗ <....Duetothemaximumsustainablecapital
stock,x∗cannotdiverge,soitconvergesmonotonicallytoalevelx
thatisstrictlyhigherthanx∗
t+1.Takingthelimitof(10)forthefixed
pointx∗astgoestoinfinity,weseethatthelimitxofx∗hasto
satisfy[ˇf](x)=1.Thusxs
min<xt+1∗ <x≤xsmaxandhencext+1∗ ∈X.
ThenCorollary1impliesthattheequilibriumwithexternality
asso-ciatedwithx∗
t+1is(x∗t+1,xt+1∗ ,...),however,Proposition15implies
thatitis(x∗t+1,x∗
t+2,...).Recallthatthefixedpointfromx∗t+1 is
unique,sowegetacontradictionasx∗
t+1<x∗t+2.
Now suppose that for some t, x∗
t+1<x∗t ≤xsmin. As 1= [ˇf](xmins )≤[ˇf](x∗t+1),(10)impliesu(c∗ t)≥u(c∗t+1).Thusweget f(x∗t)−x∗ t+1≤f(x∗t+1)−xt+2∗ ,whichyieldsx∗t+1>x∗t+2asx∗t >xt+1∗ . Hence,x∗
t+2<x∗t+1<xsmin.Inductivelyweobtain...<xt+1∗ <x∗t ≤
xs
min.Takingthelimitof(10),weseethattheonlypossibilityisx∗
convergestozero.Ifx∗convergestozero,thereexistssomelargeT
suchthat[ˇf](xt∗)>1forallt>T,because[ˇf](0)>1.Thenu(ct∗)>
u(c∗
t+1)fort>Tby(10).Hence,f(xt∗)>f(x∗t)−x∗t+1=ct∗>c∗T>0
forallt>T.However,f(x∗t)convergestozeroalongwithx∗bythe
continuityoff.Contradiction.
Thereforewehaveshownthatthefixedpointassociatedwith
x0<xsminmonotonicallyconvergestotheclosestsolutionx
s minof
theequation[ˇf](x)=1.
Thecaseofx0>xsmaxisanalogoustox0<xsmin.
Justtorecall,thefixedpointx∗of˚,namelytheequilibrium
withexternalityassociatedwithx0,solvestheproblem
maxx ∞
t=0 t s=1 ˇ(x∗ s) u(f(xt)−xt+1) subjectto 0≤xt+1≤f(xt), x0>0,given.Wewillnowprovethatsuchafixedpointisindeeda
competi-tiveequilibrium.
Theorem 1. Assume that f(0)ˇ>1. Define q∗=1,p∗ t =
(
ts=1ˇ(x∗s))u(f(x∗t)−xt+1∗ ), c∗t =f(x∗t)−x∗t+1,
∀
t. Then, (c∗, x∗,Proof. First, we prove that p∗ is in 1 +\
0
. Clearly p∗t >
0. We need to prove that
∞
t=0 p∗ t = ∞ t=0 ( t s=1 ˇ(x∗ s))u(f(x∗t)−x∗t+1)is bounded. We have shown that x∗ convergesto some xs∈X.
[ˇf](0)>1suggestsxs
min>0,hencethesequenceu(f(x∗t)−x∗t+1)
isbounded.Thus,thesum
∞
t=0
p∗
t existsinR+.
Since x∗ ∈[0, A(x0)]∞ we get x∗∈ ∞+. By the constraints of
theproblem(PE),f(A(x0))≥f(x∗t)≥x∗t+1≥0.Hence,f(A(x0))≥c∗t =
f(xt∗)−x∗
t+1≥0,i.e.,c∗∈ ∞+.
Theproofsthatx∗isasolutiontotheproblem(PP)andc∗solves
theconsumer’sproblem(CP)arestandard(see,e.g.,LeVanand
Dana,2003).
Finally,onecaneasilynotethatthemarketclearingcondition,
c∗
t +x∗t+1=f(xt∗)issatisfied.
ByProposition18,theoptimalpathsconvergetoapointx∗with
f(x∗)ˇ(x∗)=1. (11)
Clearly,suchapointisuniqueif[ˇf](· )isstrictlydecreasing.
