MIT LIBRARIES 3
9080 03130 2398
|
HB31 .M415working
paper
department
of
economics
CHARACTERIZING
PROPERTIES
OF
STOCHASTIC
OBJECTIVE
FUNCTIONS
Susan Athey 96-1 Oct.1995massachusetts
institute
of
technology
50
memorial
drive
Cambridge, mass. 02139
CHARACTERIZING
PROPERTIES
OF
STOCHASTIC
OBJECTIVE
FUNCTIONS
Susan
Athey1
INTRODUCTION
Thispaperstudies optimizationproblems wherethe objectivefunctioncan bewritteninthe form V(8)
=
\jt{s)dF{s\6), where;risapayofffunction,F
is a probabilitydistribution, and6
andsare real vectors. For example,thepayofffunctionn
mightrepresentanagent'sutilityora firm'sprofits, thevectorsmightrepresentfeaturesofthecurrent stateoftheworld,andtheelementsof6 mightrepresent an agent's investments, effort decisions, other agent's choices, or the nature of the
exogenousuncertainty in the agent'senvironment.
The economic
problem underconsideration often determinessome
propertiesofthe payofffunction; forexample,autilityfunctionmight be
assumed
tobe nondecreasing andconcave, whileamultivariate profit function might have sign restrictions
on
cross-partial derivatives. These assumptionsthen determine asetIT of admissiblepayofffunctions.We
mightthenwishtoanswerquestionssuch as: Isasetof investments
6
isworthwhile?Are
there decreasingreturns tothose investments?Does
one investment increase thereturns to anotherinvestment?To
answerthose questions,we
needtoknow
whetherV(Q) satisfiestheappropriate properties,i.e., nondecreasing, concave, or supermodular.1The
goalofthispaperistodevelopmethods suchthat,given aparticularpropertyP
(suchasnondecreasing),asetof admissiblepayoff functionsII,andaparameterizedprobability distributionF,
we
can determinewhetherthefollowing statementistrue:J7i(s)dF(s;6)
satisfiesproperty
P
in6
foralln
inIL (1.1)Thatis,given
P
andII,we
wishtodescribethesetofprobability distributionswhichsatisfy(1.1).Thispaperrestrictsattention tothecase
where
the propertiesP
andsets IIare"closedconvexcones."
We
definea propertyP
tobea"ClosedConvex Cone" (CCC)
propertyifthesetof functionswhichsatisfy
P
isclosed(under an appropriate topology),positivecombinations of functionsintheset are also intheset,andconstant functionsare in theset. ImportantexamplesofCCC
propertiesinclude nondecreasing,concave,supermodular,anypropertywhichisdefinedby
placingasign restrictiononapartialderivative,and combinations oftheseproperties. Thus,ifthepayofffunction representsa
consumer'sutility,IIcouldbethe setofunivariate,nondecreasing,concave payofffunctions;ifthe
1Intuitively,
afunctionissupermodularif.givenanytwoarguments ofthe function, increasingoneincreases the returns toincreasing the other.
When
thefunctionisdifferentiable,thisamountstopositive cross-partial derivativesbetween every pairof arguments. Supermodularityisimportantinthecontext of comparativestatics(seeTopkis(1978),payoff functionrepresentsafirm's profits asa functionof
two
complementaryqualityinnovations,n
couldbethesetofbivariate,nondecreasing,supermodularpayofffunctions.Forsets IIandforproperties
P
whichareCCC,
thispaperdevelops a characterization ofwhichprobability distributions satisfy(1.1). Checking(1.1) directlyisdifficultbecause,ingeneral,Ifmight bea verylargeset.
Thus
we
ask,When
isitpossibletofinda smallersetofpayofffunctions,denotedT,sothatforanyprobability distributionF, statement(1.2)belowwillbetrueifandonlyif(1.1)is
true?
J
7t(s)dF(s;8)satisfiesproperty
P
in6forall%
inT
( 1.2)We
canthinkofT
asa"testset" forII: ideally,T
isasetofpayofffunctionswhichissmallerandeasier tocheckthanII,butitcan beusedtotestwhether
J7C(s)dF(s;0)satisfiesproperty
P
in6,givenonlytheinformationthat
K
isinthe largersetII.Usingthese ideas,
we
canrestatethegoalofthepaper:we
wanta theorywhichhelps us determinethebest"testset"fora givenII.
We
proceedintwo
steps. First,we
examinethecasewheretheproperty
P
is"nondecreasing"; second,we
study otherCCC
properties.The
firstcasehasbeenthefocusofthe literatureonstochasticdominance, wheredifferentauthorshavestudieddifferentsets II. 2
Thispaperunifies andextends the existing literatureonstochasticdominance,providing an exact characterizationof themathematical structureunderlyingall stochastic
dominance
theorems.We
furtherprovideanalgorithmforgenerating
new
stochasticdominance theorems whichrelaxesthead hocdifferentiabilityandcontinuityassumptions whicharecommon
inthis literature.3 In thesecondpartofthepaper
we
show
thatthemethodsfromstochasticdominance can beappliedtocharacterizewhen
V{6)satisfiesotherpropertiesP,includingconcavityandsupermodularity.We
now
provideanoverview of ourresults. In thefirstpartofthispaper,we
show
thatfortheproperty nondecreasing,ifIIisa closedconvex cone andcontains constantfunctions,thenthebest
T
isthe setof"extremepoints"ofII. Just asabasisgenerates alinearspacevia linearcombinations, so asetofextremepointsgeneratesa closedconvex conevia positivelinearcombinations andlimits.
Thus,
we
show
thatifwe know
onlythatapayofffunctionk
liesintheclosedconvex coneII,andwe
wanttoknow
ifV(6)isnondecreasing,itisequivalenttocheckthatV(0)isnondecreasingon
a smaller2
Inparticular,the univariate stochasticdominanceproblem has beenstudiedby Rothschild andStiglitz(1970,1971)and
HadarandRussell (1971); notable contributionstothe multivariateproblemincludeLevyand Paroush(1974),Atkinson and Bourguignon(1982),andMeyer(1990). Shakedand Shanthikumar (1994) provide areferencebookonthe subjectof stochasticordersandtheirapplicationsineconomics,biology,andstatistics.
3Brumelle and Vickson (1975)
takeafirststeptowardsrelaxing theseassumptionsandidentifying themathematical structurebehind stochasticdominance; in contrast to theirwork, whichprovides only sufficientconditions fora stochasticdominancerelationship,thispaperprovidesanexact characterizationofallstochasticdominancetheorems.
setofpayofffunctions, theextremepointsof n. Itwilloftenbe
much
easier to verifymonotonicityof V(6) forthe setofextremepointsofn
thanforn
itself. Inthispaper, theprocedureof usingtheextreme pointsof
n
asatestsetfor II willbe referredtoas the "closedconvex cone"method
of provingstochasticdominancetheorems.Examples
fromthe existing stochasticdominanceliterature,whichare specialcasesofthis result,includeFirstOrderStochastic
Dominance (FOSD),
wheren
isthe setofunivariate,nondecreasing payofffunctions,andSecond OrderStochasticDominance
(SOSD), whereTIisthe setofunivariate,concave payofffunctions. In the case of
FOSD,
the setofextreme pointsisthe setofone-step functionswhicharezerouptosome
constant,and onethereafter. ThesearepicturedinFigure 1. Inthecase of
SOSD,
thesetofextremepointsisthe setof "angle," or"min"functions,picturedinFigure2,where eachfunctiontakes the
minimum
ofitsargumentandsome
constant.The
closedconvex conemethod
for stochasticdominancehasbeenrecognizedandexploredinthecontextofparticular setsofpayofffunctions (Topkis,1968;Brumelle andVickson, 1974; Gollierand
Kimball, 1995).4 However,thispapergoes
beyond
the existing literature intwo
respects. First,by
developing appropriateabstract definitions todescribestochasticdominancetheorems,we
areabletomake
general statements aboutthe entire class ofstochasticdominance
theorems, including thosewhich havenotyetbeenconsideredintheeconomicsliterature. Second and
more
importantisthe fact thatwe
proveanew
resultaboutthisclassof theorems:we
proveformallythatthe"closedconvex cone" approachtostochasticdominanceisexactlythe rightone.By
thiswe mean
that (1.1)and(1.2) returnthesame
answerforeveryprobability distributionF
ifandonlyiftheclosedconvex coneofT
(uniontheconstantfunctions,ifthesearenotinT)isequaltoIT;
no
otherT'swillalwaysmake
(1.1)and(1.2) equivalent. Inthiscase,clearlythesmallestsetwhichgeneratesIIisthebestsettocheck.
