TheEconomics

ofRisk
TheEconomics
ofRisk
DonaldJ.Meyer
Editor
2003
W.E.UpjohnInstituteforEmploymentResearch
Kalamazoo,Michigan
LibraryofCongressCataloging-in-PublicationData
Theeconomicsofrisk/DonaldJ.Meyer,editor.1sted.
p.cm.
Includesbibliographicalreferencesandindex.
ISBN0-88099-268-9(hardcover:alk.paper)ISBN0-88099-267-0
(pbk.:alk.paper)
1. Risk.2. Economics.3. Riskmanagement. I.Meyer,DonaldJ.,
1957-
HB615.E33872003
368dc22
2003020210
2003
W.E.UpjohnInstituteforEmploymentResearch
300S.WestnedgeAvenue
Kalamazoo,Michigan490074686
Thefactspresentedinthisstudyandtheobservationsandviewpointsexpressedarethe
sole responsibility of the authors. They do not necessarily represent positions of the
W.E.UpjohnInstituteforEmploymentResearch.
CoverdesignbyAlcornPublicationDesign.
IndexpreparedbyDianeWorden.
PrintedintheUnitedStatesofAmerica.
Printedonrecycledpaper.
Contents
Introduction1
DonaldJ.Meyer
WesternMichiganUniversity
1 RiskandRiskManagement 9
KeithJ.Crocker
UniversityofMichiganBusinessSchool
2 StatesoftheWorldandtheStateofDecisionTheory 17
MarkJ.Machina
UniversityofCalifornia,SanDiego
3 GamblingwiththeFuture 51
WilliamR.Eadington
UniversityofNevada,Reno
4 CommonValueAuctionsandtheWinnersCurse 65
JohnH.Kagel
OhioStateUniversity
5 SharingVeryHighRisks 103
KatherineSwartz
HarvardSchoolofPublicHealth
6 RiskandAgriculture 127
RulonD.Pope
BrighamYoungUniversity
Index 169
AbouttheInstitute 185
1
Introduction
DonaldJ.Meyer
WesternMichiganUniversity
Warning
UndertheMichiganEquineActivityLiabilityAct,anEquinePro-
fessionalisnotliableforaninjurytooreventhedeathofapartic-
ipant in an equine activity resulting from an inherent risk of the
equineactivity.
MichiganFarmBureau
FarmBureauInsurance
Thissignispostedatthehorsestableswheremytwoyounger
daughtersridehorsesonSaturdaymornings.Thesigncommunicatesat
leasttwodistinctmessagestoitsreader.First,ridinghorsesisarisky
activity.Eventhoughitisasmallpercentageofridersoverall,anum-
berdogetinjuredinhorse-ridingincidentswiththepossibilityofeven
sustainingaseriousorlife-threateninginjury.Second,whenyou
decidetorideahorseatthesestables,youtakeonandassumethisrisk;
thatis,thestableownerisnotinanywayliableforaninjurythatyou
mayincurwhileparticipatinginequineactivities.
Ridinghorsesisjustoneofalmostcountlesssituationsinlifein
whichyouencounterriskwhereyouencounterthechanceofinjury,
damage,loss,orofmakingadangerouschoice.Otherexamples
includetheriskoflosingyourjob,theriskofcontractingsomedebili-
tatingdisease,thechanceofgettinginanautomobileaccidentonthe
waytotheshoppingmall,andtheriskofbeingstruckbylightningina
summerthunderstorm.Riskisallaroundusandaffectsusall.Noone
canescapefromitsclutchesorattainperfectimmunitytoit.Youmay
decidetonotacquiretheadditionalriskofridingahorsebynotengag-
inginthatactivity,butonecannotsoeasilyavoidtheriskofgetting
cancerortheriskofbeinghitbyacarwhilecrossingthestreetonthe
waytoschool.
1
2 Meyer
Riskissomethingthatmostofusdislikeandtrytoavoid.Econo-
mistsandotherresearchershavestudiedriskandhaveobtainedconsid-
erableamountsofevidencethatindicatethatpeoplearegenerallyrisk
averseintheirattitudetowardsrisk.Thismeansthat,everythingelse
heldconstant,peoplechoosethelessriskyalternativeandtheywill
takemeasurestoreduceorshifttherisktootherswhenfeasible.Pur-
chasinghealth,life,orautomobileinsuranceisonewaytodothis.
Diversifyingyourinvestmentportfoliobyinvestingindifferentkinds
ofstocks,bonds,andmoneymarketsecuritiesisanother.Thesemea-
sureshelpreducetheamountofriskthatwemustfacetoamorecom-
fortableandmanageablelevel.Butwecannotcompletelyinsulate
ourselvesfromthemanyrisksthatonecancomeacross.
Onoccasion,peoplewillinglyseektoengageinariskyactivityor
situationinsteadofselectingtheusualriskavoidancestrategy.Aslot
machineorsomeothergameofchanceinacasinoenticessomepeople
togambleandtakeonrisk.Otherpeopleinvestaportionoftheir
incomesintheriskystockmarket.Thisisnotnecessarilyevidence
contrarytoriskaversion,becausegamblingcanprovideentertainment
valueorgiveanadrenalinerushtothegambler.Also,historically,
theaveragereturnofthestockmarkethasbeenhigherthanthereturn
ofsafermoneymarketassets,andthishigherexpectedreturncancom-
pensateforthehigherriskofthestockmarketassets.
Thisbookcontainschaptersthataddressvariousaspectsofrisk.
Twochaptersdealwithriskdirectlybylookingatriskmanagement
and how it is applied to decision making, or by assessing what
researchershavelearnedoverthelastfewdecadesintheirtheoretical
investigationsofrisk.Theotherchapterslookatriskindirectlyby
examiningmarketsinwhichriskhasasignificantpresence.Casino
gamblingenterprises,agriculturemarkets,auctions,andhealthinsur-
ancemarketsareplaceswhereriskmakesaconsiderableimpact.A
numberofproblemsthatresultfromriskinthesemarketsandinthe
economywillalsobeaddressed.Auctionparticipantsmayfeelthe
stingofthewinnerscursewhentheobjecttheyarebiddingonhas
uncertainvalue.Significanthealthissuesandpotentialproblemsface
thosewhoarewithouthealthinsurance.Riskincentiveproblemshave
plaguedtheFarmBill,governmentsresponsetofarmerstodecrease
theriskinagriculture.Problemandpathologicalgamblersmakeupa
percentageofthosewhoenteracasinoestablishment.
3 Introduction
ThefirstchapterinthisbookisRiskandRiskManagement:Basic
Concepts,byKeithJ.Crocker.Themainfocusofthischapterison
riskmanagementforthebusinessfirmandforthegeneralconsumer.
Crockerdiscusseshowoneidentifiesriskandthenhowtodealwithit.
Thisisafittingfirstchapterbecauseitisanaturalstartingplaceinthe
investigationofrisk,anditisverybasicandapplicabletoeveryreader,
regardlessofhowandwhereonefitsintotheeconomy.
Crockergoesthroughanumberofstepsinvolvedwithriskman-
agement,startingwithriskidentificationandthenquantifyingthemag-
nitudeoftheexistingrisk.Riskmitigationandcontrolfollowsinthe
process.Crockerexaminesbothlosspreventionandlossreduction
measuresandlooksatthedecisionofwhichriskstoretainandwhich
totransfer.
Crockerhighlightsthisprocesswithanumberofinterestingappli-
cationsandexamples.HiscentralbackdropistheFebruary1999natu-
ralgasexplosionoccurrenceattheFordMotorCompanyRiverRouge
powerplantinDetroit,Michigan.Detailedinformationassociatedwith
eventsleadinguptoandimmediatelyfollowingthiscatastropheare
usedtoexemplifythepresenceofriskandofriskcontrolandmanage-
ment.Asbestosexposureanditssubsequentcleanupandthetaintingof
TylenolcapsuleswithcyanidealsoserveasillustrationsinCrockers
discussion.
Thesecondchapter,byMarkJ.Machina,isentitledStatesofthe
WorldandtheStateofDecisionTheory.Asthetitlesuggests,
Machinaassessesthestateoftheprofessionregardingitstheoretical
investigationofriskanduncertainty.Hereviewsanddiscusseswhere
weareintermsofthemodelinganddevelopmentofriskanalysis.The
majortheoreticalriskresearchofthelastseveraldecadesisdivided
intotwomajorapproaches.
Choiceunderobjectiveuncertaintywasthefirsttheorydeveloped
aboutdecisionmakingunderriskandcomprisesMachinasinitialdis-
cussion.ItsrootsgobackseveralcenturiestoearlyworkbyPascaland
Fermat.Objectsofchoicecanbedescribedasobjectivelotteriesin
whichalloutcomesandtheobjectiveprobabilitiesoftheseoutcomes
areknown.Anindividualspreferencefunctionovertheselotteriesis
generallyassumedtofollowtheobjectiveexpectedutilityformfor
somevonNeumann-Morgansternutilityfunction.Violationsofthe
expectedutilityhypothesis,includingthefamousAllaisParadox,are
4 Meyer
noted,andMachinadiscussessomenonexpectedutilitymodelsthat
havedevelopedasaresponsetotheseviolationsandparadoxes.
Choiceundersubjectiveuncertaintyistheothermainbranchof
theoreticalriskresearch.Machinatracesthistheorybacktoworkby
Savage,whowasinstrumentalintheformulationandearlydevelop-
mentofthesubjectiveuncertaintymodel.Themaintenetsofthis
approachincludestatesofnature,events,andoutcomesorconse-
quences.Individualsareallowedtohaveprobabilisticbeliefs,and
thesebeliefscandifferacrossindividuals.Thus,underthistheoryyou
andyourfriendcandifferastotheassessedchancethatthestockmar-
ketwillriseinthenextfewweeks.Machinathenaddressesviolations
ofthistheoryincludingtheEllsbergParadoxandsubsequentmodeling
adjustmentsinresponsetotheviolations.
Aftercarefuldiscussionofeachofthesetheories,Machinaoffers
hispersonalinsightsintothesimilaritiesanddifferencesbetweenthem
andhowtheapproachesactuallyaremorerelatedthanonemightini-
tiallyassess.Includedinhisobservationisanintuitivediscussionof
hisrecentlycompletedworkonthissubject.Formoredetails,theinter-
estedreaderisinvitedtoconsultMachinasfascinatingdevelopments
andfindings.
ThethirdchapterisGamblingwiththeFuture:Economicand
SocialPerspectivesonCasinosinAmerica,byWilliamR.Eadington.
Strongdemandforgamblingactivitiesforentertainmentorrisk-seek-
ingvaluehasexistedaslongasmankindhas.Gamblinghasgonethe
gambitfromanactivitythathasbeenlargelybannedandviewedby
manyasimmoraltoamorewidelyaccepted,legalized,andcontrolled
settinginwhichspecialinterestgroupscompetefortheindustryreve-
nues.Casinogamblingisasignificantandfast-growingindustry,one
whosegrowthoftenoccursineconomicdownturnswhentaxrevenue
generationandjobcreationarehighlydesirable.Inthischapter,Ead-
ingtonfocusesontheeconomicsofcasinos,establishmentsthathouse
numerousvarietiesofgamesofchance.
EadingtonstartsbytracingthroughU.S.casinohistoryofthelast
century,beginningwithNevadascasinolegislationin1931.Heexam-
inesthecurrentstatusoftheindustryaswellaslookingatprojected
trendsintothefuture,includingnewgamblingformssuchasInternet
gambling.Differentcasinomarketsaredescribed,includingdestina-
tionresortcasinossuchasLasVegas,AtlanticCity,orBiloxi,urbanor
5 Introduction
suburbancasinoslocatedinmajormetropolitanareas,andruralcasinos
whichincludemosttribalcasinos.Gamblingisanever-changing
industry,asevidencedbythecreationofracinoshorseracetracks
thathavebecomeequippedwithslotmachinesorothergaming
devices.
Eadingtonexploresasocialaswellaseconomicperspectiveon
casinosbylookingatcostbenefitanalysisofcasinos.Thebenefitside
iswellfoundedinconsumersurplustheoryineconomics.Consumers
benefitbyhavingaccesstoalegalizedcasinoestablishment,andmany
wouldbewillingtopaymoneytodoso.Thecostsideismuchless
developedandishardertoquantify.Problemandpathologicalgam-
blersimposeacostonthemselvesandothers.Otherresearchershave
triedtolinkcasinosorgamblingwithincreasedcrimerates,gambler
financialtroublesandhigherdivorceandsuiciderates,andageneral
declineinthenationsmoralfiber.
ThefourthchapterisbyJohnH.KagelandisentitledCommon
ValueAuctionsandtheWinnersCurse:LessonsfromtheEconomics
Laboratory.Inthischapter,Kageldiscussestheriskofexperiencing
thewinnerscurseinacommonvalueauction,wheretheauctioned
itemsvalueisthesametoallbiddersbutisunknown(risky)atthe
timeofthebid.Biddingonanoffshoreoiltractisafittingexample,as
theprecisevalueofthehydrocarbonsbeneaththeoceanfloorisuncer-
tainatthetimethebidsareplaced.
Generallybiddersobtaindistinctprivateindicatorsorsignalsofthe
objectsvalue.Someofthesignalswillbehigherthanthisvaluewhile
otherswillbelower.Thewinningorhighbidderlikelyhasthehighest
oroneofthehighestsignals.Ifthebidderdoesntrecognizethisand
factoritaccordingly,thesubmittedbidmaybemorethantheuncertain
valueeventhoughitislessthanthesignalitisbasedon.Ifthishap-
pens,thewinnerissaidtobecursed,andbelow-normalprofits,even
lossesandbankruptcies,canresult.
Kageltracesthroughthehistoryandearlyreportingofthewinners
curseinauctionsandinothermarkets.Itisimportanttorecognizethat
thewinnerscurseisnotatheoretical,equilibriumconceptorresultbut
ratherisahypotheticalempiricalphenomenon,indicatingthatbidders
donotproperlyaccountforreceivingprivatesignalsinthebidding
process.Kageldiscussestheconsiderableevidenceofthecurse,focus-
ingonthatofsealedbidauctions.Healsoexaminesalternativehypoth-
6 Meyer
esesforexplainingtheoverlyaggressivebiddingbehaviorsuchas
limitedliabilityforlosses.Auctionexperienceisfoundtobeanimpor-
tantlearningvariable,asinexperiencedbiddersaregenerallymostsus-
ceptibletothecurse.
KagelthenlooksforthewinnerscurseinthecontextofEnglish
auctionsandfirstpriceauctionswithinsiderinformation.Thesemarket
typeswereselectedbecausetheyareenvironmentsthatmayeliminate
oratleastsharplyreducethewinnerscurseeffect.Kagelcomparesthe
dataresultsinthesemarketswithearlierfindingsofsealedbidauc-
tions.Heconcludesthatevidenceforthewinnerscurseispresentin
thesemarkets,althoughthemagnitudeseemstobelesssevere.
Chapter5isSharingHighRisks:HowGovernmentcanmake
HealthInsuranceMarketsmoreEfficientandmoreAccessible,by
KatherineSwartz.InthischapterSwartzlooksatthecharacteristicsof
thoseintheeconomywhoarewithouthealthinsuranceandthensug-
gestshowtomakethesehealthinsurancemarketsmoreaccessibleand
moreefficient.Recallfromearlierdiscussionthatthetypicaldecision
makerinoureconomyisriskaverseanddesiresinsurancetohelp
reducetheriskofundesirableevents.Tobewithouthealthinsuranceis
notonlybadfortheuninsuredindividualbutforthegeneraleconomy
aswell.
Swartzbeginsbydescribingthetraitsoftheuninsured,including
thatofage,income,laborforcestatus,andhealthstatus.Thetypical
uninsuredpersonisyoung,haslowincome,isworking,andisingood
health.Shegoesontoexaminetheinsurancecompanyorsellersideof
themarket.Healthinsurancemarketscanbedividedintothreediffer-
enttypes:largeemployergroup,smallgroup,andindividual.The
smallgroupandindividualmarketssignificantlydifferfromthelarge
employergroup,andthisdifferenceiscriticalfortheoccurrenceofthe
uninsured.
Thesmallandindividualmarketssufferfromwhatisknowninthe
riskliteratureastheadverseselectionproblem.Thepoolofpeople
seekinghealthinsuranceiscomprisedofthosewhoareathigherrisk
forillness,disease,andsizablemedicalbills,andthosethatareof
lowerrisk.Theinsuranceapplicantgenerallyknowsmuchmoreabout
thesizeofthisriskthandoestheinsurancecompany.Insurancecompa-
niesfearthatthoseapplyingforcoveragearedisproportionatelycom-
posedofthehighriskorhighcostgroup.Thiscanleadtosignificant
7 Introduction
lossesfortheinsurancecompanies.Thisforcescompanies,Swartz
suggests,tocompeteforthelowerriskorrelativelyhealthyindividu-
als,andthehighriskorsicklypeoplesubsequentlygetleftoutandtend
tobeuninsured.
Swartzproposesthatthegovernmentshouldstepinandcoverthe
23percentofinsuredcaseswiththehighestmedicalcosts.This
extremecostsegmentmakesupalargefractionofthecostspaidoutby
insurancecompanies.Bytakingthesecasesover,thegovernmentfrees
upresourcesexpendedbyinsurancecompaniestotrytoscreenoutthe
casesexpectedtobehighcost.This,Swartzargues,willenhanceover-
allinsurancemarketefficiencyandalsoaccessibilityforthoseseeking
healthinsurancecoverage.Shemodelsherplanafterarecentlydevel-
opedandimplementedplaninNewYorkstate.
ThesixthandfinalchapterisbyRulonD.Popeandisentitled
RiskandAgriculture:SomeIssuesandEvidence.Adverseweather,
disease,anddamaginginsectpestsarejustsomeoftherisksfarmers
faceduringgrowingseasonthatcouldresultinuncertaincropyields.
Productpriceisalsorisky,asmanyagriculturalmarketsareatomistic
innatureandcroppricesaresubjecttochangingmarketsupplyand
demandconditions.Inhischapter,Popehighlightssomecentralcon-
ceptspertainingtoriskinagriculture.
OneofPopesprimaryconceptsofinterestisthatofdiversification
asaresponsetorisk.Diversificationcanbeaccomplishedinanumber
ofways;forexample,cropdiversification,areadiversification,orthe
diversificationoflaborincomeawayfromthefarm.Popealsolooksat
riskreductionandinputusage.Hediscusseshedgingandforwardmar-
ketsasameansofriskreduction.Alloftheseactionsareconsistent
withfarmerswhoexhibitriskaversion,andPopecitesempiricalevi-
denceinsupportofthisposition.Finally,foodsafetyisaddressedasa
concerntobothfarmersandconsumersalike.
PartofPopesanalysisisdevotedtogovernmentsupportprograms
andcropinsurance.Thegovernmentprogramshavenotproventobe
financiallysuccessful,becausethereisgenerallynotenoughmoney
collectedininsurancepremiumstocoverthecroplosspayments.Two
problemsadverseselectionandmoralhazardcanhelpexplainwhy
thegovernmentmayhavefinancialdifficulty.KatherineSwartzdis-
cussesinChapter5theadverseselectionproblemasitpertainsto
healthinsurancemarkets,whichapplieshereaswell.Moralhazardis
8 Meyer
theideathatfarmersmaytakefewerprecautionstoavoidriskwhen
theyholdinsurancethanwhentheydonot.Thiscanresultinmorecrop
lossesthanexpectedandmorepayoutsbythegovernment.
Thesixchaptersinthisbooklookatabroadarrayofresearchrelat-
ingtorisk.Theauthorsconveymuchinformationaboutriskasitper-
tainstothevariousmarketsthattheyaddress.Acomprehensivelistof
referencesaccompanieseachchaptertoguidetheinterestedreader
whowantstopursuesomeparticularfacetinmoredetail.Learning
moreaboutwhatriskisandhowitaffectsusreducessomeofthe
uncertaintyinlifethatweallface,andthenhelpsusmakemore
informeddecisions.Wewillneverbeabletoeliminatealltheriskthat
wecanpotentiallyencounter,butwecanstrivetobetterunderstandthe
risksinvolvedandthendealwiththeminthebestlightpossible.
1
RiskandRiskManagement
BasicConcepts
KeithJ.Crocker
UniversityofMichiganBusinessSchool
Thishasgottobetheworstdayofmylife,observedWilliam
ClayFordJr.,FordMotorCompanyChairman,ashecontemplatedthe
February1999naturalgasexplosioninboilernumbersixthathadjust
leveledpartoftheRiverRougepowerhouseinDetroit,Michigan.The
disasterkilled6peopleandseriouslyinjured14,andcutoffpowerto
the1,100acrefacility.
WhileFordJr.sremarksweredirectedtowardthehumandimen-
sionofthetragedy,fromacorporatestandpointtheprognosismust
haveappearedequallysobering.TheRougecomplexpowerhousethe
centerpieceofHenryFordsdreamofbuildingentirecarsinasingle
locationhadsuppliedelectricity,compressedair,millwater,and
steamtosixassemblyandpartsplantsemploying10,000workers,and
alsototheindependentlyownedRougeSteelplant.Althoughanengi-
neeringmarvelofitstime,theconcentrationofproductionatRiver
Rougehadprecipitatedariskmanagersworstnightmare,astheeffects
oftheintegratedplantsshutdownrippledthroughFordsinternalsup-
plynetwork.
FirsthitwasRougesownMustangassemblyplant,whichhad
beenworkingovertimewithtwo10-hourshiftsdailycrankingoutthe
popularsportscompact.NextcameRougesmetalstampingplant,sup-
plyingmetalparts(fendersandsimilarproducts)to16ofFords20
NorthAmericanplants.Resultswerepredictable.Shiftswerecutfrom
8hoursto4hoursatthreeMidwestassemblyplants,andlostproduc-
tionatRougesframeplantresultedintheeliminationofscheduled
overtimeattruckplantsinKansasCity,Missouri;Norfolk,Virginia;
andOakville,Ontario.
9
10 Crocker
Evenatthesereducedlevels,productionwassupportedonlybythe
buffersofexistinginventoriesandsuppliesintransitthat,once
exhausted,wouldnecessitateplantshutdowns.Andapreviously
planned$240millionreplacementpowerhousewouldnotbecom-
pletedforatleastayear(FinancialTimes1999).
Riskisendemictoourpersonal,aswellasprofessional,experi-
ences.Everytimewedecidetocrossthestreetorascendthestairsin
ourhomes,wearemakingpersonaldecisionsinvolvingrisksandtheir
management.Howwehandlethesesituationshasanimportantimpact
onthequality(and,inmanycases,thelength!)ofourlives.
WHATISRISK?
Webstersdictionarydefinesriskasthechanceofinjury,damage,
orloss.Unlike,say,aportfolioofstocks,whichhasapotentialfor
gain,riskspresentonlyadownside.Ariskisachanceofsomething
badoccurringand,hence,tobeavoided.Ofcourse,evenbadthings
canprovideaprofitopportunitytosomebodythecitytaxesmeto
haulawaymygarbage,therebyprovidingemployment,andtheOrkin
Manishappytofumigatemyhouse,forafee.ButIdonotgenerally
bringhomeextragarbageorencouragetermitestoinfestmyhouse.
Nordosensiblepeopleseekoutrisk.However,riskcanbemanaged.
Thischapterlaysoutthekeyelementsofriskmanagement:identifica-
tionandquantification,mitigationandcontrol,financing,andcatastro-
pheplanning.
IDENTIFICATIONANDQUANTIFICATION
Giventhatrisksareendemicinouruncertainworld,adopting
appropriatestrategiestodealwithriskexposuresandtheirconse-
quencesisaneverydaytask.Considerthecaseofthepedestriancon-
templatingcrossingabusystreet.Thefirststepistoidentifytherisk
(speedingautomobileswithdistracteddriverschattingoncellphones?)
andtoquantifyitsmagnitude(scrapes?bruises?brokenbones?fatali-
RiskandRiskManagement 11
ties?).Thismundanetaskisthecriticalpointofdepartureforonecraft-
ingariskmanagementstrategyremembertheoldaphorismthat
forewarnedisforearmed,whichisprobablythebestpieceofcheap
advicethatariskmanagercangive.
1
Inthebusinesssetting,manykindsofriskareidentifiable,evento
themostuninitiated.Dangerousmachineryorexposedelectricalwir-
inginafactorysetting,orslipperyfloorsinanofficeorretailestablish-
ment(squashedgrapesonthefloorareagrocersnightmare)are
obviousexamples.Othertypesofriskexposuresmaybelessapparent
anddiscernibleonlytothosewithexperienceinaparticularareaofrisk
analysis.Muchasstandingunderatreeduringathunderstormmay
seemreasonabletothoseunfamiliarwithlightning,riskexposuresmay
notbeapparenttoanuntrainedeye.
InthecaseoftheFordRougepowerplant,forexample,therewere
certainlyengineeringadvantagesassociatedwiththeconsolidationof
productionoftheelectricity,steam,andhigh-pressureairrequiredby
theentireRougecomplex.Buttherisksofthisapproachalsoturned
outtobesubstantial,astheeventsofFebruary1999attest.
Perhapsthemostinsidiousrisksfacingbusinessesthesedays,
however,comefromevolvinglegalrules,aswehaveobservedinthe
caseofenvironmentalliabilityandasbestosexposure.TheCompre-
hensiveEnvironmentalResponse,CompensationandLiabilityAct,the
1980Superfundhazardoussubstanceclean-uplegislation,introduced
strictliabilitythatmayinvolveseveralentitiesjointlyforcleaningup
hazardouswastesites.Asaconsequenceofthisnewlegalreality,a
businesscouldhavebeeninfullcompliancewithallapplicablelawsat
thetimeofthewastedisposal,orsimplybethecurrentownerofan
existingsite,yetstillbestrictlyliableforthecostsofclean-up.Even
partialcontributorstothesitearefullyliablefortheentirecostof
clean-up,duetojointandseveralliability,
2
leadingtothepredictable
prospectingfordeeppocketsbyenterprisingtortattorneys.These
liabilitiesalsomaybeinherited,whichmakesmergersandacquisitions
problematicthesedays.
Asbestosexposurealsoprovidesaninstructiveexample.Fifty
yearsago,mostpeoplehadlittleunderstandingofthehealthrisksasso-
ciatedwithairborneasbestosfibersintheworkplace,andexposure
standardsreflectedthis.Overtime,however,itbecameincreasingly
clearthatasbestosis(aclosecousinoftheblacklungdiseasesuffered
12 Crocker
bycoalminers)andmesothelioma(anuntreatablecancerofthelungor
stomachliningthatisbothswiftandinvariablyfatal)wereassociated
withworkplaceexposures.Theresulthasbeenanexplosionoflitiga-
tion(estimatedpotential:1.3to3.1millionclaims)withexpected
asbestosliabilitiesof$200billion,ofwhich$78billionwillbeborne
bytheaffectedcompaniesandtherestbytheirinsurers(Parloff2002).
Litigationhasalreadydestroyedtheprimaryproducersofasbes-
tosJohns-Manville,Unarco,andRaybestosManhattanalldeclared
bankruptcylongagoandhasmovedontobankruptcompaniesthat
merelypurchasedasbestosproducts,includingBabcock&Wilcox,
OwensCorning,GAF,andW.R.Grace.Currentlyinthecrosshairof
asbestoslitigationareGeorgia-Pacific(involvinggypsumproducts),
3M(forallegedlyfailingtowarnthatthedustmaskswouldntworkif
improperlyused),andFord(forexposuresrelatedtotheasbestosused
inbrakes).Federal-MogulCorp.,anautomotivesupplier,recently
soughtChapter11bankruptcyprotectionbecauseofanasbestosliabil-
ityinheritedfromits1998acquisitionofT&NPLCofManchester,
England,acompanythathadusedasbestosinaseparatebuildingsup-
pliesbusiness.Atthetimeoftheacquisition,Federal-Mogulsetaside
$2.1billionincashtocovertheanticipatedclaims,asumthatinretro-
spectseemstohavebeennowherenearenough.
DanielS.Sobczynki,theformerDirectorofCorporateInsurance
forFord,putitbest:Thehighestpotentialrisksarethosethatareuni-
dentifiedandunmanaged.Itiscriticaltoevaluateyourrisksandto
learnfromthelessonsofothers,hesays.Theproblemoflearning
frompersonalexperienceisthatitgivesyouthelessonafterthetest
hasbeenadministered(FinancialTimes1999).
MITIGATIONANDCONTROL
Aftertheriskexposurehasbeenassessed,thenextstepistocon-
siderhowonedealswithit.Continuingwithourstreet-crossingexam-
ple,onepossibilitywouldbetoavoidtheriskentirelyandnotcrossthe
streetatall(awisestrategyiftheroadinquestionwere,say,Interstate
94atrushhour).Alternatively,ifwedecidetoproceed,thequestion
mightbethefollowing:dowejaywalkandcrossthestreetnow,or
RiskandRiskManagement 13
strolldowntothetrafficsignalandwaitforthegreenlight?Eachof
thesealternativesrepresentsaneconomicdecision,weighingthecost
ofthestrategyagainstthepotentialbenefits.
Generically,mitigatingariskexposureentailstheidentificationof
tacticseithertoreducetheprobabilityofabadoutcome,ortoreduce
themagnitudeofaloss,shouldabadoutcomeoccur.Theformertypes
ofactivities,referredtoaslosspreventionmeasures,wouldincludethe
cross-at-the-intersectionoptiondiscussedabove,or,inamoremun-
daneindustrialsetting,theinspectionofelectricalwiringtoreducethe
probabilityofanelectricalfire.Indeed,mostoftheriskmitigation
strategiesthatcomeeasilytomindaredesignedtokeepusoutoftrou-
bleinthefirstplacedontputthegasolinecannexttothefurnace,
dontsmokeinbed,lockyourdoorsbeforeyouretireforthenight.
Lossreduction,ontheotherhand,describestheclassofriskmitigation
activitiesdesignedtoreducethemagnitudeofaloss,shouldoneoccur.
Thestandardexampleherewouldbetheinstallationofsprinklersina
warehouse,whichdoesntreducetheprobabilityofafirestartingbut,
rather,mitigatesthedamagesthatresultfromthefire.
TheexplosionofboilernumbersixattheRiverRougepowerhouse
occurredduringamaintenanceshutdown.Asfarascanbedetermined,
avalveunintentionallyleftopenallowednaturalgastoflowintothe
boiler,whichwasquicklyignitedbytheelectrostaticscrubberslocated
intheboilerschimney.
Inretrospect,itappearsthatthetragedystemmedfromalackof
attentionpaidtoissuesofriskmitigationduringroutineepisodesof
maintenance.Notonlywastheactofshuttingdowntheboilersrare,
butapparentlytherewerenowrittenproceduresorcheckliststoguide
theprocess.Employeeswhohadnotbeentrainedinshuttingoffthe
boilersandwhohadlastreceivedanequipmentmanualin1997,hadto
shutoffover30(unlabeled)naturalgasvalvesthroughoutthepower-
housecomplex.Theymissedone,andtherestishistory.
Wemaketrade-offsinourpersonalandbusinesslivesbetweenthe
burdenofriskexposureandthecostofriskmitigation.Financingthe
costsassociatedwithabadoutcomebecomesthequestion.Inpersonal
settings,theriskfinancingstrategygenerallyadoptedisthatofrisk
shiftingtoathirdparty,usuallyaninsurancecompany(thinkaboutthe
collisionandliabilityinsuranceonyourcar,homeownersinsurance,
orthewarrantyonanewappliance).Theproblemwiththistypeofrisk
14 Crocker
transfer,though,isthatitcreateswhatisknownineconomicsasa
moralhazard.
Acolleagueofminekeptasailboatmooredofftheendofhisdock
onLongIslandSound.Oneday,duringcasualconversation,Iasked
abouthisstrategyfordealingwithstormsandthelikeasaboat
ownermyself,Iwasaware(riskidentificationandquantification)of
theeffectsofheavywaveactiononaboatbangingagainstadock.He
respondedthathewasntworriedbecausehehadinsuranceandhe
nevertooktheboatoutofthewateruntiltheendoftheseason.The
problemhere,ofcourse,isthatifoneisfullyinsuredagainstaloss,
thenonehasnoincentivetotake(privatelycostly)actionstoreduce
onesriskexposure.Insurancecompanies,notsurprisingly,havefig-
uredthisout.
Whenmyteen-agedsonfinallymadeenoughmoneytopurchasea
car,itturnedoutthatthemachineofhisdreamswasa1994Camaro
Z28,witha5.7literV-8engineand270horsepower.Youmightthink
thatnoinsurerintheirrightmindwouldwritecoverageinasituation
likethis,butyouwouldbewrong.AnautomobileinsurerinMichigan
waswillingtoprovideliabilitycoverageatafinitepremium.But,there
wasacatchnocoverageforcollisiondamage.
3
Effectively,hehasa
100percentdeductibleifhewrapsthecararoundatree.
Thisretainedriskhasincentivizedmysontodrivecarefully.
Thisisgenerallythetrade-offthatyouwillfindinyourpersonaland
professionalriskfinancingdecisionsincreasedinvestmentinrisk
eliminationreducesthepremiumsyoupayperdollarofcoverage,but
thedownsideisthatyouareexposedtomorerisk.
CATASTROPHEPLANNING
Accidentsdohappendespitethebestintentionsandmosteffective
effortstoforestallsucheventualities.Andtheresponsetothebadnews
isprobablythemostcriticalcomponentofanylossreductionstrategy.
IntheimmediateaftermathoftheRougeRiverpowerhousecatas-
trophe,WilliamClayFordJr.dispatchedhispersonalaide,withcredit
cardinhand,totrackdownthevictimsfamiliesanddowhateverwas
requiredtohelpout.Thecompanyworkedwithitssupplierstoprocure
RiskandRiskManagement 15
electricalswitchingequipmentandtoobtainportableboilersforsteam.
DetroitEdisonbuiltanoutdoorsubstationinaweektosupplythe
powernecessarytogettheRougeRivercomplexbackonline.The
resultwasatriumphinlossreductionapotentiallycatastrophicbusi-
nessinterruptionscenariotruncatedtoaone-weekhiccuponthepro-
ductionline.
Therearemanyotherexamplesoftheimportanceofcatastrophe
planning,goodandbad.Forexample,backin1986,whenastilluni-
dentifiedindividualreplacedthepainkillerinseveralbottlesofTylenol
capsuleswithcyanide,theresultwasthedeathofaninnocentcon-
sumer.Johnson&Johnson,themakerofTylenol,didntattemptto
deflectblame(afterall,theyhadntadulteratedthecapsules)orother-
wisetemporize.Theyimmediatelyrecalledallthecapsulesfromstore
shelveseventhosethatwereclearlyuntaintedandthendesigned
thegenerationoftamper-proofcontainersstillinusetoday.Thisisa
textbookloss-reductionstrategytimely,aggressive,and(whilecostly
intheshortrun)effective.
Incontrast,considerthestrategyofJohns-Manville,oncethe
worldsbiggestproducerofasbestos,which,aswenotedearlier,col-
lapsedundertheweightoflitigationfromasbestosclaimsin1982.
Johns-Manvillesapparentdecisiontoignoretherisksofasbestos
exposuretoitsworkers,longaftertheevidenceindicatedthatmanage-
mentmayhavesuspectedalinkbetweenasbestosexposuresinthe
workplaceandworkerhealth,resultedinlivesruinedandlost.Thecost
toManvilleanditsshareholderswasultimatelythatofcorporatebank-
ruptcy.
DanSobczynskiofferssomesoundadvice:Eithermanagethe
risk,oritwillmanageyou,hesays,and,whenitdoes,thelosswill
happenwhenyouareleastprepared(FinancialTimes1999).
Notes
1. Students of history will recall that Winston Churchill was almost killed by a
speedingtaxiinNewYorkCityduringthe1930s.Accustomedtocarsdrivingon
theleftsideoftheroad,helookedthewrongwaywhilecrossingthestreet,aclear
failureinriskidentificationandquantification.
2. Joint and several liability means, in practice, that even a 1 percent ownership
stakeinthepropertycanleadtoliabilityfor100percentoftheclean-upcostsif
16 Crocker
the owners of the other 99 percent interest are financially unable to pay their
share.
3. Actually,theywouldprovidesuchcoverage,butatanannualpremiumeffectively
equaltothebookvalueofthecar!
References
FinancialTimes.1999.ManagingRiskbeforeitManagesYou.TheMaster-
ingStrategyseries,October11,pp.89.
Parloff,Roger.2002.The$200BillionMiscarriageofJustice.Fortune
145(5):154196.
2
StatesoftheWorldandthe
StateofDecisionTheory
MarkJ.Machina
UniversityofCalifornia,SanDiego
TWOEXTRAORDINARYNONSCIENTISTS
Almost20yearsago,IbrieflyknewamanbythenameofCraig.
AlthoughhediedaboutayearafterImethim,Ivethoughtabouthim
eversince.Craighadthisuncannyabilitytoconversewithapersonfor
a few minutes, and then announce what make and model of car they
drove.NeitherI,noranyoneIeverspoketo,hadeverseenhimgetit
wrong.Craigwasneverabletoexplainhowhedidit,andhisunique
abilityfollowedhimtothegrave.
What Craig had perfected was an impressive skillperhaps even
anartbutitwasnotscience.Itwasnotsciencebecauseitwasnota
procedure that he could verbally communicate or write down, so that
otherpeopleinotherplacesorothertimescoulddoitalso.Oneofthe
defining features of scientific activity is that it generates a body of
knowledge and techniques that can be communicated and utilized by
othersinthisway.
IalsoknewawomannamedTulawithanequallyimpressiveabil-
ity.Tulawasabletopredicthowwellapersonsdaywouldgo,based
ontheshape,size,andcoloroftheauratheyemittedinthemorning.
And in contrast to Craig, she could even explain the specifics of her
method.Forexample,ifyouraurawasroundandblue,youwouldhave
goodluckallday.Butifitwassquareandyellow,thenyoudbestgo
back home and stay in bed. Tula had prepared a chart with the com-
plete relationship between properties of your aura and the upcoming
featuresofyourday,soifyouhadacopyofthechart,youjustneeded
adailyreadingofyouraura.AlthoughTulassuccessratewasntper-
17
18 Machina
fect(likeCraigswas),itstillcomparedfavorablywithstandardmedi-
cal, meteorological, and macroeconomic predictions, and most of her
friendswouldstopbyeachmorningforaquickreadingoftheiraura,
andthengoawaytoconsulttheirchart.
By constructing and distributing her chart, Tula had codified and
communicated features of her technique in a way that Craig never
could.ButsinceTulawastheonlyonewhocouldseetheseauras,what
shewasdoingstillwasnotscience.Anactivityisnotscienceunlessit
involvestechniquesthatotherscanalsoapplyaswellasvariablesthat
otherscanobserve.
Thepurposeofthischapteristoexamineoneofthemostimpor-
tant theoretical constructs of modern decision theorynamely, the
concept of states of the world or states of naturefrom the point of
viewoftheseandsimilarscientificconsiderations.Arestatesofnature
inherently descriptive or prescriptive objects? Do individuals making
choicesunderuncertaintyfacethesestatesofnature,ordotheycreate
them? Are states external and independently observable, like an indi-
vidualscommoditydemandlevels,oraretheyinternalandnotdirectly
observable, like utility or marginal utility levels? In addressing these
questions,Iwillofferanoverviewofhowresearchershavesoughtto
represent the concept of uncertainty, from the original formulation of
probabilities and objective uncertainty in the seventeenth century,
through Leonard Savages twentieth century formulation of states of
nature and subjective uncertainty, to current work which seeks to
eliminateoratleastredefinethedistinctionbetweenobjectiveand
subjective uncertainty. The following section presents some scientific
issues common to all theories of choice, whether under certainty or
uncertainty. The next two sections sketch out the current theories of
choiceunderobjectiveandsubjectiveuncertainty.Afterthat,Iaddress
thequestionofwhetherstatesofnatureshouldbeconsidereddescrip-
tive or prescriptive constructs, and then I consider scientific issues
related to the observability and measurement of states of nature. The
finalsectionconcludeswithcurrentworkontherelationshipbetween
subjectiveandobjectiveuncertainty.
StatesoftheWorldandtheStateofDecisionTheory 19
SCIENTIFICCONSIDERATIONSINTHETHEORY
OFCHOICE
ScientificModelingFromtheOutsideIn
Thehumandecision-makingprocessmaywellbeoneofthemost
complicated systematic phenomena in the universe. In terms of the
point of view of the scientific observer, it is certainly unique. On the
onehand,ascientisttryingtomodelthisprocessislikeananatomistin
thedaysbeforeanesthesiaandvivisectionscientistscanobserveand
to some extent even control external influences on a system, and can
observetheresultingbehaviorofthesystemasawhole,buttheycan-
not get inside to observe its constituent parts at work. On the other
hand, every scientist is a human decision maker with powers of self-
consciousnessandself-reflection.However,self-reflectionofourdeci-
sion-making processes has not produced that much more hard sci-
ence than has, say, self-reflection of our breathing or digestive
processes.
While advances in neuroscience may ultimately do for decision
theory what vivisection did for anatomy, decision theory currently
remainsverymuchablackboxscience.Althoughdecisiontheorists
can(anddo)useintrospectiontosuggesttheoriesandhypotheses,the
rigorous science consists of specifying mutually observable indepen-
dent variables (in particular, the objects of choice available for selec-
tion), mutually observable dependent variables (the selected
alternative),andrefutablehypotheseslinkingthetwo.Inotherwords,
choice theory attempts to explain why particular alternatives are
selectedfromasetofavailablechoices.
IssuesofObservability
Because decision scientists cannot perform dissection, they are
subject to a greater scientific discipline than that required of anato-
mists. If a decision scientist tried to account for an individuals pur-
chasesofbananasasthedirectresultofsomethinglikeanappetitefor
fruit, we would not know how to test this hypothesisthat is, we
would not know how to independently look for such an appetite,
even if we had a scalpel and an open, anesthetized brain. Such unob-
20 Machina
servable constructs like appetites, utility, and preferences canand
doplayaroleinscientificdecisiontheory,butonlyasinsidelinksin
acausalchainthatultimatelystartswithfullyobservableindependent
variables and ultimately ends with fully observable dependent vari-
ables.Forexample,giventhejointhypothesisthatwell-definedcom-
moditypreferencesexistandarealsostablefromdaytoday,standard
consumer theory allows us to infer enough information about these
preferences from an individuals past demand behavior to be able to
make refutable predictions about their future demand behavior, even
forsomecombinationsofpricesandincomeneverbeforeobserved.
Inthefollowingsections,weshallseethatinpassingfromchoice
over certain commodity bundles to choice over uncertain prospects
(eitherobjectivelotteriesorsubjectiveacts),hypothesesinvolving
theunobservableconstructsofcommoditypreferencesandutilityfunc-
tions can be replaced by hypotheses involving the unobservable con-
structs of risk preferences and beliefs, which also link observable
independenttoobservabledependentvariables.Whetherthenotionof
statesofnaturecansimilarlyserveremainstobediscussed.
IssuesofClassification
In order for a variable or phenomenon to satisfy the criterion of
scientificobservability,itisnotenoughthatmorethanonescientist
beabletoseeititisnotevenenoughthatacamerabeabletorecord
it. Rather, a variable is only scientifically observable if independent
observers can agree on their description of what they have just
observed. Thus, while a scientist can photograph facial expressions,
they cannot be said to have photographed expressions of emotion
unlessthereisawell-definedspecificationofwhichexpressionscorre-
spond to each emotion, and independent observers predominantly
agree in their assignment of emotions to each photograph. In other
words, scientific observability requires well-defined and commonly
accepted classification schemes for the observations, sufficient for
groupingandcomparingsuchobservations,andrelatingthemtogen-
eralhypothesesandtheories.
Just as different types of variables can have different degrees of
observability, different classification schemes will have different
degrees of common agreement. Thus, in regular consumer theory, we
StatesoftheWorldandtheStateofDecisionTheory 21
aremuchmorepronetoclassifycommoditiesanddefinepreferencesin
termsofcategoryschemeslike{fruits,vegetables,grains}com-
paredtoschemeslike{deliciousfoods,fillingfoods,unpleasant
foods}. Although the latter scheme is in some sense much more
directly connected to any given individuals preferences than the
former scheme, the latter scheme cannot be defined independently of
theparticularconsumerbeingstudied.Sincefoodscannotbeclassified
accordingtothislatterschemepriortoobservationoftheconsumers
(verbalorchoice)behavior,itcannotbeusedasaclassificationscheme
forindependentvariables.Categorieslikedeliciousfoods,unpleas-
antfoods,etc.canbedefinedfordependentvariables,however,either
on the basis of the consumers verbal expressions, or on the basis of
their past purchases or consumption behavior. Thus, whether a given
classificationschemedoesordoesnotsatisfythecriterionofscientific
observabilitymaywelldependuponwhethertheschemeisintendedto
beappliedtotheindependentvariablesortothedependentvariablesof
atheory.
IssuesofMeasurability
Theaboveexampleofclassifyingfacialphotographsintodifferent
categoriesofemotionsisanexampleofaqualitativeclassificationof
thebasicobservations.Althoughqualitativecategoriesandqualitative
variablesareperfectlyvalidinthephysical,biological,andsocialsci-
ences, theories and hypotheses are most powerful when they involve
quantitative independent and dependent variables. Many economists
are of the opinion that economics has a more impressive scientific
trackrecordthananthropologybecauseeconomistsworkwithnumeri-
cal variables such as prices, quantities, and income, rather than with
qualitative variables like trust, group identification, or loyalty. Most
theoriesandhypothesesinvolvingquantitativeindependentanddepen-
dentvariablesareeasiertotest,tofinetune,andifnecessary,torevise,
thanmosttheoriesandhypothesesinvolvingqualitativevariables.
Isuncertaintyaninherentlyqualitativeorquantitativeconstruct?In
thefollowingsectionsweshallseethatoneofthetwoprimarymeth-
odsofrepresentinguncertaintytheso-calledobjectiveapproach
represents uncertainty quantitatively, via numerical probabilities. On
the other hand, the other primary methodthe so-called subjective
22 Machina
approachhas traditionally represented uncertainty in a qualitative
manner, via an unstructured set of states of nature. However, in the
finalsectionofthispaper,weseethattakingameasurable,quantitative
approachtosubjectiveuncertaintycanenhanceitspower,andinmany
sensescanserveasanalmostcompletesubstituteforwhatmaybecon-
sidered the more ad hoc assumptions made about the world in the
objectiveapproach.
CHOICEUNDEROBJECTIVEUNCERTAINTY
Outcomes,Probabilities,andObjectiveLotteries
The earliest formal representation of uncertainty came from
founders of modern probability theory such as Pascal and Fermat. In
thisapproach,theuncertaintyattachedtoanyeventisrepresentedbya
numerical probability p between 0 and 1. Because probability theory
derivedfromthestudyofgamesofchancethatinvolvedvirtuallyiden-
ticalrepeatedevents,suchprobabilitieswereheldtobeintrinsicprop-
erties of the events in the sense that an objects mass is an intrinsic
property of the object. These probabilities could either be calculated
fromtheprinciplesofcombinatorics,foraneventsuchasbeingdealta
royal flush, or measured by repeated observation, for an event like a
bentcoinlandingheadsup.
For an individual making a decision under objective uncertainty,
the objects of choice are objective lotteries of the form P =
(x
1
,p
1
;...;x
m
,p
m
),whichyieldoutcomex
i
withobjectiveprobabilityp
i
,
wherep
1
++p
m
=1.Thetheoryofchoiceunderuncertaintytreats
lotteries in a manner almost identical to the way it treats commodity
bundles under certainty. That is, each individuals preferences over
such lotteries can be represented by a real-valued preference function