Forsimplicity,letusassumeaCobb–Douglasproduction
tech-nology,f(x)=Ax˛,whereA>0denotesthetotalfactorproductivity.
Theremayexistuniqueorcontinuumofsteadystatesinthismodel
withendogenoustimepreferencedependingonthecharacteristics
ofthediscountfunction.
Example2. Letˇ(x)=−e−(x+)ε,where0<,0<ε<1,0<<
(1−˛)eε
,and(e−ε/1−˛)<<1.Notethatˇsatisfiesallofour
assumptions.Wewillnowshowthatˇ(x)f(x)isstrictlydecreasing.
Wehave
ˇ(x)f(x)
=−e−(x+)εA˛x˛−1=A˛x˛−2
e−(x+)ε1−˛+εx(x+)ε−1−(1−˛).
Then,forx>0,
ˇ(x)f(x)<0ifandonlyif
h(x):=exp
−(x+)ε1−˛+εx(x+)ε−1−(1−˛)<0.
Itisclearthathisdifferentiable,henceobtainsitsmaximumatthe
boundariesorcriticalpoints.Attheboundaries,
h(0)=exp
−ε[1−˛] −(1−˛) =(1−˛)
exp
−ε−=−(1−˛)ˇ(0)<0,and lim x→∞h(x)=x→∞lim(1−˛)e −(x+)ε+ lim x→∞e −(x+)εεx(x+)ε−1 −(1−˛)=−(1−˛)<0.
Soatanycriticalpointx,wehaveh(x)=0.Accordingly,
e−(x+)εε(x+)ε−1
−1−˛+εx(x+)ε−1+1−(1−ε) x
x+
=0.
Hence,(1− ˛+εx(x+)ε−1)=1−(1−ε)(x/x+).Thenwehave
h(x)=e−(x+)ε
1−(1−ε) xx+
−(1−˛)
atanycriticalpointx.Itisclearthattheexpressione−(y+)ε[1−
(1−ε)(y/y+)] is maximized at y=0. Therefore, we obtain
thath(x)≤e−ε
−(1−˛).Recallthat(e−ε/1−˛)<,hence
e−ε
−(1−˛)<0,implyingthatatanycriticalpointx,h(x)<0.
Thus, h is a negativevalued function, implyingthat ˇ(x)f(x) is
strictlydecreasing.
Anydiscountfunctionˇunderwhichˇ(x)f(x)turnsouttobea
strictlydecreasingfunctionimpliesauniquesteadystateandglobal
convergence.Theuniquesteadystateisbothlocallyandglobally
determinate.Moreover,themodelmayevenpossessacontinuum
ofsteadystates,anditdependsontheinitialconditionastowhich
oneisrealizedinthelongrun.Hence,theeconomyexhibitsno
tendencytowardglobalconvergence;animportantdeparturewith
respecttotheneoclassicalgrowthmodel.
Example3. Considerthefollowingdiscountfunctionthatconsists
ofthreeparts: ˇ(x)=
⎧
⎪
⎪
⎨
⎪
⎪
⎩
mx+n, x<a; 1 f(x) = 1 A˛x1−˛, a≤x≤b; −x, b<x.where A>0, 1>˛>0, (A˛/2−˛)1/1−˛>b>a>0, m=1−˛/A˛a−˛,
n=(1/A)a1−˛,=(2−˛/A˛)b1−˛,and=(1−˛/A˛)b2−˛.Letscheck
theassumptions.ˇisalreadydifferentiableintheregions(0,a),(a,
b),(b,∞).Somesimplealgebraforcheckingtheleftandrightlimits
ofˇandˇatthepointsaandbshowsthatthisspecificationofm,
n,andmakesˇisdifferentiableataandb.Thereforeˇis
differ-entiable,alsoimplyingcontinuity.ˇ(0)=n>0,henceˇrangesinto
R++.Also,asm>0,A>0,˛∈(0,1),and>0,ˇisstrictlyincreasing.
Asˇisstrictlyincreasing,thesupremumofwhichis,isstrictly
lessthanoneby(A˛/2−˛)1/1−˛>b.Moreover,thesupremumof
thederivativeofˇisalsoboundedwhichiseasytoseebysome
algebra.