The
secondpartofthispaper showsthatthe"closedconvex cone"method
canalsobeappliedto characterizeotherpropertiesofV(6).We
asktwo
questions:First,forwhatpropertiesP
istheclosedconvex cone approachvalid?
And
second,forwhatpropertiesP
istheclosedconvex cone approachexactly therightone, as in thecase ofstochastic
dominance?
We
firstshow
thatifP
isaCCC
property,thentheclosedconvex cone approach can always be usedtocharacterize
when
V(6)satisfies P.We
thenfinda subsetofCCC
properties,whichwe
call"Linear DifferenceProperties," forwhichwe
canshow
thattheclosedconvex cone approachisexactlythe rightoneforcheckingwhetherV(0)satisfies P.
Examples
of LinearDifference Properties include monotonicity, supermodularity, concavity,andpropertieswhichplace signrestrictionson
partial derivatives. Combinations oftheseproperties,however,arenotingeneralLinear DifferenceProperties,althoughsuch combinationsare
4Independently, Gollier
CCC
properties. Table Isummarizesproperties whichareCCC
propertiesandLinear DifferenceProperties.
TABLE
ILINEAR DIFFERENCE
PROPERTIES
AND
CLOSED
CONVEX
CONE
PROPERTIES
Property Closed
Convex Cone
Linear Difference PropertyNondecreasing Yes Yes
Supermodular Yes Yes
Concave/Convex(Multivariate) Yes Yes
SignRestrictiononaPartial Yes Yes
Derivative
Constant Yes
No
Nondecreasingand Concave Yes
No
ArbitraryCombinationsof Yes
No
CCC
PropertiesArbitraryCombinationsof Yes
No
LDP
PropertiesFor
many
sets ofpayofffunctions,n,
whicharecommonly
studiedineconomics, the existing literatureon
stochasticdominancehasimplicitly identified theextremepointsof thosesets. Insuchcases, theproblemof characterizing a Linear Difference Property (suchassupermodularity)
becomes
quitestraightforward:simply looktothe the existingliteraturetofind theappropriatetest set,T, for the setofpayofffunctionsIIunderconsideration. Then,usingthe resultsofthispaper,
we know
that (1.2) characterizes the setof parameterizedprobability distributions forwhichthe stochastic objectivefunctionwillbe supermodularforallpayoff functionsinII.
We
illustratethis techniqueby
showinghow
the closedconvex cone approach can be usedtocharacterize theproperty supermodularityforseveralimportantclassesof payofffunctions. These
resultscaninturnprovidesufficient(andsometimesnecessary) conditionsforcomparativestatics
conclusions. Inparticular,
we
examineapplications inprincipal-agent theory,welfareeconomics,andThis paper proceedsas follows. InSection2,
we
introducea motivatingexample,theproblemofa risk-averse agent's choice ofeffort. Inthisproblem,
we
characterizethepropertiesnondecreasing, concave,and supermodularforthe agent'sexpectedutilityfunction.Our
generalanalysisofstochasticdominancetakes placeinSection3.
We
provideanexact characterization ofstochasticdominancetheorems, highlightingtheimportantroleplayed
by
linearityofthe integral.We
furtherextendourresulttoincorporate "conditionalstochasticdominance." Section4developscharacterizationsof other
propertiesof f7t(s)dF(s;d)andprovides applicationsoftheproperty supermodularity. InSection5,
we
analyze conditions underwhich
functions oftheform
J
K(x,s)dF(s;6) are supermodularor
concave in (x,d), showing
how
to apply stochasticmonotonicity resultsto this problemas well.Section6concludes.
2
MOTIVATING EXAMPLE:
A
RISK-AVERSE AGENT'S
CHOICE OF
EFFORT
Inthissection,
we
present a motivatingexample, wherewe
analyzehow
a risk-averse agent's choice ofeffort affects herexpected payoffs, andhow
that choice ofeffort interacts with exogenousparameters
which
describethe probability distribution. Formally,we
characterize the properties nondecreasing, concave,and supermodularfortheagent'sexpectedutilityfunction. Theseresultsare specific examples—
exampleswhere
the payofffunctionis nondecreasing and concave—
ofastochasticdominancetheorem, a"stochasticconcavity theorem,"anda"stochasticsupermodularity theorem." These examplesillustratethe parallel structureunderlyingthethree classesof theorems.
Considerarisk-averse agent
whose
utility {it)dependsontheoutputofastochasticproduction technology,wheretheoutputisdenoteds. Supposethattheagent'seffort(c)affectsthe probability distributionofs. Further,consider a parametertwhichrepresentsexogenous changesinthe stochasticproduction technology. For example,achangeintmightrepresentaworker
moving
from one jobto another. Then,we
canwrite theagent'sproblemasfollows:max
fk(s)•dF(s;e,t)-c(e)
Firstobservethatitisnottrivialto verify thatV(e,t)
=
\it(s) dF(s;e,t)isnondecreasingineffort:anincrease in effortwhichisproductiveonaveragemightnotincreaseexpectedutility ifthe effort also
increasesthe riskinessofthe distribution. Second,notethatifV(e,t)-c(e)isconcaveineffort,then
thefirstorder conditions
(FOCs)
characterizetheoptimum,a propertywhichisusefulintheanalysissome
economicproblems,suchasprincipalagent problems. Further,ifV(e,i)failstobe concavein e,thenthere existssome
linearcostfunction c(e)=
ae
suchthattheFOCs
failtocharacterizethenondecreasingint. If V(e,t) failstobesupermodular, thenevenif
V
isconcave, there existssome
linear costfunction c(e)
=
ae
suchthat e*{i) fails to be nondecreasing in t (this follows fromMilgrom
andShannon
(1994);seeTheorem
A.lintheAppendix). Thus,therequirementthatV(e,t) satisfiesthepropertysupermodular(orconcave,respectively)issufficient for thedesiredeconomicconclusion, andfurtheritcannot berelaxedaslongas
we
requirethattheconclusion holdsforall linear cost functions.Letus firstidentifyconditionsunder whichV(e,i)isnondecreasingine.
The
followingwell-known
resultisadaptedfromtheRothschildandStiglitz(1970,1971)work
onstochasticdominance (wherewe
suppresstinthe notation):Proposition 2.1 Thefollowing twoconditions are equivalentforallprobabilitydistributions F(-,e):
(i)
For
allK
nondecreasingand
concave, \lt(s) dF(s;e) isnondecreasingine. (ii) Thefollowing aresatisfied:(a) \sdF(s;e) isnondecreasingine. a
(b)
For
all a, -\F(s;e) isnondecreasingine.Intuitively,forarisk-averseagent
who
likesincome,effortwillincreaseexpectedutility ifandonlyif effortincreasesthemean
income(condition(ii)(a))andreducesthe "riskfunction" (condition(ii)(b)).Thisresultisoftenusedinthefinanceliterature;ithasbeencalledSecond Order MonotonicStochastic
Dominance
(SOMSD).
In this paper,
we
willwork
with a restatement of Proposition 2.1,which
emphasizes thatconditions(i)and(ii)are actuallysymmetricconditions.
The
following propositionisequivalenttoProposition2.1 (where
A
1bethespaceofprobability distributionsdefinedonSR, and Si indicates the
extendedrealline,thatis, SR
u
{-00,00}):Proposition 2.1' Thefollowingtwoconditionsare equivalentforall F(-;e)eN:
(i)
For
alln
inthe set YlSOM=
[ti\k:SR-»%
nondecreasing, concave},J7t(s) dF(s;e) is nondecreasingine.
(ii)
For
allyinthe setT
SOM=
{y\y(s)=
min(a,s),ae
9t}, jy(s) dF(s\e) isnondecreasingConditions (i) and(ii)ofthisPropositionsimply restateconditions(i)and (ii)ofProposition2.1,
except that
we
have replaced -JF(s;e) with jmin(a,s)dF(s;e) in condition (ii). It isstraightforwardto verify(usingintegrationbyparts) that theformer termisnondecreasingifandonly
ifthelattertermis. This conditionisnot usually associated with
SOMSD
(seeBrumelle and Vickson(1975) foranexception). This
way
ofwriting theSOMSD
theoremillustratesthe mathematicalstructureunderlyingthe stochasticdominancetheorem.
As
written inProposition2.1',the resultsaysthatinsteadofcheckingthat j n(s) dF(s;e) isnondecreasingin X for allpayofffunctionsin the
relativelylargeset, II50M,itisequivalenttocheckthat \y(s)dF(s;e) isnondecreasingine forall
payoff functionsin thesmallerset,
T
S0M.The
twosetsofpayoff functionsarepicturedinFigure2.The
relationshipbetweenthetwo
setsof payoff functionscanbedescribedas follows: T1S0M isequaltotheclosedconvex coneofthesetwhichistheunion of
T
S0Mandthefunctionsy{s)
=
1andy(s)=
-1. Thatis,by
takingpositivecombinationsandlimitsof sequences(or nets)of elements ofthelatter set,
we
cangenerateanyfunctionin TISOM,whichis itselfaclosedconvex cone. In Section3,
we
willshow
that this"closedconvex cone"relationshipbetweenTlSOM and
r
SOM holdsforallstochasticdominancetheorems.