V(), in the sense that for any pair of lotteries P* =(
x
1

,p;...; ,p
1
x
m m
)
andP=(x
1
,p
1
;...;x
m
,p
m
),theindividualprefersP*overPifandonlyif
V(P*)=
V( , p

;...;
x

m
,p

m
)
exceedsV(P)=V(x
1
,p
1
;...; x
m
,p
m
),andis

x
1
1
indifferent between the two lotteries if and only if V(P*) =

(
x
1
,p
1

;...;
x
m
,p
m
)
exactlyequalsV(P)=V(x
1
,p
1
;...;x
m
,p
m
).
1
StatesoftheWorldandtheStateofDecisionTheory 23
TheExpectedUtilityHypothesis
In standard consumer theory, the preference function over com-
modity bundles is typically assumed to have certain mathematical
properties but is typically not hypothesized to take any specific func-
tionalform,suchastheCobb-DouglasorConstantElasticityofSubsti-
tution form. Specific functional forms are typically only used when
absolutely necessary, such as in empirical estimation, calibration, or
testing.
In contrast, the standard theory of choice under objective uncer-
tainty typically does assume (or does assume axioms sufficient to
imply)aspecificfunctionalformfortheindividualspreferencefunc-
tion over lotteries, namely the objective expected utility form
V
EU
(x
1
,p
1
;...;x
m
,p
m
) = U(x
1
) p
1
+ + U(x
m
) p
m
for some von Neu-
mann-MorgensternutilityfunctionU().Mathematically,thecharacter-
isticfeaturesofthisfunctionalformarethatitisadditivelyseparablein
the distinct (x
i
,p
i
) pairs, and also that it is linear in the probabilities.
The term expected utility arises since it can be thought of as the
mathematicalexpectationofthevariableU(x)(theindividualsutility
ofwealth)ifwealthxhasdistributionP=(x
1
,p
1
;...;x
m
,p
m
).Thelitera-
tureonchoiceunderuncertaintyhasgeneratedanumberoftheoretical
resultslinkingtheshapeoftheutilityfunctiontoaspectsoftheindivid-
uals attitudes toward risk, such as risk aversion or comparative risk
aversionforapairofindividuals.Excellentdiscussionsofthefounda-
tionsandapplicationsofexpectedutilitytheorycanbefoundinstan-
dardgraduatelevelmicroeconomictextssuchasKreps(1990,Chapter
3), Mas-Colell, Whinston, and Green (1995, Chapter 6), and Varian
(1992,Chapter11).
ViolationsoftheExpectedUtilityHypothesis
Althoughtheexpectedutilitymodelissometimesviewedasbeing
quite flexible (since the von Neumann-Morgenstern utility function
couldhaveanyshape),itdoesgeneraterefutablepredictions.Unfortu-
nately,thereisagrowingbodyofevidencetosuggestthatindividuals
preferencesoverlotteriestendtosystematicallyviolatesomeofthese
predictions. Risk preferences tend to systematically depart from the
expected utility property of linearity in the probabilities. The most
3
24 Machina
notable example of this is the well-known Allais Paradox (Allais
1953),whichasksindividualstorankeachofthefollowingpairsoflot-
teries(where$1Mdenotes$1,000,000):
0.10chanceof $5M
0.89chanceof $1M
0.01chanceof $0

{
1.00chanceof $1M : : a
1
versus a
2
0.10chanceof $5M 0.11chanceof $1M

: :

a versus a
4
0.90chanceof $0 0.89chanceof $0

Experiments by Allais and others have found that the modal (and in
somestudies,themajority)choicesarefora
1
overa
2
inthefirstpair,
anda
3
overa
4
inthesecondpair.However,apreferencefora
1
inthe
first pair implies that the utility function satisfies the inequality
0.11U($1M)>0.10U($5M)+0.01U($0),whereasapreferencefora
3
inthesecondpairimplies0.11U($1M)<0.10U($5M)+0.01U($0),
whichisacontradiction.
Although the Allais Paradox was originally dismissed as an iso-
latedexample,subsequentworkbyMacCrimmonandLarsson(1979),
Kahneman and Tversky (1979), and others have uncovered a qualita-
tivelysimilarpatternofdeparturefromtheexpectedutilityhypothesis
of linearity in the probabilities, over a large range of probability and
payoffvalues(seeMachina1983,1987forsurveysofthisevidence).
NON-EXPECTEDUTILITYMODELSOFRISK
PREFERENCES
Responsestotheabove-mentionedviolationsoftheexpectedutil-
ityhypothesishavetakentwoforms.Onebranchoftheliteraturehas
proceededbypositingmoregeneralfunctionalformsforthepreference
function (Edwards 1955, 1962; Kahneman and Tversky 1979; Chew
1983; Fishburn 1983; Quiggin 1982; and Yaari 1987). Such forms
accommodate most of the observed departures from linearity in the
probabilities, and, given the appropriate curvature assumptions, can
StatesoftheWorldandtheStateofDecisionTheory 25
alsoexhibitstandardfeatureslikeriskaversion,comparativeriskaver-
sion,etc.
Asecondlineofworkinnon-expectedutilitytheoryproceedsina
mannerclosertothatofstandardconsumertheoryratherthanadopt-
ingsomenewfunctionalform,itgeneralizestheexpectedutilityprop-
erty of linearity in the probabilities to its natural extension of
smoothnessintheprobabilities(e.g.,Machina1982).Thatis,ittreats
thepreferencefunctionV(x
1
,p
1
;...;x
m
,p
m
)asageneralsmoothfunction,
and studies how properties of its probability derivatives relate to atti-
tudestowardrisk.Thisapproachfindsthatmuchofexpectedutilitythe-
ory is analytically robust to departures from linearity in the
probabilities.
2
CHOICEUNDERSUBJECTIVEUNCERTAINTY
States,Events,Outcomes,andActs
Fromamathematicalperspective,therepresentationofuncertainty
by means of additive, numerical probabilities allows us to apply the
tremendous body of analytical results of modern probability theory
(e.g., Feller 1968, 1971; Billingsley 1986). But from a modeling per-
spective, the assumption that uncertainty comes prepackaged with
well-defined, measurable objective probabilities is unrealistic. Out-
side of the gambling hall, most economic decisions and transactions
involving uncertaintyinvestment decisions, search decisions, insur-
ance contracts, financial instrumentsare defined in terms of uncer-
taineventsratherthannumericalprobabilities.
This approach to representing uncertainty and uncertain pros-
pectsformalizedbySavage(1954)andnowknownasthesubjective
approachinvolvesthefollowingbasicconstructs:

={..., x, ...} anarbitraryspaceofoutcomesorconse-
quences.
26 Machina
S ={..., s, ...}
E ={..., E, ...}
()=[x
1
onE
1
;...;x
m
A ={...,(),...}
W()and

aspaceofmutuallyexclusiveand
exhaustivestatesofnature,representing
allpossiblealternativeunfoldingsofthe
world.
analgebraofevents,eachasubsetofS.
onE
m
] subjectiveactyieldingoutcomex
i
in
eventE
i
,forsomepartition{E
1
,...,E
m
}of
S(orequivalently,yieldingoutcome(s)
instates).
thesetofallsuchsubjectiveacts.
theindividualspreferencefunctionand
correspondingpreference
relationoverA.
Awonderfulexampleoftheuseofthisframeworktorepresentan
uncertaindecisionwasprovidedbySavage(1954,pp.1315):Sayyou
aremakingomeletsandhavealreadybrokenfiveofyoursixeggsinto
amixingbowl.Thedecisionyoumustmakeis:Doyoubreakthesixth
egg? The uncertainty arises from the fact that this sixth egg has been
around for some time and might be rotten. You can either break this
eggintothebowlwiththeothereggs,breakitintoaseparatesaucerto
inspectit,orthrowitawayunbroken.Savagerepresentsthisproblem
intermsofstates,acts,andoutcomesbymeansofthefollowingtable:
State
Act Eggisgood Eggisrotten
Breakintobowl Six-eggomelet Noomelet,andfivegood
eggsdestroyed
Breakintosaucer Six-eggomelet,and Five-eggomelet,and
asaucertowash asaucertowash
Throwaway Five-eggomelet,and Five-eggomelet
onegoodeggdestroyed
TheHypothesisofProbabilisticSophistication
Althoughthesubjectiveapproachdropstheassumptionthatuncer-
taintyisdefinedintermsofnumericalprobabilities,itstillallowsfor
StatesoftheWorldandtheStateofDecisionTheory 27
individuals to possess probabilistic beliefs, with the feature that such
beliefs may now differ across individuals. Formally, an individual is
said to be probabilistically sophisticated, with a subjective (or per-
sonal) probability measure () over the events E, if their preference
functionW()oversubjectiveactstakestheform
W
PS
(())=W
PS
(x
1
onE
1
;...;x onE
m
)=V(x
1
, (E
1
);...; x
m
, (E
m
))
m
forsome(notnecessarilyexpectedutility)preferencefunctionV(P)=
V(x
1
,p
1
;...;x
m
,p
m
)overlotteries.Thatistosay,anindividualisproba-
bilistically sophisticated if their uncertain beliefs can be completely
summarizedbyasubjectiveprobability(E)attachedtoeacheventE,
andtheindividualevaluateseachsubjectiveact()=[x
1
onE
1
;...;x
m
on E
m
] solely on the basis of its implied probability distribution
(x
1
, (E
1
);...;x
m
, (E
m
))overoutcomes.ThisrepresentationofW
PS
()as
thecompositionofapreferencefunctionV()overlotteriesandasub-
jective probability measure () over events is now referred to as the
classicalseparationofriskpreferencesfrombeliefs.
ViolationsoftheHypothesisofProbabilisticSophistication
Savages(1954)jointaxiomatizationofexpectedutilityriskpref-
erences and probabilistic beliefs, employing an expected utility func-
tion for the risk preference function, has been justly termed the
crowning glory of choice theory (Kreps 1988, p.120). However, the
violations of expected utility first observed by Allais were soon
matchedbyviolationsofprobabilisticsophistication,eveninsituations
involving the simplest forms of subjective uncertainty. The most
famous of these examples, known as the Ellsberg Paradox (Ellsberg
1961, 2001), involves drawing a ball from an urn containing 30 red
ballsand60blackoryellowballsinanunknownproportion.Thefol-
lowing table illustrates four subjective acts defined over the color of
thedrawnball,whentheentriesinthetablearepayoffsoroutcomes:
28 Machina
30balls 60balls

red black yellow

1
() $100 $0 $0

2
() $0 $100 $0

3
() $100 $0 $100

4
() $0 $100 $100
Whenfacedwiththesechoices,mostsubjectspreferact
1
()over
2
(),
onthegroundsthattheprobabilityofwinning$100in
1
()isguaran-
teedtobe1/3,whereasin
2
()itcouldrangeanywherefrom0to2/3.
Similarly,mostsubjectsprefer
4
()over
3
(),onthegroundsthatthe
probabilityofwinning$100in
4
()isguaranteedtobe2/3,whereasin

3
() it could range anywhere from 1/3 to 1. Although this reasoning
maywellbesound,itisinconsistentwiththehypothesisofprobabilis-
ticbeliefs.Thatis,thereisnotripleofsubjectiveprobabilities{(red),
(black),(yellow)}thatcansimultaneouslygenerateapreferencefor

1
()over
2
()andfor
4
()over
3
(),sinceaprobabilisticallysophisti-
catedindividualwouldonlyexhibittheformerrankingwhen(red)>
(black),andonlyexhibitthelatterrankingwhen(red)<(black).
Ellsberg also presented what many feel to be an even more fatal
example,involvingtwourns:
50balls 50balls 100balls