Noticethatthroughout thewholeinterval[a,b],ˇ(x)f(x)=1,
thereforewehaveacontinuumofsteadystates[a,b].Aswehave
proventhatanyequilibriumwithexternalitywithinitialcapital
x0∈[a,b]willbeconstantovertime,theinitiallevelofcapitalstock
lowerthanawillleadtoanincreasinglymonotonicconvergence
towardsawhereastheinitialtheinitiallevelofcapitalstockhigher
thanbwillresultindecreasinglymonotonicconvergencetowards
b.
Remark 2. Existenceofthecriticalvaluexc andcontinuumof
equilibriawithstrictlyconcaveproductionfunctionarepeculiarto
ourmodelofendogenoustimepreference.
4. Conclusion
Inthispaper,wepresentthedynamicimplicationsof
endoge-noustimepreferencedependingonthestockofwealthinaone
sectorgrowthmodel.Weprovewithoutanyassumptiononthe
curvaturesoftheproductionandthediscountfunctionsthat
opti-malpolicyissinglevaluedanddifferentiablealmosteverywhere
in theplanner’sproblemand theoptimalpathsaremonotonic.
We consider the optimal path dynamics in the long run and
showthatourmodelcanexhibitglobalconvergenceevenunder
aconvex–concavetechnologyandmultiplicityofoptimalsteady
states even under a convex technology. Indeed, we show that
there existsa critical stockof capital belowwhich theoptimal
programleadsonlytoforeverdiminishingcapitalandgradually
convergingtodevelopmenttrapandabovewhich theeconomy
convergestothehighsteadystate.Wealsoshowthatifitso
hap-pensthattheinitiallevelofcapitalstockequalsthecriticalstock,
whichisnotanoptimalornonoptimalsteadystate,an
indetermi-nacyemerges.Moreover,themultiplierssystemassociatedwith
anoptimalpathisproven tobethesupportingpricesystemof
concerningthepropertiesofoptimal(equilibrium)pathsare pro-vided.
Extensionstoourmodelareobviouslypossibleandincludethe
considerationsofuncertainty,heterogenousagentsandstrategic
interactions.Theseareinourresearchagenda.
References
Amir,R.,Mirman,L.J.,Perkins,W.R.,1991.One-sectornonclassicaloptimalgrowth: optimality conditions and comparative dynamics. International Economic Review32(3),625–644.
Askenazy,P.,LeVan,C.,1999.Amodelofoptimalgrowthstrategy.Journalof Eco-nomicTheory85(1),24–51.
Askri,K.,LeVan,C.,1998.Differentiabilityofthevaluefunctionofnonclassical optimalgrowthmodels.JournalofOptimizationTheoryandApplications97, 591–604.
Araujo,1991.Theoncebutnottwicedifferentiabilityofthepolicyfunction. Econo-metrica59(5),1383–1393.
Azariadis, C., Stachurski, J., 2005. Poverty Traps. In: Aghion, P., Durlauf, S. (Eds.), In: Handbook of Economic Growth, vol. 1. Elsevier, Amsterdam, pp.295–384.
Barro,R.J.,1997.DeterminantsofEconomicGrowth.MITPress,Cambridge,MA. Barro,R.J.,Sala-i-Martin,X.,1991.Convergenceacrossstatesandregions.Brookings
PapersonEconomicActivity1,107–182.
Beals,R.,Koopmans,T.C.,1969.Maximizingstationaryutilityinaconstant technol-ogy.SIAMJournalofAppliedMathematics17(5),1001–1015.
Becker,G.S.,Mulligan,C.B.,1997.Theendogenousdeterminationoftimepreference. TheQuarterlyJournalofEconomics112(3),729–758.
Benhabib,J.,Farmer,R.E.A.,1996.Indeterminacyandsectorspecificexternalities. JournalofMonetaryEconomics37,21–443.
Benhabib,J.,Perli,R.,1994.Uniquenessandindeterminacy:transitionaldynamics withmultipleequilibria.JournalofEconomicTheory63,113–142.
Benveniste,L.M.,Scheinkman,J.A.,1979.Onthedifferentiabilityofthevalue func-tionindynamicmodelsofeconomies.Econometrica47,727–732.
Clark,C.W.,1971.Economicallyoptimalpoliciesfortheutilizationofbiologically renewableresources.MathematicalBiosciences17,245–268.