Now,
letusask adifferent,butrelated, question:When
isJx(s) dF(s;e)concavein effort? This
problem
was
addressedby
Jewitt (1988),who
analyzes conditions underwhich
the FirstOrder Approach(FOA)
toanalyzing principal-agentproblemsisvalid.The
FOA
requires thatiftheagent'sFOCs
aresatisfied,thentheagent'schoiceofeffortmust beoptimal. ExtendingJewitt'sanalysis,we
canshow
thatthe sufficient conditionshe derivesare in factnecessary.5The
followingresult isanalogoustoProposition2.1:
Proposition 2.2 Thefollowing twoconditions are equivalentforall
F(;e)e
A
1:
(i)
For
all%
inthe set YlS0M, \k(s)•dF(s;e) isconcaveine. (ii)
For
allyinthe setT
SOM',\y(s) dF(s;e) isconcaveine.
Proof: Thiswillbeestablished asa simplecorollaryof
Theorem
4.1below, together with Proposition2.1.
Q.E.D.
5Jewittfirstderives conditionsunderwhichthe agent'sutilityfunctionisincreasingand concaveinoutput,giventhe optimalcontract;hethenaddresses the questionof concavity oftheexpectedutilityfunctionin effort. It isonlythelatter
Proposition2.2saysthatexpected payoffsareconcaveine forallpayoff functionsin
U
SOM ifandonlyifexpected payoffs areconcaveineforallpayofffunctionsin
T
SOM. Again,this theoremisusefulbecausecondition (ii)is easier to checkthat condition(i): the set
r
SOM ismuch
smaller.Condition (ii) can be interpreted asrequiring thatthere are decreasing returns to e in terms of increasingthe
mean
anddecreasingthe "riskfunction." Notethatthe pairofsetsof payofffunctions,(
n
SOM
,
T
SOM
),isthe
same
inbothpropositions.Now
we
turn toask afinalquestion:When
istheoptimal choice ofeffortmonotone
nondecreasingint,whichparameterizesthe stochasticproduction technology?
More
precisely,how
can t affecttheprobability distributionover outputin sucha
way
that monotonicity oftheoptimal effort int isensured, without any additional information about the cost of effort function or the agent's preferences? This question has not been answeredin the existingliterature; thus,the following proposition provides a
new
insight into thecomparativestaticsproblem. Notethatthispropositionistrue irrespectiveofwhetherexpectedutilityismonotoneineffort.
Proposition 2.3 Thefollowing three conditions are equivalentforall F(-;e,f)e
A
1:
(MCS) For
allcostfunctions cand
allK
in IlSOM, e'(t)=
argmax\K(s)dF(s;e,t)-c(t) ise Js
monotone
nondecreasingint.6(i)
For
alln
e
TlS0M, \K{s)dF{s;e,t) issupermodularin{e,t\ (ii)For
allye
T
S0M, jy(s) dF(s;e,t) issupermodularin(e,t).Proof:
The
equivalence ofparts(MCS)
and(i)followsdirectlyfromMilgrom
andShannon
(1994), as stated in
Theorem
A.1intheAppendix.The
equivalence of(i)and(ii)willbeestablished asacorollaryof
Theorem
4.1belowtogetherwith Proposition2.1.Q.E.D.
The
formaldefinitionof supermodularity (andthecomparativestaticstheoremwhichreliesuponit) canbefoundintheAppendix. Intuitively, V(e,t)issupermodulariftincreasesthe returns toeffort.Proposition2.3providesnecessaryandsufficientconditionsfor
monotone
comparativestaticsinthisproblem. If(ii)isviolated,then
we
canconstructpayofffunctionsandcost functionssuchthatthechoiceofeffortisnotmonotonicint. Thus,
we
haveidentified theexact conditionswhichensuremonotone
comparativestatics. Condition (ii)requiresthateandtarecomplementaryinterms of increasingthemean
ofthe probability distributionandinterms of reducingtherisk.The
intuitionisstraightforward:sincearisk-averse,income-lovingagentlikeshigh expectedreturnsandlowrisk (as
"In general, theoptimal emaybe aset. Then,thistheoremrequires that thesetbe nondecreasingintheStrongSet Order, as definedintheAppendix.
shown
inSOMSD),
variableswhicharecomplementaryinincreasingthemean
anddecreasingthe risk arecomplementaryinincreasingexpectedutilityofsuch anagent. Notethatifeitheroneofeand/ does notaffectthemean
orthe riskinessof the agent'sincome,thecorrespondingcomplementarityconditionsare satisfiedtrivially.
Notice that Propositions 2.2 and 2.3 have astructure
which
is very similarto the existingstochasticdominanceresult,as stated inProposition2.1. In Section3,
we
willbuild aframeworkforanalyzing Proposition2.1 andotherstochasticdominancetheorems. In Section4,
we
willformalizethe relationshipbetweenPropositions2.1through2.3.
3
MONOTONICITY OF STOCHASTIC
OBJECTIVE
FUNCTIONS
The
goal ofthis section is to provide a unifiedframework
for analyzing stochasticdominance
theorems. In Section 3.1,we
introduce aframework which
incorporates the existing stochasticdominanceliterature,arguingthateachstochasticdominance theoremdescribesarelationshipbetween twosetsofpayofffunctions. InSection3.2,
we
provearesultwhich
characterizesamathematicalrelationshipbetweentwosetsofpayoff functionswhichisequivalenttothe relationshipdetermined
by
the stochasticdominancetheorem. Section3.3extendsthe result tothecaseofconditionalstochastic
dominance.
3.1
A
UnifiedFramework
forStochasticDominance
In this section,
we
introducetheframework which
we
willuse to discuss stochasticdominance
theoremsasanabstract classof theorems,andtodrawpreciseparallelsbetweenstochasticdominancetheoremsandothertypesof theorems.
Letusfirstconsideranotherwell-known exampleof astochasticdominancetheorem,FirstOrder
Stochastic
Dominance
(FOSD). Thistheorem can bestated asfollows(where IA(s) istheindicator
functionforthesetA):
Proposition 3.1 Thefollowingtwoconditions are equivalentforall
F(;0)e
A
1:
(i)
For
alln
e
n
TOs
{k\k:9i-»%
nondecreasing},JK(s)dF(s;6)isnondecreasinginft
(ii)
For
ally
e
T
FOm
{y\y(s)
=
IM
(s),ae
Sfi}, jy(s)dF(s;6) isnondecreasinginftThistheoremhasthe
same
structure as theSOMSD
theorem, Proposition2.1
'.
The
setsofpayoff functionsare illustrated inFigure 1. Thistheoremsaysthatinsteadof checkingthatexpected payoffsarenondecreasing in
8
forallnondecreasing payofffunctions,IF
,itisequivalenttocheckthatfunctionsofupperintervals.
The
latter setofpayoff functionsismuch
smaller,andso condition(ii)iseasier tocheckthatcondition(i); condition(ii)can be reducedto a restrictionwhichrequiresthat
1
-
F(a;6),thecomplement
ofthecumulativedistributionfunction,isnondecreasingin8
forae9?.The
latterrequirementisthemore
standardway
ofstatingFOSD.