red black red black
g
1
() $100 $0 g
3
() $100 $0
g
2
() $0 $100 g
4
() $0 $100
In this example, most subjects are indifferent between g
1
() and g
2
(),
areindifferentbetweeng
3
()andg
4
(),butstrictlyprefereitherofg
1
()
org
2
()toeitherofg
3
()org
4
().Itisstraightforwardtoverifythatthere
existnopairofsubjectiveprobabilities{(red),(black)}fortheright-
handurn50:50orotherwisethatcangeneratethissetofpreference
rankings.Suchexamplesillustratethefactthatinsituations(evensim-
plesituations)wheresomeeventscomewithprobabilisticinformation
StatesoftheWorldandtheStateofDecisionTheory 29
andsomeevents(termedambiguousevents)donot,subjectiveproba-
bilitiesdonotalwayssufficetofullyencodeallaspectsofanindivid-
ualsuncertainbeliefs.Sincemostreal-worldeventsdonotcomewith
such probabilistic information, Ellsbergs Paradoxes and related phe-
nomenondealaseriousblowtothehypothesisofprobabilisticsophis-
tication.
Non-probabilisticallySophisticatedModelsofRisk
PreferencesandBeliefs
JustastheAllaisParadoxandsimilarevidenceledtothedevelop-
ment of non-expected utility models of risk preferences, Ellsbergs
Paradoxes and similar phenomena have inspired the development of
non-probabilistic models of preferences over subjectively uncertain
acts.Suchworkhasalsoprogressedalongtwolines.Onelinereplaces
the subjective expected utility function with more general functional
forms.
3
The second line of research on non-probabilistic models treats
W(x
1
onE
1
;...;x onE
m
)asageneralsmoothfunctionoftheeventsE
1
,
m
..., E
m
, and show how properties of W()s event-derivatives relate to
featuresofbothbeliefsandattitudestowardrisk,againtakingexpected
utilityasitsbasecase.Appendix2Apresentsmathematicalfeaturesof
thislineofresearch.
ARESTATESOFNATUREPRESCRIPTIVEOR
DESCRIPTIVE?
The second section in this chapter argued that the scientific suit-
abilityofaparticulartheoreticalconstructinthatcaseitwasapartic-
ular classification scheme for foodcould depend on whether the
constructwasmeanttobeappliedtotheindependentvariablesofathe-
ory or its dependent variables. This section addresses a similar issue,
namely that certain criteria for suitable specification of the states of
naturecandependuponwhetherthestatesaretobeusedforpositive
(thatis,descriptive)versusnormative(thatis,prescriptive)purposes.
30 Machina
Sinceitsinception,expectedutilitytheoryhasalwaysstraddledthe
boundarybetweenbeingadescriptiveandaprescriptivemodelofdeci-
sion making under uncertainty. Even its original presentation by Ber-
noulli (1738) as a solution to the St. Petersburg Paradox can be
alternatively interpreted as either a description of why people dont
assignaninfinitecertaintyequivalenttothePetersburgGame,orapre-
scription for why an individual shouldnt assign an infinite certainty
equivalent to the game. Two centuries later, proponents of objective
expectedutilitytheorydefendeditagainsttheAllaisParadoxbyshift-
ingtheiremphasisfromtheallegeddescriptivepowerofthetheoryto
itsallegednormativepower.
The same points can be made about the particular component of
subjective expected utility theory that forms the central topic of this
chapternamelythenotionofstatesofnature.Itisonethingtoassert
thatthestatesofnatureapproachoffersausefulnormativeframework
for decision making. It is quite a different thing to assert that, for the
most part, this is how individuals actually do go about making deci-
sions in the absence of probabilistic information. We shall consider
eachofthesetwodomainsinturnineachcase,withthegoalofiden-
tifyingtheproperscientificcriteriaforstates.
CriteriaforNormativeApplications
Savages omelet example effectively shows how representing
natures underlying uncertainty by a set of states, then representing
ones alterative courses of action as acts that map these states into
their respective consequences, can serve to organize a decision prob-
lemandmakeiteasiertoseeexactlyhowonesbeliefs(thestatelikeli-
hoods)andriskpreferencesshouldenterintotheproblem.Forproper
normativeapplication,thisfirststepnamely,thespecificationofthe
statesmustsatisfythreeproperties:
1)Thealternativestatesmustbemutuallyexclusivethatis,notwo
distinctstatescansimultaneouslyoccur.Thus,itwouldnothave
beencorrecttolisteggisrottenandfive-eggomeletastwo
distinctstates,sinceisitpossiblethatthesecouldsimultaneously
occur.
StatesoftheWorldandtheStateofDecisionTheory 31
2) The family of states must be exhaustivethat is, whatever hap-
pens, at least one of the states can be said to have occurred.
Althoughitisatthesamelogicallevelasthepreviouscriterion
(mutual exclusivity), the exhaustiveness criterion is much more
difficultandsomewouldargue,actuallyimpossibletoguar-
anteeinpractice.Forexample,ifyoucrackedthesixthegginto
thebowlandfoundthatitwasactuallyhollow,thenneitherofthe
two states in the Savage table could be said to have occurred,
since neither of the first-row consequences would be realized
(you would not have a six-egg omelet, nor would you have
destroyedtheotherfiveeggs).Whenthedecisionmakerhasrea-
son to expect the unexpected, the exhaustivity requirement
cannotnecessarilybeachieved,andthebestonecandoisspec-
ifyafinal,catch-allstate,withalabellikenoneoftheabove,
andaveryill-definedconsequence.
3)Thestatesmustrepresentnaturesexogenousuncertainty,sotheir
likelihoods cannot be affected by the individuals choice of act.
This issue can be illustrated by a simple example involving the
decisionwhetherornottoinstallalightningrodononeshouse.
Naturally, the relevant occurrences are the two mutually exclu-
siveresults{houseburnsdown,housedoesntburndown}.
Butsinceinstallingalightningrodwillclearlyaltertherespec-
tivelikelihoodsoftheseoccurrences,canwereallyspecifystates
of nature that are independent of the decision makers action?
The answer is illustrated in the following table, which makes it
clear that house burns down and house doesnt burn down
arenotthestatesatall,butrather,partoftheconsequences,and
clarifiesthattheeffectofinstallingalightningrodaswithany
subjectiveactistheoutcomeofaninteractionbetweentheact
andanexogenousstateofnature.
State
Biglightning Smalllightning Nolightning
Act strike strike strike
Lightning Houseburnsdown,Housedoesnt Housedoesntburn,
rod paidforrod burn,paidforrod paidforrod
Nolightning Houseburnsdown,Houseburnsdown, Housedoesntburn,
rod didntpayforrod didntpayforrod didntpayforrod
32 Machina
DoDecisionMakersSeetheStateSpaceorDoTheyConstruct
theStateSpace?
Is an individual who uses states of nature in normative decision
making working with exogenous objects that they observe, or with
endogenous objects that they construct? In one sense, this question is
eithersubsidiaryto,orequivalentto,thequestionofwhethertheyare
selecting from a menu of alternatives (subjective acts) that they
observeasbeingavailabletothem,orfromamenuofalternativesthat
theyhavethoughtupordevised.Viewedinthislargersense,theques-
tionofwhetherthealternativesareobservedorconstructedisseento
havenothingtodowithwhetherthechoicehappenstoinvolveuncer-
taintyatall,andindeed,thequestionmaybeequivalenttotheclassic
questionofwhetherAlexanderGrahamBelldiscoveredtheideafora
telephoneorinventedthisidea.Inanycase,Icannotderiveanyimpli-
cationsofthisissuethatpertaintotheuseofstatesofnaturefornorma-
tivepurposes.
ISSUESOFOBSERVABILITY,CLASSIFICATION,AND
MEASUREMENT
IndependentObservabilityandtheExogeneityofStates
Although the question of whether states are exogenous and
observedversusendogenousandconstructeddoesnotseemtomat-
terinacontextofnormativedecisionmaking,itmattersagreatdealfor
their relevance in descriptive science, for the types of reasons dis-
cussedintheScientificConsiderationsintheTheoryofChoicesec-
tion. There we argued that economics had made greater scientific
achievementsthan,say,anthropologybecausevariableslikepricesand
incomewereeasiertomeasurethanvariablesliketrustorgroupidenti-
fication.Butifforsomereasonitshouldturnoutthatthefullpriceof
apples only exists in the eye the consumer and is not independently
observable, then this advantage is lost. This might be the case if the
acquisitionofacommodityinvolvesatimecost,set-upcost,ortrans-
action cost that is observable to the consumer, but not to the outside
StatesoftheWorldandtheStateofDecisionTheory 33
observer.Notethataninabilitytoobservethetruepriceanindepen-
dent variableposes scientific problems even if we can still observe
the exact amount purchasedthe dependent variablesince it
impedes our ability to observe the relationship between the two (the
truedemandfunction).
Inthecontextofchoiceunderuncertainty,suchaproblemwould
arisewheneverthestatespaceusedbythedecisionmakerdidnotcor-
respond to the state space hypothesized by the scientific observer. In
some sense, this is less likely to happen if the states are exogenous
objectsthatareobservedthaniftheyareendogenousobjectsthatmust
beconstructed.Butevenintheformercase,thereisthepossibilitythat
thedecisionmakerobserveseitherafineroracoarsersetofstatesthan
doesthescientist.Ultimately,thequestionreducesthescientistsabil-
itytoviewthesetofactionsavailabletothedecisionmakertheleft-
hand columns in the above decision tablesand correctly predict the
decisionmakersspecification(beitanobservationorconstruction)of
boththeupperrowandcellentries.Wherethiscanandcannotbedone
isanempiricalquestion.
Ex AnteObservabilityversusEx PostObservabilityofStates
Distinctfromthequestionofwhetherthescientistcanobservethe
setofstatesusedbythedecisionmakeristhequestionofwhetherthe
scientist can observe the realized state, or exactly when the scientist
canobservetherealizedstate.Forexample,inthecaseofchoiceunder
certaintysay,thedemandforapplesitisclearlymoreimportantto
beabletoobservethepriceofapplesthattheconsumeractuallyfaces
upon arriving at the supermarket, than to know the consumers prior
expectationsofwhatthispricemighttobe.Butinterestinglyenough,
for choice under uncertainty, the ability to observe the state space
beforethefactisofmuchgreaterimportancethantheabilitytoobserve
therealizedstate.Thereasonisthatchoiceunderuncertaintyisbydef-
initionexante,andonlydependsuponexantefeaturesofthedecision
problem,namelythestatespaceandthesetofavailablesubjectiveacts
overthisspace.Ascientistwhocorrectlygleansthedecisionmakers
formulation of these concepts, who knows his beliefs over the likeli-
hoods of the states, and who knows his attitudes toward risk, will be
able to correctly predict his decisiona decision that by definition
34 Machina
must be made before, and hence cannot be influenced by, the actual
realizationofthestate.Inboththeomeletandthelightningrodexam-
ples,expostknowledgeoftherealizedstateisofnofurtherpredictive
useforthescientist,exceptforpossiblefuturedecisions,viaitseffect
onthespecificationofastatespaceforsomesubsequentdecision,and/
orlikelihoodbeliefsoverthisspace.
IssuesofClassificationandMeasurement
Arequalitativestatespaceslikelytobemoreorlesssubjecttothe
above types of observability issues than quantitative state spaces? As
theexampleoftheunobservedapplepriceillustrates,evenreal-valued
independent variablesreal-valued commodity prices or real-valued
statesofnaturearesubjecttotheseissuesinprinciple.Ontheother
hand,decisionswherethestatespaceismorenaturallyquantitativeare
probably less subject to these specification difficulties than decisions
wherethestatespaceismorenaturallyqualitative.Forexample,com-
pare the uncertainty related to investing in a domestic farming com-
pany compared to the uncertainty related to investing in a similar
companylocatedinapoliticallyunstableforeigncountry.Intheformer
case, the state space probably only has few dimensions, all of which
arequantitative:theaveragetemperatureoverthegrowingseason,the
averagerainfallovertheseason,andaverageoutputpricesatharvest.
In the latter case, the most significant sources of uncertainty may be
subjectivethe particular political party that comes to power and its
subsequent choice of expropriation policy. There is every reason to
thinkthatthescientistwilldoamuchbetterjobofmodelingthedeci-
sion makers problem formulation in the first case than in the second
case.Indeed,inthefollowingsectionweshallseethatmeasurable,as
opposed to qualitative, state spaces can actually serve to bring some
mathematicalstructureofobjectiveuncertaintyintoapurelysubjective
setting.
StatesoftheWorldandtheStateofDecisionTheory 35
ISSUESOFSTRUCTURE:ALMOST-OBJECTIVE
UNCERTAINTY
As noted, an important feature of objective uncertainty is that it
allows us to apply the analytical tools of probability theory, such as
combinatorics,theCentralLimitTheorem,theLawofLargeNumbers,
andChebyshevsInequality.Furthermore,sinceobjectiveprobabilities
arepartoftheobjectsofchoicethemselves,thesetypesofresultscan
beinvokedindependentlyof,andpriorto,anyknowledgeoftheindi-
viduals attitude toward risk. Thus, for example, the conditions under
which the sum of two independent objective lotteries will have the
samedistributionas(andthuspresumablybeindifferentto)somethird
lotterywillbethesameforallindividuals.Suchresultshavethesame
character as arbitrage results in portfolio theory, which hold indepen-
dentlyofriskpreferencesandhenceyieldextremelypowerfulresults.
Butinsomesense,thisstrengthoftheobjectiveframeworkisalso
itsgreatestweakness:itimposestoomuchuniformityofbeliefsacross
individuals,andinmanycases,toomuchstructureoneachindividuals
ownbeliefs.Incontrastwithpreferencesoverobjectivelotteries,pref-
erencesoverreal-worldsubjectiveprospectsaresubjecttothefollow-
ingthreephenomena:
1) Individuals may have different subjective likelihoods for the
sameevent(diversebeliefs).
2) Individualsbeliefsmaynotberepresentablebyprobabilitiesat
all, with some (or all) events being considered ambiguous
(absenceofprobabilisticsophistication).
3) Individualsoutcomepreferencesmaydependuponthesourceof
uncertaintyitself(outcomepreferencesmaybestate-dependent).
Nevertheless,itturnsoutthatifthestatespacehasaEuclideanstruc-
tureandpreferencesaresmoothintheeventsinthesensedescribedin
Appendix 2A, then features of objective uncertainty will emerge
eveninapurelysubjectivesetting.InAppendix2B,wesketchoutthe
intuition of these resultsreaders wishing a formal development are
referredtoMachina(2001).
36 Machina
We can summarize the scientific implications of almost-objective
uncertaintyasfollows:Inthemoretraditionalapproachtouncertainty
(e.g.,AnscombeandAumann1963),theworldpresentedtwoqualita-
tivelydifferenttypesofuncertainty:uncertainprocesses(suchasper-
fectly balanced roulette wheels) that only generated idealized, purely
objectiveeventsforwhichallagentsheldcommonbeliefsandprobabi-
listically sophisticated betting preferences; and uncertain processes
(suchastomorrowstemperatureorrainfalllevel)thatonlygenerated
purely subjective events, where individuals typically differed in their
likelihood beliefs, or had no likelihood beliefs, and could be state-
dependent. On the other hand, according to the concepts presented in
Appendix2B,onceapurelysubjectivestatespaceisgivenaEuclidean
structureandpreferencesareassumedtobesmoothintheevents,there
exist events that arbitrarily closely approximate all the properties of
classical objective events for all decision makers, in spite of any
interpersonaldifferencesinbeliefs,lackofprobabilisticsophistication,
orstate-dependence.Furthermore,oncestandardobjectiverandomiz-
ingdevicesarereexamined,theyareseentodependpreciselyonthese
typeofalmost-objectiveevents.
Given the traditional (e.g., Savage 1954) approach of positing an
almost completely unrestricted subjective state space and no event-
smoothness,theEuclideanstatespace+event-smoothnessapproach
advocatedinthepreviousparagraphmightseemoverlystrong.Butin
fact, it is well within standard economic practice. Standard consumer
theoryundercertaintyrequiresnostructureatallonafamilyofobjects
ofchoiceinordertoaxiomatizeanordinalutilityfunctionoverthese
objects. Debreus (1954) original topological assumptions were later
shown to be unnecessary by Kreps (1988, pp. 2526). But the work-
horse concepts of competitive prices, marginal rates of substitution,
demand functions, and the Slutsky equation do not emerge until we
assume a Euclidean structure for these objects (vector commodity
bundles and a Euclidean commodity space) and/or smooth prefer-
ences over this space. Under uncertainty, restricting ourselves to a
Euclidean state space amounts to nothing more than restricting our-
selves to subjective uncertainty that appears in the form of random
variables(suchastemperatureorrandomprices).Andforthetypesof
reasonsdiscussedearlierinthischapter,real-orvector-valuedstatesof
nature are much more likely to be commonly observable and com-
StatesoftheWorldandtheStateofDecisionTheory 37
monlymeasurablethanarestatesofnaturethatareelementsofsome
moreabstractspace.
Justasscienceingeneralhasprogressedmostrapidlywhenithas
beenabletoquantifyandmeasurethenaturalworld,researchinuncer-
tain preferences and beliefs will further progress most rapidly to the
extentweareabletoquantifyandmeasuretheobjectswecallstates
ofnatureorstatesoftheworld.
Notes
IwouldliketothankTedGroves,DonaldMeyer,andJoelSobelforhelpfulcomments.
ThismaterialisbaseduponworksupportedbytheNationalScienceFoundationunder
GrantNo.9870894.Allopinionsanderrorsaremyown.
1. If the outcomes x describe monetary payoffs, then the standard monotonicity
assumptions are that V(x
1
,p
1
;...;x
m
,p
m
) is increasing in each of the variables
x
1
,...,x
m
,andalsoincreasingwheneverp
i
isincreasedattheexpenseofp forsome
j
pairofoutcomesx
i
>x (thatis,wheneverprobabilitymassisshiftedfromalower
j
toahigheroutcome).
2. For example, an expected utility preference function U(x
1
) p
1
+ + U(x
m
) p
m
willberiskaverseifandonlyifitscoefficientwithrespecttoprob(x)(thatis,the
value U(x)) is a concave function of wealth. Correspondingly, a non-expected
utilitypreferencefunctionV(P)=V(x
1
,p
1
;...;x
m
,p
m
)willberiskaverseifandonly
ifitspartialderivativewithrespecttoprob(x)(thatis,thevalueV(P)/prob(x))
isaconcavefunctionofwealth.
3. SuchastheChoquetexpectedutilityform
m
W
Choquet
(x
1
onE
1
;...; x onE
m
) U(x ) C E E
i
)
C E
1
E
i1
)

(
1
(

i=1
for some utility function U() and capacity (monotonic non-additive measure)
C(), where the outcomes are labeled so that x
1
< < x (e.g., Gilboa 1987;
m
Schmeidler 1989; Wakker 1989, 1990; Gilboa and Schmeidler 1994), or the
maxminexpectedutilityform
s

s
m
U(x ) (E ) W
maxmin
(x
1
onE
1
;...; x
m
onE
m
)
min

S
U(( )) d ( )
min

i=1
i i
T T
forsomeutilityfunctionU()andfamily
{
T
}
ofprobabilitymeasureson
S(e.g., Grdenfors and Sahlin 1982, 1983; Cohen and Jaffray 1985; Gilboa and
Schmeidler1989).

( )
Appendix2A
PropertiesoftheSmoothFunctionApproachto
Non-probabilisticallySophisticatedModels
Thisapproachstartsbyequivalentlyreexpressingeachact( )=[x
1
onE
1
;
...;x
n
onE
n
]intheform( )=[;xon
1
(x);]=[;xonE
x
;],asxrang-
es over all possible outcomes
x .
The preference functionsW
SEU
( ) and
)canthenbeexpressedintheevent-additiveforms W
SDEU
(
W
SEU
(...; xonE
x
;...)
x

x
(E
x
) where (E)
def
U( ) (E) x
x
| ( ) W
SDEU
(...; xonE
x
;...)
x

x
(E
x
) where (E)
def
U(x s ) d s
x

E
wheretheeventE attachedtoeachoutcomexisevaluatedbyanadditiveeval-
x
uationmeasure
x
( ),whichisthesubjectiveanalogueofobjectiveexpected
utilitysprobabilitycoefficientU(x).
Just as linearity in a set of variables implies linearity in their changes,
event-additivefunctionslikeW
SEU
(;xonE
x
;)=
x

x
( ) andW
S-
E
x
DEU
(; x on E
x
; ) =
x

x
(E
x
) will also be additive in event changes
(growthandshrinkagesets).Thatis,theirrankingoftwoacts( )=[;x
onE
x
;]versus*( )=[;xonE
x
*;]isdeterminedbytheadditivefor-
mulas
W
SEU
(*()) W
SEU
(())

x
(E
*
)

x
(E )

x
(E
*
E )

x
(E E
*
)
x x x x x x
x x x x
W
SDEU
(*()) (E
*
E )

x
(E E
*
) W
SDEU
(())

x
(E
*
)

x
(E )

x x x x x x x
x x x x
whereforeachx,itsgrowthsetingoingfrom( )to*( ),namelythesetE *
x
E
x
,isevaluatedpositivelybyxsevaluationmeasure
x
( ),anditsshrinkage
set,namelythesetE E
x
*,isevaluatednegativelyby
x
( ).
x
Justasdifferentiabilityinobjectiveprobabilitiescanbedefinedaslocal
linearity in probability changes, smoothness in subjective events can be de-
finedaslocaladditivityineventchanges.Thatis,onecandefineageneralpref-
erencefunctionW(;xonE
x
;)tobeevent-differentiableifateachact( )
itpossessesafamilyoflocalevaluationmeasures{
x
( ;)|x

}suchthat
39
40 Machina
W()evaluatessmalleventchangesfrom()inthefollowinglocallyadditive
manner:
W(())

x
(E
*
E ;)

x
(E E
*
;) + o( (*(),())) W(*())
x x x x
x x
wherethedistancefunction(*(),())betweenactshasthepropertythatit
shrinkstozeroasthechangesetsE *E andE E
x
*allshrinktozero,and
x x x
(as with any definition of differentiability) o() denotes a function that is of
higherorderthanitsargument.InMachina(2002),Ihaveshownhowthiscal-
culusofeventscanbeappliedtoestablishtherobustnessofmostofclassical
state-independent and state-dependent subjective expected utility theory and
subjectiveprobabilitytheorytogeneralevent-smooth(butnotnecessarilyei-
ther expected utility or probabilistically sophisticated) preference functions
W()oversubjectiveacts.
Appendix2B
Almost-ObjectiveUncertainty
PROPERTIESOFPURELYOBJECTIVEEVENTS
Webeginbycontrastingthethreepropertiesofsubjectiveeventsdiverse
interpersonalbeliefs,possibleabsenceofprobabilisticsophistication,andpos-
sible state-dependencewith the following four characteristic properties of
idealized,exogenouspurelyobjectiveevents:
Unanimous, outcome-invariant revealed likelihoods: In contrast with
the above-listed properties of subjective events, all individuals exhibit
identical, outcome-invariant revealed likelihoods over purely objective
eventscorrespondingtotheirobjectiveprobabilities.
Independence from subjective realizations: In the presence of joint
objectivesubjectiveuncertainty,purelyobjectiveeventsareindepen-
dentoftherealizationofsubjectiveevents.Thus,theeventsgenerated
by an exogenous objective coin, die, or roulette wheel are invariant to
whetheranygivensubjectiveeventEdoesordoesnotoccur.
Probabilisticsophisticationoverobjectivelotteries:Itisalmostatruism
that all individuals evaluate objective lotteries P = (x
1
, p
1
; ...; x
m
, p
m
)
solely according to their outcomes and corresponding objective likeli-
hoods,viasomepreferencefunctionV(x
1
,p
1
;...;x
m
,p
m
).
Reduction of objective subjective uncertainty: Standard reduction of
compounduncertaintyassumptionsimplythatindividualsevaluateany
objectivemixtureofsubjectiveacts ()+(1)*()= [x
1
onE
1
;
* * *
...;x
m
onE
m
]+(1)[x
1
onE
1
*
;...;x
m *
onE
m *
]solelyaccordingto
* *
itsinducedmap[;(x
i
,;x
j
,1)onE
i

E
j
;]fromeventstolot-
teries.
Theabovefeaturesofobjectiveuncertaintyapplytoallindividuals,whetheror
nottheyareexpectedutility,state-independentorprobabilisticallysophisticat-
ed.Thefollowingpropertiesadditionallyholdforprobabilisticallysophisticat-
edindividualsandexpectedutilityindividuals:
41
42 Machina
Underprobabilisticsophistication,independenceofobjectiveandsub-
jective likelihoods: If the individual is probabilistically sophisticated
with probability measure () over subjective events, these likelihoods
areindependentofexogenousobjectiveevents,andviceversa.
Underexpectedutility,linearityinobjectivelikelihoods:Expectedutil-
ity is linear in objective probabilities (V
EU
(x
1
, p
1
; ...; x
m
, p
m
)
m

i=1
U(x
i
) p
i
)andinobjectivemixturesoflotteries(V
EU
( P+(1)
P*) V
EU
(P) + (1) V
EU
(P*)). Under objective subjective
uncertainty,expectedutilityislinearinobjectivemixturesofsubjective
acts: W
SEU
( () + (1) *()) W
SEU
(()) + (1) W
SEU
(*())),andsimilarlyforW
SDEU
().
ALMOST-EQUALLY-LIKELYEVENTSANDALMOST-FAIRBETS
Astheabovebulletlistsindicate,thepropertiesofpurelysubjectiveand
purelyobjectiveeventslieinstarkcontrast.Nevertheless,inaEuclideanstate
spaceS = [ , ] R
1
,somesubjectiveeventsareclosertobeingobjectivethan s s
others.Weillustratethisbyanexamplewhichapproximateswhatissurelythe
canonicalobjectiveevent:namely,theflipofanexogenous,faircoin.
Denotingtheeventsimpliedbythiscoinbywithstandardnotation{H,T},
theircharacteristicpropertyisthat,foranypairofprizesx*>x,allindividuals
willbeindifferentbetweenthebets[x*onH;xonT]and[xonH;x*onT].In
contrast,foranysubjectiveeventE,rankingofthebets[x*onE;xon~E]ver-
sus[xonE;x*on~E]candifferacrossindividuals(duetodiversebeliefs),or
canreverseiftheprizesx*>xarereplacedbyy*>y(duetostate-dependence).
However,considertheeventE
n
obtainedbydividingthestatespaceS=
[ , ] intonequal-lengthintervals,anddefiningE astheunionoftheodd- s s
n
numberedintervals(thecomplementaryevent~E
n
thusbeingtheunionofthe
even-numbered intervals). As the following diagram indicates,regardless of
an individuals particular subjective probability measure () over the state
space S (indicated by its density function m() in the Figure 2.B1), as n ap-
proachesinfinity,theindividualwillassignequalsubjectiveprobabilitiesof1/
2toeachoftheeventsE and~E
n
,andhencebevirtuallyindifferentbetween
n
thebets[x*onE
n
;xon~E
n
]versus[xonE
n
;x*on~E
n
].State-dependentindi-
vidualswillbesimilarlyindifferent,andasshowninMachina(2001),asn

, all event-smooth individualswhether or not they are expected utility,

stateindependent,orevenprobabilisticallysophisticatedwillrevealE and
n
~E
n
tobeequallylikely,viatheirindifferencebetweenanytwobetsoftheform
[x*onE
n
;xon~E
n
]versus[xonE
n
;x*on~E
n
].Inotherwords,asn

,
StatesoftheWorldandtheStateofDecisionTheory 43
thepurelysubjectiveeventsE and~E
n
bothsubsetsofthepurelysubjective
n
statespaceStakeonthepropertiesofexogenousobjective50:50events.
Figure2B.1ExampleofaSubjectiveProbabilityDensityFunction
S
m()
ALMOST-OBJECTIVEEVENTS,ACTSANDMIXTURES
s s ItisclearthatbydividingthestatespaceS= [ , ] intoalargenumberof
equal-length intervals and taking the union of every third interval, we could
create an subjective event that approximates the properties of an exogenous
eventofprobability1/3,etc.Wecanextendandformalizethisideaasfollows:
Givenanysufficientlyregular(e.g.,finiteintervalunion)subsetoftheunit
interval[0,1]andanylargen,partitionSintonequal-lengthintervals[0,1/n),
[1/n, 2/n) [(n2)/n, (n1)/n), [(n1)/n,1], and define the almost-objective
event S Sby
n
n1
(i )
s s
S =

i =0
{
s+ +

}
n
n
thatis,astheunionofslinearimagesintoeachofSsnequal-lengthinter-
vals.Thus,theeventE
n
illustratedinthepreviousfigureissimplyalmost-ob-
jectiveevent[0,1/2]

S.
n
Bytakingapartition{
1
,...,
m
}oftheunitintervalwecancreateal-
most-objectivepartitions{

S,...,
m

S}ofthestatespaceS,andinturn
1
n n
definealmost-objectiveacts[x
1
on
1

S;...;x
m
on

S].Thealmost-
m
n n
fairbetsoftheprevioussubsectionareseentobethealmost-objectiveacts[x*
on[0,1/2]

S;xon(1/2,1]

S]and[xon[0,1/2] S;x*on(1/2,1]

S].
n n
n n
Finally,giventwosubjectiveacts()=[x
1
onE
1
;...;x onE
m
]and*()=
m
44 Machina
[x
1
* * * *
onE
1
;...;x
m *
onE
m *
]andany,~

[0,1],wecandefinetheal-
most-objectivemixture[()on

S;*()on~

S]of()and*().
n
n
BELIEFSANDBETTINGPREFERENCESOVER
ALMOST-OBJECTIVEEVENTS
As with the almost-equally likely events defined above, as n all
event-smoothindividualswillexhibitidenticalrevealedlikelihoodbeliefsover
anyalmost-objectiveevent
S
essentiallytreatingitasanexogenousobjec-

n
tiveevent,withaprobabilitygivenbythetotallength()ofthesubset
[0, 1]. That is to say, given any event-smooth preference function W() over
subjective actswhether or not it is expected utility/non-expected utility,
state-independent/state-dependent, or probabilistically sophisticated/non-
*
probabilistically sophisticatedoutcomes x > x, disjoint subsets
[0,1]
with >

( ) ( ), and subjective act (), W() will exhibit

limW x *on S; xon S;f( )elsewhere)> (
n

s
n
n
limW x on S;x*on S ;f( )elsewhere);

s (
n n
n
that is, holding the payoffs elsewhere constant, all event-smooth individuals
areunanimousintheirpreferenceforstakingthegreateroftwoprizesonthe

event S andthelesseron S ,ratherthantheotherwayaround.Thus,

n
n
whilewehaveseenthattypicalsubjectiveeventsneednothaveprobabilitiesat
all,muchlessunanimouslyagreed-uponprobabilities,asn therewillbe

such unanimous agreement on the comparative likelihoods of
n

ver-
sus

S .
n
The idea that some subjective events come close to exhibiting objective
propertiesisnotnew,andprecursorsofthealmost-equal-likelihoodexample
datebackatleasttoPoincar(1912).Norarealmost-objectiveeventsmerelya
technicalcuriosuminfact,mostreal-worldobjectiverandomizationdevic-
esareactuallyexamplesoftheuseofalmost-objectiveeventstoconvertnon-
probabilisticsubjectiveuncertaintyto(almost-)objectiveuncertainty.Tosee
this,considerthesimpleexampleofagameshowspinnerdividedintoalarge
numberofalternatingredandblacksectorsofequalangularsize.Isitcorrect
to say that the spin of such a wheel is an objective process? If so, then it
wouldfollowthatallindividualswouldhavethesamebeliefsoverallevents
StatesoftheWorldandtheStateofDecisionTheory 45
definedoverthisprocess.Buthowmuchagreementwilltherebeonthelikeli-
hoodoftheeventthatthewheelspinsmorethan20revolutionsbeforefinally
stopping?
Viewedfromthisperspective,thebehaviorofthewheelitsexactnumber
ofrevolutionsandthereforethecolorofthesectorthatfinishesoppositethe
pointerisasubjectiveprocess,wherethestateofnatureistheamountofforce
appliedtothespin.Individualswillsurelydisagreeontheirsubjectiveproba-
bilitiesofaneventliketheforcewillbeenoughtogenerateatleast20revo-
lutions,andsomemaynotbeabletoattachanysubjectiveprobablyatallto
thisevent.Butifweplotthestate(theinitialforceofthespin)onthehorizontal
axis of the previous diagram, then an event such as the force will lead the
wheeltostopwiththepointeroppositeablacksectorisseentobeanalmost-
objectiveeventofthetypeillustratedinthefigure,whichiswhyevenindivid-
ualswhodisagreeonthelikelihoodofmorethan20spinswillnevertheless
agreeonthelikelihoodofblack.Inotherwords,itisnottheprocessofspin-
ningthewheelthatiseithersubjectiveorobjective,butratherthedifferent
eventsdefinedonthisprocessthatareeithersubjectiveor(almost-)objective.
Alittlethoughtwillrevealthatvirtuallyallstandardphysicalrandomizationde-
vices used to generate objective likelihoods share this property of being
basedonasubjectivelyuncertain(andhencenon-probabilistic)statevariable
(or variables), but working with periodic, almost-objective events defined
overthestatevariable.
Theaboveargumentshowsthatwithastructured(essentiallyEuclidean)
statespaceandthepropertyofevent-smoothpreferences,thereexistsasubstra-
tumofeventsthatarbitrarilycloselyapproximatethefirstofthefourabove-
listedpropertiesofpurelyobjectiveevents,namelythepropertyofunanimous,
outcome-invariantrevealedlikelihoods.InMachina(2001)Ihaveshownthat
suchevents,andtheactsandmixturesbasedonthem,alsoarbitrarilyclosely
approximate the other three listed properties of idealized purely objective
events.Thatis,asn

:
Eachindividual(probabilisticallysophisticatedorotherwise)willview
all almost-objective events as independent of each purely subjective
,

event, in the sense that for all disjoint

[
0,1
]
and each E S,
theywillhavethesamerevealedlikelihoodrankings(i.e.,bettingpref-
erences)overthejointevents(

S)

Eversus
(

S )E
asthey
n
n

doovertheevents Sversus

S(ineachcase,correspondingto
n
n

( ) ). therelativevaluesof ()versus


46 Machina
Although individuals neednt be probabilistically sophisticated over
subjective acts in general, they will be probabilistically sophisticated
overalmost-objectiveacts.Thatis,eachW()willhaveacorresponding
preferencefunctionV
W
()overlotteriessuchthat
lim W x
1
onS ;...; x
m
on S
)
= limV
W
(
x ,
(

)
;...;x ,
(

m
))
.
1 1 m (
n
1
n
m
n
n
Each individual (probabilistically sophisticated or otherwise) satisfies
the reduction of compound uncertainty property for almost-objective
mixtures of acts. Thus if

f
1
()on S ;...;f
m
()on S

and
n
m
n

1
( )on S ;...; f

m
( )on S

inducealmost-objectivelyequivalent

1
n
m
n

subacts over each event in the common refinement of
f
1
(),...,f
m
(),f

1
(),...,f

m
(),
then
(
limW f
1
()on S ;...;f
m
()on S
)
=
n
1
n
m
n
limW f

1
()on S;...;f

m
()on S
)
.
(

1
n

m
n
n
Thefollowingpropertiesofalmost-objectiveuncertaintyadditionallyholdfor
probabilisticallysophisticatedindividualsandexpectedutilityindividuals:
Eachprobabilisticallysophisticatedindividualwithsubjectiveprobabil-
itymeasure()willviewallpurelysubjectiveeventsasindependentof
eachalmost-objectiveevent,inthesensethatforall E,E

S andeach
[0,1],theywillhavethesamerevealedlikelihoodrankings(bet-
tingpreferences)overthejointevents( S ) E versus ( S ) E

as
n n
theydoovertheeventsEversus E

(ineachcase,correspondingtothe
relativevaluesof(E)versus(
E

)).
Eachexpectedutilitymaximizerwillbelinearinalmost-objectiveprob-
abilitiesandalmost-objectivemixturesofsubjectivelotteries,i.e.,
m
lim W
SEU
(
x
1
onS ;...;x
m
on S
)
=
i=1
( ) W
SEU
(
x
i
onS
)
i
n
1
n
m
n
lim W
SEU
(
( ) on S ; ... ; ( ) on S
)
=
i=1
( )W
(
( )
)