Das,M.,2003.Optimalgrowthwithdecreasingmarginalimpatience.Journalof EconomicDynamicsandControl27,1881–1898.
Dechert,W.D.,Nishimura,K.,1983.Acompletecharacterizationofoptimalgrowth pathsinanaggregatedmodelwithnon-concaveproductionfunction.Journalof EconomicTheory31,332–354.
Duran,J.,LeVan,C.,2003.Simpleproofofexistenceofequilibriuminaone-sector growthmodelwithboundedorunboundedreturnsfrombelow. Macroeco-nomicDynamics7,317–332.
Epstein,L.,1987.Asimpledynamicgeneralequilibriummodel.JournalofEconomic Theory41,68–95.
Frederick,S.G.,Loewenstein,G.,O’Donoghue,T.,2002.Timediscountingandtime preference:acriticalreview.JournalofEconomicLiterature40,351–401. Hamada,K.,Takeda,Y.,2009.OntheRoleoftheRateofTimePreferencein
Macroe-conomics:ASurvey.In:InternationalTradeandEconomicDynamics:Essaysin MemoryofKojiShimomura,.Springer,Berlin/Heidelberg,pp.393–420. Harrison,G.W.,Lau,M.I.,Williams,M.B.,2002.Estimatingindividualdiscountrates
inDenmark:afieldexperiment.AmericanEconomicReview92,1606–1617. Ikeda,S.,Ohtake,F.,Tsutsui,Y.,2005.Timediscountrates:ananalysisbasedon
eco-nomicexperimentsandquestionnairesurveys.InstituteofSocialandEconomic ResearchDiscussionPaperNo.638,OsakaUniversity.
Iwai,K.,1972.Optimaleconomicgrowthandstationaryordinalutility—afisherian approach.JournalofEconomicTheory5,121–151.
Kurz,M.,1968.Optimaleconomicgrowthandwealtheffects.International Eco-nomicReview9,348–357.
Lawrance,E.C.,1991.Povertyandtherateoftimepreference:evidencefrompanel data.JournalofPoliticalEconomy99,54–75.
LeVan,C.,Dana,R.A.,2003.DynamicProgramminginEconomics.KluwerAcademic Publishers.
Lucas,R.E.,Stokey,N.,1984.Optimalgrowthwithmanyconsumers.Journalof Eco-nomicTheory32,139–171.
Majumdar,M.,Mitra,T.,1982.Intertemporalallocationwithanon-convex technol-ogy:theaggregativeframework.JournalofEconomicTheory27,101–136. Mantel,R.R.,1998.Optimaleconomicgrowthwithrecursivepreferences:
decreas-ingrateoftimepreferences.EstudiosdeEconomia25,161–178.
Mitra,T.,Ray,D.,1984.Dynamicoptimizationonnon-convexfeasibleset:some generalresultsfornon-smoothtechnologies.ZeitschriftfurNationaokonomie 44,151–175.
Nishimura,K.,Venditti,A.,2006.Indeterminacyindiscrete-timeinfinite-horizon models.In:Dana,R.A.,LeVan,C.,Mitra,K.,Nishimura,K.(Eds.),Handbookon EconomicGrowth:Vol.1:TheDiscreteTimeHorizon.Kluwer.
Obstfeld,M.,1990.Intertemporaldependence,impatienceanddynamics.Journalof MonetaryEconomics26,45–75.
Quah,D.T.,1996.Convergenceempiricsacrosseconomieswith (some)capital mobility.JournalofEconomicGrowth1,95–124.
Samwick,A.,1998.Discountratehomogeneityandsocialsecurityreform.Journal ofDevelopmentEconomics57,117–146.
Skiba,A.K.,1978.Optimalgrowthwithaconvex–concaveproductionfunction. Econometrica46,527–539.
Stern,M.L.,2006.Endogenoustimepreferenceandoptimalgrowth.Economic The-ory29,49–70.
Stokey,N.L.,Lucas,R.,Prescott,E.,1989.RecursiveMethodsinEconomicDynamics. HarvardUniversityPress.
Uzawa,H.,1968.Timepreference,theconsumptionfunction,andoptimumasset holdings.In:Wolfe,J.(Ed.),Value,CapitalandGrowth,Chicago.Chicago Univer-sityPress.