Thereare
many
otherexamplesofstochasticdominancetheorems,some
with multiplerandom
variables;
we
will discussotherexamples belowin Section4.5. Ingeneral, stochasticdominance theoremsallhaveaparallel structure, as illustrated inPropositions2.1'and3.1. However,different stochasticdominance theoremsarecharacterizedby
different(11,0pairs.The
SOMSD
theoremstates thatthe pair(U
SOM,T
SOM
)satisfiesaparticular relationship; the
FOSD
theoremstatesthatthe pair(IT™,
r
F0)satisfiesthe
same
relationship.To make
thisrelationship precise,we
introduce a formaldefinitionofthestatement
"A
pairofsetsofpayofffunctions,(n,D,satisfiesastochasticdominancetheorem." Ifthatstatementistrue,then
we
willsaythat(11,0isa stochasticdominance
pair.We
willusethe abstract definition tomake
statementsaboutthe classofstochasticdominancetheorems,and
how
thatclassoftheoremsrelatestootherclassesof theorems.We
allowformultidimensionalpayofffunctions andprobabilitydistributions, usingthe followingnotation:the setofprobability
distributions
on
SR" is denoted A", with typical elementF:SR
n—
»[0,1]. Further, for a given parameterspace 0,
we
will use the notation A"etorepresent the setofparameterized probabilitydistributions
F
:9?"x
-»[0,1] suchthatsuchthat F(;0)eA" forallds 0.Definition 3.1 Consider a pair ofsets of payofffunctions(TI,r), with typical elements
K
:9T
—
>SRand
y
:9?"—
> SR. Thepair(II,T) isastochasticdominance
pair ifconditions (i)and
(ii)are equivalentforallparameter spaces withapartialorderand
allF
€
A
n e:(i)
For
allttgU,
jn(s)dF(s;6)is
nondecreasing
in 6.(ii)
For
ally eT,
Jy(s)dF(s;6)
is
nondecreasing
in 6.Further,
we
definetheset I.SDT tobethe setofall(11,0pairswhichare stochasticdominancepairs,as follows:2
5Dr=
{(II,r)|(n,r)isastochasticdominancepair}Thus,
when
agiven (11,0pairisastochasticdominancepair,we
write (II,T)e
Z
SDr.Definition3.1 clarifiesthestructureofstochastic
dominance
theorems. Stochasticdominance theoremsidentify pairs ofsets of payofffunctionswhich
havethe following property: givenaparameterized probability distribution F, checking that all of the functions in the set
(fy(s)dF(.s;0)|y
er|
are nondecreasing. Stochasticdominance
theorems are useful because, in general, thesetT
issmaller thantheset II.We
canthinkofthisdefinition asa statement aboutthe equalityoftwosets. First, letus definethe set ofadmissible parameter spaces for the property "nondecreasing" together with probabilitydistributionsparameterized
on
those spacesas follows:V
ND=
{(F,0)|©has apartialorderandF
eA
n
ej
Now,
we
canrephrasethe definition as follows:(11,0isastochasticdominancepairifj(F,0)€
V
nN
M
7r(s)dF(s;d)isnondecreasingon V/reIli=
|(F,0) e
V
ND \JY(s)dF(s;0)n isnondecreasingonVy
e
Ij
Definition3.1 differsfromthe existing literature(i.e.,Brumelleand Vickson, 1975),inthatthe existingliteraturegenerallycomparestheexpected valueof
two
probability distributions,sayF
andG,viewing
J7t(s)dF(s) andJ7t(s)dG(s) as
two
different linearfunctionalsmapping
payofffunctionsto Si. In contrast,byparameterizingthe probability distributionand viewing
Jn(s)dF{s\6)asabilinear
functional
mapping
payofffunctionsand(parameterized)probability distributions tothereals,we
areableto createan analogy betweenstochastic
dominance
theorems andstochasticsupermodularity theorems,an analogywhichwould
notbeobvious usingthestandardconstructions.The
utilityofthis definition willbecome more
clearwhen
we
formalizethe relationshipbetweenstochasticdominance andstochasticP
theorems,forotherpropertiesP
(suchassupermodularity).To
provide specific examples ofpairs ofsets ofpayoff functionswhich
satisfy stochasticdominancetheorems,
we
summarizethree univariate stochasticdominance theoremsinTableIt. Thereare potentially
many
otherunivariate stochasticdominance theorems(forexample,theorems wherethe setofpayofffunctionsimposesrestrictionson
the third derivativeof thepayofffunction);however,we
willsimplyreportthe threemostfamiliar univariate stochasticdominance theoremshere.TABLED
COMPONENTS
OF UNIVARIATE STOCHASTIC
DOMINANCE
THEOREMS"
SetsofPayoffFunctions,II SetsofPayoffFunctions,
T
(i)
U
F0=
{n\n:SR->%
nondecreasing}r
F°={
r
\y(s)=
I [a„)(s),aeX}
(ii)n
5°s
{n\n:SR->%
concave}r
S0={Y\Hs)
=
-s} iu{y|y(.y)
=
min(a,s),aeSRJ
(iii)
YlSOM
=
\n
k
:SR—
» SR, nondecreasing, 1 concaver«w
_
|y|y
(j)_
min(fl,s), ae9?}Each(II,r)pairinTableII isa(univariate)stochasticdominancepair.
TableII (iii)correspondstoa
SOMSD
theorem,asshown
inProposition2.1
',whileTableII(i)
correspondstoa
FOSD
theorem, Proposition3.1. Let usnow
illustratethe interpretationof Definition3.1 andTableIIwith athirdexample:Second OrderStochastic
Dominance
(SOSD),shown
inTableII (ii). Itisknown
thatJK{s)dF{s;8)
isnondecreasingin
6
forallunivariate,concave payofffunctionsifand onlyif \min(a,s)dF(s;d) is nondecreasingin
8
forallaeSi,
and\sdF(s;6)
does notdepend on 6. Observethatboth y(s)
=
s and y(s)=
-s
areincludedinT
so;thisforcesthemean
ofthe distribution tobeboth nonincreasingandnondecreasing,and henceconstantin6.
Thereare
many
otherstochasticdominance theoremsinadditiontothe univariateexamplesgiven above.Levy
and Paroush(1974) deriveresultsforbivariate functions,whileMeyer
(1990) extends theseresultsand examinessome
multivariatestochasticdominance theoremsas well.We
will reportsome
of theseresults inSection4.5,wherethemain
objectiveistoapply theseresults toproblemsofstochasticsupermodularity.
3.2 ExactConditionsfor aStochastic
Dominance Theorem
Inthissection,
we
studynecessaryandsufficientconditionsforthe pair(11,1")tobeastochasticdominancepair.
We
wanttospecify theexactmathematicalrelationshipwhichisequivalenttothestatementthat (II,T)
e
Z
sor.We
willfirstdiscussourresultanditsimplications;then, inSections3.2.1and3.2.2,
we
willprovidethemathematicalargumentsunderlyingtheresult.The main
resultofthissectionisthat (II,T)e
I,SDT ifandonlyifthefollowing statementistrue: theclosure(underanappropriate topology) oftheconvex cone ofIIu
{1,-1} isequaltotheclosure(underthattopology) oftheconvex coneof
Tu
{1,-1}, where {1,-1} denotesthesetcontainingthetwoconstantfunctions, {n{s)
=
1}u
{it{s)=
-1}.We
formalizethisusingthefollowingnotation:cc(nu
{1,-1})=
cc(ru
{1,-1}) (3.1)In thecontext ofspecific setsofpayofffunctions
n,
the existing literature identifies similar,but stronger sufficientconditions for thecorresponding stochasticdominance
theorems, usingamore
restrictivenotion of closure(i.e.,atopology with
more
opensets)thantheone whichwe
will identifybelow. For example, Brumelle and Vickson(1975) arguethat (3.1)issufficientforthe {Yl,T) pairs
shown
inTableIItobestochasticdominancepairsunderthetopology ofmonotone
convergence. In thispaper,usingthe abstract definitionwe
have developedfora"stochasticdominancepair,"we
are able toformallyprovethat the sufficientconditionshypothesizedby
Brumelleand Vickson(1975)are in fact sufficient foranystochasticdominance
theorem, notjust particularexamples. Further, the resultthat (3.1)isalsonecessaryfor (Tl,T) tobeastochasticdominancepairisanew
contributionofthispaper.
We
now
arguethat this resulttellsuswhen we
cannotdo
betterthan theclosedconvex cone method. First,observethatunlessT
isa subsetofn,thereisno
guaranteethatstochasticdominance theoremsprovide conditionswhichare easier tocheckthanJ7C(s)dF(s;0)nondecreasing in
6
foralljrell. For example,
n
might beasetwhichisnot aclosedconvexcone. Itsclosedconvex cone might bemuch
larger,anditmightnotbepossibletofindasubset,T,ofthatclosedconvex coneforwhichiseasier tocheckthat expectedpayoffsare nondecreasingin 6. Thus, (3.1) indicates that stochasticdominancetheorems,asdefinedinDefinition3.1, aremostlikely tobeuseful
when
IIisa closedconvexcone. For example,inthecaseofFOSD,
we
considerthe setof payofffunctions IIF0.Itiseasytoverify thatpositive scalarmultiplesand convex combinationsofnondecreasing functions
arenondecreasingfunctions, as are limitsofsequencesornetsofnondecreasingfunctions. Finally,
constant functionsare also in ITF0.
When
ITcontainstheconstant functionsandisa closedconvexcone,(3.1)becomes:n
=
cc(ru{l,-l}) (3.2)In principle,the mostuseful
T
is the smallestsetwhose
closedconvex
coneisII.However,
in general, there willnotbeauniquesmallestset.To
seethis,consider the caseofFOSD,
wherethe setr
FOcontains indicator functions ofupperintervals.