m

n
1 1
n
m m
n
i SEU i
andsimilarlyforW
SDEU
().
StatesoftheWorldandtheStateofDecisionTheory 47
References
Allais,M.1953.LeComportementdelHommeRationneldevantleRisque,
CritiquedesPostulatsetAxiomesdelEcoleAmricaine.Econometrica
21(4):503546.
AnscombeF.,andR.Aumann.1963.ADefinitionofSubjectiveProbabil-
ity.AnnalsofMathematicalStatistics34(1):199205.
Bernoulli,D.[1738]1954.SpecimenTheoriaeNovaedeMensuraSortis,
CommentariiAcademiaeScientiarumImperialisPetropolitanae[Papersof
theImperialAcademyofSciencesinPetersburg]V,175-192.English
translation:ExpositionofaNewTheoryontheMeasurementofRisk.
Econometrica22(1):2336.
Billingsley,P.1986.ProbabilityandMeasure.2ded.NewYork:JohnWiley
andSons.
Chew,S.1983.AGeneralizationoftheQuasilinearMeanWithApplications
totheMeasurementofIncomeInequalityandDecisionTheoryResolving
theAllaisParadox.Econometrica51(4):10651092.
Cohen,M.,andJ.Jaffray.1985.DecisionMakingintheCaseofMixed
Uncertainty:ANormativeModel.JournalofMathematicalPsychology
29(4):428442.
Debreu,G.1954.RepresentationofaPreferenceOrderingbyaNumerical
Function.InDecisionProcesses,R.Thrall,C.Boombs,andR.Davis,eds.
NewYork:JohnWileyandSons,pp.159165.
Edwards,W.1955.ThePredictionofDecisionsamongBets.Journalof
ExperimentalPsychology50(3):201214.
.1962.SubjectiveProbabilitiesInferredFromDecisions.Psycho-
logicalReview69(2):109135.
Ellsberg,D.1961.Risk,Ambiguity,andtheSavageAxioms.Quarterly
JournalofEconomics75:643669.
.2001.Risk,AmbiguityandDecision.NewYork:GarlandPublishing,
Inc.
Feller,W.1968.AnIntroductiontoProbabilityTheoryandItsApplications,
VolumeI. 3dEd.NewYork:JohnWileyandSons.
.1971.AnIntroductiontoProbabilityTheoryandItsApplications,
VolumeII.2dEd.NewYork:JohnWileyandSons.
Fishburn,P.1983.TransitiveMeasurableUtility.JournalofEconomicThe-
ory31(2):293317.
48 Machina
Grdenfors,P.,andN.Sahlin.1982.UnreliableProbabilities,RiskTaking,
andDecisionMaking.Synthese53(3):361386.
.1983.DecisionMakingwithUnreliableProbabilities.BritishJour-
nalofMathematicalandStatisticalPsychology36(2):240251.
Gilboa,I.1987.ExpectedUtilitywithPurelySubjectiveNon-AdditiveProb-
abilities.JournalofMathematicalEconomics16(1):6588.
Gilboa,I.andD.Schmeidler.1989.MaximumExpectedUtilityWithaNon-
UniquePrior.JournalofMathematicalEconomics18(2):141153.
.1994.AdditiveRepresentationsofNon-AdditiveMeasuresandthe
ChoquetIntegral.AnnalsofOperationsResearch52:4365.
Kahneman,D.,andA.Tversky.1979.ProspectTheory:AnAnalysisof
DecisionUnderRisk.Econometrica47(2):263291.
Kreps,D.1988.NotesontheTheoryofChoice.Boulder,Colorado:Westview
Press.
.1990.ACourseinMicroeconomicTheory.Princeton:PrincetonUni-
versityPress.
Machina,M.1982.ExpectedUtilityAnalysisWithouttheIndependence
Axiom.Econometrica50(2):277323.
.1983.GeneralizedExpectedUtilityAnalysisandtheNatureof
ObservedViolationsoftheIndependenceAxiom.InFoundationsofUtil-
ityandRiskTheorywithApplications,B.StigumandF.Wenstop,eds.
Dordrecht:D.ReidelPublishingCo.,pp.263293.
.1987.ChoiceUnderUncertainty:ProblemsSolvedandUnsolved.
JournalofEconomicPerspectives1(1):121154.
.2001.Almost-ObjectiveUncertainty.Workingpaper2001-12.San
Diego:UniversityofCalifornia,DepartmentofEconomics.
.2002.RobustifyingtheClassicalModelofRiskPreferencesand
Beliefs.Workingpaper2002-06.SanDiego:UniversityofCalifornia,
DepartmentofEconomics.
MacCrimmon,K.,andS.Larsson.1979.UtilityTheory:AxiomsVersus
Paradoxes.InExpectedUtilityHypothesesandtheAllaisParadox,
AllaisandHagen,eds.Dordrecht:D.ReidelPublishingCo.,pp.333409.
Mas-Colell,A.,M.WhinstonandJ.Green.1995.MicroeconomicTheory.
Oxford:OxfordUniversityPress.
Poincar,H.1912.CalculdesProbabilits.2ded.Paris:Gauthiers-Villars.
Quiggin,J.1982.ATheoryofAnticipatedUtility.JournalofEconomic
BehaviorandOrganization3(4):323343.
Savage,L.1954.TheFoundationsofStatistics.NewYork:JohnWileyand
Sons.
Schmeidler,D.1989.SubjectiveProbabilityandExpectedUtilitywithout
Additivity.Econometrica57(3):571587.
StatesoftheWorldandtheStateofDecisionTheory 49
Varian,H.1992.MicroeconomicAnalysis.3ded.NewYork:W.W.Norton&
Co.
Wakker,P.1989.ContinuousSubjectiveExpectedUtilityWithNon-Addi-
tiveProbabilities.JournalofMathematicalEconomics18(1):127.
.1990.UnderStochasticDominanceChoquet-ExpectedUtilityand
AnticipatedUtilityareIdentical.TheoryandDecision29(2):119132.
Yaari,M.1987.TheDualTheoryOfChoiceUnderRisk.Econometrica
55(1):95115.
3
GamblingwiththeFuture
EconomicandSocialPerspectives
onCasinosinAmerica
WilliamR.Eadington
UniversityofNevada,Reno
CommercialgamingbecameasubstantialindustryintheUnited
Statesoverthesecondhalfofthe20thcentury,generatingrevenuesin
2001inexcessof$64billion,andhavingalegalpresencein48states.
OverhalfofgamingrevenuescomefromcommercialandIndiancasi-
noslocatedinmorethan30states.
Fromstrictprohibitionsinmoststatesonlyagenerationago,laws
governingcasino-stylegamblinghavebeenentertainedorenactedby
stategovernmentsinterestedinnewsourcesoftaxrevenues,newcata-
lystsforjobcreationandcapitalinvestment,newreasonsforattracting
touristspending,andoccasionallyinresponsetocitizensdesires
toparticipateincasino-stylegamblingforthefunofit.Thetypesof
gamblingauthorizedincludeNevada-stylecasinos,slotmachinesat
racetracks,andvideopoker,videolotteryterminals,andotherelec-
tronicgamingdevicesinbarsandtaverns.Besidesstate-authorized
gaming,nearly200Indiantribeshaveopenedtribalcasinosandgam-
ingcentersthroughoutthecountry,underthegeneralguidanceofthe
IndianGamingRegulatoryActof1988.
Thischapterexaminesthemajorpolitical,social,andeconomic
dynamicsthathaveresultedintherapidproliferationofpermitted
gamblingespeciallycasinosandcasino-stylegamblinginthe
UnitedStatesoverthepastquartercentury.Thisprocessoflegalization
andderegulationhascreatedgamingindustriesofincreasingsize,
sophistication,andpresence,whichhavebecomeorarequickly
51
52 Eadington
becomingpartofthemodernmainstreamofcommercialentertain-
ment,leisure,andtourismindustriesinvariouspartsofthecountry.
Economicbenefitsnotwithstanding,permittedcasino-stylegaming
remainsahighlychargedpoliticalissue.Casinogamingisstillconsid-
eredbysometobeanessentiallyunhealthyactivitythathasnotlostits
previousstatusasaperniciousvice.Statesthathaveauthorizedcasino
gaminghaveoftendonesounderconditionsoflimitedcompetitionor
byusingregionalmonopolystructures.Thisapproachcreateseco-
nomicrentsormonopolyprofitsthatbecometheobjectiveofongoing
rent-seekingbehaviorbyvariousspecialinterestsintheireffortsto
capturetherents.Suchdesignedmarketstructuresaretypicallyaby-
productofdesiresatthelegislativeleveltocontrolgamingssocial
impactsthroughregulatoryconstraints,geographicisolation,or
plannedundersupply.Nonetheless,pressurestoexpandthescopeof
permittedgamingarefoundinmanyjurisdictions,especiallywhen
needsfortaxrevenuegenerationorjobcreationaresubstantial.
Itisdifficulttomakeanunambiguouscaseeitherinfavoroforin
oppositiontopermittingcasino-stylegamingintoanycommunitythat
previouslydidnothavesuchactivities.Nonetheless,inthefirstyears
ofthe21stcentury,itappearsthatgamingindustrieswillcontinueto
expandinnewanddiversewaysinmanyjurisdictions.Atminimum,
expandingpermittedcasino-stylegamingisnowactivelyonthe
agendainmanystatelegislatures,anditislikelytoremainsoforsome
timetocome.
TRENDSINGAMINGINTHEUNITEDSTATES
Eventsofrecentyearsareacontinuationofprocessestowardlegal-
izationandagreaterpresenceofpermittedgamingthatbeganwith
Nevadascasinolegislationin1931andNewHampshireslottery
legalizationin1963.However,themainspreadoflegalcommercial
casinosoccurredinthefirsthalfofthe1990s.Priorto1988,casinos
hadbeenauthorizedonlyinNevadaandinAtlanticCity,NewJersey.
AtlanticCityitselfwasarelativelynewaddition,withitscasinosopen-
ingtheirdoorsforthefirsttimein1978.Between1988and1996,a
totalofninestates
1
authorizednewcasinoindustries,someasriverboat
GamblingwiththeFuture 53
casinos,someaslimited-wagercasinosinformerminingtowns,and
someasurbancasinos.Indiantribalcasinoswereeffectivelylegalized
byaSupremeCourtdecisionin1987
2
andwereprovidedastatutory
frameworkwiththepassageoftheIndianGamingRegulatoryActin
1988.
3
Indiancasinosspreadtonearly30statesbytheearly21stcen-
tury,withthemostsignificanttribalcasinosfoundinsuchstatesas
Connecticut,Minnesota,Michigan,NewYork,andCalifornia.
TheproliferationofpermittedcommercialcasinosintheUnited
Statessloweddownafter1993,coincidingwiththeimprovementsin
performanceofthenationaleconomy.However,anothertrendsoon
emerged:theauthorizationofgamingdevicesatracetracksinvarious
states,purportedlytoprovidetheracingindustrywithalevelplaying
fieldagainstnewlyauthorizedformsofgaming,andacompetitive
edgeovertracksinotherstatesinattractingpursesandhigh-quality
racehorses.Theeffectofthisdevelopmentwastocreateanumberof
racinos,wherethepresenceofslotmachineswouldtransformrace
tracksintodefactocasinos,andtypicallyleadtoahighproportionof
totalrevenuesforsuchoperationsbeinggeneratedbythegaming
devicesratherthanwageringonracing.Suchracetrackcasinoshave
developedinIowa,Delaware,WestVirginia,NewMexico,Rhode
Island,andLouisianainthe1990s.
Changingeconomiccircumstances,especiallyrecessionandsub-
stantialfiscalshortfallsatthestatelevel,contributetothecasino
debate.In2003,theUnitedStateswentthroughanotherroundofdis-
cussionofwhethertolegalizeandexpandcasinosandcasino-style
gaming.Economiccircumstancesintheearlyyearsofthefirstdecade
ofthe21stcenturyparalleltheperiodfrom1989to1993,whenthe
nationaleconomyslowedandthenmovedintorecession,andwhen
manystatesfoundthemselvesfinanciallystrappedanddesperatefor
job-creatingstrategies.Withtheeconomicslowdownandrecessionof
20002003,anincreasingnumberofjurisdictionsintheUnitedStates
foundthemselvesinfinancialdifficulty.Assuchslowdownsoccurred,
commercialgamingwasoftenoneofthestrategiesputforthforraising
governmentrevenuesandstimulatinglocalandregionaleconomies.
Thus,in2002and2003,debatesoncasinos,slotmachinesatrace
tracks,andevenslotmachinesinbarsandtavernstookplaceinlegisla-
turesandamongpoliticalleadersinHawaii,Kentucky,Maryland,
54 Eadington
Pennsylvania,NewHampshire,RhodeIsland,Wisconsin,andMinne-
sota,aswellasinotherstates.
Othereventscanalsohaveimpactsonthedebatesofwhetheror
nottoexpandthepresenceofcasinos.InOctober2001,shortlyafter
theSeptember11terroristattacks,theStateofNewYorkauthorizedsix
newtribalcasinosandslotmachinegamingateightracetracks.This
wasmotivatedinnosmallmeasurebytheneedtoclosethegapagainst
largeimpendingstatedeficits,relatedbothtotheeconomicslowdown
andtotheanticipatedeconomicconsequencesoftheterroristactions
andthesubsequentwaronterrorism.Thedebatewashastenedbythe
realitythatby2001,NewYorkwassurroundedbysuccessfulcasino
gamingoperationsinAtlanticCity,EasternConnecticut,andOntario,
Canada.
COMMERCIALGAMINGANDCONTROVERSY
Between1982and2001,totalgamingrevenuesofcommercial
gamingindustriesintheUnitedStatesgrewfrom$10.2billionto$65.8
billion,withmorethanhalfofthe2001totalcomingfromcommercial
andtribalcasinos.Lotteries,pari-mutuelwageringonracing,andchar-
itablegambling,includingbingo,alllostmarketshareascasinosand
electronicgamingdevicesincreasedtheirpresenceandpopularityover
thepasttwodecades.
However,inspiteofrapideconomicexpansions,generalattitudes
towardtheacceptanceofpermittedcasinosremainedatbestlukewarm
inmostjurisdictions.Thereisgrowingsentimentinanumberofstates
thatatleastinsomesituationsgovernmentshaveauthorizedtoo
muchgaming.Insuchlocales,therearepressurestoreversesomeof
thetrendsthathavecharacterizedcommercialgamingindustriesinthe
pastthreedecades.
Insomesituations,substantialcommercialgamingindustrieshave
seentheirlegalstatusesrevoked.Thisoccurredwhenauthorizationfor
videopokermachinesinSouthCarolinawasallowedtoexpirein2000,
eliminatinganindustrythatwasgeneratinggrossgamingrevenuesin
excessof$500millionperannum.In1996,localelectionsreversedthe
GamblingwiththeFuture 55
legalstatusofvideopokermachinesin34ofthe66parishesinLouisi-
ana.
Theissueofwhengamblingoverextendsitspoliticalwelcomecan
beseeninrecenteventsabroad.Suchdevelopmentsmightprovide
insightsintowhatmaylieaheadforAmericanjurisdictionswantingto
fullyexploittheeconomicrentsfromcasino-stylegambling.Following
publicationofa1999ProductivityCommissionReportonGambling
(AustralianProductivityCommissionReport1999),Australiaadopted
anumberofrestrictionsonelectronicgamingdevicesafterdeclaring
problemgamblingtobeapublichealthissue,underadeclaredstrategy
ofharmminimization.Thisfollowedadecadewherethenumberof
electronicgamingdevicesinAustraliaexpandedfromabout70,000to
approximately190,000,90percentofwhichwerelocatedoutsideof
casinos(Monaghan2001).TheProductivityCommissionReport
claimed,amongotherfindings,thatthe2.1percentofadultAustralians
whowereproblemgamblersmadeup10percentofregularplayerson
gamingmachines,andgenerated42percentofspendingongaming
machinesinAustralia.
Newtechnologicaldevelopmentsinthegamingindustrieshave
alsobecomepartofthepoliticalcontroversysurroundinggambling.
PerhapsthemostdramaticoftheseisInternetgambling,whoselegal
statushasbeenactivelydebatedinmanycountriesthroughoutthe
world,withnoclearresolutionintheearly21stcenturyingeneral
trendsanddirections.Internetgamblinghasaverylargepotentialmar-
ketandhasthecapabilitytobringhighlysophisticatedgamingprod-
uctsintohouseholdseverywhere.Basedonthespottyevidencethat
existsonthisstilllargelygrayareaactivity,thesizeoftheglobal
Internetgamblingmarketisalreadymeasuredinthebillionsofdollars
(see,forexample,Cabot2000).
Ontheotherhand,Internetgamblingraisessocialconcernsabout
thepotentialadverseimpactssuchubiquitousgamingopportunities
mightbringabout,especiallyintheareasofunderagegamblingand
problemandpathologicalgambling.Theactivityalsoposesinteresting
challengesforjurisdictionsonhowtoregulateandtaxtheactivity,cre-
atingadilemmaforgovernmentsthataretolerantofpermittedgaming
primarilybecauseoftheirabilitytoextracteconomicrentsfromexcise
taxesontheactivity.BecauseInternetgamblingoperateswithlittle
concernfornationalborders,andbecausesomejurisdictionshave
56 Eadington
decidedtoencourageInternetgamingsitestolocatewithintheirbor-
dersthroughofferinglowtaxrates,otherjurisdictionswillhaveto
matchorcomeclosetothosetaxratestoremaincompetitive.Further-
more,theUnitedStates,atboththecongressionallevelandinvarious
states,hasdemonstratedlittledesiretomoveforwardtofullyexploit
theeconomicopportunitiesofInternetgambling.
ItispossiblethatInternetgamblingisjustthetipofthetechnolog-
icaliceberg.Interactivetelevisionbettingandtheuseofvarioushand-
heldcomputerdevicesforplayinggamesandmakingwagersareper-
hapsthenextmajorgamblingdevelopments.However,theywillcon-
tinuetobepoliticallycontroversialbecauseofthedifficultiesin
exercisingsocialcontrolsovertheactivitieswheretheytakeplace,and
becauseoftheabilityofnewtechnologiestooutstriplegislative
attemptstoconstrainthepresenceoravailabilityofgamblingingen-
eral.Aby-productofthenewworldofInternetandotherlow-costand
virtuallyinstantaneouscommunicationsislikelytobetheinabilityto
significantlyconstraingamblingactivitiesthattakeplacethroughthose
media,regardlessofthewishesanddesiresoflegislativeandparlia-
mentarybodies.
AFRAMEWORKFOREVALUATINGTHEBENEFITS
ANDCOSTSOFEXPANDEDGAMBLING
Oneaspectoflegalizingnewformsofpermittedgamblingisthat
suchactionscreatebenefitsthatimpacteconomiesespeciallylocalor
regionaleconomiesinwaysthataregenerallytangible,measurable,
andeconomic.Butanexpandedpresenceofpermittedgamblingalso
generatessocialcoststhataffectindividualsandhouseholdsinways
thatarefarlesstangible,measurable,andvisible.Itisextremelychal-
lengingtopolicymakersandsocialscientiststoconceptualize,identify,
andmeasurethesocialcoststhataccompanygamblinginanymean-
ingfulway(see,forexample,WalkerandBarnett1999andEadington
2003).Furthermore,becauseoftherelativelackofattentiontothe
costsandbenefitsofgamblingpriortothemid1990s,littleserious
effortwasundertakentoaddresstheseissues.
4
Itislikelythatthese
dimensionsofbenefitsandcostsassociatedwithgamblingwillremain
GamblingwiththeFuture 57
attheheartofthedebateoverthewisdomofexpandingorcontracting
theavailabilityofpermittedgamblingforsometimetocome.
Nonetheless,anumberofobservationscanbemadeabouttheben-
efitsandcostsofpermittedgamblingincomparisontoreasonable
alternativestatesofnature.Forthosejurisdictionsthatarestilldebating
thestatusofgamingwithintheircontrol,suchobservationsshould
proveuseful.
First,itshouldbenotedthattheprimarybenefitassociatedwith
permittedgamblingisthecreationofconsumersurplus,theincremen-
talvaluetoconsumersfrombeingabletoparticipateinanactivitythat
waspreviouslyprohibited.Consumersurplusisgenerallydefinedas
thedifferencebetweenwhatconsumerswouldbewillingandableto
payforanactivityversuswhattheyactuallyhavetopayforthatactiv-
ity.Suchgainsaccruepredominantlytotheconsumersofgambling
servicesratherthantoproducersorthegovernmentswhoauthorizethe
activity.
However,whenpermittedgamblingisauthorizedinamannerthat
preventsthemarketfromexpandingtoitsdemandpotential,orwhen
themarketstructureisdesignedtoresultinmonopolyorotherwise
restrictedcompetition,thenthepriceoftheactivityincreases.Asa
result,aportionofpotentialconsumersurplusisdivertedawayfrom
consumersandbecomesvalueforsomeoneelse.Thedivertedcon-
sumersurpluscanbereferredtoaseconomicrents.Economicrentscan
becapturedbygovernmentthroughtheimplementationofexcisetaxes
ontheactivity,orbyoutrightownershipofthegamingfranchise.Other
economicrentsmightbecapturedbycompaniesororganizationsthat
offergamblingservicesthroughexclusiveorlimitedfranchises.Only
whenthemarketisallowedtoexpandtoitsdemandpotential,orwhen
competitionfromrelatedsubstituteactivitiesbiddownthepriceofthe
primaryactivitytocompetitivelevels,aretheeconomicrentsbid
away.
5
Aswithotheractivities,mostofthecostsandbenefitsassociated
withpermittedgamblingareinternaltotheconsumersandproducers
ofthegamblingactivities.Undertheassumptionofrationaleconomic
actors,consumerschoosetospendmoneyongamblingbecausethey
derivegreatervaluefromparticipationthantheexpectedorrealized
cost.Producersprovidegamblingservicesbecauseitprovidesagreater
returnontheirresourcesthanthenextbestalternatives.Asprivateben-
58 Eadington
efitsandcosts,thereislittlereasonforpublicpolicyconsiderationsto
affectthedecisionprocessesthatgeneratetheseallocations.
Publicpolicyinterventionistypicallyjustifiedwhennegative
externalitiesareassociatedwiththeactivity.
6
Withgambling,thepri-
marynegativeexternalitiesarelinkedtoproblemandpathological
gambling.Generallyspeaking,therehavebeentwomajordriving
forcesthathaveinfluencedsocietaldecisionstoliberalizegambling
lawsandregulations:1)adesireonthepartofgovernmentstocapture
economicrentsthroughpermittingapreviouslyprohibitedactivity;
and2)adesiretomitigatethenegativesideeffects(realorperceived
negativeexternalities)associatedwiththeactivitybyconstrainingitin
waysthatwouldallowforgreatercontroloftheadversesideeffects.
Thecombinationofthesetwosomewhatconflictingforceshasled
toavarietyofeccentriclawspassedinvariousjurisdictionsthroughout
theworld.IntheUnitedStates,riverboatgamblingwithmandatory
sailing,orminingtowncasinoswithlosslimitsandrestrictionstohis-
toricbuildingsonly,reflectstateseffortstocaptureeconomicrents
whileprovidingprotectionsagainstpeoplewhomightoverindulgein
gamblingactivities.VoterratificationofIndiangaming,asintheState
ofCaliforniain1998and2000,wasavalidationofthedistributionof
economicrentstotribesandtribalmembers;Californiaslegislature
andvotershavebeenreluctanttobestowsimilareconomicrentson
otherrentseekers,suchasthecardclubindustryortheracingindustry.
Aswithothervicessuchastobacco,alcohol,illicitdrugs,and
commercialsex,gamblingisperceivedasanactivitythathasastrong
realizedandlatentdemandthatemanatesfromaportionofthepopula-
tion.Aswiththeothervices,italsopossessesavarietyofnegativeside
effectsperceivedorrealthatareviewedasimmoralorotherwise
sociallydamagingby(typically)anothersubsetofthepopulation.Such
sideeffectshaveservedastheimpetusforconstraintsonthepermitted
offeringsofgamblingservices.Aswiththeothervices,thereisno
clearconsensusonthebestapproachtoregulatingandconstrainingthe
availabilityofgambling,andasaresult,therehasnotbeenmuchsta-
bilityonthemannerinwhichlegalgamblinghasbeenpermittedand
constrainedfromonepoliticaljurisdictiontoanother.
Theextentofdemandforgamblingthatisrealizedasopposedto
remaininglatentispartlyafunctionofgamblingslegalstatus.If
casinogaming,throughtheprocessoflegalizationorderegulation,is
GamblingwiththeFuture 59
mademoreattractiveandavailabletoasocietyspopulationcenters,
thenthedemandforgamblingingeneral,andthetotalamountof
incomespentongambling,willincrease.Thegreatertheavailabilityof
gambling,andthefewertheconstraintsthatareappliedtogaming
activities,thelargertherealizeddemandwillbe.Furthermore,the
morethatpermittedgamblingisofferedinacompetitivemarketcon-
text,themorethatdemandforgamblingwillincrease.Increasedcom-
petitionwillresultinlowerrealizedpricestoconsumersforgambling,
andcompetitionwillenhancethepriceandavailabilityofcomplemen-
tarynongamingactivitiesaswell.Therecentexperienceofcompetitive
venuessuchasLasVegasandMississippiincomparisontomore
supplyconstrainedormonopolisticjurisdictionssuchastheurbancasi-
nosinDetroitandNewOrleans,orriverboatjurisdictionsinIllinois,
Indiana,orLouisianaclearlydemonstratetheseeffects.
Whentryingtoevaluatesocialbenefitsandcostsassociatedwith
gambling,itisimportanttoevaluatethealternativestothestatusof
permittedgamblingunderconsideration.Ifajurisdictioncurrentlypro-
hibitsgamblingbuthasasubstantialamountofillegalgamblingtaking
placewithinitsborders,removaloftheprohibitionswilllikelydimin-
ishtheadverseeconomicimpactsoftheillegalindustry,andquitepos-
siblywilldiminishtheseverityofsomeofthesocialcostsassociated
withsuchillegalactivities.
Itisalsousefultolookatthegenerallocationalstructureunder
whichcasinoandcasino-stylegamblingisoffered,intermsofits
potentialfordeliveringbenefitsandcosts.Thoughitisarguedelse-
wherethatbenefitsandcostsofpermittedgamblingshouldbedoneat
thenationallevel(GrinolsandMustard2001),mostpolicyanalysis
concentratesonlocalandregionaleconomicbenefitsassociatedwith
permittedcasinosandcasino-stylegaming.Usingthatasastarting
point,onecancreatethefollowingcategoriesofcasinosandnear-casi-
nos:
Destination resort casinos located away from population centers
(suchasLasVegas,Reno,andLakeTahoe,Nevada;Biloxi,Mis-
sissippi;orAtlanticCity,NewJersey).
Rural casinos, located away from population centers (such as
Foxwoods in Connecticut and most tribal casinos in the United
60 Eadington
States;andthecasinosinDeadwood,SouthDakota;andinCen-
tralCity,Blackhawk,andCrippleCreek,Colorado).
Urbanorsuburbancasinoslocatedinornearmajormetropolitan
areas(suchasthosefoundinDetroit,orinandaroundSt.Louis,
Kansas City, and Cincinnati), as well as most race track casinos
(racinos).
Neighborhood casino-style gaming (such as video poker
machines, video lottery terminals, and other gaming devices
foundinbarsandtavernsinsuchstatesasNevada,Montana,Ore-
gon,andSouthDakota).Thisissometimesreferredtoasconve-
niencegambling.
Ifwecomputebenefitsandcostsforgamblinginthetraditional
manner,anddiscounttheimportanceofconsumersurplus,
7
wefind
thatjurisdictionsthatexportgamblingtocitizensofotherjurisdictions
tendtocaptureasubstantialamountofeconomicbenefitintheformof
economicrentsandvalueaddedbyproducersandownersoflocal
resources(i.e.,thebenefitsofincreasedlocalemployment),whereas
thesocialcostsassociatedwithproblemgamblinginparticulartendto
getexportedtothejurisdictionswherethegamblingconsumersreside.
Insuchcases,theratioofbenefitstocostswithinthejurisdictionisrel-
ativelyhigh.
Inasimilarfashion,benefit/costratiosforruralcasinosarealso
fairlyhigh,especiallyiftheregionforwhichtheimpactsarebeing
evaluatedincludesonlytheruralarea.Thisisoftenthecasewith
Indiantribalcasinos,wheretheprimarygroupofinterestisthetribe
itself,andmostofthecasinocustomersarenottribalmembers.
Ontheotherhand,ifurbanorsuburbancasinosareevaluatedin
thismanner,thebenefit/costratioisconsiderablylower.Mostofthe
gamingactivitiesprovidedbysuchcasinoscatertodemandinthelocal
market.Insuchacase,spendingongamblingdoesnotstimulatethe
localeconomyinthesamemanneritwouldifgamblingactivitieswere
exported.Furthermore,socialcoststypicallyremainwithinthecom-
munitywherethegamingfacilitiesarelocated.Thus,measuredbene-
fitswillbelowerandsocialcostswillbehigherthanineitherofthe
firsttwocases.Nonetheless,suchurban/suburbancasinoscancreate
significantregionalinvestmentandmightserveasefficientmecha-
nismsfortaxrevenuegeneration.Furthermore,theymightbringabout
GamblingwiththeFuture 61
considerableimportsubstitutionbehavior,encouraginglocalresidents
whootherwisemighttraveloutoftheregiontopursuegamblingactiv-
itiestospendtheirgamblingbudgetsinlocalcasinosinstead.
Ifweconsiderthesituationofconveniencegamblinggaming
devicesinbarsandtavernslocatedinneighborhoodsthegeneralten-
dencyisforbenefitstobelowerandsocialcoststobehigherthanin
anyoftheprevioussituations.Sincesuchfacilitiesgeneratelittleinthe
wayofnewinvestmentorjobcreationassociatedwiththegambling
activities,economicbenefitstendtobelower.
8
Becausecasino-style
gamingisofferedinmoreaccessiblesurroundingsthanistypicalfor
site-specificcasinos,theremightbeagreaterincidenceofimpulsive
gamblingand,asaresult,ofproblemandpathologicalgambling.
Theratioofbenefitstocostsforaregionorjurisdictionisabell-
wethertotheextentofcontroversyassociatedwiththevarioustypesof
permittedgambling.Inlightofthisframework,especiallywhencon-
sumersurplusisgivenrelativelylittlestanding,itisnotsurprisingto
seeconveniencegamblingasthemostpoliticallyvulnerableofthe
alternativesconsidered.Thisthesisisconsistentwiththerecentexperi-
encesinAustralianotedpreviously,aswellasjurisdictionssuchas
SouthCarolinaandLouisiana,whereconveniencegamblingwaselim-
inatedorthreatenedwitheliminationbecauseofthepoliticalback-
lashesassociatedwithit.
Thisframeworkalsocarriesimplicationsforthenewformsof
gambling.Unlessconsumersurplusisgivengreaterstanding,Internet
gamblingandinteractivetelevisiongambling,forexample,willlikely
provetobeverylowonperceivedeconomicbenefitsandveryhighon
socialcosts.Furthermore,thecompetitiveandglobaldimensionsof
Internetgamblingmakeitverydifficultforgovernmentstocapture
economicrents,especiallyintheformoftaxesongrossgamingreve-
nues.Also,theregulatorychallengesofpermittedgamblinginthe
home,especiallygamblingbyyouthorbythosepronetooverindulge,
implythatthesocialcostsassociatedwithsuchactivitiesaregoingto
bebothsociallydangerousandveryhardtocontrolwithoutviolating
otherdimensionsofpersonalprivacy.Thus,thesenewestformsof
gamblingmightprovetobethemostcontroversialofall.
62 Eadington
CONCLUSION
Insummary,theongoingdynamicsoftheeconomicandsocial
impactsofgamblingandofpermittedgamingindustriespointouta
numberofimportantdimensionscharacteristicoftheactivity,the
industry,andofpublicpolicyprocessesregardinggambling.Most
importantoftheseare:
Gamblingisoneofthelargestindustrieswhosefundamentaleco-
nomic characteristics are substantially determined by political
decisions.
Political decisions regarding gambling are largely influenced by
the ability of competing special interestsincluding state gov-
ernmentstocaptureeconomicrentsassociatedwithliberalizing
permittedgamingactivities.Thisisoftencounteredbyperceived
or real social costs associated with problem and pathological
gamblingandwithanincreasedavailabilityofgamblinginsoci-
ety.
Thereisastronglatentdemandforcasino-stylegaming(includ-
ing gaming within casinos and with electronic gaming devices
located outside of casinos), which is manifested when the legal
statusofgamblingisliberalized.
Technologieshavedevelopedoverthepasttwodecadesthathave
broadenedtheappealof,andthemarketfor,commercialgaming.
The same technologies have raised concerns over some adverse
social impacts that such an increased presence of gambling in
societymightbringabout.Manyoftheseadversesocialimpacts
arerelatedtoproblemandpathologicalgamblingbehavior.
Benefit/costanalysisappliedtopermittedgamingactivitiesisstill
arelativelyprimitivescience,primarilybecauseofthedifficulties
in conceptualizing, observing, and measuring social costs.
Becauseofitslingeringstatusasavice,consumersurplusassoci-
ated with gambling consumption is often discounted in policy
discussions.
Some types of permitted gambling raise greater social concerns
overtheirimpactsthandoothers.Somecategoriesofvenuesfor
GamblingwiththeFuture 63
casinosandcasino-stylegamblingaremorevulnerabletopoliti-
calcontroversyandpossiblyreversalofliberalizationofpermit-
ted activities than are others. The forms of gambling with the
greatestpotentialforcontroversyincludeconveniencegambling,
Internetgambling,andinteractivetelevisiongambling.
Thedebateovertheproperroleofpermittedgamblinginsocietyis
farfromover,thoughtherearesomeclearlong-termtrendsvisible
formuchofthepasthalf-centurythathavesupportedincreasedlegal-
izationandderegulationinmanyjurisdictions.Inmanyrespects,these
trendsreflectsocietysincreasedacceptanceofgamblingasaproper
formof(adult)leisureandentertainment.
However,ashasbeendemonstratedinvarioussituations,public
attitudestowardgamblingcanbefickle.Shouldsignificantproblems
arisesuchascorruptionscandals,thepresenceoforganizedcrime,or
evensensationalincidentsinvolvingpathologicalgamblersgambling
mightonceagaincomeunderfire.Iftheperceptionsofsocialcosts
associatedwithgamblingbecomesubstantialrelativetotheeconomic
benefitsthatitiscreating,thenthepoliticalwindscanquicklyshift
harshlyagainstitspermittedstatus.Unlessanduntilrespectforgam-
blingasaconsumptionactivityachievesalevelcomparablewithother
consumptionactivities,newertypesofpermittedgamblingwillcon-
tinuetoraisepublicpolicydebatesandremainatthecenterofpolitical
controversies.
Notes
1. The states were South Dakota, Iowa, Illinois, Colorado, Louisiana, Mississippi,
Missouri, and Indiana. Only Michigan, where voters authorized three casinos in
Detroitin1996,wasaddedtothislistbetween1994and2001.
2. Californiaetal.v.CabazonBandofMissionIndiansetal.,No85-1708,February
25,1987;reprintedinEadington(1990).
3. P.L.100-497(1988),100thCongress;reprintedinEadington(1990).
4. UntilcasinosspreadbeyondNevadaandAtlanticCityintheUnitedStates,there
waslittleinthewayofinstitutionallyfundedresearchongambling.Similarcir-
cumstancesprevailedinothercountries.Sincethe1990s,therehavebeenanum-
berofmajornationalstudiesundertakeninvariouscountries,includingtheFinal
Report(NationalGamblingImpactStudyCommission1999)intheUnitedStates,
theGamblingReviewReport(DepartmentforCulture,MediaandSport2001)in
64 Eadington
theUnitedKingdom,andAustraliasGamblingIndustries(ProductivityCommis-
sion1999)inAustralia.
5. It should also be noted that the idea of consumer surplus has seldom been an
important factor in deliberations regarding legalizing or deregulating gambling.
This is probably because of long-standing prejudices that gambling is a tainted
activity, and people who participate in gambling are themselves exercising poor
judgmentintheirconsumptionchoices,andshouldthereforenotbegivenmuch
considerationindeliberations.Asaresult,mostpolicydeliberationreliesprima-
rilyonthemagnitudeanddistributionoftheeconomicrents.
6. Negativeexternalitiesarisewhenthemarkettransactionsbetweentwopartiescre-
atecostsforthirdpartieswhoarenotinvolvedinthetransactions.Withoutpolicy
intervention, this shifting of costs results in overproduction of the activity that
createsnegativeexternalities.
7. Seenote6.
8. It should be noted, however, that such gaming devices might be extremely effi-
cienttaxcollectors.
References
Cabot,Anthony.2000.TheInternetGamblingReport,4thed.LasVegas:
TracePublishing.
DepartmentforCulture,MediaandSport.2001.GamblingReviewReport.
Norwich,U.K.:StationeryOffice.Availableat:<www:culture.gov.UK/>.
Eadington,WilliamR.1990.IndianGamingandtheLaw.Reno:University
ofNevada.
.2003.MeasuringCostsfromPermittedGaming:ConceptsandCate-
goriesinEvaluatingGamblingsConsequences.JournalofGambling
Studies19(2):185212.
Grinols,EarlL.,andDavidB.Mustard.2001.BusinessProfitabilityVersus
SocialProfitability:EvaluatingIndustrieswithExternalitiesTheCaseof
Casinos.ManagerialandDecisionEconomics22(1):143162.
Monaghan,Sean.2001.AustralianGamingMachineReview:Assessingthe
PaceandCostofHarmMinimization.Analystreport,Sydney:Burdett,
BuckeridgeYoung.
NationalGamblingImpactStudyCommission.1999.FinalReport.Washing-
ton,DC:GovernmentPrintingOffice.
ProductivityCommission.1999.AustraliasGamblingIndustries.Melbourne;
Canberra:ProductivityCommission.Availableat:<http:www.pc.gov.au/
inquiry/gambling/index.html>.
Walker,DouglasM.,andA.H.Barnett.1999.TheSocialCostsofGambling:
AnEconomicPerspective.JournalofGamblingStudies15(3):181212.
4
CommonValueAuctions
andthe
WinnersCurse
LessonsfromtheEconomicsLaboratory
JohnH.Kagel
OhioStateUniversity
Auctionsareofconsiderablepracticalandtheoreticalimportance.
Inpracticalterms,thevalueofgoodsexchangedinauctionseachyear
is huge. Governments routinely use auctions to purchase goods and
services,tosellgovernmentassets,andtofundthenationaldebt.Pri-
vate sector auctions are common as well, and are of growing impor-
tance in areas such as deregulated utility markets, allocation of
pollutionrights,andthelargevarietyofitemsnowbeingsoldviaInter-
netauctions.Auctionsarecommonlyemployedwhenonepartytothe
exchange (for example, the seller) is uncertain about the value that
buyers place on the item; they provide a mechanism, absent middle-
men, to establish value in such situations. Auctions play a prominent
roleinthetheoryofexchange,astheyremainoneofthesimplestand
mostfamiliarmeansofpricedeterminationintheabsenceofinterme-
diate market makers. In addition, auctions serve as valuable illustra-
tions, and one of the most prominent applications, of games of
incomplete information, as bidders private information is the main
factoraffectingstrategicbehavior(Wilson1992).
Thereareatleasttwodistincttypesofriskinauctions.Inprivate
valueauctions,wherebiddersknowthevalueoftheitemtothemselves
with certainty, there is uncertainty regarding other bidders values. In
first-pricesealed-bidauctions,inwhichbuyerssimultaneouslysubmit
sealedbidswiththehighbidwinningtheitematthepricebid,bidders
65
66 Kagel
face a strategic trade-off: the lower their bid the higher their surplus
conditional on winning, but the lower their probability of winning.
1
Further,thefamous(withineconomiccircles,atleast)Vickreyauction
(Vickrey1961),inwhichthehighbidwinsbutpaysthesecond-highest
bid price, was designed with the specific purpose of eliminating this
strategic uncertainty. Bidders in the Vickrey auction have a dominant
strategyofbiddingtheirvaluations,sotheydonothavetoconsiderthis
strategic trade-off. (Similarly, in an open outcry English auction in
which bidding starts out low and the auctioneer gradually raises the
price,biddershaveadominantstrategytoremainactiveuntiltheprice
reachestheirvaluation.Hence,therearenostrategictrade-offshereas
well.)
Commonvalueauctions,theothercanonicaltypeofauction,intro-
duceawholenewriskdimension.Inapurecommonvalueauction,the
value of the item is the same to all bidders. What makes the auction
interestingisthatbiddersdonotknowthevalueatthetimetheybid.
Insteadtheyreceivesignalvaluesthatarecorrelatedor,moretechni-
cally, affiliated (Milgrom and Weber 1982)with the value of the
item, so that bidders must estimate the common value based on their
private information signals, while still wrestling with the strategic
issuesassociatedwithprivatevalueauctions.Mineralrightsauctions,
particularlythefederalgovernmentsoutercontinentalshelf(OCS)oil
leaseauctions,aretypicallymodeledaspurecommonvalueauctions.
Thereisacommonvalueelementtomostauctions.Forexample,bid-
dersforanoilpaintingmaypurchasefortheirownpleasure,constitut-
ingaprivatevalueelement,buttheymayalsobidforinvestmentand
eventualresale,reflectingthecommonvalueelement.
Therearenoefficiencyissuesinpurecommonvalueauctions,as
allbiddersplaceequalvalueontheitem.
2
Whathasbeenofoverriding
concern to both theorists and practitioners for these auctions are the
revenue-raising effects of different auction institutions. A second key
issue, one that has provided much of the focus for both experimental
andempiricalworkoncommonvalueauctions,isthewinnerscurse,
anunpredictedeffectthatwasinitiallypostulatedonthebasisoffield
data,andwhoseexistencehasoftenbeenhotlydebatedamongecono-
mists.
ThewinnerscursestorybeginswithCapen,Clapp,andCampbell
(1971),threepetroleumengineerswhoclaimedthatoilcompanieshad
CommonValueAuctionsandtheWinnersCurse 67
suffered unexpectedly low rates of return in the 1960s and 1970s on
OCSleasesalesyearafteryear.
3
Theyarguedthattheselowratesof
returnresultedfromthefactthatwinningbiddersignoredtheinforma-
tional consequences of winning. That is, bidders naively based their
bids on the unconditional expected value of the item (their own esti-
mates of value), which, although correct on average, ignores the fact
thatyouonlywinwhenyourestimatehappenstobethehighest(orone
ofthehighest)ofthosecompetingfortheitem.Butwinningagainsta
numberofrivalsfollowingsimilarbiddingstrategiesimpliesthatyour
estimateisanoverestimateofthevalueoftheleaseconditionalonthe
eventofwinning.Unlessthisadverseselectioneffectisaccountedfor
informulatingabiddingstrategy,itwillresultinwinningbidsthatpro-
ducebelownormalorevennegativeprofits.Thesystematicfailureto
accountforthisadverseselectioneffectiscommonlyreferredtoasthe
winnerscurse:youwin,youlosemoney,andyoucurse.
Terminological aside: When discussing the winners curse, many
economists,particularlytheorists,unfortunatelyusethetermtoreferto
the difference between the expected value of the item conditional on
theeventofwinningandthenaiveexpectation(notconditioningonthe
eventofwinning).Further,theiruseofthetermtypicallyreferstoplay-
erswhofullyaccountforthiswinnerscurse,ratherthanthosewhofall
preytoit.
TheideathatoilcompaniessufferedfromawinnerscurseinOCS
lease sales was greeted with skepticism by many economists, as it
impliesthatbiddersrepeatedlyerr,violatingbasiceconomicnotionsof
rationalityandcontrarytoequilibriumpredictions.
4
Analternativeand
simplerexplanationastowhyoilcompaniesmightclaimthattheyfell
preytoawinnerscurseliesincarteltheory,asresponsivenesstothe
winnerscurseclaimcouldserveasacoordinationdevicetogetrivals
to reduce their bids in future sales. Nevertheless, claims that bidders
fellpreytothewinnerscursehaveariseninanumberoffieldsettings.
In addition to the oil industry (Capen, Clapp, and Campbell 1971;
LorenzandDougherty1983andreferencescitedtherein),claimshave
been made in auctions for book publication rights (Dessauer 1981),
professionalbaseballsfreeagencymarket(CassingandDouglas1980;
Blecherman and Camerer 1998), corporate takeover battles (Roll
1986),andinrealestateauctions(AshenfelterandGenesore1992).
68 Kagel
It is exceedingly difficult to support claims of a winners curse
using field data because of reliability problems with the data and
because alternative explanations for overbidding are often available.
For example, Hendricks, Porter, and Boudreau (1987) found that in
early OCS lease sales, average profits were negative in auctions with
sevenormorebidders.Theynotethatonepossibleexplanationforthis
outcome is the increased severity of the adverse selection problem
associated with more bidders. However, they note that the data could
alsobeexplainedbybidderuncertaintyregardingthenumberoffirms
competingonagiventract(theirpreferredexplanation).Thatis,since
mosttractsreceivedlessthansixbids,itseemslikelythatfirmswould
expectthisnumberorless.Asaresult,althoughfirmsmighthavefully
accountedfortheadverseselectioneffectbasedontheexpectednum-
beroffirmsbiddingonatract,theywouldneverthelessbeincorrectfor
tractsthatattractedaboveaveragenumbersofbidders,andoverbidon
thosetracts.
Theambiguityinherentinusingfielddata,inconjunctionwiththe
controversialnatureofclaimsregardingawinnerscurse,providedthe
motivationforexperimentalstudiesofthewinnerscurse.Earlylabo-
ratory experiments showed that inexperienced bidders are quite sus-
ceptibletothewinnerscurse(BazermanandSamuelson1983;Kagel
andLevin1986;Kageletal.1989).Infact,thewinnerscursehasbeen
suchapervasivephenomenoninthelaboratorythatmostoftheseini-
tialexperimentshavefocusedonitsrobustnessandthefeaturesofthe
environment that might attenuate its effects. Additional interest has
focused on public policy issuesthe effects of public information
regarding the value of the auctioned item and the effects of different
auctioninstitutionsonsellersrevenue.
This survey begins with a brief analysis of the first experimental
demonstrationofthewinnerscurse(BazermanandSamuelson1983).
Thisisfollowedbysummariesofexperimentsinvestigatingbiddingin
common value auctions using an experimental design that I helped
develop. These experiments also demonstrate the existence of a win-
ners curse even when allowing for extensive feedback and learning
from past auction outcomes. They also address policy issues such as
theeffectsofpublicinformationanddifferentauctioninstitutions(e.g.,
first-pricesealed-bidauctionsversusopenoutcryEnglishauctions)on
sellersrevenue.Iconcludewithabriefsummaryoftheempiricalfind-
CommonValueAuctionsandtheWinnersCurse 69
ings from the experimental literature and the role experiments have
played in the successful sale of government airwave rights (the spec-
trumauctions).Inreviewingtheexperimentalworkoncommonvalue
auctions,Ihopetoshowhowexperimentsproceedbysuccessivelynar-
rowing down plausible explanations for the question at hand. This is
done through a series of experiments rather than any single critical
experiment;itisbasedonsortingoutbetweencompetingexplanations,
andonfollowinguponthelogicalimplicationsofbehaviorobserved
inearlierexperiments.
ANINITIALEXPERIMENTDEMONSTRATINGTHE
WINNERSCURSE
Bazerman and Samuelson (1983) conducted the first experiment
demonstratingawinnerscurse.UsingM.B.A.studentsatBostonUni-
versity, the experiment was conducted in class, with students partici-
patinginfourfirst-pricesealed-bidauctions.Biddersformedtheirown
estimates of the value of each of four commoditiesjars containing
800pennies,160nickels,200largepaperclipseachworthfourcents,
and400smallpaperclipseachworth$0.02.Unknowntosubjects,each
jarhadavalueof$8.00.(Subjectsbidonthevalueofthecommodity,
not the commodity itself.) In addition to their bids, subjects provided
their best estimate of the value of the commodities and a 90 percent
confidenceboundaroundtheseestimates.Aprizeof$2.00wasgiven
fortheclosestestimatetothetruevalueineachauction.Thenumberof
bidders varied between 4 and 26. Their analysis focused on bidder
uncertaintyaboutthevalueofthecommodityandthesizeofthebid-
dingpopulation.
Theaveragevalueestimateacrossallfourcommoditieswas$5.13
($2.87belowthetruevalue).Astheauthorsnote,thisunderestimation
should reduce the likelihood and magnitude of the winners curse. In
contrast to the mean estimate, the average winning bid was $10.01,
resultinginanaveragelosstothewinnerof$2.01.
5
Theaveragewin-
ningbidgeneratedlossesinoverhalfofalltheauctions.
Estimatedbidfunctions,usingindividualbidsastheunitofobser-
vation, showed that bids were positively, and significantly, related to
70 Kagel
individualestimatessothatbiddersindeedfacedanadverseselection
problem,onlywinningwhentheyhadhigherestimatesofthevalueof
theitem.Bidswereinverselyrelatedtotheuncertaintyassociatedwith
individual estimates, but this effect was small (other things equal, a
$1.00 increase in the 90 percent confidence interval reduced bids by
$0.03). Numbers of bidders had no significant effect on individual
bids.
In contrast, regressions employing the average winning bid
showedthatthesebidswerepositively,andsignificantly,relatedtothe
winningbiddersestimateofuncertaintyandtothenumberofbidders
in the auction. This suggests that winning bidders are substantially
moreaggressivethanotherbidders.Indeed,BazermanandSamuelson
note that average winning bids were sensitive to a handful of grossly
inflatedbids.
Theresultsofthisexperimentshowthatthewinnerscurseiseasy
to observe. However, many economists would object to the fact that
subjects had no prior experience with the problem and no feedback
regardingtheoutcomesoftheirdecisionsbetweenauctions,sothatthe
resultscouldbeattributedtothemistakesoftotallyinexperiencedbid-
ders. The robustness of these results is even more suspect given their
sensitivitytoahandfulofgrosslyinflatedbids,whichonemightsup-
pose would be eliminated as a result of bankruptcies or learning in
responsetolossesincurredinearlierauctions.Commonvalueauction
experimentsconductedbyKagelandLevin(1986)andtheirassociates
exploretheseissues,alongwithanumberofpublicpolicyimplications
ofthetheory.
SEALED-BIDAUCTIONS
Kagel and Levin and their associates conducted experiments in
whichbiddersparticipatedinaseriesofauctionswithfeedbackregard-
ing outcomes. Bidders were given starting cash balances from which
lossesweresubtractedandprofitswereadded.Bidderswhosecashbal-
ancesbecamenegativeweredeclaredbankruptandwerenolongerper-
mittedtobid.UnliketheBazermanandSamuelsonexperiment,Kagel
and Levin controlled the uncertainty associated with the value of the
CommonValueAuctionsandtheWinnersCurse 71
auctioneditemratherthansimplymeasuringit.Theydidthisbycon-
ductingauctionsinwhichthecommonvalue,x ,waschosenrandomly
o
each period from a known uniform distribution with upper and lower
bounds[ , x x].Inauctionswithasymmetricinformationstructure,each
bidder is provided with a private information signal, x, drawn from a
uniformdistributionon[x ,x
o
+],whereisknown.Infirst-price
o
sealed-bid auctions, bids are ranked from highest to lowest with the
highbidderpayingtheamountbidandearningprofitsequaltox b
1
,
o
whereb
1
isthehighestbid.Losingbiddersneithergainnorlosemoney.
Inthisdesign,thestrategyofbidding,max[x ,x],isarisk-free
strategy that fully protects a bidder from negative earnings since it is
thelowerboundestimateofx
o
.Thislowerboundestimateforx was
o
computedforsubjectsalongwithanupperboundestimateofx ,(min
o
[x +,x]
).Bidderswereprovidedwithillustrativedistributionsofsig-
nal values relative to x
o
, and several dry runs were conducted before
playingforcash.Followingeachauctionperiod,bidderswereprovided
withthecompletesetofbids,listedfromhighesttolowest,alongwith
thecorrespondingsignalvalues,thevalueofx ,andtheearningsofthe
o
highbidder.
Surviving bidders were paid their end-of-experiment balances in
cash.Toholdthenumberofbiddersfixedwhilecontrollingforbank-
ruptcies, m > n subjects were often recruited, with only n bidding at
anygiventime(whobidsineachperiodwasdeterminedrandomlyor
byafixedrotationrule).Asbankruptciesoccur,m shrinksbut(hope-
fully)remainsgreaterthanorequaltothetargetvaluen.
SomeInitialExperimentalResults:InexperiencedBidders
Auctions with inexperienced bidders show a pervasive winners
curse that results in numerous bankruptcies. Table 4.1 provides illus-
trative data on this point. For the first nine auctions, profits averaged
$2.57, compared to the risk neutral Nash equilibrium (RNNE) pre-
diction of $1.90, with only 17 percent of all auctions having positive
profits.Note:thisisafterbiddershadparticipatedintwoorthreedry
runs,withfeedbackofsignalvalues,x ,andbidsfollowingeachauc-
o
tion, so that the results cannot be attributed to a total lack of experi-
ence. The negative profits are not a simple matter of bad luck either,
or a handful of grossly inflated bids, as 59 percent of all bids and 82
Table4.1 ProfitsandBiddinginFirstNineAuctionsforInexperiencedBidders
7
2