By
taking limitsof sequences of convexcombinations of elements of
T
FOu
{1,-1},andappropriatelyscalingthesefunctions,
we
cangenerateanynondecreasingfunction. However,
we
canalsodefine thesett
FO=
{y\y(s)=
I[a„,($), a
e
q\, whereQ
representsthe rationals,andnotethatn
F0=
cc(fF0u
{1,-1}) .While
f
F0
c
T
FOpracticethesmallerset
f
F0 isnot anyeasier tocheck. Thus, stochasticdominance
theorems aregenerallystatedso that
T
isthesmallestclosedsetwhose
closedconvex coneis II;we
willcallsucha setthe"extremepoints"ofII.Finally,because(3.2)isnecessaryandsufficientfor(H,T)tobeastochasticdominancepair
when
n=cc(IIu{l,-l}),we know
thatwe
cannotdo anybetterthanlettingT
bethe setofextremepoints ofII: thereisnosmaller or easier-to-check closedsetof payofffunctions, f, suchthat (II,f) is a stochasticdominancepair. Thisiswhatwe
mean
when
we
saythatwe
have provedthattheclosedconvex cone methodisthe "right"approach.
In the next
two
subsections,we
prove that (3.1) characterizes stochasticdominance
pairs.Considertheproblemof orderingtwoprobability distributions,
F
1and
F
2,where
we
saythatasetofpayoff functionsIIorders
F
1higherGower)than
F
2if \7r(s)dF\s)>
(<)\x(s)dF2(s) forallK
inII; ifneither inequalityholdsforalln
inII,thenwe
saythatthe distributionsarenotorderedby
II. InSection 3.1.1,
we
firstshow
that II orderstwo
distributions exactly thesame
as the setcc(IIu{l,-l}).
We
thenshow
thattwo
sets, II and T, order arbitrary pairs of probabilitydistributions in the
same
way
ifand
onlyif(3.1) holds, thatis,theirclosedconvex
conesarethesame. InSection3.2.2,
we
use theseresults toprovethe characterizationofstochasticdominancepairsaccordingto(3.1).
3.2.1
The Main
Mathematical
ResultsWe
beginby
provingsome
generalmathematicalresults aboutlinearfunctionalofthe form\Kdfi.
Our
results are variationsonstandardtheoremsfromthe theoriesoflinearfunctional analysis,topologicalvector spaces, andlinearalgebra; the
main
contributionofthis sectionistodefinetheappropriatefunction spacesandtopologyandrestatetheprobleminsuch a
way
thatwe
canadaptthesetheoremstosolvethe stochasticdominanceproblem.
We
willwork
with aclassof objectsknown
as finitesignedmeasureson
9t",denotedW.
Any
finite signedmeasure\ihas a"Jordandecomposition" (see
Royden
(1968), pp. 235-236), sothatfi
=
/i+—
it
,where eachcomponent
isapositive,finitemeasure.We
willbeespecially interested in finitesignedmeasureswhich havethepropertythat \d/i=
0,sothat \d/i+=
\dfi~. Denotethesetofallnon-zerofinitesignedmeasures which havethispropertyby^»;
we
are interested inthis setbecause elements ofthis setcan alwaysbewritten as \i=
—
[F
1-
F
2],whereitisapositive scalar,andF
1and
F
2areprobability distributions; likewise, forany
two
probability distributionsF
1and
F
2 , themeasure
F
1Inthissection,
we
willproveresultswhichinvolveinequalitiesof theform\n
d/i>
0, wherefieg*. Sincepositive scalarmultipleswillnotaffectthisinequality,and becausethe integraloperator islinear,
we
canwithoutlossofgenerality interpretthisinequality as jJt(s)dF1(s)
>
J7t(s)dF2(s) for theappropriatepairofprobabilitydistributions.The
formernotationwillbeeasier towork
withinterms of provingour
main
results, andfurtheritwillbeusefulinprovingresultsinSection4aboutstochastic
P
theoremsforpropertiesP
other than "nondecreasing."Now
letus begin our formalanalysis. LetP* bethe setofbounded, measurable payofffunctionson 9?". Definethe bilinearfunctional f5:P'
x%"
-»SR by
/J(7T,/i)=
jady..Then
f}{P",7X«)isaseparated duality: thatis,forany
&*
(J?,thereisan
eP*
suchthat j3(7T,/i,)*
Pix,^),
and forany it
x
*
n
2,thereis a \ieW*
suchthat fi{Kvii)*
/J(7T2,/0-7Our
choiceof(P",3£") issomewhat
arbitrary:allof ourresultsholdif
we
letA" bea subset ofmeasurable payofffunctionson9?" andwe
let
B
nbe anysubsetofTff,solongasp\AnJBn)isa separatedduality. Forconsistency
we
willusethe pair{P*,WP)inour formalanalysis.We
now
construct ourtopology,where
would
like to findthe coarsesttopology (thatis,the topology withthefewestopensets)suchthatthesetofallcontinuouslinearfunctionalson
P*(thedualof P*)isexactly the set {/J(-,//)|/ie2£*}; this is the
weak
topology a{P',tKH)
on
P*.By
Bourbaki
(1987, p. 11.43), this topology uses as a basisneighborhoods
of theform
N(7z;£,(Hi,—,Hk))
=
\ftmax|/J(7T-7t,fl,)\
<
£
k
wherethereisaneighborhoodcorrespondingtoeachfiniteset (/^,...,^t )<z
W
and each e>
0.We
will returnto clarify the relationshipbetweenthistopologyandothertopologiesinthediscussionfollowing
Theorem
3.4,below.Let cc(A) denote theconvex cone ofasetA,andlet
A
denote theclosure ofA
(where the topologyisunderstoodtobe a^P*,?^)inthediscussionbelow,unlessnoted).We
now
usethe fact thatthefunctional P(7t,^i)=
\ndiiislinearandcontinuousinitsfirstargumenttoprovethefollowing simplelemma.
The
proof ofthislemma
is elementary, butwe
state ithere because all ofthemathematicalresults inthispaperbuilduponit.
Lemma
3.2 Considera setof payofffunctionsIIcP».Then
thefollowing twoconditionsareequivalentforall jl&g>:
(i)
For
alln
e
II,\nd\i>
0.7Theboundedness assumption guarantees
thatthe integralofthepayofffunctionexists. Itispossible toplaceother
restrictionsonthepayofffunctionsandthespaceoffinitesignedmeasures sothatthe pairisaseparated duality,inwhich case theargumentsbelowwouldbe unchanged;forexample,itispossible torestrictthepayofffunctionsandthesigned measuresusinga"boundingfunction."Formorediscussionsofseparateddualities,seeBourbaki(1987,p.n.41).
(ii)
For
all7tecc(Uu
{1,-1}),jnd/i>0.
Proof: First,
we
show
(i)implies(ii).Fix ameasure fieg'.Then
thefollowing implicationshold:
For
allK
ell,jndn>0
=>
Forall;rellu
{1,-1},J;rd/i>0
(Since J d[i
=
-j
dji=
0).=>
Forallff€cc(nu
{l.-l}),J^d//>0
(Since [tt, rf/z>
and Jn
2dfi>0
implies J[a,^,
+
a
27T2]d/i=
a^TTid\i+
a
2\n
2dfi>0
when
a,,a2>0).
=>
For
a//K
ecc(IIu{l,-l}), J7rd//>
(Recall that foracontinuousfunction/,
/(A)
c
/(A).The
implicationthen followsbecausethehalf-space [x
e
9^|jc>
0}isclosed,andthe linear functional P(;/l)iscontinuousforall/*.)That(ii)implies(i)followsbecauseIT
c
cc(IIu
{1,-1}). £?.£.£>.Lemma
3.2canberestatedanotherway:thesetofmeasures//e^»
forwhich jndfi>
forall7Te
IIisexactlythe
same
as the setofmeasures [ie£*
forwhichJtt^
>
for all ite
cc(Ilu
{1,-1}).Formally,
{/i€^-|J^d/x>0
V;renj
=
{//Gf*IJ^rd/t>0
V;recc(IIu{l,-l})}As
discussedabove,becausepositive scalarmultiplesdo
not reversethe inequalitiesand becausethefunctionalfiislinear initssecondargument,thelatterequalityisequivalenttothefollowing:
FKF^A^JTrdF^JKdF
2Vn
e
III=
If^F
2eA^TtdF
1ZJndF
2V/r
e
cc(IIu
{!,-1})Thatis,forany
two
probabilitydistributions, IIordersthetwo
distributions exactlythesame
asBuilding
from
thislemma,
we
turn toprove themain
mathematical theorem underlyingthe characterizationofstochasticdominancetheorems,aswellasthe characterizationsof otherproperties inSection4. Thistheoremmakes
use ofthe linearityofthefunctional /J(tt,jU) in n.The
proofthat(ii) implies (i)relies on
Lemma
3.2,while theproofthat (i) implies (ii)makes
useofa standard separating hyperplane argument. Notethatthechoiceof topology,whichdeterminesthemeaningofclosure,iscriticalforthe applicationofthe separatinghyperplane theorem.