Percentof Average Percentof

auctions Average predicted Percentofall auctionswon Percentageof Percentageof
withpositive actualprofits profitsunder bids byhighsignal highbids subjectsgoing
Experiment profits (t-statistic) RNNE(S
m
) b>E[x
o
/X=x
ln
] holder b
1
>E[x
o
/X=x
ln
] bankrupt
a
1 0.0 4.83 0.72 63.4 55.6 100 50.0
(3.62)***
(0.21)
2 33.3 2.19 2.18 51.9 33.3 88.9 16.7
(1.66)
(1.02)
3 11.1 6.57 1.12 74.6 44.4 88.9 62.5
(2.80)**
(1.19)
4 11.1 2.26 0.85 41.8 55.6 55.6 16.7
(3.04)***
(0.43)
5 33.3 0.84 3.60 48.1 44.4 88.9 50.0
(1.00)
(1.29)
6 22.2 2.65 2.55 67.3 66.7 100 33.3
(1.53)
(1.17)
7 11.1 2.04 0.57 58.5 88.9 66.7 50.0
(2.75)**
(0.25)
8 11.1 1.40 1.59 51.9 55.6 55.6 16.7
(2.43)**
(0.34)
9 44.4 0.32 2.37 35.2 88.6 66.7 16.7
(0.30)
(0.76)
7
3

10 0.0 2.78 3.53 77.2 66.7 100 20.0

(3.65)***
(0.74)
11 11.1 3.05 1.82 81.5 55.6 88.9 37.5
(3.53)***
(0.29)
Average 17.2 2.57 1.90 59.4 59.6 81.8 41.1
NOTE:S
m
=standarderrorormean.**significantatthe5%level,two-tailedtest;***significantatthe1%level,two-tailedtest.
a
Forallauctions.
SOURCE:FromKageletal.(1989).ReprintedwithpermissionfromWesternEconomicAssociationInternational.
74 Kagel
percentofthehighbidswereaboveE[x |X=x
1n
];theexpectedvalue
o
ofx
o
conditionalonhavingthehighestsignalx.Further,40percentof
all subjects starting these auctions went bankrupt. In short, the win-
ners curse is a genuinely pervasive problem for inexperienced bid-
ders. It is remarkably robust being reported under a variety of
treatment conditions (Kagel et al. 1989; Lind and Plott 1991; Goeree
and Offerman 2000) and for different subject populations, including
professional bidders from the commercial construction industry
(Dyer,Kagel,andLevin1989).
AuctionswithModeratelyExperiencedBiddersandtheEffectsof
PublicInformationonSellersRevenue
Kagel and Levin (1986) report auctions for moderately experi-
enced bidders (those who had participated in at least one prior first-
pricecommonvalueauctionexperiment).Treatmentvariablesofinter-
estwerethenumberofrivalbiddersandtheeffectsofpublicinforma-
tion about x on revenue. Table 4.2 reports some of their results. For
o
smallgroups(auctionswith34bidders),thegeneralpatternwasone
of positive profits averaging $4.32 per auction, which is significantly
greaterthanzerobutstillwellbelowtheRNNEpredictionof$7.48per
auction. In contrast, for these same bidders bidding in larger groups
(auctionswith67bidders),profitsaveraged$0.54perauctioncom-
pared to the RNNE prediction of $4.82. Thus, the profit picture had
improved substantially compared to the inexperienced bidders dis-
cussedintheprevioussection.
However,comparinglargeandsmallgroupauctions,actualprofit
decreasedsubstantiallymorethanprofitopportunitiesasmeasuredby
the RNNE criteria. This implies that subjects were bidding more
aggressively, rather than less aggressively, as the number of rivals
increased, contrary to the RNNE prediction. This is confirmed in
regressions using individual subject bids as the dependent variable.
Higher individual bids in response to increased numbers of rivals is
oftenconsideredtobethehallmarkcharacteristicofawinnerscurse.
Thus, although bidders had adjusted reasonably well to the adverse
selection problem in auctions with 34 bidders, in auctions with 67
bidders,withitsheightenedadverseselectioneffect,thewinnerscurse
reemerged as subjects confounded the heightened adverse selection
CommonValueAuctionsandtheWinnersCurse 75
effectbybiddingmoreaggressivelywithmorebidders.Thisresultalso
suggests that the underlying learning processes are context-specific
ratherthaninvolvingsomesortoftheoryabsorptionthatreadilygen-
eralizestonewenvironments.
6
Public information was provided to bidders in the form of
announcingthelowestsignalvalue,x
L
.FortheRNNE,publicinforma-
tion about the value of the item raises expected revenue. The mecha-
nism underlying this outcome works as follows: All bidders evaluate
theadditionalpublicinformationassumingthattheirsignalisthehigh-
est since, in equilibrium, they only win in this case. Evaluating addi-
tional information from this perspective, together with affiliation,
inducesallbiddersotherthanthehighestsignalholderto,onaverage,
revise their bids upward after an announcement of unbiased public
information.Thisupwardrevisionresultsfromtwofactors:
1)Bidderswithoutthehighestsignaltreatthepublicinformationas
goodnews.Thesebiddersformulatedtheirbidsontheassump-
tion that they held the highest private information signal and
wouldwintheauction.Assuch,withaffiliation,thepublicinfor-
mationtellsthemthat,onaverage,theexpectedvalueoftheitem
is higher than they had anticipated (i.e., the private information
signaltheyareholdingissomewhatlowerthanexpected,condi-
tionalonwinning,forthisparticularauction),whichleadsthem
toincreasetheirbids.
2) Bidders respond to this anticipated increase in bids from lower
signalholdersbyraisingtheirbids.Thebidderwiththehighest
signal is not, on average, subject to this first force. Thus, she
doesnot,onaverage,reviseherestimateofthetruevalue.Never-
theless,sheraisesherbidinanticipationofotherbiddersraising
theirbids;thedominoeffectofbidderswithlowersignalsrais-
ingtheirbids.
These strategic considerations hold for a wide variety of public
information signals (Milgrom and Weber 1982). There are, however,
several methodological advantages to using x
L
. First, the RNNE bid
functioncanbereadilysolvedforx
L
,providedlowsignalholdersare
restricted to bidding x
L
, so that the experimenter continues to have a
benchmarkmodeloffullyrationalbehavioragainstwhichtocompare
7
6

Table4.2 ProfitsandBiddingbyExperimentandNumberofActiveBidders:PrivateInformationConditions
(profitsmeasuredindollars)
Numberof Averageprofit Percentofauctions
Auctionseries
(no.ofperiods)
active
bidders
Averageactualprofit
(t-statistic)
a
underRNNE
(standarderrorofmean)
wonbyhigh
signalholder
Percentofhighbids
b
1
>E[x
o
/X=x
1n
]
6 3.73 9.51
(31) 34 (2.70)** (1.70) 67.7 22.6
2 4.61 4.99
(18) 4 (4.35)*** (1.03) 88.9 0.0
3small 7.53 6.51
(14) 4 (2.07) (2.65) 78.6 14.3
7small 5.83 8.56
(19) 4 (3.35)*** (2.07) 63.2 10.5
8small 1.70 6.38
(23) 4 (1.56) (1.21) 82.6 39.1
1 2.89 5.19
(18) 5 (3.14)*** (0.86) 72.2 27.8
3large 2.92 3.64
(11) 57 (1.49) (0.62) 81.8 63.6
7large 1.89 4.70
(18) 6 (1.67) (1.03) 72.2 22.2
4 0.23 4.78
(25) 67 (0.15) (0.92) 69.2 48.0
7
7

5 0.41 5.25
(26) 7 (0.44) (1.03) 42.3 65.4
8large 2.74 5.03
(14) 7 (2.04) (1.40) 78.6 71.4
Smallmarket 4.32 7.48
average 34 (5.55)*** (0.77) 75.2 19.0
Largemarket 0.54 4.82
average 67 (0.87) (0.50) 62.9 53.9
NOTE:**Significantatthe5%level,two-tailedt-test;***significantatthe1%level,two-tailedt-test.
a
Testsnullhypothesisthatmeanisdifferentfrom0.0.
SOURCE:FromKagelandLevin(1986).ReprintedwithpermissionoftheAmericanEconomicAssociation.
78 Kagel
actualbidding.Second,x
L
providesasubstantialdoseofpublicinfor-
mationaboutx
o
(itcutsexpectedprofitinhalf),whilestillmaintaining
aninterestingauction.Assuchitshouldhaveasubstantialimpacton
prices,regardlessofanyinherentnoiseinbehavior.Finally,theexperi-
mentercanalwaysimplementfiner,moresubtleprobesofpublicinfor-
mationafterseeingwhathappenswithsuchastrongtreatmenteffect.
7
KagelandLevin(1986)foundthatinauctionswithsmallnumbers
ofbidders(34),publicinformationresultedinstatisticallysignificant
increases in revenue that averaged 38 percent of the RNNE models
prediction.However,inauctionswithlargernumbersofbidders(67),
publicinformationreducedaveragesellersrevenueby$1.79perauc-
tion, compared to the RNNE models prediction of an increase of
$1.78.KagelandLevinattributethisreductioninrevenuetothepres-
enceofarelativelystrongwinnerscurseinauctionswithlargenum-
bersofbidders.Ifbidderssufferfromawinnerscurse,thehighbidder
consistently overestimates the items value, so that announcing x
L
is
likelytoresultinadownwardrevisionofthemostoptimisticbidders
estimate. Thus, out of equilibrium, public information introduces a
potentiallypowerfuloffsettotheforcespromotingincreasedbidsdis-
cussedearlier,andwillresultinreducedrevenueifthewinnerscurse
isstrongenough.Thishypothesisisconfirmedusingdetaileddatafrom
auctions with 67 bidders, which shows that the RNNE models pre-
diction of an increase in sellers revenue is critically dependent on
whetherornottherewasawinnerscurseinthecorrespondingprivate
informationmarket.
IstheWinnersCurseaLaboratoryArtifact?LimitedLiability
forLosses
Results of experiments are often subject to alternative explana-
tions. These alternative explanations typically provide the motivation
for subsequent experiments that further refine our understanding of
behavior.Thissectiondealswithonesuchalternativeexplanationand
theresponsestoit.
IntheKagelandLevin(1986)design,subjectsenjoyedlimitedlia-
bility as they could not lose more than their starting cash balances.
Hansen and Lott (1991) argued that the overly aggressive bidding
reportedinKagelandLevinmayhavebeenarationalresponsetothis
CommonValueAuctionsandtheWinnersCurse 79
limitedliabilityratherthanaresultofthewinnerscurse.Inaone-shot
auction,ifabidderscashbalanceiszero,sothattheyarenotliablefor
any losses, it indeed pays to overbid relative to the Nash equilibrium
biddingstrategy.Withdownsidelosseseliminated,theonlyconstraint
on more aggressive bidding is the opportunity cost of bidding more
thanisnecessarytowintheitem.Inexchange,higherbidsincreasethe
probability of winning the item and making positive profits. The net
effect,inthecaseofzeroorsmallcashbalances,isanincentivetobid
more than the Nash equilibrium prediction. Hansen and Lotts argu-
mentprovidesapossiblealternativeexplanationtotheoverlyaggres-
sive bidding reported in Kagel and Levin (1986) and in Kagel et al.
(1989).
Responses to the limited-liability argument have been twofold.
First,KagelandLevin(1991)reevaluatedtheirdatainlightofHansen
and Lotts arguments, demonstrating that for almost all bidders cash
balanceswerealwayslargeenoughsothatitneverpaidtodeviatefrom
the Nash equilibrium bidding strategy in a one-shot auction. Second,
subsequentempiricalworkhasdemonstratedawinnerscurseinexper-
imental designs where limited liability for losses could not logically
account for overbidding. This provides experimental verification that
limited-liabilityforcesdonotaccountfortheoverlyaggressivebidding
reported.
Kagel and Levins design protects against limited-liability prob-
lems since bidding x insures against all losses and bidders have
theirownpersonalestimateofthemaximumpossiblevalueoftheitem
(min[x+,
x
]).Thelatterimpliesthatitisneverrational,limitedlia-
bilityornot,tobidabovethismaximumpossiblevalueinafirst-price
auction.Further,cashbalancesonlyhavetobeafractionofthemaxi-
mumpossiblelossforthelimited-liabilityargumenttoloseitsforcein
afirst-priceauction.Forexample,KagelandLevin(1991)reportsimu-
lationsforauctionswith4or7bidders,with=$30andcashbalances
of$4.50(which48outofthe50biddersalwayshad),forwhichunilat-
eral deviations from the RNNE bid function were not profitable even
when fully accounting for bidders limited liability. Further, limited-
liability arguments imply more aggressive bidding in auctions with
fewerratherthanlargernumbersofbidders,justtheoppositeofwhat
the data show.
8
As such, overbidding in the Kagel and Levin experi-
80 Kagel
mentmustbeexplainedonsomeothergrounds,suchasthejudgmental
errorunderlyingthewinnerscurse.
Empiricalworkonthisissuehasproceededonseveralfronts.Lind
andPlott(1991)replicatedKagelandLevinsresultsinauctionswhere
bankruptcyproblemswerealmostcompletelyeliminated.Oneexperi-
mental treatment involved conducting private value auctions where
subjects were sure to make money simultaneously with the common
value auctions, thereby guaranteeing a steady cash inflow against
whichtochargeanylossesincurredinthecommonvalueauctions.A
second treatment involved sellers markets in which bidders tendered
offers to sell an item of unknown value. Each bidder was given one
item with the option to keep it and collect its value or to sell it. Lind
andPlottsresultslargelyconfirmthose reported by Kagel and Levin
andtheirassociates.
Cox, Dinkin, and Smith (1998) conducted auctions using Kagel
andLevinsdesigninwhich,underonetreatment,theyreinitializebid-
derscashbalancesineachauctionperiod,withbalanceslargeenough
thatsubjectscouldnotgobankruptevenifbidingwellabovetheirsig-
nal values. In contrast to this unlimited-liability treatment, their other
treatments employed procedures where cash balances fluctuated, bid-
derscouldgobankrupt,andinsometreatments,bidderswithnegative
cash balances were permitted to continue to bid. Using data for all
treatmentsandalllevelsofbidderexperience,Cox,Dinkin,andSmith
findnosignificantdifferencesinindividualbidpatternsintheunlim-
ited-liability treatment, contrary to Hansen and Lotts argument. Fur-
ther, restricting their analysis to experiments with experienced
subjects,anddroppingdatafromanentireexperimentifevenonesub-
ject adopted a pattern of high bids when having a negative cash bal-
ance,Cox,Dinkin,andSmithfindthattheunlimited-liabilitytreatment
significantly increased individual bids, the exact opposite of Hansen
and Lotts hypothesis. This unexpected outcome is, however, consis-
tent with Kagel and Levins (1991) argument that in a multi-auction
setting, where cash balances carry over from one auction to the next,
there is a potentially powerful offset to any limited-liability forces
present in a one-shot auction: Overly aggressive bidding due to low
cash balances may be offset by the risk that such bids will result in
bankruptcy,therebypreventingparticipationinlaterauctionswiththeir
positiveexpectedprofitopportunities.Unfortunately,itisalsoconsis-
CommonValueAuctionsandtheWinnersCurse 81
tentwiththeartifactualexplanationthatbecausesubjectswerepaidoff
inonlyafewoftheunlimited-liabilityauctions(inordertokeepscosts
toamanageablelevel),subjectstreatedtheseauctionsdifferentlythan
thoseinwhichtheywerepaidasaresultofeachoutcome.
9
Summary
Even after allowing for some learning as a result of feedback
regarding past auction outcomes, a strong winners curse is reported
forinexperiencedbiddersinsealed-bidcommonvalueauctions.High
bidders earn negative average profits and consistently bid above the
expectedvalueoftheitemconditionalonhavingthehighsignalvalue.
Further,thisisnottheresultofahandfulofoverlyaggressivebidders
butappliesratherbroadlyacrossthesamplepopulation.Similarresults
are reported in low-bid wins, supply auctions with both student sub-
jectsandprofessionalbiddersdrawnfromthecommercialconstruction
industry (Dyer, Kagel, and Levin 1989). Arguments that these results
can be accounted for on the basis of limited liability for losses have
been shown to be incorrect. Further, a clever experiment by Holt and
Sherman(1994)(alsoseeAveryandKagel1997)isabletoruleoutthe
ideathatthewinnerscurseisaresultofanaddedthrill,orextrautility,
fromwinning.
Notethattheoverbiddingassociatedwiththewinnerscurseisnot
simply a matter of miscalibrated bidders, it is associated with funda-
mentalbreakdownsofthecomparativestaticpredictionsoftherational
biddingmodel:Withawinnerscursepublicinformationreducesreve-
nue, contrary to the theorys prediction, as the additional information
helps high bidders to correct for overly optimistic estimates of the
items worth. In second-price sealed-bid auctions, increased numbers
of bidders produce no change in bidding, contrary to the robust Nash
equilibrium prediction that bids will decrease (Kagel, Levin, and
Harstad1995).
Wearestillleftwiththepuzzle,firstexpressedbyLindandPlott,
thatalthoughmanyexperimentsreportaclearwinnerscurse(negative
profits), comparing between the symmetric RNNE and totally naive
biddingmodelsofferedintheliterature(allplayerstreattheirsignalsas
iftheyareprivatevaluesandgoontobidasifinaprivatevalueauc-
tion; Kagel and Levin 1986), bidding is closer to the RNNE. One
82 Kagel
promisingexplanationforthisphenomenonappearstobethatbidders
are cursed to different degrees. That is, agents may make partial, but
incomplete, adjustments for the adverse selection effect associated
withcommonvalueauctions,withtheperfectlyrationalandperfectly
naive bidding models being polar cases. Depending on the extent to
whichplayersarecursed,theymaysufferlosses,butbiddingcan,in
fact, still be closer to the symmetric RNNE bidding model than the
totallynaivebiddingmodel.(SeeEysterandRabin2000foraformal
modelofthissort.)
ENGLISHAUCTIONSANDFIRST-PRICEAUCTIONSWITH
INSIDERINFORMATION
MycolleaguesandIhavealsostudiedEnglishauctionsandfirst-
priceauctionswithinsiderinformation(onebidderknowsthevalueof
theitemwithcertaintyandthisiscommonknowledge).Theseexperi-
mentswereinitiallymotivatedbyeffortstoidentifyinstitutionalstruc-
tures that would eliminate, or mitigate, the winners curse for
inexperiencedbidders.Theexperimentsalsoinvestigatethecompara-
tive static properties of Nash equilibrium bidding models for very
experiencedbidders.Inbothinstitutionalsettings,thewinnerscurseis
alive and well for inexperienced bidders, although it is clearly less
severeinEnglishthaninfirst-priceauctions.
Incontrast,comparativestaticpredictionsoftheNashequilibrium
bidding model are largely satisfied for more experienced bidders.
However, in the case of English auctions, the information processing
mechanism that the Nash bidding model specifies is not satisfied.
Rather,biddersfollowarelativelysimpleruleofthumbthatresultsin
almostidenticalpricesandallocationsastheNashmodelspredictions
forthedistributionofsignalvaluesemployedintheexperiment.Inthe
insider information auctions, less-informed bidders (outsiders) have
some proprietary information (i.e., the insider knows the value of the
item with certainty but does not know the outsiders signals). This
results in marked differences in predicted outcomes compared to the
standard insider information model in which the insider has a double
informationaladvantagesheknowsthevalueoftheitemandthesig-
CommonValueAuctionsandtheWinnersCurse 83
nalstheoutsidershave(Wilson1967;Weverbergh1979;Englebrecht-
Wiggans, Milgrom, and Weber 1983; Hendricks, Porter, and Wilson
1994).Mostnotably,inourmodeltheexistenceofaninsidergenerates
higheraveragerevenuethaninauctionswithasymmetricinformation
structure,apredictionthatissatisfiedinthedataforexperiencedbid-
ders. In contrast, in the double informational advantage model the
existenceofaninsiderreducesaveragerevenue.
EnglishAuctions
Levin, Kagel, and Richard (1996) implement an irrevocable exit,
ascending-price(English)auction.Pricesstartatx,thelowestpossible
valueforx
o
,andincreasecontinuously.Biddersarecountedasactively
biddinguntiltheydropoutoftheauctionandarenotpermittedtoreen-
teroncetheyhavedroppedout.Thelastbidderearnsaprofitequalto
x
o
lessthepriceatwhichthelastbidderdroppedout.Biddersobserve
thepricesatwhichtheirrivalsdropoutofthebidding.Auctionsofthis
sorthavebeenruninJapan(MilgromandWeber1982;Cassady1967).
Theirrevocableexitprocedure,inconjunctionwiththepublicposting
of drop-out prices, insures that in equilibrium bidders can infer their
rivalssignalvaluesfromtheirdrop-outprices.
In a symmetric RNNE, the bidder with the low signal value (x
L
)
dropsoutoftheauctiononcethepricereacheshissignalvalue.
10
The
priceatwhichthelowbidderdropsoutoftheauctionrevealshissignal
value to the remaining bidders. Thus, the public information, x
L
, that
wasprovidedbytheexperimentersinKagelandLevin(1986)ispro-
videdendogenouslyhere(atleastintheory)bythefirstdrop-outprice.
Giventheuniformdistributionofsignalvaluesaroundx
o
,inasymmet-
ricequilibrium,foranyremainingbidderj,(x
L
+x )/2providesasuffi-
j
cient statistic for x conditional on x
j
being the highest signal, so that
o
drop-out prices other than x
L
contain no additional information and
shouldbeignored.Thissufficientstatisticistheequilibriumdrop-out
priceforj(d
j
)inthesymmetricRNNE
d
j
=(x
L
+x
j
)/2.
Thisrepresentsthemaximumwillingnesstopay,conditionalonallthe
informationrevealedbyearlierdrop-outpricesandconditionalonwin-
84 Kagel
ning. As in first-price auctions with x
L
publicly announced, expected
profitintheEnglishauctionissharplyreduced(byaboutahalf)com-
paredtofirst-priceauctionswithstrictlyprivateinformation(aslongas
n>2).Assuch,inequilibrium,theEnglishauctionispredictedtosig-
nificantlyraiseaveragesellersrevenuecomparedtofirst-pricesealed-
bidauctions.
The key difference between the English auction and a first-price
sealed-bid auction with x
L
publicly announced is that in the English
auction information dissemination is endogenous, rather than exoge-
nous.Highersignalholdersmustbeabletorecognizeandprocessthe
relevantinformation,andlowsignalholdersmustrecognizethefutility
ofremainingactiveoncethepriceexceedstheirsignalvalue.Assuch,
we would expect the information dissemination process to be noisier
than with x
L
publicly announced. Nevertheless, if bidders are able to
correctlyrecognizeandincorporatethepublicinformationinherentin
other bidders drop-out prices, we would predict two results: 1) for
inexperienced bidders, contrary to the Nash equilibrium bidding
modelsprediction,Englishauctionswillreduceaveragesellersreve-
nues compared to first-price sealed-bid auctions, as losses will be
sharplyreduced,orevenbeeliminated,onaverage,intheEnglishauc-
tions, and 2) for more experienced bidders, where negative average
profits have been largely eliminated in the sealed-bid auctions, the
Englishauctionswillraiseaveragerevenue,asthetheorypredicts.The
secondpredictionisthestandard,equilibriumprediction.Thefirstpre-
diction follows directly from our experience with first-price auctions
withx
L
publiclyannounced.
Table4.3showsaveragesofpredictedandactualchangesinreve-
nuebetweenEnglishandfirst-priceauctionsforinexperiencedbidders,
aswellasaveragesofpredictedandactualprofit,withtheresultsclas-
sifiedbynumbersofbiddersand.
11
Averagerevenueispredictedtobe
higherintheEnglishauctionsinallcases,forthesetofsignalvalues
actuallydrawn,withsignificantlyhigheraveragerevenuepredictedfor
all values of with n = 4 and for = $12 with n = 7.
12
However, for
theseinexperiencedbidders,withtheexceptionofn=4and=$24,
actualrevenueislowerintheEnglishauctionsinallcases,withsignif-
icantlyloweraveragerevenueforn=4and7with =$6,andwiththe
reduction in revenue barely missing statistical significance (at the 10
percent level) with n = 7 and = $12. Further, the revenue increase
8
5