Theorem
3.3 Consider a pair ofsetsof payofffunctions(II,I*),whereIIand
T
aresubsetsofP
t. Thenthefollowingtwoconditions are equivalent:
(i)
lii€$'\J7tdn>0 V;rell} =
j/ie£-|Jy^>0 Vyerj.
(ii)
cc(U
u
{1,-1})=
cc(ru
{1,-1}).Proof: Firstconsider(ii)implies(i). Supposethat
cc(IIu
{1,-1})=
cc(Tkj{1,-1}).Then
we
have: j//e2'|Jffd!/i£
VTrellj
=
|/X6f*|J^d//>0
V;recc(IIu{l,-l})j (ByLemma
3.2.)=
In e -f
^jydfi>
V76cc(ru{l,-1})}
(Byassumption.)=
{nep\jydfi>0
Vyer}
(ByLemma
3.2.)Now
we
provethat(i)implies(ii). Define II=
cc(U
u
{1,-1})andF =
cc(Tu
{1,-1}).Suppose(withoutlossofgenerality) that there existsa
y
e
f
suchthaty
e
fl.We
know
thatthe
(XP'&P)
topologyisgeneratedfromafamilyofopen,convexneighborhoods. Recallfrom abovethatthesetof continuouslinearfunctionalson P»isexactlythe set
{p(-,H)\fisTX'}. Usingthesefacts,acorollary tothe
Hahn-Banach
theorem8impliesthat8SeeDunford
and Schwartz (1957,p.421),Kothe(1969,p.244)fordiscussionsofthe relevanttheorems aboutthe separationofconvexsets. SeealsoMcAfeeandReny(1992)forarelated applicationoftheHahn-Banachtheorem, where theseparatinghyperplanealsotakes theform of an element ofJ*.
since flisclosedandconvex,there existsa constant candafi.
e
7JF (aseparatinghyperplane)sothat P(x,n.)
>
cforall iteft,and /J(y,//,)<
c.Since {1,-1}
€
flandis flconvex, €fl as well. Thus, /?(0, /x.)=
>
c.Now
we
willargue
we
cantakec=
withoutlossofgenerality. Supposenot.Then
there existsan
e
flc
suchthat c
<
/?(#,//.)=
c<
0.Choose
anypositive scalarp
suchthatp
>
—
>
1(which cimpliesthat
pc<c).
Since flisa cone, pit€
ft. But, f}(pk,iL.)= pc<c,
contradicting thehypothesisthat/J(7T,/i.)
>
cforallK
eft. So,we
letc=
0.Because {1,-1}
e
ft,and fi{n,^u)> foralln
e
ft,we
concludethat /3(1,//.)=
-p\l,/i.)=
0,andthus jJ^,
=
0. So, fi.eg', andwe
haveshown
that \7tdfi,>
forall 7Tgft,butJyrfyU,
<
0,whichviolatescondition(i). Q.E.D.The
proofofTheorem
3.3 is analogous tothe proofofthe "bipolartheorem"from
the theory of topological vector spaces(seeSchaefer, 1980,p. 126). ThisresultisdifferentbecauseW,
not£",isthedualofthespaceP";thisaccountsfor theinclusionoftheconstant functionsincondition(ii) of
Theorem
3.3.As
above,we
canrestatecondition(i)ofTheorem
3.3as follows,withoutlossofgenerality:If
,F
2e A" I\k dF
l>
\izdF
2 Vtte
n
j= If
1 ,F
2e
A" IjydF
1>
jydF
2Vy
e
T
jThe
theoremfirststatesthatiftwosets ofpayofffunctionshavethesame
closedconvexcone,then theywillorderanypairofprobability distributions thesame
way. Second,thetheoremstatesthatiftwosetsofpayoff functions orderallpairsofprobability distributionsthe
same
way,thosetwosetsof payoff functionsmust havethesame
closedconvexcone.3.1.2 Exact Characterization of Stochastic
Dominance
Theorems
Building
from
themathematicalresultsofthelastsubsection,we
now
statethemain
resultof Section3. ThistheoremappliesTheorem
3.3 togivenecessaryandsufficientconditionsforapair (Il,r)to satisfyastochasticdominancetheorem.Theorem
3.4 Considera pair ofsetsofpayofffunctions(II,T),whereIIand
T
are subsetsofP
1. Thenthefollowingtwoconditionsare equivalent: (i) Thepair(II,T)isastochasticdominancepair.
Proof:
To
seethat(ii)implies(i),fixapair(11,0and assumethat(ii)holds. Consideranarbitrary probability distribution F(;0) eA",and choose 8H
>
6L. DefineF
1=
F(;6
H) andF
2=
F(;0
L). Let//=
F
1-
F
2,andnotethat/I6£* .
Now,
notethatTheorem
3.3impliesthatFor
allTen,
\itd(F
l-
F
2)
>
<=>
For
allyer, \yd{F
x-
F
2)
>
Sincethismust betrueforall
6
H>6
L,then(II,r)isastochasticdominancepair.To
see that(i)implies(ii),notethatif(ii)fails,then (withoutlossofgenerality) there existssome y
ef
suchthaty
efl. Butthen,byTheorem
3.3there existsa iue
£* sothatjndfi,
>
forall iteft,but Jyd/x.<
0. Let fc=
jd^:
=
jdfc
. Let=
{6L,6H},where
6
H>6
L,anddefineaparameterizedprobability distributionas follows:F(;6
H)= j/C
andF(;0
L)=
j/z;. Butthen,Jn(s)dF(s;9)isnondecreasingin forall
^
6
II,while\y{s) dF(s;d)isstrictlydecreasingin6. Q.E.D.
Theorem
3.4provides an algorithm forgenerating stochasticdominance
theorems, and forchecking whether (11,0pairs satisfystochasticdominancetheorems. Itgives theweakestpossible
sufficientconditionsonaparticular(n,r)pair toguaranteethatitisastochasticdominancepair:if cc(II
u
{1,-1})*
cc(Tu
{1,-1}),thenthere willalwaysexistaparameterizedprobability distributionsuchthat
fy(s)dF(s;d)isnondecreasingin
8
forally
e
T,but Jn(s)dF(s;6)isstrictlydecreasingin6
forsome 7ren.
However,
Theorem
3.4makes
useoftheweak
topology,which might belessfamiliarthansome
others;forthisreason,
we
willnow
discussthe relationshipbetweenclosureintheweak
topologyandclosureinothertopologies. Inparticular,
some
existing stochasticdominance
theoremshavebeen provedby showingthattheclosureundermonotone
convergence oftheconvex cones ofthetwosets are equal (Brumelle and Vickson, 1975; Topkis, 1968). This raises the question,what
is therelationshipbetweenthoseresults,andresultsusing the notion of closureunderthe
weak
topology?The
followingcorollaryanswersthat question.Corollary 3.4.1 LetT beanytopology
on P>
suchthat thejunctionals {/J(-,//)|/l€
w\,
are continuous linear Junctionals. Supposethat(11,0isa pair ofsetsofpayofffunctions, eachinp:
Ifcc(Ilu{l,-l})T
=cc(ru{l,-l})
T
(closuretaken with respectto t),then(11,0
«
astochasticdominancepair.
Proof: Tisfinerthano(p*pK?) (by Bourbaki,
TVS
n.43). For anysetA
in"P*,we know
AcA
rcA
ff. Takingclosuresin
(RP'W)
ofeachset,we
concludethatA
c
=
A
T. Thus,if
cc(nu{l,-l})
r=cc(ru{l,-l})
T,then
cc(nu{l,-l})
a=cc(ru{l,-l})
ff. This implies
that (Il,r)isastochasticdominancepairby
Theorem
3.4. Q.E.D.To
seethe intuitionbehindthis result, firstnotethat,giventwo
topologies T, and r2,wherez, iscoarserthan 12 (written T,
c
T2 ),theclosureof asetA
under %2 (-A*2)iscontainedinthe closureof
A
undert,(
A
T
>).Thisistruebecausethe finertopology has
more
closedsets;A
Tl isa closedsetunder T2,butsincex2 isfiner,theremight bea closedsetwhichisastrictsubsetofA
T
< butthatstillcontains
A.