Table4.3 InexperiencedBidders:ActualversusTheoreticalRevenueChangesandProfitLevels
a
inEnglishversus
First-PriceAuctions
n=4 n=7
Averagechangeinrevenue:
Averageprofit
Averagechangeinrevenue:
Averageprofit
Englishlessfirst-price First-price English Englishlessfirst-price First-price English
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
Actual Theo- Differ- Actual Theo- Actual Theo- Actual Theo- Differ- Actual Theo- Actual Theo-
retical ence retical retical retical ence retical retical
$6 1.54** 1.54*** 3.08*** 2.13 2.76 0.58 1.23 1.98** 0.10 2.08** 3.85 0.99 l.87 0.89
(0.72) (0.49) (0.71) (0.52) (0.38) (0.50) (0.30) (0.87) (0.34) (0.78) (0.71) (0.19) (0.51) (0.29)
[29] [28] [18] [18]
$12 0.54 2.76*** 3.30*** 1.32 5.01 0.78 2.25 1.95* 1.08 3.03*** 3.75 2.76 1.80 1.68
(1.25) (0.92) (0.84) (0.79) (0.60) (0.95) (0.69) (1.19) (0.65) (0.92) (0.89) (0.53) (0.77) (0.40)
[41] [45] [30] [43]
$24 1.09 8.10*** 7.01** 1.20 9.83 0.11 1.73 ND ND ND ND ND
(3.29) (2.32) (3.05) (1.93) (1.25) (2.64) (2.14)
[25] [13]
NOTE:Standarderrorsareinparentheses.Bracketedtermsarethenumberofauctionperiods.ND=nodata.*Thenullhypothesisthat
thevalueisgreaterthanorequaltozerocanberejectedatthe10%significancelevel;**thenullhypothesisthatthevalueisgreaterthan
orequaltozerocanberejectedatthe5%level;***thenullhypothesisthatthevalueisgreaterthanorequaltozerocanberejectedat
the1%level.
a
Allvaluesreportedindollars.
SOURCE:FromLevin,Kagel,andRichard(1996).ReprintedwithpermissionoftheAmericanEconomicAssociation.
86 Kagel
withn=4and=$24isstatisticallyinsignificant,andiswellbelow
thepredictedincrease.
These perverse revenue effects in terms of Nash equilibrium bid-
dingtheoryareassociatedwithnegativeaverageprofitinboththefirst-
price and English auctions. The negative average profits reported in
Table4.3indicatethatinexperiencedbidderssufferedfromawinners
curseinbothauctioninstitutions,butthatthecursewasrelativelystron-
ger in the first-price auctions. These results serve to generalize those
reportedforfirst-pricesealed-bidauctionswithx
L
publiclyannounced:
Givenarelativelystrongwinnerscurseinsealed-bidauctions,public
informationreducesratherthanraisessellersaveragerevenue.Thetwo
major differences between the present results and the first-price auc-
tionswithx
L
publiclyannouncedare:1)here,publicinformationisgen-
erated endogenously in the form of drop-out prices, and 2) average
profits in the English auctions were negative, but with the exogenous
releaseofpublicinformationinthefirst-priceauctionstheywereposi-
tive. This last result suggests that information dissemination in the
Englishauctionisnoisierthanwithx
L
publiclyannounced.
13
For more experienced bidders, English auctions are capable of
raisingaveragesellersrevenue,asthedatainTable4.4demonstrate.
With n = 4, actual revenue is higher in the English auctions for both
valuesof,withastatisticallysignificantincreasefor=$18.How-
ever,forn=7,thereisessentiallynodifferenceinrevenuebetweenthe
first-priceandEnglishauctions.Thesignificantincreaseinrevenuein
Englishauctionswithn=4and=$18isassociatedwithelimination
oftheworsteffectsofthewinnerscurseinthefirstpriceauctions,as
biddersearnedasubstantialshare(morethan50percent)ofpredicted
profit.Theimportanceofeliminatingthewinnerscurseforthereve-
nue-raisingpredictionofthetheorytoholdisreinforcedbytheabsence
ofanyrevenueincreasewithn=7,inconjunctionwiththerelatively
lowshareofexpectedprofit(21percent)thatwasearnedinthesefirst-
priceauctions.
Levin, Kagel, and Richard (1996) develop an econometric model
to characterize how bidders process information in the English auc-
tions. As noted, the Nash bidding model predicts that bidders with
highersignalvalueswillaveragetheirownsignalvaluewiththefirst
drop-out price observed, ignoring all intermediate drop-out prices.
WhatLevin,Kagel,andRichardfound,however,isthatbiddersplaced
8
7

Table4.4 Super-ExperiencedBidders:ActualversusTheoreticalRevenueChangesandProfitLevels
a
inEnglish
versusFirst-PriceAuctions
n=4 n=7
Averagechangeinrevenue:
Averageprofit
Averagechangeinrevenue:
Averageprofit)
Englishlessfirst-price First-price English Englishlessfirst-price First-price English
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
Actual Theo- Differ- Actual Theo- Actual Theo- Actual Theo- Differ- Actual Theo- Actual Theo-
retical ence retical retical retical ence retical retical
$18 2.21** 3.96*** 1.75** 3.37 6.77 1.16 2.82 0.25 2.85*** 3.10*** 0.76 3.86 1.01 1.01
(0.95) (0.73) (0.68) (0.50) (0.48) (0.88) (0.53) (0.86) (0.61) (0.59) (0.65) (0.50) (0.56) (0.37)
[163] [107] [75] [96]
1.20 2.98 1.78 8.45 11.27 7.25 8.29
$30 (3.10) (2.30) (2.19) (1.28) (1.34) (2.76) (l.93) ND ND ND
[31] [33]
NOTE:Standarderrorsareinparentheses.Bracketedtermsarethenumberofauctionperiods.ND=nodata.**Thenullhypothesisthat
thevalueisgreaterthanorequaltozerocanberejectedatthe5%significancelevel;***thenullhypothesisthatthevalueisgreaterthan
orequaltozerocanberejectedatthe1%significancelevel.
a
Allvaluesreportedindollars.
SOURCE:FromLevin,Kagel,andRichard(1996).ReprintedwithpermissionoftheAmericanEconomicAssociation.
88 Kagel
weight on their own signal value and the immediate past drop-out
price,ostensiblyignoringx
L
andanyearlierdrop-outprices.Further,as
morebiddersdroppedout,subjectsplacedlessandlessweightontheir
ownsignalvalue,andmoreweightonthelastdrop-outprice.Thispat-
tern,althoughinconsistentwiththeNashmodel,isconsistentwithbid-
ders acting as if they were averaging their own signal value with the
signal values underlying the drop-out prices of all earlier bidders.
Levin,Kagel,andRichardattributetheadoptionofthissignalaverag-
ingruleinfavoroftheNashruletothefactthatitiseasyandquitenat-
ural to use, and that it yields results similar to the Nash rule without
requiringthatbiddersexplicitlyrecognizetheadverseselectioneffect
ofwinningtheauctionand/orknowinganythingaboutsufficientstatis-
tics.
AuctionswithInsiderInformation
Kagel and Levin (1999) investigate bidding in first-price sealed-
bid auctions with an asymmetric information structure (AIS). The
asymmetry is introduced by choosing one bidder at random in each
auctionperiodtheinsider(I)toreceiveaprivateinformationsignal
xequaltox
o
andbeingtoldthatx=x
o
.Eachoftheotherbidders,the
outsiders(Os),receiveaprivateinformationsignalfromauniformdis-
tributionon[x ,x
o
+],asintheauctionswithasymmetricinfor-
o
mation structure (SIS). The insider does not know the realizations of
Osprivateinformationsignals.OsknowthattheyareOs,thatthereis
asingleIwhoknowsx
o
,andthewaythatallotherOsgottheirprivate
signals.
Note that this information structure differs substantially from the
standardinsiderinformationmodelemployedintheeconomicsliter-
ature in which the insider has a double informational advantageI
knows x
o
and Os only have access to public information about x
o
(Engelbrecht-Wiggans,Milgrom,andWeber1983;HendricksandPor-
ter1988).Incontrast,inourdesignOshavesomeproprietaryinforma-
tion, which permits them to earn positive expected profit in
equilibrium. In the double informational advantage model, Os earn
zeroexpectedprofitinequilibrium.
Thisexperimentaldesignhasanumberofinterestingcomparative
static predictions that contrast sharply with the double informational
CommonValueAuctionsandtheWinnersCurse 89
advantagemodel.Firstandforemost,theexistenceofaninsiderbene-
fitsthesellerbyincreasingexpectedrevenuerelativetoauctionswith
an SIS. In contrast, in the double informational advantage model, the
existenceofaninsiderunambiguouslyreducessellersexpectedreve-
nue.
14
Second,increasesinthenumberofOsresultsinIsbiddingmore
aggressively in our model. In contrast, in the double informational
advantagemodel,Isbiddingstrategyisunaffectedbyincreasesinthe
number of Os. Finally, both models imply that Is earn substantially
largerexpectedprofitthanOs(zeroprofitforOsinthedoubleinforma-
tionaladvantagemodel),andthatIsearnhigherexpectedprofit,condi-
tionalonwinning,thaninSISauctions,althoughthepredictedincrease
inprofitisrelativelysmallinourdesign.
KagelandLevin(1999)conjecturethatforinexperiencedbidders
theexistenceofaninsidermightattenuatethewinnerscurse.Osinthe
AIS auctions who win against better informed Is face a stronger
adverse selection effect than in SIS auctions. However, it is entirely
plausible that the need to hedge against the existence of an insider is
more intuitive and transparent than the adverse selection problem
resultingfromwinningagainstsymmetricallyinformedrivals.Thus,at
leastforinexperiencedbidders,havinganinsidermayactuallyreduce
theseverityofthewinnerscurse.Thiswouldbetrue,forexample,if
Os view the situation as similar to a lemons market (Akerlof 1970),
whereitseemsreasonablyclearthereisnorampantwinnerscurse(our
culturewarnsustobewareofusedcarsalesmen).Ontheotherhand,
inexperienced subjects may bid higher in order to make up for their
informationaldisadvantage,thusexacerbatingthewinnerscurse.
KagelandLevinemploytwoalternativedefinitionsofthewinners
curseforOsintheAISauctions.Thefirst,veryconservativedefinition
concerns bidding above the expected value conditional on having the
highestsignalvalueamongOs(ignoringIsbid).IfallOsbidthisway,
andIsbestrespondtothesebids,thenOswouldearnaveragelossesof
morethan$1.50perauction,conditionalonwinning.Theseconddefi-
nition accounts for Is best responding to Os bids, and solves for the
zeroexpectedprofitlevelforOs.Table4.5reportsresultsforinexperi-
encedbiddersintheseauctions.Thedataclearlyindicatethatthewin-
ners curse is alive and well for inexperienced Os. Consider auctions
with=$6,whichwereusedtostarteachsession.Withn=4,almost
60percentofthehighOsbidswereabovetheconservativemeasureof
9
0

Table4.5 InexperiencedBidders:AuctionswithAsymmetricInformationStructure(AIS)
Outsidersbids Insidersbids
Frequencyofwinnerscurse(rawdata)
Frequencyhigh
Average
Againstoutsiders Againstoutsiders
outsiderbid Average
earnings Frequency
only andinsiders
fromhigh earnings
Number conditionalon ofoutsiders High High Average outsidersignal conditional Average
of winning winning(%) outsider Allbids outsiderbid Allbids bidfactor
a
holder(%) onwinning bidfactor
bidders (S
m
) (rawdata) bid(%) (%) (%) (%) (S
m
) (rawdata) (S
m
) (S
m
)
1.68 70.6 58.8 39.2 94.1 70.6 1.16 52.9 0.71 1.46
b
4 6 (0.93) (12/17) (10/17) (20/51) (16/17) (36/51) (0.62) (9/17) (0.35) (0.26)
1.40 65.2 39.1 23.2 65.2 47.8 6.00 73.9 2.74 2.25
12 (0.50)** (15/23) (9/23) (16/69) (15/23) (33/69) (0.77) (17/23) (0.77)** (0.35)
6.56 71.4 28.6 14.3 85.7 57.1 11.61 100 5.05 5.09
24 (3.07) (5/7) (2/7) (3/21) (6/7) (12/21) (2.78) (7/7) (3.50) (1.27)
3.68 100 100 85.2 100 92.6 0.61
c
66.7 1.09
b
7 6 (0.61)*** (9/9) (9/9) (46/54) (9/9) (50/54) (0.62) (6/9) (0.29)
2.47 78.9 89.5 69.7 89.5 79.8 4.85 73.7 1.93 1.91
b
12 (1.03)** (15/19) (17/19) (78/112) (17/19) (91/114) (1.03) (14/19) (0.61)*** (0.33)
NOTE:S
m
=standarderrorofthemean.**Significantlydifferentfrom0atthe5%level,two-tailedtest;***significantlydifferentfrom
0atthe1%level,two-tailedtest.
a
Highbidsonly.
b
Asingleoutlierbidlessthanx wasdropped.
o
c
Inthistreatment,highOsactuallybidabovetheirsignalvalues,onaverage.
SOURCE:FromKagelandLevin(1999).ReprintedwithpermissionoftheEconometricSociety.
CommonValueAuctionsandtheWinnersCurse 91
thewinnerscurse,sothatthesebidswouldhavelostmoney,onaver-
age,justcompetingagainstotherOs.Further,consideringthebehavior
ofbothIsandOs(thesecondwinnerscursemeasure),94percentof
thehighObidsweresubjecttothewinnerscurse.Withn=7,thereis
anevenstrongeradverseselectioneffect,withtheresultthatthewin-
ners curse was more pervasive: 100 percent of the high O bids and
85.2percentofallObidsfellpreytothewinnerscurse,evenwithno
accountingforIsbids.Thenetresult,inbothcases,waslargenegative
profitsforOswhentheywon($1.68perauctionwithn=4;$3.68
with n = 7). Although somewhat diminished in frequency, a strong
winnerscurseisalsoreportedforhighervaluesofasOscontinuedto
earn negative profits throughout, with at least 47 percent of all bids
subjecttothewinnerscurseforanyvalueof (whenaccountingfor
bothIsandOsbids).Finally,regressionscomparingbidfunctionsfor
inexperiencedOsinAISauctionsversusinexperiencedbiddersinSIS
auctions show no significant difference between the two treatments.
Thus,contrarytoKagelandLevinsoriginalconjecture,theintroduc-
tion of an insider did not induce significantly less aggressive bidding
forinexperiencedOscomparedtoSISauctions.
Table4.6reportsdataforsuper-experiencedbidders(subjectswho
hadparticipatedinatleasttwopriorfirst-pricesealed-bidauctionses-
sions).Forthesebiddersthewinnerscursehasbeenlargelyeliminated
andthecomparativestaticpredictionsofthetheoryaregenerallysatis-
fied.Isearnedsignificantlygreaterprofitsconditionalonwinningthan
didOs.Forexample,withe=$18andn=7,Osearnedaverageprofits
of around $0.50 per auction conditional on winning. In contrast, Is
earned around $3.25 per auction, conditional on winning. Further, Os
earnedsubstantiallylowerprofitsthanincorrespondingSISauctions,
forwhichprofitsaveragedaround$2.25perauction.Also,asthethe-
ory predicts, Is increased their bids in the face of greater competition
frommoreOs.
Last, but not least, as the theory predicts, for more experienced
bidders, auctions with insider information consistently raised average
sellers revenue compared to SIS auctions (Table 4.7). The intuition
underlyingthispredictionforourmodelisasfollows:Thesellerwould
beunambiguouslyworseoffintheAISauctionrelativetotheSISauc-
tionifIsintheAISauctionwonallthetimewhilebiddingaccordingto
theprescribed(AIS)equilibrium.However,Isdonotwinallthetime,
9
2

4
Table4.6 Super-ExperiencedBidders:AuctionswithAsymmetricInformationStructure(AIS)
Outsidersbids
Frequencyofwinnerscurse:
Frequencyhigh
Average Frequencyof
Againstoutsidersandinsiders
outsiderbidfrom
earnings outsiders
(rawdata)
Average highoutsider
Numberof conditionalon winning(%) Highoutsiderbid Allbids bidfactor
a
signalholder(%)
bidders winning(S
m
) (rawdata) (%) (%) (S
m
) (rawdata)
Insidersbids
Average
earnings
conditional Average
onwinning bidfactor
(S
m
) (S
m
)
0.65 53.7 9.3 4.9 10.05 92.6 3.30 3.60
c
12 (0.43) (29/54) (5/54) (8/162) (0.23) (50/54) (0.23)*** (0.19)
0.87 63.3 3.3 1.1 15.29 93.3 4.13 5.80
c
18 (0.68) (19/30) (1/30) (1/90) (0.26) (28/30) (0.37)*** (0.50)
3.67 42.1 5.3 3.5 27.04 94.7 7.94 8.24
30 (2.32) (8/19) (1/19) (2/57) (0.65) (18/19) (0.69)*** (0.61)
0.52 64.5 22.4 17.2 15.86 86.8 3.24 4.35
7
b
18 (0.34) (49/76) (17/76) (77/453) (0.26) (66/76) (0.36)*** (0.26)
3.90 41.7 16.7 19.4 26.95 83.3 4.95 5.98
30 (3.07) (5/12) (2/12) (14/72) (0.85) (10/12) (0.80)*** (0.67)
NOTE:S
m
=standarderrorofthemean.***Significantlydifferentfrom0atthe1%level,two-tailedt-test.
a
Highbidsonly.
b
Includesseveralauctionswithn=6.
c
Asingleoutlierbidlessthanx wasdropped.
o
SOURCE:FromKagelandLevin(1999).ReprintedwithpermissionoftheEconometricSociety.