Sincethe
weak
topologyiscoarser thananytopologywhichmakes
the linear functionals /?(7T,/x)continuousinn,theclosureof asetof payoff functionsunder
monotone
convergenceiscontainedin theclosureofthesetundertheweak
topology. Thus,iftheclosed(undermonotone
convergence)convex conesoftwosetsofpayoff functionsarethesame, thentheclosed (underthe
weak
topology)convex conesof those
two
setsofpayofffunctionswillbethesame. Thistellsusthatcheckingthat theclosed(undermonotoneconvergence)convexcones oftwosetsofpayoff functionsarethesameissufficient to establish that the
two
setsofpayofffunctions satisfyastochasticdominancetheorem.Thus, in practice,
when
checking whether (TIT) satisfy sufficientconditionsforastochasticdominancetheorem,itispossibletocheck whethertheclosureoftheconvex conesofthetwosetsare equivalent,usinganytopologywhichisconvenient (providedof coursethatthetopology guarantees continuity of the functional /?(•,/*))•
The
topologies ofmonotone
convergence, dominatedconvergence, and uniform convergence are examples oftopologies
which
might be useful in applications.Remark
1We
havecharacterizedthe relationshipbetweenY1S0M andT
SOM asfollows:n
sou=
cc(n
soMu
!!_!})=
cc(T
somu
[i-i])Since
much
ofthe existing literatureinvolvesintegrationby
parts,itisusefultoillustratethe relationshipbetweenintegrationby
partsandourcharacterization.We
can viewthe integrationby
partsasameans
of discoveringtheextremepointsof asetofpayoff functionswhich formsa closedconvexcone.
In thecasewherethedistributionfunctioniscontinuousanddifferentiablewith respectto
support [0,J], both Propositions 2.1' and 3.1 can be proved directly
by
first rewritingexpectedprofits using integration
by
parts, and then taking the partial derivative ofthisexpression withrespect to theparameter(s). It isthe integration
by
partswhichdeterminesthe relationshipbetweenU
S0M andF
SOM,andestablishingthis relationshipplays adirect role in
the proofs of both Propositions 2.1'
and
3.1.We
can rewrite the functional P(7Z,F)=
J' 7t(s)dF(s) as follows,usingintegration
by
parts:f
7C(s)dF(s) Js=0=
\k(s)-
n\s)
s+
s f' n"(s) ds]•f
dF(s) (3.3) Js=0 J Js=0+
7r\s)\
SsdF(s)-\
Sk"{s)\
S min(s,v)dF(v)ds Jj=0 Jj=0 Jv=0Notethatthefunctions y(s)
=
s and y(s)=
min(a,s),which
determinethe setoffunctionalT
SOM,appearinthisexpression,asdothefunctions n' and 7t",whichcharacterizethe setof
payofffunctions, TlSOM. Infact,
we
canseethatthefunctionalf K(s)dF(s)isthe limitof a
Js=0
sequenceofpositivecombinationsoftheexpected value ofthefunctionsin
T
SOM,pluslinearcombinationsof
J
dF(s).
The
set TlSOM isdefined sothat %' and-n"
are positive; thus, the functions y{s)=
sand y(s)=
min(a,.s) arepositive inT
S0M.The
constant functions {1,-1}areincludedbecause f dF(s)isconstantat1,butthe coefficientJj=0
\k(s)
-
7t'(s) s+
sJ* iz"(s) ds
canbepositiveornegative.
To
seehow
this isused in the study ofmono
tonicity, note thatwe
canevaluate thefunctional given in (3.1) at a parameterized probability distribution F(;0) (i.e., take
P(n,F(;d))) andthendifferentiate,asfollows:
-§}Jlo n(s)dF(s;d) (3.4)
=
k'(s)-^jsdFisid)
-H
o n"(s) JqJJ min(s,v)dF(v;6)dsThis expression canbe usedtoprovethe
SOMSD
theoremdirectly. Ifthemean
and j^Tmn(v,s)dF(v;0) are nondecreasingin 6, thenfor any nondecreasing,concave payofffunction, this expression is clearly positive.
On
the other hand, if themean
orj^min(v,s)dF(v;9) weredecreasingin
8
somewhere,we
couldconstructa nondecreasing,concave payofffunctionwhichputalloftheweightonthefailures.
What
theintegrationby
parts
shows
usis thatinsteadof checkingthat /J(tt,F(;0)) is nondecreasing in6
forallit
6
n
S0*f
(the left-handsideof(3.4)),
we
cancheckthat p(y,F(;6)) isnondecreasingin6
secondsetof conditionsiseasier tocheck.
The
constant functions {1,-1}from
abovedo
not appearinT
SOMbecause [1-dF(s;&)=
1isalwaysconstantin0.We
emphasizethe fact that theJj=0 sets
U
S0M andT
50Maredeterminedby the integration
by
parts (equation 3.3),notby
thesubsequentdifferentiationwithrespect to theparametersofthe distribution.
Inananalogous way,
we
canevaluatethe functionalf n(s)dF(s)attheparameterized
Js=0
probability distribution
F(;0
X,62),using equation(3.3),andtake themixed
partialderivative:The
proofofthe stochasticsupermodularity theoremfollows similar argumentsto thatofSOMSD.
Thisexampleclarifiesthe relationshipbetweenthenew
stochasticsupermodularitytheoremandthe existing stochasticdominanceresult.
3.3 Conditional Stochastic
Dominance
Inthissection,
we
show
thatTheorem
3.4 can be extendedtothe caseof"conditionalstochasticdominancetheorems."
We
willusethefollowingnotation:f dF(t;6)
G(
S;6,K)=
±£^
J dF(t;6) jx(s)dF(s,6) \dF(s,6) seKWe
now
introduceanobjectwhichisanalogoustoastochasticdominancetheorem:Definition 3.2 Consider apairofsets of payofffunctions(II,T), with typical elements
it:SR"->SR
and
y
:9?"-»SR. LetK
beacollectionofsubsetsof9?". Thenthepair (TIX)isaK-conditionalstochastic
dominance
pair ifconditions(i)and
(ii)are equivalentforall withapartialorder
and
allF
:9T
x
->[0,1]suchthatsuchthat F{;6)e
A":(i)
For
allneU.
and
allKeK,
k{G\K)isnondecr
easingin 6.(ii)
For
allyeT
and
allKeK,
y(6\K)isnondecreasing
in 6.Theorem
3.5 Consider apairofsetsof payofffunctions(Yl,T),whereIIand
T
aresubsets ofP
1. Let
K
heacollection ofsubsetsof9t\ where 9?"€ K. Thenthefollowingtwoconditionsare equivalent:
(i) The pair (TIT)isa ^-conditionalstochastic
dominance
pair.(ii) cc(II
u
{1,-1})= ccOTu
{1,-1}).Proof:
We
canapplytheproofofTheorem
3.4almostexactly. Letn^
=
{n
IK\neII},andlikewisefor Tk.
Then
notethatcc(Ilu{l,-l})=
cc(Tu
{1,-1})impliesthatcc(U
Ku{I
K,-IK})=
cc(TKu
{IK,-IK})forallK. Then,foreveryK
we
applyTheorem
3.4,(ii)implies(i).
To
show
that(i)implies(ii),theargumentsofTheorem
3.4canbeusedtoshow
thatif(ii)fails,then(i)mustfailforthecasewhereK
=
9?". Q.E.D.One
exampleof a conditionalstochasticdominance theorem whichhasappearedinvariousformsin theliterature(seeWhitt(1982)) involvesthe setofnondecreasingpayofffunctions, as follows:Theorem
3.6 The following conditions are equivalent:(i)
For
all it:9?"—
»9? nondecreasingand
allsublattices#c
91", k(6\K)isnondecreasingin6.
(ii)
For
allincreasingsetsAc
9?"and
allsublattices isfc9?", f dG(s;6,K) is JAnKnondecreasingin6.
Proof: Apply
Theorem
3.5together withthe fact thatcc({k\k nondecreasing}
u
{1,-1})=
cc({IA\IA(s)nondecreasing}
u
{1,-1}). Q.E.D.Thistheoremisusuallyprovedusing algebraic arguments; but usingthe resultsofthispaper,
we
see thatitfollowsasanimmediatecorollaryofTheorem
3.5. First, take thecase wheren-1.
Then
condition(ii)isequivalenttorequiringthatF(s;6)satisfiestheMonotone
Likelihood Ratio(MLR)
Order(foraproof,seeWhitt(1980)),definedasfollows:Definition 3.2 The parameter
6
indexes theprobability distribution F(;6)e
A
1accordingto the
Monotone
Likelihood RatioOrder
(MLR)
if,forall6
H>
L, thereexistnumbers
-oo<a<b<°°
and
a nondecreasingfunctionh:[a,b]-»9? suchthat F(a;6H)
=
0, F(b;6L)
=
1,and
jKdF(s;dH)
=
jKh(s)dF(s;GL)forall K^[a,b].When
thesupportofF
isconstantin6,andF
has a density/, then an equivalent requirementisthat/satisfiesthe
Monotone
Likelihood Ratio Property(MLRP),
as follows:f(sH;6)
f(sL;6)
isnondecreasingin
6
foralls11>
Milgrom
(1981) introduced theMLRP
to theeconomics literaturein thecontext of signaling problems.The
density/satisfies theMLRP
ifandonlyifthelogofthedensityissupermodular.The
MLR
impliesFOSD
(butnotthe reverse); thatis,ifF(s;6)satisfiesMLR,
thenthe distribution F(s;6)isdecreasingin flpointwise.