CommonValueAuctionsandtheWinnersCurse 93
Table4.7 ChangeinSellersRevenue:AISversusSISAuctionswith
Super-ExperiencedBidders
n=4 n=7
Changein
revenue:
Meanprofits
(
2
) ( )2 Changein
Meanprofits
(
2
)
AISless revenue:
SIS AISSIS
auctions AIS SIS auctions AIS SIS
1.759 2.063 3.822 0.739 1.492 2.231
=$18 (2.057)
**
(8.561) (49.972) (1.573)
*
(6.770) (19.221)
2.734 6.148 8.876 0.919 4.517 5.436
=$30 (1.097) (24.334) (59.731) (0.425) (17.978) (15.839)
NOTE:t-statisticsareinparentheses.*Significantlydifferentfrom0atp<0.10,one-
tailedtest;**significantlydifferentfrom0atp<0.05,one-tailedtest.
SOURCE:FromKagelandLevin(1999).ReprintedwithpermissionoftheEconomet-
ricSociety.
andwhenOswin(withtheirequilibriumbid),theywinwithrelatively
high signal values, yielding more revenue than when Is win. Further,
theexistenceoftheinsiderhelpstoprotectthesellersrevenuecom-
paredtoanSISauctionwhenOswouldhavewonwithrelativelylow
signalvaluesintheSISauction,sinceinthiscaseIwinsandpaysmore
thanOwouldhavepaidintheSISauction.Thenetresultishigherrev-
enueforthesellerandreducedvarianceinsellersrevenue(holdingx
o
constant)comparedtoSISauctions.
15
The increase in revenue resulting from an insider in our model is
counterintuitiveforthosewhoseintuitionhasbeenhonedonthedou-
ble informational advantage model. This reversal of the double infor-
mational advantage models prediction rests critically on the fact that
lessinformedbiddershavesomeproprietaryinformation.Manyreal
worldcasesaremorerealisticallymodeledwithOshavingsomepro-
prietaryinformationandnotjustpublicinformation.Inthesecircum-
stances,itmaywellbethecasethattheintroductionofasinglewell-
informed insider increases average sellers revenue, and that both Is
and Os earn economic rents. This potential for insider information to
raiseaveragesellersrevenuehadnotbeenexplicitlyrecognizedinthe
auctionliteraturepriortothis.
16
94 Kagel
CONCLUSION
SummaryandPolicyImplications
Experimental studies of common value auctions have been going
onformorethan15yearsnow,parallelingtheprofessionsinterestin
thetheoreticalandpracticalpropertiesoftheseauctions.Thisresearch
hasestablishedseveralfactsaboutbehaviorrelativetothetheory.
For inexperienced bidders, Nash equilibrium bidding theory does
not predict well. Inexperienced bidders suffer from a winners curse,
earning negative average profits and with relatively large numbers of
bidders going bankrupt. Overbidding here represents a fundamental
breakdown in the theory, resulting in the reversal of a number of
importantcomparativestaticpredictions:Biddingdoesnotdecreasein
response to increased numbers of bidders in second-price auctions as
thetheorypredicts,andpublicinformationaboutthevalueoftheitem
reduces,ratherthanraises,revenueinthepresenceofawinnerscurse.
Thisperverseeffectofpublicinformationinthepresenceofawinners
curse extends to the endogenous release of public information in
Englishclockauctions.
Experienced bidders in the lab eventually overcome the worst
effectsofthewinnerscurse,rarelybiddingabovetheexpectedvalue
oftheitemconditionalonwinningandearningpositiveaverageprof-
its.Super-experiencedbiddersalsosatisfykeycomparativestaticpre-
dictions of the theory: Release of public information in sealed-bid
auctionsraisesrevenue,andEnglishclockauctionsraisemorerevenue
than do sealed-bid auctions. Further, average revenue increases in an
experimentaldesignwheretheexistenceofaninformedinsiderispre-
dicted to raise revenue compared to auctions with symmetrically
informed bidders. Nevertheless, these super-experienced bidders still
earnwellbelowequilibriumprofitsand,intheoverwhelmingmajority
of cases, are not best responding to rivals bids (they are bidding far
moreaggressivelythantheyshould;KagelandRichard2001).
It is worth noting that these very experienced bidders in the lab
havelearnedhowtoovercometheworsteffectsofthewinnerscurse
in an environment with strong information feedback, substantially
strongerthanislikelytobepresentinfieldsettings.Assuch,learning
CommonValueAuctionsandtheWinnersCurse 95
might not proceed as quickly in field settings. Further, there are
dynamics of interactions within organizations that may retard adjust-
menttothewinnerscurse.Theseincludepaymentsoflargesalariesto
petroleum geologists to estimate likely reserves, and then having to
recognize that these estimates still have a very large variance and are
not very precise; transfers of personnel within the firm and between
firms prior to receiving feedback about the profitability of bids; and
gamingthatgoesonwithinorganizations.
17
Finally,evenassumingthat
thewinnerscursewillbeeliminatedinthelongruninfieldsettings,it
often takes some time before this happens, so this out-of-equilibrium
behaviorisimportantinitsownright.
Thewinnerscurseextendstoanumberofothersettingsaswell:
bilateralbargaininggames(SamuelsonandBazerman1985;Ball,Baz-
erman,andCarroll1991),blind-bidauctions(Forsythe,Isaac,andPal-
frey1989),marketswherequalityisendogenouslydetermined(Lynch
etal.1986,1991),andvotingbehavior(theswingvoterscurse;Fedder-
senandPesendorfer1998,1999).
18
Experimentalstudiesofauctionmarketshaveplayedasignificant
role in the design and execution of the recent wave of spectrum (air
waverights)auctionscarriedoutinthiscountryandabroad.
19
Auction
experimentshaveservedtwoprinciplefunctionsinthiswork:1)asa
windtunneltotestouttheauctionsoftware,whichimplementsarel-
ativelycomplicatedsetofbiddingrules(see,forexample,Plott1997),
and2)asatestbedagainstwhichtocomparetheorywithbehavior.In
thelatterrole,acentraldesignelementhasbeentouseascending-price
auctions (with price feedback for bidders) to both minimize the pres-
ence of the winners curse and to generate increased revenue in the
absenceofawinnerscurse,centralinsightsderivedfromtheinterac-
tionbetweencommonvalueauctiontheoryandexperiments:
Anascendingauctionoughttoremoveanothercommonproblem
withauctions,thewinnerscurse.Thisstrikeswhenasuccess-
ful bidder discovers too late that his prize is not worth what he
paid for it. Some critics of the scale of the bids seem to see the
curseatwork[inBritainsthirdgenerationsales].Yetthewinners
curse is much likelier in sealed-bid auctions, where bidders lack
animportantpieceofinformationaboutthevalueoftheasset:the
valuations of other, perhaps better-informed, bidders. In an
96 Kagel
ascending auction, however, that information is clearly revealed.
(TheEconomist2000,p.21)
...byallowingbidderstorespondtoeachotherbids,[anascend-
ing-priceauction]diminishesthewinnerscurse:thatis,theten-
dencyfornaivebidderstobidupthepricebeyondthelicensess
actualvalue,orforshrewdbidderstobidcautiouslytoavoidover
paying.(McAfeeandMcMillan1996,p.161)
Notes
ResearchsupportfromtheEconomicsandDRMSDivisionsofNSF,theSloanFoun-
dation,andtheRussellSageFoundationaregratefullyacknowledged.Specialthanks
to my colleagues and my coauthors, especially Dan Levin, who have taught me so
much.MuchofthematerialhereistakenfrommypaperwithDanLevintitledBid-
dinginCommonValueActions:ASurveyofExperimentalResearch,whichappears
asChapter1inthecollectionofourpublishedpapersinvestigatingcommonvalueauc-
tions:JohnH.KagelandDanLevin,CommonValueAuctionsandtheWinnersCurse,
PrincetonUniversityPress.
1. Here, I am assuming that buyers are competing to purchase an item. Similar
remarksholdforprocurementauctionsinwhichsellerscompetetoofferservices
atthelowestcost.Inthiscase,however,thetrade-offisinverted;thehighertheir
bids,thelargerthesurplusconditionalonwinning,butthelowertheprobability
ofwinning.
2. However, once the seller uses a minimum bid requirement, and/or we consider
entry to be determined endogenously, different auctions may induce different
probabilitiesofanactualsale.Thus,efficiencymaybecomeanissue(Levinand
Smith1994).
3. Unless,ofcourse,onearguesthattheGrouchoMarxstatementIdonotwishto
joinanyclubthatacceptsme,isanearlierrecognitionofthewinnerscurse.
4. See,forexample,theexchangebetweenCoxandIsaac(1984,1986)andBrown
(1986).
5. Winning bidders paid these losses out of their own pockets or from earnings in
otherauctions.
6. Thereisawholebodyofpsychologicalliteratureindicatingthedifficultyoflearn-
ing generalizing across different contexts (see, for example, Gick and Holyoak
1980;PerkinsandSalomon1988;SalomonandPerkins1989).
7. KagelandLevin(1986)didnotrestrictlowsignalholderstobiddingx
L
,failingto
recognizethatwithoutthisrestrictionthereisnopurestrategyNashequilibrium,
butamuchmorecomplicatedmixedstrategyequilibriumsothattheirbenchmark
calculationsareincorrect.However,thecorrectbenchmarkyieldsanevenhigher
increaseinrevenuefromannouncingx
L
sothattheconclusionsreachedregarding
CommonValueAuctionsandtheWinnersCurse 97
public information receive even stronger support with the correct benchmark
(Campbell,Kagel,andLevin1999).
8. Thegreaterthenumberofrivals,thelowertheprobabilityofwinningasaresult
ofmoreaggressivebidding;hence,thelesslikelyitistopaytodeviatefromthe
Nash strategy even with limited liability. See also the calculations reported in
KagelandRichard(2001).
9. For a completely different approach to the limited liability problem, see Avery
andKagel(1997).
10. Theintuitionisroughlyasfollows:Givensymmetry,thelowsignalholderknows
thatthoseremainingintheauctionhavehighersignalvalues.Butthelowsignal
holdercantprofitfromthisadditionalinformationsinceitisonlyrevealedonce
thepriceisgreaterthantheseremainingsignalvalues;i.e.,priceisalreadygreater
thantheexpectedvalueoftheitemtothelowsignalholder.Theanalysisiscon-
finedintheinterval x+ . x x
11. Commonvalueauctionsinvolvepuresurplustransferssothatrevenuedifferences
arecalculatedas:[
F
]where
E
and
F
correspondtoprofitsinEnglishand
E
first-priceauctions,respectively.Inthiswaywehaveeffectivelynormalizedfor
samplingvariabilityinx
o
bysubtractingitfromtheprice.
12. t-testsareconductedforpredictedrevenueincreasestomeasurethereliabilityof
the prediction for Levin, Kagel, and Richard sample data. One-tailed t-tests are
usedheresincethesymmetricRNNEmakesunambiguouspredictionsregarding
revenue increases. Two-tailed t-tests are used for determining statistical signifi-
canceofactualrevenuechanges,sinceinpracticethereareforcepromotinglower
revenuesinEnglishauctionsandweoftenobservethisoutcome.
13. Tofurtherinvestigatethisquestion,wehaveconductedsomeadditionalsessions
withinexperiencedbiddersinwhichx
L
waspubliclyannouncedpriortobidding
intheEnglishauction.Inauctionswithsixbiddersand =$12,averageprofitsin
the standard English auction (where x
L
was not announced) were $1.55, with
average profits in auctions with x
L
announced of $1.56 (t = 1.46, d.f. = 30, p<
0.10,one-tailedtest;KagelandLevin2002).
14. Although one can readily demonstrate that increased revenue is not a general
characteristicofAISauctionsinwhichOshavesomeproprietaryinformation,it
isanaturalelementinourdesignandcanbefoundinotherAISstructuresaswell
(CampbellandLevin2000).
15. Inourdesign,theincreaseinrevenuegoingfromSIStoAISvarieswithn,with
revenuedifferencesincreasingstartingfromlown,reachingamaximumrevenue
differentialforintermediatelevelsofn,anddecreasingthereafter.
16. TheseresultsmotivatedCampbellandLevin(2000)tofurtherinvestigatetherole
ofinsiderinformationinfirst-priceauctionscomparedtohomogeneousinforma-
tionenvironments.Thischapterconnectstherevenueraisingeffectsofaninsider
to more general propositions regarding the revenue raising effects of increased
bidderinformationfoundinMilgromandWeber(1982).
17. AfriendofmineinHoustonwhowasageologistforamajoroilcompanytoldme
thattherewassuchabroadrangeoflegitimatevalueestimatesformosttractsthat
98 Kagel
whenthebiddingdepartmentstartedreducingbidsrelativetovalueestimatesto
thepointthattheywerewinningveryfewauctions,thegeologistssimplyraised
theirestimates.(Geologistslovetodrill,andfailuretowintractsmeanstheycant
drill.)
18. See Kagel and Levin (2002) for reviews of this work, or better yet, consult the
originalpublications.
19. LedbytheFederalCommunicationsCommission,theU.S.governmenthascon-
ductedanumberofsalestodateraisingatotalof$23.9billionandsellingover
10,000licensesbetweenJuly1994andJuly2000.Evenmorespectacular,inan
auctionendinginApril2000,theBritishgovernmentraised22.5billion($35.53
billion)fromthesaleofthirdgenerationmobilephonelicenses.SeeKlemperer
(2000) and McAfee and McMillan (1996) for reviews and evaluations of these
auctions.
References
Akerlof,G.1970.TheMarketforLemons:QualitativeUncertaintyandthe
MarketMechanism.QuarterlyJournalofEconomics84(3):488500.
Ashenfelter,O.,andD.Genesore.1992.TestingforPriceAnomaliesinReal
EstateAuctions.AmericanEconomicReview:PapersandProceedings
82(2):501505.
Avery,C.,andJ.H.Kagel.1997.Second-PriceAuctionswithAsymmetric
Payoffs:AnExperimentalInvestigation.JournalofEconomicsandMan-
agementStrategy6(6):573604.
Ball,S.B.,M.H.Bazerman,andJ.S.Carroll.1991.AnEvaluationofLearn-
ingintheBilateralWinnersCurse.OrganizationalBehaviorandHuman
DecisionProcesses48(7):122.
Bazerman,M.H.,andW.F.Samuelson.1983.IWontheAuctionButDont
WantthePrize.JournalofConflictResolution27(4):618634.
Blecherman,B.,andC.F.Camerer.1998.IsThereaWinnersCurseinthe
MarketforBaseballPlayers?Photocopy,BrooklynPolytechnicUniver-
sity,Brooklyn,NY.
Brown,K.C.1986.InSearchoftheWinnersCurse:Comment.Economic
Inquiry24(5):513515.
Capen,E.C.,R.V.Clapp,andW.M.Campbell.1971.CompetitiveBiddingin
High-RiskSituations.JournalofPetroleumTechnology23(6):641653.
Campbell,C.,andD.Levin.2000.CantheSellerBenefitfromanInsiderin
CommonValueAuctions?JournalofEconomicTheory91(1):106120.
Campbell,C.,J.H.Kagel,andD.Levin.1999.TheWinnersCurseandPub-
licInformationinCommonValueAuctions:Reply.AmericanEconomic
Review89(1):325334.
CommonValueAuctionsandtheWinnersCurse 99
Cassady,Ralph.1967.AuctionsandAuctioneering.Berkeley,CA:University
ofCaliforniaPress.
Cassing,J.,andR.W.Douglas.1980.ImplicationsoftheAuctionMecha-
nisminBaseballsFreeAgentDraft.SouthernEconomicJournal47(1):
110121.
Cox,J.C.,andR.M.Isaac.1984.InSearchoftheWinnersCurse.Eco-
nomicInquiry22(4):579592.
.1986.InSearchoftheWinnersCurse:Reply.EconomicInquiry
24(3):517520.
Cox,J.C.,S.H.Dinkin,andV.L.Smith.1998.EndogenousEntryandExitin
CommonValueAuctions.Photocopy,UniversityofArizona,Tucson,AZ.
Dessauer,J.P.1981.BookPublishing.NewYork:Bowker.
Dyer,D.,J.H.Kagel,andD.Levin.1989.AComparisonofNaiveandExpe-
riencedBiddersinCommonValueOfferAuctions:ALaboratoryAnaly-
sis.EconomicJournal99(394):108115.
Engelbrecht-Wiggans,R.,P.R.Milgrom,andR.J.Weber.1983.Competitive
BiddingandProprietaryInformation.JournalofMathematicalEconom-
ics11(2):161169.
Eyster,E.,andM.Rabin.2000.CursedEquilibrium.Photocopy,University
ofCaliforniaatBerkeley,Berkeley,CA.
TheEconomist.2000.Comments.356(8181):21.
Feddersen,T.,andW.Pesendorfer.1998.ConvictingtheInnocent:TheInfe-
riorityofUnanimousJuryVerdictsunderStrategicVoting.American
PoliticalScienceReview92(7):2336.
.1999.Elections,InformationAggregation,andStrategicVoting.
ProceedingsoftheNationalAcademyofScience96(19):1057210574.
Forsythe,R.,R.M.Isaac,andT.R.Palfrey.1989.TheoriesandTestsof
BlindBiddinginSealed-BidAuctions.RandJournalofEconomics
20(2):214238.
Gick,M.L.,andK.J.Holyoak.1980.AnalogicalProblemSolving.Cogni-
tivePsychology12(3):306355.
Goeree,J.K.,andT.Offerman.2000.EfficiencyinAuctionswithPrivateand
CommonValues:AnExperimentalStudy.Photocopy,UniversityofVir-
ginia,Charlottesville,VA.
Hansen,R.G.,andJ.R.Lott,Jr.1991.TheWinnersCurseandPublicInfor-
mationinCommonValueAuctions:Comment.AmericanEconomic
Review81(1):347361.
Hendricks,Kenneth,andRobertH.Porter.1988.AnEmpiricalStudyofan
AuctionwithAsymmetricInformation.AmericanEconomicReview76:
865833.
100 Kagel
Hendricks,K.,R.H.Porter,andB.Boudreau.1987.Information,Returns,
andBiddingBehaviorinOCSAuctions:19541969.TheJournalof
IndustrialEconomics35(4):517542.
Hendricks,K.,R.H.Porter,andC.A.Wilson.1994.AuctionsforOilandGas
LeaseswithanInformedBidderandaRandomReservationPrice.Econo-
metrica62(6):14151444.
Holt,C.A.Jr.,andR.Sherman.1994.TheLosersCurse.AmericanEco-
nomicReview84(3):642652.
Kagel,J.H.,andD.Levin.1986.TheWinnersCurseandPublicInformation
inCommonValueAuctions.AmericanEconomicReview76(5):894920.
.1991.TheWinnersCurseandPublicInformationinCommon
ValueAuctions:Reply.AmericanEconomicReview81(1):362369.
.1999.CommonValueAuctionswithInsiderInformation.Econo-
metrica67(5):12191238.
.2002.BiddinginCommonValueAuctions:ASurveyofExperimen-
talResearch.InCommonValueAuctionsandtheWinnersCurse,J.H.
KagelandD.Levin,eds.Princeton:PrincetonUniversityPress,pp.184.
Kagel,J.H.,D.Levin,andR.M.Harstad.1995.ComparativeStaticEffects
ofNumbersofBiddersandPublicInformationonBehaviorinSecond-
PriceCommon-ValueAuctions.InternationalJournalofGameTheory
24:293319.
Kagel,J.H.,D.Levin,R.Battalio,andD.J.Meyer.1989.First-PriceCom-
monValueAuctions:BidderBehaviorandtheWinnersCurse.Economic
Inquiry27(2):241258.
Kagel,J.H.,andJ.F.Richard.2001.Super-ExperiencedBiddersinFirst-Price
CommonValueAuctions:RulesofThumb,NashEquilibriumBiddingand
theWinnersCurse.ReviewofEconomicsandStatistics83(3):408419.
Klemperer,P.2000.WhatReallyMattersinAuctionDesign.Photocopy,
NuffieldCollege,OxfordUniversity,Oxford,U.K.
Levin,D.,J.H.Kagel,andJ.F.Richard.1996.RevenueEffectsandInforma-
tionProcessinginEnglishCommonValueAuctions.AmericanEconomic
Review86(3):442460.
Levin,D.,andJ.L.Smith.1994.EquilibriuminAuctionswithEntry.Amer-
icanEconomicReview84(3):585599.
Lind,B.,andC.R.Plott.1991.TheWinnersCurse:ExperimentswithBuy-
ersandwithSellers.AmericanEconomicReview81(1):335346.
Lorenz,J.,andE.L.Dougherty.1983.BonusBiddingandBottomLines:
FederalOff-ShoreOilandGas.SPE12024,58thAnnualFallTechnical
Conference,SanFrancisco,October58.
Lynch,M.,R.M.Miller,C.R.Plott,andR.Porter.1986.ProductQuality,
ConsumerInformationandLemonsinExperimentalMarkets.InEmpir-
CommonValueAuctionsandtheWinnersCurse 101
icalApproachestoConsumerProtectionEconomics,P.M.Ippolitoand
D.T.Scheffman,eds.Washington,DC:FTCBureauofEconomics,pp.
251306.
.1991.ProductQuality,InformationalEfficiency,andRegulationsin
ExperimentalMarkets.InResearchinExperimentalEconomics,R.Mark
Isaac,ed.Greenwich,CT:JAIPress,p.4.
McAfee,R.P.,andJ.McMillan.1996.AnalyzingtheAirwavesAuction.
JournalofEconomicPerspectives10(1):159176.
Milgrom,P.,andR.J.Weber.1982.ATheoryofAuctionsandCompetitive
Bidding.Econometrica50(5):10891122.
Perkins,D.N.,andG.Salomon.1988.TeachingforTransfer.Educational
Leadership46(1):2232.
Plott,C.1997.LaboratoryExperimentalTestBeds:ApplicationtothePCS
Auction.JournalofEconomicsandManagementStrategy6(3):605638.
Roll,R.1986.TheHubrisHypothesisofCorporateTakeovers.Journalof
Business59(2,Part1):197216.
Salomon,G.,andD.N.Perkins.1989.RockyRoadstoTransfer:Rethinking
MechanismsofaNeglectedPhenomenon.EducationPsychologist24(2):
113142.
Samuelson,W.F.,andM.H.Bazerman.1985.TheWinnersCurseinBilat-
eralNegotiations.InResearchinExperimentalEconomics,Vol.3,V.L.
Smith,ed.Greenwich,CT:JAIPress,pp.105138.
Vickrey,W.1961.Counterspeculation,Auctions,andCompetitiveSealed
Tenders.JournalofFinance16(1):837.
Weverbergh,M.1979.CompetitiveBiddingwithAsymmetricInformation
Reanalyzed.ManagementScience25:291294.
Wilson,R.1967.CompetitiveBiddingwithAsymmetricInformation.Man-
agementScience13(11):816820.
.1992.StrategicAnalysisofAuctions.InHandbookofGameThe-
orywithEconomicApplications,Vol.1,R.J.AumannandS.Hart,eds.
Amsterdam:ElsevierSciencePublishers.
5
SharingVeryHighRisks
HowGovernmentCanMakeHealthInsurance
MarketsMoreEfficientandMoreAccessible
KatherineSwartz
HarvardSchoolofPublicHealth
Between40and44millionAmericansoneinsixnonelderlydo
nothaveanyformofhealthinsurance,accordingtothe2001Census.
Why they do not have health insurance involves a variety of reasons,
many of which are often present in any particular person who lacks
coverage.Wecanmaketwogeneralizations,however.First,amajority
simply cannot afford to purchase health insurance unless it is heavily
subsidized, which currently means subsidized by an employer that
sponsorsgroupcoverage.Abouttwo-thirdsoftheuninsuredhavefam-
ily incomes below $35,000, which is generally too low to be able to
afford health insurance unless an employer pays a large share of the
group premium. The second generalization is that health insurance
markets,especiallythesmallgroupandindividual(nongroup)markets,
aresubjecttomarketfailure.Themarketfailureiscausedbyinsurers
fear of adverse selection. Carriers know from experience that people
whoknoworsuspecttheywillhaveexpensivehealthcareneedsinthe
comingyeararealsomorelikelytoapplyforinsurancecoveragethan
peoplewhodonotexpectsuchexpenses.Suchpeoplemakeupadis-
proportionatefractionofthepeoplewhoapplyforcoverageeveryyear.
As a result, insurers are especially likely to either refuse to insure an
applicant or set a high premium for anyone who they perceive to be
likely to incur higher medical expenditures. People who fall into this
categoryaregenerallyovertheageof45,female,workinginparticular
typesofoccupations,andhavehadmedicalproblemsinthepast.For
these people, health insurance is also either unaffordablegiven the
highpremiumsrelativetotheirincomesorsimplyunavailable.
103
104 Swartz
Both of these explanations of why people lack health insurance
provide rationales for government taking a role in health insurance.
MedicaidandtheStateChildrensHealthInsuranceProgram(SCHIP)
werecreatedlargelytodealwiththefactthatlow-incomepeoplecan-
not afford to purchase health insurance at existing prices. Current
effortstoincrementallyexpandeligibilityforMedicaidandnew,subsi-
dizedbuy-inprogramsaresimilarlygroundedinthetraditionthatwe
use government to redistribute resources in our society to make sure
thatlow-incomeorotherwisedeservingpeoplereceivegoodsandser-
vicesdeemednecessities.
When markets break down in the absence of full information (as
with adverse selection), economic theory argues for government to
intervenetocountertheproblemwiththeobjectiveofmakingthemar-
ket competitive and thereby efficient.
1
In this chapter, I develop the
ideathatinthecaseofhealthinsurancemarkets,governmentinterven-
tionintheformofbeingresponsiblefortheveryhighest-costindividu-
alseveryyearwouldreduceinsurersfearofadverseselection.Inturn,
thiswouldreduceinefficiencycausedbyinsurersspendingenormous
efforttopredictwhetherornotanindividualwillbelikelytohavehigh
medicalcosts,andpremiumsoughttobelowerasaresult.Inaddition,
if insurers do not need to bear the risks of very high-cost people
because such risks have been shifted to governmentand society at
largethenaccessibilitytohealthinsuranceshouldbegreater.
Theplanofthechapterisasfollows.Inthenextsection,Ibriefly
describe who lacks health insurance in the United States. In the third
section,Idescribehowhealthinsurancemarketsworkandhowinsur-
erscompeteintheindividual,nongroupmarket.Inthefourthsection,I
discusstheproposaltohavegovernmentshifttheriskofveryhigh-cost
people from insurers to the general population and how it could
increase efficiency and accessibility in individual and small group
insurancemarkets.Ialsoprovidesomeexamplesofgovernmenttaking
the role of reinsurer and backstopper of markets so that they func-
tion.Finally,Ioffersomeconcludingcomments.
SharingVeryHighRisks 105
WHODOESNOTHAVEHEALTHINSURANCE?
TheuninsuredareacrosssectionofAmericanschildren,young
adults, and middle-aged people who generally work full time but do
not earn more than $30,000 per year (in part because they have no
more than a high school diploma and do not have specific skills).
Because they have low incomes and no health insurance, they fre-
quently cannot afford their share of health insurance premiums when
an employer does sponsor coverage and have debts for emergency
medicalcarethattheyareworkingtopaydown.Someoftheadultsare
widowed or divorced, with young children, so the income they earn
doesnotenablethemtopayfornongrouphealthinsurance.Manyunin-
sured adults are self-employed or working in small, family-run busi-
nesses that cannot afford to sponsor health insurance. About 9.2
millionoftheuninsuredarechildren,andperhapsasmanyas3million
of these children are eligible for Medicaid or the SCHIPs. However,
parentseitherdonotrealizetheirchildrenareeligiblefortheprograms
ortheyfindtheprocessofapplyingforpubliccoverageunpleasant
(KaiserCommissiononMedicaidandtheUninsured2000a).Amajor-
ityofuninsuredarewhite,butAfricanAmericansandHispanicscom-
priseadisproportionateshareoftheuninsured.
TheHenryJ.KaiserFamilyFoundationsCommissiononMedic-
aid and the Uninsured has conducted lengthy interviews with seven
families and one 52-year-old grandmother (Kaiser Commission on
Medicaid and the Uninsured 2000b). Two common threads run
throughouttheirstories.One,theadultsworkhardbutdonotearnhigh
incomes,soevenwhentheyhavetheoptionofobtaininghealthinsur-
ance through an employer, they feel that they cannot afford the
employeeshareofthepremium.Second,alloftheuninsuredfamilies
haveincurredmedicaldebtsasaresultofbeinguninsured.Thedebts
areforverytreatablemedicalproblemsthatwouldnotcauseaninsured
person to think twice about seeing a physician or going to the emer-
gencyroomwithasickchild.Buttheuninsuredbillsforsuchcare
running between $1,000 and $6,000leave the uninsured families
bothstrappedforcashtopayforhealthinsuranceandindailyfearof
furthermedicalbills.
106 Swartz
Whenweexaminedemographicandsocioeconomiccharacteristics
oftheuninsured,themultidimensionalstoriesofrealpeopleareoften
overshadowed. Nonetheless, knowing more about the distributions of
characteristicsoftheuninsuredhelpswhendevelopingpublicpolicies
toincreaseaccesstohealthinsurance.IwilldrawupontheMarch1999
CurrentPopulationSurvey(CPS)formostofwhatfollows.TheMarch
1999CPSshowedthattherewerealmost44millionnonelderlyAmeri-
cans without any form of health insurance. According to the March
2001 CPS, the number of uninsured declined to about 38.4 million,
largelyasaresultoftheboomingeconomyandsmallincreasesinthe
number of people with employer-sponsored coverage. However, the
mildrecessionin2001throughearly2002,combinedwiththeincrease
inunemployment,hasmostanalystsbelievingthatthenumberofunin-
suredinearly2003willbeclosertothenumberin1999,soIwilluse
1999data.
Age
The uninsured are generally young64 percent are younger than
35makingthemrelativelyinexpensiveintermsofexpectedmedical
careuse(Table5.1).Aquarteroftheuninsuredarechildrenunderthe
ageof18.The11millionuninsuredchildrenaccountfor15.4percent
of all children. Two decades ago, about a third of the uninsured were
children,andcloseto20percentofallchildrenwereuninsured,sothe
decline in the number of uninsured children is a reflection of the
impact of the expanded Medicaid eligibility criteria for children.
Youngadults(1824)andadultsbetween25and34havemuchhigher
chances of being uninsured30 percent of young adults and 24 per-
centof2534-year-oldslackcoverage.
Income
Justoverhalfoftheuninsuredin1999hadfamilyincomesinthe
previousyearofunder$25,000(Figure5.1).(Forcomparison,in1999
the median household income for all Americans was $42,100.)
Another15percenthadfamilyincomesbetween$25,000and$35,000.
Thus,two-thirdsoftheuninsuredin1999hadincomesbelow$35,000.
Anotherwayoflookingatfamilyincomeistoadjustitforfamilysize
SharingVeryHighRisks 107
Table5.1 UninsuredbyAgeCohort,1999
Agecohort Number %ofuninsured %ofagecohort
<18 11.073 25.0
1824 7.776
2534 9.127
3544 7.708
4564 8.239
65+ 0.358
Total 44.281
15.4
17.6 30.0
20.6 23.7
17.4 17.2
18.6 14.2
0.8 1.1
100.0 16.3
P
e
r
c
e
n
t