FOSD
does not placeanyrestrictionsonthemovements
of F(s;6)apartfromthe restriction thatchangesin6 donotcausethe distribution to cross.
The
MLR,
ontheother hand,requires that foranysublatticeK,the distribution conditionalonK
satisfiesFOSD.
Now,
supposethatn>
1. Ifwe
assumethatthevectorsisa vectorofaffiliatedrandom
variables(see
Milgrom
andWeber
(1982),who
defineaffiliationandshow
thatifthe distributionhas adensity, thelogofthatdensitymust be supermodular9),then condition(ii)requires thateachmarginal, Ffaid),
satisfies
MLR.
The
mostgeneralcase,wherethevectorsisanarbitraryvectorofrandom
variables,has nottoourknowledge beenanalyzedintheliterature.
We
havepresentedTheorem
3.6 toillustratethe relationshipbetweenseveral existingresults.The
framework developedinthispapershowsthattheproof ofTheorem
3.6isanimmediateconsequence; previous analyseshavereliedonmore
complicated arguments. But,Theorem
3.6isjustoneexample;inhisanalysisof firm entryandexit inadynamicenvironment,
Hopenhayn
(1992) derives anotherconditional stochasticdominanceresult,which heterms
monotone
conditionaldominance. Thisresultconsidersthecase of
n
=
IlFOandK
isthe setofindicatorfunctionsofupperintervals.Our
theorem showshow
togenerateother conditional stochasticdominance theoremsasapplicationsarise.4
OTHER
PROPERTIES
OF
STOCHASTIC OBJECTIVE FUNCTIONS
Thissectionderivesnecessaryandsufficientconditionsfortheobjectivefunction, j 7c(s)dF(s;&), to
satisfyproperties
P
otherthan"nondecreasing in 0,"forexample,the propertiessupermodularorconcavein8.
We
asktwo
questions:(i)For whatpropertiesP
istheclosedconvex conemethod
ofproving stochastic
P
theoremsvalid? (ii)For whatpropertiesP
canwe
establish that the closedconvex coneapproachisexactly the right one, as in thecaseofstochasticdominance?
We
proceedas follows. InSection4.1,we
introduce notationanddefinitionswhichextend ourstochasticdominance frameworkto"stochasticsupermodularity theorems,"andfor arbitrary properties
P,"stochastic
P
theorems." Section 4.2answersquestion(i),showingthattheclosedconvex coneapproachtoprovingstochasticdominance theoremsisvalid forall"stochastic
P
theorems,"whereP
isa
CCC
property. InSection 4.3we
introduceanew
classofproperties,Linear Difference Properties(LDPs),whichisa subset of
CCC
properties.We
show
thatexamplesofLDPs
includeconvexity,9In fact,Milgrom and Weber
(1982) also showthat, for general distributions, affiliationis equivalentto log-supennodularity oftheSR"derivativeon a product measure.
supermodularity,andallpropertieswhichplacea signrestriction
on
aderivative (refer toTableIforasummary
ofCCC
propertiesand LDPs). Section 4.4 addresses question(ii),provingthattheclosedconvex cone approachisexactlythe rightoneforallstochastic
P
theoremswhen
P
isanLDP.
Section 4.5 applies the resultsofthis sectionto thecaseofsupermodularity, providingexamples of(11,1") pairsfromthe stochasticdominanceliterature(whichare also stochasticsupermodularitypairs,byourmainresult)anddiscussingapplicationsofstochasticsupermodularity theorems.
4.1 StochasticSupermodularity
Theorems
and
StochasticP
Theorems
We
now
introduce a constructwhich
we
will callstochasticP
theorems,which
are preciselyanalogous to stochastic
dominance
theorems. Let us begin witha specific example, stochasticsupermodularity theorems. Using
two
parallel definitions,we
will be abletoidentifythe closerelationshipbetweenthe stochastic
dominance
theoremsandstochastic supermodularity theorems. Stochasticdominance
theorems focuson
monotonicity of expectedprofitswith respecttoasingleparameter, whilestochastic supermodularitytheoremsexamine whether increasingone parameter
increases the returns(inexpectedprofits) toincreasingtheotherparameter(recallthatsupermodularity
canbecheckedpairwise). Proposition2.3isanexample ofastochasticsupermodularity theorem.
The
followingdefinitionispreciselyanalogoustoDefinition3.1:Definition 4.1 Consider a pair ofsetsofpayofffunctions(II,F), with typical elements
it:
9T
—
>9?and
y
:9?"—
»91. The pair(II,T)isastochastic supermodularity pairifconditions(i)
and
(ii)are equivalentforalllattices©
and
allF
e
A" e:(i)
For
alln
sU,
fJt(s)dF(s;6) is
supermodular
in 6.Js (ii)
For
allyeT,
[y(s)dF(s;6) is
supermodular
in6.Further,
we
definethe setZ
J5r tobethe setofall(II,r)pairs
which
arestochasticsupermodularity pairs,analogousto "Lsm
.To
placeourexamplefrom
Proposition 3.3 in thisframework,thepair(U
SOM,r
SOM)ismI.
!lsr.When
comparingDefinition3.1 (stochasticdominance theorem) andDefinition4.1 (stochasticsupermodularity theorem), itishelpfulto recall that stochastic
dominance
theoremsandstochasticsupermodularity theorems are bothcharacterized
by
(11,0pairscorresponding to sets oflinear functionalsoftheformf5(K,-),whicharethencomposed
with parameterizedprobability distributions.Thus, itis possibleto
compare
thetwo
types of theoremsdirectly, even thoughin a stochasticdominancetheorem,theparameter spaceisanysetwith apartialorder,whileinastochasticdominance
More
generally,we
can definea classoftheoremscalled StochasticP
Theorems, which aredefinedforanarbitrarypropertyP.
We
are interested inpropertiesP
togetherwithparameterspacesQ
pwhicharedefined sothat,givenafunction h:P
—
> 9t,thestatement "h(6)satisfiespropertyP
on
0/'
iswell-definedandtakesonthefollowingvalues: "true"or"false."LetQ
p denotethesetof allsuch parameter spaces 0,,. Further, fora givenproperty P,we
willdefinethe setof admissibleparameter spacestogetherwithprobability distributionsparameterizedonthosespaces:
V
nP
=
{(F,e
p)\epee
pandF
eA%
p}.
Definition 4.2 Consider apairofsetsof payofffunctions (IT,r), withtypical elements
K
:SK"—
>9?and
7:9?"—
>9?. Thepair(II,O
w
astochasticP
pairifconditions(i)and
(ii)are equivalentforall(F,Q
P)eV
p:(i)
For
all 7teTl,j7t(s)dF(s;6) satisfies property
P
on©
p.(ii)
For
allyeT,
fy(s)dF(s;6)satisfies property
P
on 0„.Js
For example,forastochasticconcavitytheorem, 0,, can be any convexset, andcondition (i)of Definition 4.2 isinterpreted, "Forall tte
n,
Jn(s)dF{s;6) is concave in 6."
As
in the case ofstochasticdominance,
we
let I.SPT bethe setofallstochastic
P
pairs foragiven property P. Alsoanalogousto stochasticdominance,
we
canrewrite therequirementof Defintion 4.2as follows:l(F,e) e
V
p \J7t(s)dF(s;6)satisfiesP
on Vttenj
=
|(F,0)ev;\jy(s)dF(s;G)satisfiesP
onV/
€rl
Inthenextsection,
we
show
thattheclosedconvex conemethod
of provingstochasticdominancetheoremscanbeextendedtoallstochastic
P
theorems,ifP
isa closedconvex coneproperty.4.2 The Closed
Convex
Cone Approach
isValidforallCCC
PropertiesInthissection,
we
identifyaclassofproperties,whichwe
call"ClosedConvex Cone" (CCC)
properties,suchthatthefollowing statementistrue:forallproperties
P
whichareCCC,
ifapair(IIJ)
is astochastic
dominance
pair, then(tl.DisastochasticP
pair;thus, forallCCC
properties, theclosedconvex cone