o
f

u
n
i
n
s
u
r
e
d

SOURCE:March1999CPS.
Figure5.1 IncomeDistributionofNonelderlyUninsured,1999
25.0%
20.0%
15.0%
10.0%
5.0%
0.0%
< $10,000 $10,000- $20,000- $25,000- $35,000- > $50,000
$19,999 $24,999 $34,999 $49,999
Family income (1998)
SOURCE:March1999CPS.
108 Swartz
andcomputeitrelativetothepovertylevelbyfamilysize.Indatanot
shown here, two-thirds of the uninsured had incomes below 250 per-
centofthepovertylevel.Theseincomesaresimplytoolowforpeople
toaffordtopurchasehealthinsuranceunlessitisheavilysubsidizedby
anemployerthatsponsorsgroupcoverage.The21percentoftheunin-
suredwhohavefamilyincomesabove$50,000reflectstwochangesin
the uninsured over the 1990s. One is the growing economy and tight
labor market by the end of the 1990s. This caused many people with
part-timeorpart-yearjobs(thatdonotincludehealthinsuranceaspart
ofthecompensation)toworkmorehoursperweekand/ormoreweeks
during the year, enabling them to earn incomes above $50,000. This
wasespeciallytrueintwo-earnerfamilieswhereeachadultmighthave
earned less than $20,000 in weaker economic times. A second factor
thatexplainssomeoftheuninsuredwithincomesabove$50,000isthat
a little more than half of these people live with family members who
are not part of their nuclear or insurance family unit. That is, they
livewithparents,grownchildren,orsiblings,andbecausetheyareall
relatives,theirfamilyincomeishigherthanitwouldbeforaninsur-
ancedefinitionoffamily.Nonetheless,itisworrisomethatanincreas-
ingnumberofuninsuredpeoplehavefamilyincomesthatwethinkof
assolidlyinthemiddle-classsectionoftheincomedistribution.Wedo
notknowhowmuchofthisgrowthreflectspeoplebeingofferedhealth
insurance where they work but declining it for themselves or their
dependentsbecausetheycannotaffordtheemployeeshareofthepre-
mium.
LaborForceStatusofAdults
More than two-thirds of uninsured adults are in the labor force,
with 60 percent of uninsured adults working and another 8 percent
unemployedandlookingforwork.Whenwecountallthedependents
of working uninsured adults, a little more than four out of five unin-
sured live with someone who works (71 percent live with someone
whoworksfulltimeand12percentlivewithsomeonewhoworkspart
time, according to the Urban Institutes analysis of CPS data for the
KaiserCommissiononMedicaidandtheUninsured2000c).
SharingVeryHighRisks 109
HealthStatus
The uninsured are in relatively good health, with only 7 percent
sayingtheyareinfairhealthandanother2percentsayingtheyarein
poorhealth.Onereasonmoreoftheuninsuredarenotinpoorhealth
is that some of the population in poor health qualify for Medicaid or
Medicare (the latter by virtue of long-term disability). Moreover, the
vastmajorityofyoungpeopleandpeoplewhoworkgenerallydonot
have serious medical conditions. The vignettes of the uninsured col-
lected by the Kaiser Commission on Medicaid and the Uninsured
showedpeoplewhowerenotinpoorhealthinspiteofthefactthatthey
oftenhadmedicaldebtsof$1,000ormore.Themedicalbillswerefor
treatablemedicalepisodesthatoccurredinemergencyroomsbecause
the people were uninsured (e.g., strep throat, childhood asthma
attacks),oreventssuchasunexpectedcaesariansectiondeliveries.
BasicPolicyDilemma
This picture of the uninsured illuminates a basic policy dilemma.
On the one hand, health insurance coverage in the United States is
based on employer-sponsored coverage, and we assume that working
people will obtain insurance through an employer group. Employer
competitionforhigh-skilllaborhasforcedcompensationforhigh-skill
jobs to include higher wages and fringe benefits, including health
insurance. On the other hand, we have an economy where many jobs
donotrequirehighereducationand/orspecialskills.Suchjobsgener-
allyhavelowwagesandnohealthinsurance.(Althoughlow-skilljobs
in large firms are more likely to provide health insurance as a fringe
benefit,in1999aquarterofalluninsuredadultsworkedforfirmswith
morethan500employees.)
The fact that the labor market for low-skill workers is not tight
enoughtocauseemployerstoofferhealthinsuranceforlow-skilljobs
isalargepartoftheexplanationforwhy60percentoftheuninsured
adults are working but uninsured. Most uninsured adults do not have
morethanahighschooleducationandarenotskilledenoughtobein
high-skill jobs. This problem is further compounded by the fact that
almosthalfoftheuninsuredworkersareemployedbyfirmsthathave
fewerthan25employees.
110 Swartz
AsIdiscussbelow,smallfirmsfacemuchhigherperpersonpre-
miumsthandolargefirms,whichhavemuchlargernumbersofpeople
forpoolingrisksofmedicalexpenditures.Becausesmallfirmsgener-
allyhavesmallprofitmargins,theycannotaffordtoincreasethecom-
pensation of low-skill workers with the relatively high cost health
insuranceavailabletothem.
Thus, unless we want to radically alter the labor market for low-
skillworkersandtheeconomicconditionsinwhichsmallfirmsoper-
ate,weneedtodeveloptwoconcurrentpoliciestoexpandhealthinsur-
ance coverage. One policy would provide heavily subsidized quasi-
publiccoveragetopeoplewithincomesbelowsomelevel,suchas250
percent of the poverty level, or $35,000. The second would increase
accesstoprivatehealthinsurersforhigher-incomeuninsuredindividu-
alsbyreducingtherisktoinsurersofcoveringpeoplewhodonothave
employer-sponsored coverage. Developing such a policy would pro-
videawayforprivateinsurerstocontinuetobetheprimarysourceof
healthinsuranceintheUnitedStatesandcovermoreoftheuninsured.
Toseewhyrequiresanunderstandingofhowinsurersviewtheunin-
sured and how they compete for business, the subject that we turn to
next.
HOWHEALTHINSURANCECOMPANIESCOMPETE
To understand the health insurance markets in the United States,
we start with the fact that the majority of people obtain coverage
throughemployers.Approximately63percentofthepopulation(ofall
ages) have employer-sponsored group coverage.
2
Those with
employer-sponsoredcoveragepooltheirindividualrisksofhighmedi-
calcarecosts.Almosteveryoneinlargeemployergroupsparticipates
in the employer-sponsored health insurance, so there is only a small
proportion of each group who are likely to have unexpectedly high
medicalexpenses.Butpeoplewhodonothaveaccesstosuchpooling
of risksthe uninsured and the people who obtain individual cover-
ageface insurance markets in which adverse selection is a major
problem.
SharingVeryHighRisks 111
ThreeInterconnectedHealthInsuranceMarkets
Health insurance is sold in the United States in several intercon-
nected markets. We can loosely distinguish between large employer
group, small group, and individual (or nongroup) insurance markets.
Someindemnityinsurersandmanagedcareplans(hereafterreferredto
collectivelyascarriers)activelysellcoverageinallthreemarkets,but
mostdonot.Moreoften,weobservelargecarrierssellingcoverageto
largeemployergroups,andsmallercarrierssellinginthesmallgroup
and individual markets. In addition to these three types of markets,
every state regulates how insurance is sold within its borders. The
stateshavedifferentregulationsgoverningfacetsofinsuranceranging
fromwhatbenefitsmustbecoveredbyinsurancepoliciestohowrates
are determined to requirements about financial reserves. As a result,
thereare51differentsubmarketswithineachofthethreedistinctmar-
kets.Manycarriers,particularlysmallercarriers,offerpoliciesonlyin
thosestateswithsimilarregulationssotheydonothavetokeeptrack
ofandrespondtomanyregulatorychanges.
One result of this is that in the individual markets in 1997, the
numberofcarrierssellingindividualpoliciesrangedfromonlytwoor
three(inDelaware,Idaho,andAlaska)tomorethan40(inNewYork
and Texas) (Chollet, Kirk, and Chow 2000). New Yorks relatively
large number of carriers selling individual coverage is due to the
requirement that all HMOs sell individual coverage. In 1997, just
under 700 carriers sold individual policies in the United States; by
comparison, 2,450 carriers sold policies in the large and small group
markets(Chollet,Kirk,andChow2000).Inspiteofthisdifference,the
individual and group markets are characterized by a small number of
carriershavingatleasthalfofthetotalnumberofpoliciessoldineach
typeofmarketineachstate(Chollet,Kirk,andChow2000).
Largeemployershaveavoidedstateregulationsandstatetaxeson
health insurance by self-insuring (or self-financing) their employees
healthcarecosts.TheEmployeesRetirementandIncomeSecurityAct
of 1974 (ERISA) exempts self-insured employers from state regula-
tions and taxes on policies sold within a state. Most self-insured
employerspayafeetoathird-partyadministrator(almostalwaysacar-
rier) to administer the claims from medical care providers, and the
112 Swartz
employeesareusuallyunawarethatthethird-partyadministratorisnot
theirinsureraswell.
Health coverage is sold and priced quite differently in the three
typesofhealthinsurancemarkets(ignoringforthemomentthe50dif-
ferentjurisdictionsregulations).Thesellingpracticesandpricingdif-
ferences largely reflect the extent to which carriers fear adverse
selection in each of the markets. In the large group market, adverse
selectionatthegrouplevelisuncommonsincealmostallemployeesin
alargecompanygenerallyenrollforcoverage.Ifanemployeroffersa
choiceofplans,thencarriersmaybeconcernedaboutadverseselection
iftheyarethechoiceofasmallproportionofthegroup(Buchmueller
and Feldstein 1997; Cutler and Reber 1998). Employees and their
dependents in large groups pay average premiums based on the total
expected costs of the group; a particular persons expected medical
care costs are not factored into the premium he or she pays. Usually,
the employer also negotiates with several carriers as to the out-of-
pocketcostsharingandbenefitscovered,andtrade-offsbetweenthese
andthepremiums.
Smallgroups(typically,groupswithlessthan50employees)and
individualsfaceverydifferentmarkets.Perpolicypremiumsaresub-
stantiallyhigherinthesemarketsthaninthelargegroupmarket;itis
notunusualtofindpremiumsforsingleorfamilypoliciestobemore
thantwiceasexpensiveforsmallgroupsorindividualsthanforlarge
groups.Theprimaryreasonforthesehigherpremiumsisthatpooling
ofrisksoccursovermuchsmallergroupsofpeopleinthesmallgroup
andindividualmarkets.Asaresult,thevarianceontheexpectedcosts
ismuchlarger.Thiscreatesagreaterriskthatactualcostswillexceed
expectedcostsbyawidemargin.Carriersrespondtothisintwoways.
First,theysethigherpremiumsforsmallgroupandindividualpolicies
becausetheriskperpolicyishigherandtheyneedtobecompensated
for bearing greater risk. Second, they try to insure only people who
theyexpectwillhavelowermedicalcostsandtoavoidinsuringpeople
whotheyperceivetobehigh-costusersofmedicalcare.Carriersgoto
great expense to selectively insure people who they perceive to have
low risks of high medical care costs. The costs of the risk-selection
mechanismsusedbycarriersarealargecomponentofthehigherpre-
miumsforsmallgroupandindividualpolicies.
SharingVeryHighRisks 113
InformationAsymmetryShapestheFormofCompetition
betweenCarriers
Carriers cannot discern from applicant information whether an
applicantwillhavehighmedicalcareuseinthecomingyear.Butthey
believe that people who apply for insurance coverage are dispropor-
tionately comprised of people who expect to have high medical care
useinthenearfutureperhapsbecausetheyoracloserelativehada
medicalconditioninthepast.Theproblemforcarriersisthattheyusu-
allycannotobtainthisinformation;thereisanasymmetryofinforma-
tionbetweenwhatthecarriersknowandwhattheinsuranceapplicants
know. When there is asymmetric information in a market, the market
cannotbecompetitiveandinefficiencywillresult.Inthecaseofhealth
insurance markets, the carriers have the disadvantage in terms of the
asymmetryofinformation.
Carriers fear of adverse selection among applicants in the small
group and individual markets motivates their behaviors. Carriers fear
adverse selection because it causes them to underestimate premium
revenues needed for expenditures and thus risk substantial financial
losses.Toavoidadverseselection,manycarriersadoptselectionmech-
anismstoscreenoutapplicantswhomtheysuspectwilluseexpensive
medical care (Swartz and Garnick 1999, 2000a,b; Chollet and Kirk
1998). Such mechanisms include medical underwriting practices,
3
refusingtoissueorrenewapolicy,excludingcoverageofservicesfor
preexisting medical conditions, and differentiating their policies from
theircompetitorsbygenerouslycoveringsometypesofservices(e.g.,
preventative) but limiting coverage of other services (e.g., substance
abusetreatment)(Stone1993;Franketal.1997).
4
Thus,competitionininsurancemarkets,especiallythesmallgroup
andindividualmarkets,focusesonhowwellcarriersusemechanisms
to identify which firms or individuals might be high-risk versus low-
risk.AsNewhousepointedoutinthecontextofriskadjustmentmod-
els,acarrieronlyneedstobealittlebetterthanitscompetitorsinthe
use of selection mechanisms to make more of a profit (Newhouse
1994).Whencarriersarenotconstrainedintheirabilitytosetdifferent
premiums for people who they believe have different probabilities of
using expensive medical care, then carriers compete in large part in
terms of the accuracy of their models for predicting a persons (or
114 Swartz
firms)medicalexpenses.Thesemodelsaregenerallyknownasactuar-
ialmodelsbecausetheyarebasedonactuarialtablesoflikelihoodsof
using different amounts of medical care by many different demo-
graphicandsocioeconomiccharacteristicsaswellashealthstatusand
prioruseofhealthcare.
5
Differentcarrierswillthenpricetheirhealth
insurancepoliciestopeopleandsmallfirmsbasedontheindividuals
orfirmsexpenditurespredictedbyeachcarriersactuarialmodel.Usu-
ally, the models are used to determine how the premiums might be
underwrittenforparticularindividualsorfirms.Thatis,ifasmallfirm
is predicted to have a high risk of high medical expenses in the next
yearbecauseseveralpeopleinthegrouphadhighexpensesinthelast
year, the carrier may agree to offer insurance only if the firm pays a
substantiallyhigherpremium.Theadditionalpremiumamountunder-
writesthebasicpremiumforthepolicy.
Underwriting principles might also cause a carrier to deny cover-
agecompletelyorexcludecoverageforaconditiontoagrouporper-
sononthebasisofinformationknownbythecarrier.Moststatesallow
exclusion of coverage for a preexisting condition (such as cancer,
osteoarthritis, or allergies) for a limited time periodtypically 12
months.Asaresult,carriersmoreoftensimplydenyanapplicationifa
personhashadquiteseriousconditions,suchasanginaoramyocardial
infarction(CholletandKirk1998).Insomestates,underwritingofpre-
miumsisnotpermittedbecauseitisviewedasaselectionmechanism
that discriminates against people if they are perceived to have high
risksofexpensivemedicalcare.Whenunderwritingisnotpermittedor
itsuseisrestricted,carriersturntootherselectionmechanismstoavoid
insuringhigh-riskpeople.
A frequently used mechanism for separating high- and low-risk
applicantsconsistsofdifferentiatingthebenefits(ormedicalservices)
coveredbyapolicy.Ifacarrierisabletoidentifyahealthcarebenefit
thatisparticularlyattractivetolow-riskpeoplebutnothigh-riskpeo-
ple,thenitcandesignpoliciesthatcausepeopletovoluntarilyreveal
thattheyarelikelytobelow-orhigh-riskpeople.Carriersuseofdif-
ferences in benefits packages is a mechanism for getting individuals
(or groups) to reveal information that separates them in terms of risk
levelsfornominallyunpredictableexpensivemedicalevents.Thus,for
example, if a person knows that cancer runs in his or her family
whichthecarriersdonotknowthepersonmightchooseapolicythat
SharingVeryHighRisks 115
hashighupperlimitsoncoveredexpenses,providesforcancerscreen-
ingtests,andincludesfirst-ratecancercentersinthelistofproviders.
By choosing such a policy, the person is revealing information to the
carrierregardinghisorherriskexpectations.Carriershaveinvestedin
substantial efforts to understand how differences in benefits packages
can be used to attract low-risk people to some policies and high-risk
peopletootherpolicies.
Carriers also have developed monopolistic market niches in the
smallgroupandindividualmarketsasanothermechanismforavoiding
adverseselection(SwartzandGarnick2000a,b).Intheindividualmar-
kets,forexample,somecarriersspecializeinmarketingtoindividuals
who have left the armed services; others specialize in policies attrac-
tive to very small firms of professionals (e.g., lawyers or financial
advisors) or only to individuals who are self-employed. As a result,
fewcarriersinastatemarketactivelycompeteforbusinessamongall
consumersseekingindividualpolicies,andpeoplewhominsurersper-
ceiveashigh-riskhavefew,ifany,optionsforobtaininghealthinsur-
ance(Pollitz,Sorian,andThomas2001;GAO1996).
The differences in states regulations of the insurance markets
within their borders permit the greater or lesser use of these mecha-
nisms or different combinations of the strategies to avoid insuring
high-risk people. States that have attempted to block carriers use of
suchpreferentialselectionmechanisms,particularlyinthesmallgroup
orindividualmarkets,havealmostalwayssetupregulationsthatblock
theuseofonlyoneortwoofthesemechanisms.Stateregulations,for
example, might mandate that all policies sold in the state must cover
substance abuse treatment so as to inhibit carriers ability to avoid
high-riskpeoplewhomaywantcoverageofcareforsubstanceabuse.
Some states have enacted regulations requiring carriers to accept any
applicant(guaranteedissue)soacarriercannotturndownanappli-
cantitviewsashigh-risk.
6
Ofcourse,ifastatehasonlyoneortwoof
these regulations in place, the carriers can use other mechanisms that
arenotproscribedtoaccomplishthesameobjective.Acommonexam-
pleiswhenastaterequirescarrierstoacceptanyapplicantbutdoesnot
also have a regulation governing the way in which premiums can be
set, we observe what should be a totally expected outcome: high-risk
peopleareindeedofferedcoveragebutatanextraordinarilyhighpre-
mium. Similarly, when states require community rating of premiums
116 Swartz
(say, in the small group insurance market) but do not standardize the
benefits to be covered in policies sold in the market, carriers can use
differencesinwhatbenefitsarecoveredunderdifferentpoliciestotry
toseparatehigh-riskfirmsfromlow-riskfirms.
In summation, the information asymmetries in health insurance
marketscausethemarkets,particularlythesmallgroupandindividual
markets, to be inefficient. Inefficiency reflects the fact that enormous
efforts and expense are spent in developing and applying selection
mechanismstoavoidcoveringpeoplewhoarelikelytouseexpensive
medicalcare.Carrierscompetewitheachothernotintermsofproduc-
inginsuranceperseatthelowestpossiblecost,butintermsofinsuring
as high a proportion of low-risk people as possible in order to keep
costslow.Thus,theusualcompetitivemarketforcesthatcauseproduc-
erstoseekprofitsbyreducingtheircostsofproductionandincreasing
market share have been altered by the fear of adverse selection in
insurancemarkets.Ininsurance,carriersseektominimizetheirriskof
unexpectedhighcostsbycompetingtohaveveryhighsharesoflow-
riskpeopleamongthepeopletheyinsure.Thecompetitionamongcar-
riersconsistsoftryingtodobetterthanothercarriersatselectinglow-
riskpeople,whichinvolveseffortsthatdonotcontributetoproducing
insurance.Thecostsofcreatingandusingselectionmechanismsarea
measureoftheinefficiencythatexistsinhealthinsurancemarkets.
AROLEFORGOVERNMENT:COVERVERYHIGH-COST
PEOPLEEVERYYEAR
The market failure caused by carriers fear of adverse selection
leavesuswithtwooutcomes.Oneisthatriskselectionactivitiescause
premiumstobesubstantiallyhigherinthesmallgroupandindividual
market than in the large group market, making health insurance rela-
tively unaffordable for most people who do not have access to
employer-sponsored coverage. The second outcome is that a substan-
tialnumberofpeopledonothaveaccesstohealthinsurance,especially
in the individual market, because they have some characteristic that
causesacarriertoperceivethemashigh-risk.
SharingVeryHighRisks 117
The inefficiency due to expenditures on risk selection could be
substantially reduced if government were to shift responsibility from
thecarrierstothegeneralpopulationforthecostsofpeoplewho,each
year,haveveryhighcoststhatis,peoplewhohavehealthcarecosts
inthetop13percentofthedistributionofmedicalexpenditures.Cur-
rently, if a carrier has enrollees with unexpectedly high costs, those
costsarebornebytheotherpeopleinsuredbythecarrierandwhatever
stockholders the carrier may have. If the carrier has to substantially
increase premiums to recover from losses due to unexpectedly high
costsofsomeenrollees,thereisahighprobabilitythatsomenumberof
enrolleeswhohavelowcostswillleavethecarrierinresponsetothe
premium increase. This leaves the carrier with a risk pool that has a
higher average expected cost. If the following year there are again
unexpectedlyhighcosts,thecyclewillrepeatitself;ifitcontinues,we
havewhattheinsuranceindustrycallsadeathspiral,wherethepar-
ticular policy has to be closed down and abandoned or the carrier is
forcedoutofbusiness.Thisoutcomeplacesalltheburdenofinsuring
high-cost people on the individuals who have had health insurance
fromthecarrierandwhohavetopayhigherpremiumsordroptheir
coverageandtheshareholdersofthecarrier.
If the costs of very high-cost people were shifted instead to the
governmentand thus to the entire populationcarriers fears of
adverse selection and a death spiral would be substantially reduced.
Theburdenofsuchcostswouldberedistributedfromthecarriersthat
encountered adverse selection. As a result, carriers would no longer
have an incentive to use and develop risk selection mechanisms, and
the inefficiency present in the small group and individual insurance
markets would be greatly reduced. This would also enable people to
purchasehealthinsurancepoliciesratherthanbeingdeniedcoverage.
WhatIamsuggestingisthatgovernmentmostlikelythefederal
government, but it could be state governmentstake on the role of
reinsurer for carriers that have insured people who have very high
medicalbillsinayear.Thatis,thegovernmentcouldpayaportionof
thecostsofthoseindividualswhosetotalannualmedicalcostsexceed
some thresholdsay, $30,000or an amount that places a persons
medicalexpendituresabovethe98thor99thpercentileofthedistribu-
tion of medical expenses of the entire population. Carriers often pur-
chasereinsurancetoprotectthemselvesfromtheriskthataninsureds
118 Swartz
claims will exceed $50,000. Instead, if the government acted as the
reinsurerforthehigh-costclaims,thecarrierswouldthenhavefarless
incentive to avoid insuring people they expect to have high expendi-
tures.
Examining the distribution of medical expenditures for the U.S.
population shows why this proposal would greatly reduce carriers
incentivestouseselectionmechanisms.Accordingtopreliminaryesti-
matesfromthe1996MedicalExpenditurePanelSurvey,Monheitpre-
dictsthat68percentofthepopulationhadmedicalexpendituresbelow
$1,000.
7
He further estimates that 4.5 percent of the population had
expenditures between $5,000 and $9,999, while just 4 percent of the
populationhadexpendituresabove$10,000.Itisverydifficulttopre-
dictwhowillhaveexpendituresbetween$5,000and$10,000peryear.
But so long as a carrier is not responsible for costs of people with
expendituresabove,say,$30,000,thenitisnotworththeexpensefora
carriertouseriskselectionmethodstoavoidpeoplewithexpenditures
inthe90thto96thor98thpercentileoftheexpendituresdistribution.It
is simply too difficult to distinguish between people who will have
expendituresatthe30thpercentileandthosewhowillbeinthe510
percentilesbelowthethresholdforreinsurance.Moreover,whilethere
issomecorrelationbetweenapersonsmedicalexpendituresfromone
year to the next, that correlation falls away when a longer period of
time is considered (McCall and Wai 1983; Welch 1985; Goodman et
al. 1991; Gornick, McMillan, and Lubitz 1993). Thus, we should
expect that different people each year would have very high medical
expendituresthatwouldqualifyforthegovernmentreinsurance.
Reinsurance almost always requires the original insurer (the car-
rier)tobearsomeportionofthecostsabovethethresholdwhererein-
surance picks up insuring events. This cost-sharing is built into the
reinsurancestructuresotheoriginalcarrierwillretainanincentiveto
manage the health care of high-cost people. It would be important to
maintain this incentive if the government were to reinsure the very
highmedicalcareexpenses.Inaddition,foranypersonwhohashealth
care expenditures over the reinsurance threshold level each year, the
governmentcouldcovereitheraportionofthecostsabovethethresh-
oldoraportionofallofthepersonscosts.Ineithercase,theshareof
costs that the government would cover also could vary over different
levels of expenditures. For example, the government could cover 90
SharingVeryHighRisks 119
percentofthecostsabovetheeligibilitythresholduptotwotimesthe
threshold, 80 percent of the costs from two times the threshold up to
threetimesthethreshold,andthen100percentofthecostsabovethat.
Having government take on the role of reinsurer would make the
small group and individual insurance markets function more effi-
ciently.Thiswouldimmediatelyprovidewhateconomistscallawel-
fare gain to everyone who purchases health insurance in the small
group or individual insurance markets, since the premiums for insur-
ance will decline in proportion to the reduction in use of selection
mechanisms. Moreover, high-risk people who currently cannot obtain
coveragefromallcarriersalsowouldbenefitbecausecarrierswouldno
longer deem them undesirable. High-risk people would have greater
accesstocarriersandpoliciesininsurancemarkets.
Thewelfaregainscausedbytheincreasedefficiencyintheinsur-
ancemarketsarenotfree,ofcourse.Thisrequiresgovernmentreve-
nues to pay all or some of the medical care costs of the designated
high-cost people. A political advantage of using the income tax and
sources of revenues for the general revenue funds is that they do not
requireimplementationofanewtaxtopayforeitheranewinsurance
programforhigh-costpeopleorareinsurancefundtopaycarriersfor
high-costclaims.Ontheotherhand,whenaprogramiscompetingfor
general revenue funds along with high-visibility government pro-
gramssuchaseducation,highwaymaintenanceandconstruction,or
homelandsecuritythenitisvulnerabletopressurestocutthebudget.
This is particularly true for programs that benefit everyone, but may
appear to assist only a small number of peoplein this case, those
individuals with high-cost claims. The argument has to be made that
both of the government options for high-cost individuals increase the
efficiency of insurance markets, thereby providing benefits to every-
one.
Implementing an institutional structure to permit the government
totakeresponsibilityforthehealthcareexpensesoftheveryhigh-cost
individualsalsowouldrequiresomestandardizationofhealthpolicies
soldinthesmallgroupandindividualmarkets.Standardizingthebene-
fits covered by policies would make it possible to compare medical
expenditure patterns of people and then to identify those people who
havetheveryhighestmedicalexpenses.Withoutsuchstandardization,
itwouldbequitedifficulttoknowwhetherapersonhadhighexpendi-
120 Swartz
turesbecauseofaverygenerousinsurancepolicyasopposedtobeing
quiteill.
NewYorkStatecreatedasubsidizedhealthinsuranceprogramfor
low-incomeindividualsandsmallfirmswithlow-wageworkersthatis
very close to my proposed plan. Healthy New York was developed
during2000andbeganenrollingindividualsinFebruary2001.Under
HealthyNewYork,thestatepaysasmuchas90percentofthecostsof
claimsbetween$30,000and$100,000forpeoplewhohaveclaimsina
calendar year that exceed $30,000 (Swartz 2001). The money for the
pool of funds that pay for these costs comes from the states tobacco
settlementfunds.Toensuretransparencyofwhypeoplehavehigh-cost
claims, currently there is only one standardized benefits package for
theHealthyNewYorkpolicies.PremiumsunderHealthyNewYorkfor
eligiblelow-incomeindividualsareabout50percentlessthanthepre-
miums for individual coverage in the regular individual market; for
smallfirmsthepremiumsareabout1530percentbelowpremiumsfor
comparablepoliciesinthesmallgroupmarket.
In sum, if government were to redistribute the risk of very high
medical care costs from carriers to the broader population, efficiency
would be increased in the small group and individual insurance mar-
kets, enabling more people to obtain health insurance. Premiums
wouldbereducedbecausecarrierswouldreducetheireffortstoiden-
tifyhigh-riskpeoplewhomtheydonotwanttoinsure.Asaresult,rel-
atively low-risk people would be more likely to obtain and retain
coverage.Higher-riskpeople,whocurrentlyhavegreatdifficultyfind-
ingcarrierswillingtoinsurethem,wouldhavemorechoiceofpolicies
andcarrierssincetherewouldbesharplyreducedincentivesforcarri-
erstoavoidhigher-riskenrollees.
CONCLUSIONANDPOLICYIMPLICATIONS
Whenriskispresentinmarkets,suchashealthinsurancemarkets,
market failure can be especially likely because of information asym-
metryandthepotentialforadverseselection.Riskalsocancausemar-
kets to fail to form. If government acts to take care of or remove the
worst risks in such markets, the inefficiency in the markets would be
SharingVeryHighRisks 121
greatly reduced, and markets that otherwise could not even start up
wouldbeabletofunction.
Thereareprecedentsinothermarketswithriskwherethefederal
government has taken responsibility for the worst risks, thereby
enabling markets to function and grow. A market for reinsurance for
catastropheshasdevelopedintheUnitedStatesbecausetherehasbeen
ahistory(including,mostrecently,theresponsetotheterroristattacks
of September 11, 2001) of the federal government stepping in to pay
largefractionsofthecostsofcatastrophes.Indeed,thecreationofthe
Federal Emergency Management Agency in 1978 formally acknowl-
edged the federal governments role in assisting with recovery from
catastrophes. The secondary mortgage market in the United States,
which enables lenders of mortgage money to replenish their capital,
wasestablishedbecausethefederalgovernmenthastakenresponsibil-
ity for the worst-risk mortgages since 1954. The Federal Housing
Authority (FHA) and the Veterans Administration (VA) shifted the
risk of default from mortgage lenders to the federal government for
people who otherwise would not have qualified for mortgage loans.
TheFHAmortgageinsuranceandtheVAmortgageguaranteeprogram
setminimumstandardsforwhatpropertieswereeligibleformortgages
andwhattypesoffinancialinformationwereneededfromborrowers.
This standardization of information permitted mortgages to be resold
onanationalbasisbecausestandardizedinformationmadeiteasierfor
lendinginstitutionsthatwerenotlocaltoperformduediligenceinves-
tigations of mortgages that were offered for resale in the secondary
mortgagemarket.Inaddition,veryhigh-riskmortgagesarebackedby
federal guarantees. It is unlikely that either the reinsurance market or
the secondary mortgage market would function without the govern-
mentbackstoppingthembycoveringtheworstrisks.
Similarly, if government were to reinsure the costs of those indi-
viduals with the highest medical expenditures each year, the risk of
very high costs would be shifted from carriers to the general popula-
tion.Thiswouldcausecarriersinthesmallgroupandindividualinsur-
ance markets to spend substantially less on efforts to avoid insuring
peopletheyperceivetobelikelytohavehighcosts.Inturn,thiswould
reduce the rates for health insurance faced by people who purchase
insurance in these markets and enable a much larger set of people to
obtainhealthcoverage.
122 Swartz
Havingthegovernmentactasreinsurer,alongwithbackstopcarri-
ersintheindividualandsmallgroupmarkets,willhelpaboutathirdof
the people who currently are uninsured. The remaining uninsured do
not have sufficient incomes to afford health insurance unless it is
heavilysubsidized.Aswenotedearlier,manyofthelow-incomeunin-
sured have medical debts for highly treatable episodes of care. Such
debtswouldbefarlowerifthepeoplehadobtainedmedicalcareinset-
tingsotherthanhospitalemergencydepartments.Tofacilitatetheuse
of more efficient settings for medical care by the low-income unin-
sured, government should either create more community health care
centersorextendeligibilitytoadultsforpublicprogramssimilartothe
SCHIPs.Suchgovernmentmovesalsowouldincreaseefficiencyinthe
provisionofhealthcaretotheverylow-incomeuninsured.
Finally,therisingcostsofmedicalcaremeanthathealthinsurance
premiums will also increase, along with increased cost-sharing
requiredwhenpeopleusemedicalcare.Ifthepastisanyindicationof
howthiswillaffectpeoplesdecisionstopurchaseinsuranceortakeup
employer-sponsored coverage, the rising costs will lead to greater
numbersofuninsuredasmorepeoplecometoviewhealthinsuranceas
unaffordable. As we have seen in the last decade, the uninsured are
increasingly people with lower middle-class incomes. We need to
rethinkbothhowweprovideandfinancehealthinsuranceifweareto
avoidrisingnumbersofuninsuredsuchrethinkingcouldbeginwith
thegovernmenttakingontheroleofreinsurerforsmallgroupandindi-
vidualhealthinsurancemarkets.Thegovernmentasreinsurerprovides
amechanismforpublicfundstoenableprivatehealthinsurancemar-
ketstooperateefficientlyandbeaccessibletomorepeople.
Notes
TheauthorgratefullyacknowledgesthefinancialsupportoftheRobertWoodJohnson
FoundationunderGrantNumber033818.Opinionsexpressedarethoseoftheauthor
and may not represent the opinions of the Robert Wood Johnson Foundation or Har-
vardUniversity.
1. Inthecaseofpublicgoods,theargumentisthatgovernmentshouldproducethe
goodsbecauseamarketcannotbesustained.
2. BasedonestimatesbytheCensusBureaufromtheMarch2001CurrentPopula-
tionSurvey.Theestimatesbytypeofcoveragearenotmutuallyexclusivebecause
SharingVeryHighRisks 123
peoplecanbecoveredbymorethanonetypeofhealthinsuranceduringtheyear,
andinsomecasesatthesametime(forexample,somepeoplehavebothMedi-
care and Medicaid coverage). See <www.census.gov/hhes/hlthins/hlthin01/
fig03.gif>.
3. Medical underwriting is the process by which carriers set the premium for an
applicantbasedonthepersonsexpectedmedicalcarecosts.Thus,ifapersonhas
poorhealthstatus,actuarialunderwritingpracticeswouldyieldahigherpremium
thanthatforasimilarpersoninexcellenthealth.Theunderwritingprocessessen-
tiallydetermineswhetherapersonpaysanadditionalamountplusthebasepre-
miumforthepolicy.
4. The Health Insurance Portability and Accountability Act of 1996 (HIPAA) has
been sometimes mistakenly assumed to restrict these selection practices in the
individual insurance market. HIPAA does not prohibit carriers from applying
selectionpracticestothegreatmajorityofindividualswhoseekcoverageinthe
individualinsurancemarkets.SeeNicholsandBlumberg(1998)fordetails.
5. Applicants in both the small group and individual markets generally have to
respondtoquestionnairesabouttheirhealthstatus,useofmedicationsandmedi-
calcareinthepast,andhealthriskbehaviors.Itisnotunheardofforsmallgroups
tobeofferedcoverageformostbutnotallofthemembersofthegroup,withthe
rejectedmembersbeingdeniedcoveragebecausecarriersbelievetheywillhave
highmedicalexpenditures.
6. For example, Washington State, New York, and New Jerseys individual insur-
ancemarketsarerequiredtoguaranteeissueofpoliciestoanyapplicantregard-
lessoftheapplicantshealthstatus,age,gender,orplaceofresidence.
7. CommunicationbetweenAlanMonheitandtheauthor,Spring2001.Monheitand
Marc Berk have analyzed the distribution and concentration of the populations
medicalexpenditures.See,forexample,BerkandMonheit(2001).
References
Berk,M.,andA.Monheit.2001.TheConcentrationofHealthExpenditures,
Revisited.HealthAffairs20(2):918.
Buchmueller,T.C.,andP.J.Feldstein.1997.TheEffectofPriceonSwitch-
ingamongHealthPlans.JournalofHealthEconomics.16(2):231247.
Chollet,D.J,.andA.M.Kirk.1998.UnderstandingIndividualHealthInsur-
anceMarkets:Structure,Practices,andProductsinTenStates.Menlo
Park,CA:TheHenryJ.KaiserFamilyFoundation.
Chollet,D.J.,A.M.Kirk,andM.E.Chow.2000.MappingStateHealthInsur-
anceMarkets:StructureandChangeintheStatesGroupandIndividual
HealthInsuranceMarkets,19951997.Washington,DC:Academyfor
HealthServicesResearchandHealthPolicy.
124 Swartz
Cutler,D.M.,andS.Reber.1998.PayingforHealthInsurance:TheTrade-
offbetweenCompetitionandAdverseSelection.QuarterlyJournalof
Economics(113)2:433466.
Frank,R.G.,T.G.McGuire,J.P.Bae,andA.Rupp.1997.Solutionsfor
AdverseSelectioninBehavioralHealthCare.HealthCareFinancing
Review18(3):109122.
GeneralAccountingOffice(GAO).1996.PrivateHealthInsurance:Millions
RelyingonIndividualMarketFaceCostandCoverageTrade-offs.GAO/
HEHS-97-8.
Goodman,M.J.,D.W.Roblin,M.C.Hornbrook,andJ.P.Mullooly.1991.
PersistenceofHealthCareExpenseinanInsuredWorkingPopulation.
InAdvancesinHealthEconomicsandHealthServicesResearch(Risk-
BasedContributionstoPrivateHealthInsurance),Vol.12,R.M.Scheffler
andL.F.Rossiter,eds.Greenwich,CT:JAIPress,pp.149173.
Gornick,M.,A.McMillan,andJ.Lubitz.1993.ALongitudinalPerspective
onPatternsofMedicarePayments.HealthAffairs12(2):140150.
KaiserCommissiononMedicaidandtheUninsured.2000a.MakingItSimple:
MedicaidforChildrenandCHIPIncomeEligibilityGuidelinesandEnroll-
mentProcedures.MenloPark,CA:TheHenryJ.KaiserFamilyFounda-
tion.
.2000b.InTheirOwnWords:TheUninsuredTalkAboutLivingWith-
outHealthInsurance.MenloPark,CA:TheHenryJ.KaiserFamilyFoun-
dation.
.2000c.HealthInsuranceCoverageinAmerica1999DataUpdate.
MenloPark,CA:TheHenryJ.KaiserFamilyFoundation.
McCall,N.andH.S.Wai.1983.AnAnalysisoftheUseofMedicareSer-
vicesbytheContinuouslyEnrolledAged.MedicalCare21:567585.
Newhouse,J.P.1994.PatientsatRisk:HealthReformsandRiskAdjust-
ment.HealthAffairs13(1):132146.
Nichols,L.M.andL.J.Blumberg.1998.ADifferentKindofNewFederal-
ism?TheHealthInsurancePortabilityandAccountabilityActof1996.
HealthAffairs17(3):2542.
Pollitz,K.,R.Sorian,andK.Thomas.2001.HowAccessibleIsIndividual
HealthInsuranceforConsumersinLess-than-PerfectHealth?MenloPark,
CA:TheHenryJ.KaiserFamilyFoundation.
Stone,D.A.1993.TheStrugglefortheSoulofHealthInsurance.Journalof
HealthPolitics,PolicyandLaw18(2):287317.
Swartz,K.2001.HealthyNewYork:MakingInsuranceMoreAffordablefor
Low-IncomeWorkers.Report484,NewYork:TheCommonwealthFund.
SharingVeryHighRisks 125
Swartz,K.,andD.W.Garnick.1999.CanAdverseSelectionbeAvoidedina
MarketforIndividualInsurance?MedicalCareResearchandReview
56(3):373388.
.2000a.LessonsfromNewJerseysCreationofaMarketforIndivid-
ualHealthInsurance.JournalofHealthPolitics,PolicyandLaw25(1):
4570.
.2000b.AdverseSelectionandPriceSensitivityWhenLow-Income
PeopleHaveSubsidiestoPurchaseHealthInsuranceinthePrivateMar-
ket.Inquiry37(1):4560.
Welch,W.P.1985.MedicareCapitationPaymentstoHMOsinLightof
RegressionTowardstheMeaninHealthCareCosts.InAdvancesin
HealthEconomicsandHealthServicesResearch(BiasedSelectionin
HealthCareMarkets),Vol.6,R.M.SchefflerandL.F.Rossiter,eds.Green-
wich,CT:JAIPress,pp.7596.
6
RiskandAgriculture
SomeIssuesandEvidence
RulonD.Pope
BrighamYoungUniversity
Asasubfieldineconomics,agriculturaleconomicshasanunusual
genesis and hence an unusual orientation. In production, its roots are
found in the study of agronomy and horticulture. Out of these disci-
plines grew studies and training in farm management. As Marshalls
(1920) marginal analysis reached its climax, agricultural economics
was just beginning to emerge as a discipline in land-grant colleges
throughouttheUnitedStates.Itembracedmarginaleconomicanalysis,
comparative advantage, and competition as important insights into
marketbehavior.Ahard-foughtviewbegantoemergethatthebehav-
ior of those involved in agriculture throughout the world was consis-
tentwiththesebasiceconomicconcepts(summarizednicelyinSchultz
1980). Yet, agricultural economics has always strived to help family
farms(intheUnitedStatesorabroad)understandmorefullytheireco-
nomic environment. Thus, there has always been a normative dimen-
siontoagriculturaleconomicsaswell(similar,perhaps,tofinanceina
businessschool).Inmostotherfieldsofeconomics,economistsarenot
so presumptuous as to suggest to economic agents how they should
optimizeunlessitisthegovernment.
Today,agriculturaleconomicsconsidersabroadsetofissuesand
behavior about resources, consumers, the environment, and policy
aboutfoodandfiberusingthefullrangeofcurrenteconomicconcepts
and methods. Likely second only to finance, agricultural economics
hasembracedriskconceptsasanessentialingredienttounderstanding
andprescribingbehavior.Itwasanearlyentryintoexperimentaleco-
nomicsbymeasuringindividualriskpreferencesandsubjectiveproba-
bilities across a relatively broad set of agents (see Young et al. 1979;
127
128 Pope
NelsonandBessler1989).Thepurposeofthischapteristohighlighta
fewselectivebutcentralconcepts,issues,andcontributionsaboutagri-
cultural behavior under risk. Although the central paradigm of eco-
nomicbehavioriscalledexpectedutilitymaximization,thedistinctions
among various models of behavior under risk will not be important.
Indeed, the relevant concepts and issues can be portrayed as a choice
among distributions based upon the mean (a measure of central ten-
dency) and the variance (a measure of dispersion). Such models can
oftenberationalizedasmaximizationofexpectedutility(Meyer1987).
Themainnormativeandpositiveissuesareaboutchoicesthatreduce
risk but, even more fundamentally, raise expected utility. It would be
impossible to cover all of the relevant topics, but this chapter will
address some issues central to agricultural economic research on risk
(seeRobisonandBarry1987;JustandPope2002;andCaswell1995
formorein-depthdiscussions).
Itisusefultostateatleastmyperceptionofafewgenerallyrele-
vanteconomicfactsaboutagriculturethatserveasbackground:
1) Agricultural production is atomistic and is generally placed on
international markets. However, demand for raw agricultural
productsismuchmoreconcentratedthanfinalconsumerdemand
for food products. This implies that farms and final consumers
aregenerallypricetakerswithinternationalshocksreadilytrans-
mittedtoagriculturalmarkets.Thereisoftenanunderlyingsus-
picionbymanyagriculturalproducersthatmarketsareunfairto
thembecauseofthisallegedasymmetricmarketpower.
2) Farm products have relatively price-inelastic demand and sup-
plies.Incomeelasticitiesofdemandformanyrawfoodgoodsare
relativelylowcomparedtomanufacturesandservices.Muchhas
beenmadeoftheinelasticdemandsandsuppliesinagriculture,
implying that shocks have greater price and income conse-
quencesthaninmanysectors.
3) Production is heavily constrained by biological processes that
havelonglagsbetweenthepointintimeinwhichadecisionis
madeanditsultimateconsequences.Thisisparticularlynotable
inlivestockproductionbutisprevalentthroughoutagriculture.
RiskandAgriculture 129
4) Production is heavily seasonal with definitive intra-seasonal
stagesofproduction.
5) Investment decisions tend to have long physical and economic
lives.Landoftenhaslowalternativeuses,exceptnearcities.
6) Weather,disease,andpests(vicissitudesofnature)aredirectand
pervasiveinagriculture.
7) Governmentpolicyisomnipresentandoftenintrusiveinmarket
outcomes(e.g.,theCommonAgriculturalPolicyintheEuropean
Union, target prices, and subsidies in the United States). In
developed countries, policy generally attempts to raise farm
incomes and often raises consumer prices. In developing coun-
tries,policyoftenattemptstolowerconsumerprices.
8) Most of the demands on factors of production (inputs) in crops
are inherently spiked rather than distributed uniformly through-
outtheseason.Thismayimplyanincentivetochooseproductive
activities that dont compete for resources at a given point in
time.
9) Institutions for the ownership of factors of production and the
organizationofproductionvarywidelythroughouttheworld.
10) Evidence seems to suggest that yields are generally increasing
over time but that deviations about this trend are random (not
bunchy).However,pricesarehighlycorrelated(thus,bunchy)in
adjacenttimeperiods.
ItisalsousefultobrieflystatesomestylizedfactsaboutU.S.agri-
culture.
1) Production occurs in predominantly single or family-run enter-
prises. Despite ever-increasingly larger farms, the last agricul-
tural census shows that over 85 percent of farms are family
farms, and true corporations (beyond small family-held corpo-
rations)makeuponly0.4percentofproducers(AllenandLueck
1998).
2) Structural changes in livestock production have been dramatic
andoftenresemblemanufacturingwithlargescaleandsubstan-
tialdivisionoflabor.Thereissubstantialcontractfarmingwhere
130 Pope
farmers supply only some of the inputs and are paid incentive
contractsforproducing.
3) Production is increasingly specialized during the post-war
period, but multi-output production is still common. There
appearnowtobesubstantialreturnstospecializationofproduc-
tion. There is widespread innovation with continual technical
progress,andtherearelargenumbersofstrainsofagivencropor
livestockavailableforproductionwithdifferentinherentcharac-
teristics.
4) Manyfarmsrentandownland;thus,contractingfortheservices
oflandisubiquitous.
5) Cropinsuranceanddisasterreliefhavebeenthecenterofpolicy
debatesinrecentdecades;pricesupportsandproductioncontrols
werecentralinearliertimes.
6) AninterestingaspectofU.S.agriculturaldataisthattherearelit-
tlefarmleveldataavailabletoresearchers.Thereisasmallsetof
selective (not random) panels in a few states, but the data are
oftenoflimitedvalueforthequestionsstudiedandtheyarenot
widely available. This constrains the kind of evidence that is
accumulated(JustandPope2001).
DIVERSIFICATIONASARESPONSETORISK
Since at least the early 1950s, risk reduction through diversifica-
tionhasreceivedconsiderableattention.Ontheprescriptiveside,agri-
cultural economists studied and proposed various diversification
strategies to reduce risk. However, there was a rather serious policy
aspect to this research. If farmers have significant opportunities to
reduce their risk, then perhaps some of the rationale for agricultural
policyneedstoberethought.Thebasicincentivefordiversificationis
widely known and can be discovered with a simple thought experi-
ment. Suppose that the variance of the net income from a 1,000-acre
cormfarmisanumberlabeled
2
,whiletheexpectedreturnfromthe
farmislabeled.Ifanothercropexistssay,soybeans,whichhasan
RiskandAgriculture 131
identical and independent distribution to cornthen as Samuelson
(1967)hasshown,theoptimalchoiceforarisk-averseindividualisto
plant 500 acres in each crop. Using variance to illustrate, this perfect
diversificationwillreducethevarianceofincomefrom
2
,whenspe-
cializedincorn,to
2
/2whenthefarmisdiversified.Thevarianceof
farm income is reduced in half by diversification, and a risk-averse
producer would presumably find this attractive. This is the incentive
for diversification: low returns in one enterprise may be mitigated by
highreturnsinthesecondenterprise.Indeed,ifthereareNidentically
and independently distributed enterprise returns, the variance can be
furtherreducedto
2
/Nbyputting500/Nacresineachenterprise(this
assumes that there are no economies or diseconomies of scale). In
casesofmoregeneraldistributionalsettings(unevenmeans,variances,
andnonzerocovariances),thereisamarginalbenefitfromdiversifica-
tion (reduction of the variance) and a marginal cost (reduction in
expectedoraverageincomebynotspecializingintheactivityorenter-
prisewiththelargestexpectedreturn).
The standard approach to economic behavior up to 1950 implied
specialization:choosetheenterprisewiththehighestexpectedreturn.
This is equivalent to maximizing any increasing function of expected
wealthormaximizing
(6.1)U=u(W
0
+),
where isexpectednetincomefromfarmingandW
0
isinitialcertain
wealth.Thus,ifconfrontedwiththechoiceofproducingcorn,whichis
expectedtoyield$25peracre,andhay,whichisexpectedtoyield$15
peracre,aprudentfarmermaximizingexpectedwealthwouldspecial-
ize in the production of corn. However, in 1952, E.O. Heady argued
thatfarmerslikelyhaddistasteforrisk(riskaversion)asmeasuredby
thevariabilityorvarianceofnetincome.Givenestimatesofindividual
enterprise variances and covariances, farmers can analytically choose
thecroporenterprisecombinationthatminimizestotalfarmvariance
ofincome.Thisprocedurefocusesonthebenefitsfromdiversification
andhighlyfavorsdiversificationratherthanspecializationasanopti-
mal decision. Knowing that enterprise expected returns will likely be
unequal, Heady also discussed choices that minimize variance for a
132 Pope
givenexpectedincomeappropriateforanymean-varianceutilityfunc-
tionoftheform
(6.2)U=u(W
0
+,
2
),
where is expected farm income and enters utility u positively,
2
is
thevarianceoffarmincome,andUisutility(expectedutility).Inthis
chapter,anindividualisriskresponsivewhenoneincludesthevari-
ance in a maximization such as in Equation (6.2). A person is risk-
averse when increased variance reduces utility. In the summary in
Youngetal.(1979),mostoftheindividualfarmerswhosepreferences
were elicited were not risk-neutral for some decisionsa majority
were risk-averse, and some were mixed, meaning that for some deci-
sions a person might be risk-averse and for others risk preferring or
neutral.
In Headys analysis, expected farm income is: = h
1

1
+ h
2

2
,
wherehistheproportionoftotallandorinvestmentinenterprise1and

1
and
2
areexpectedreturnsperunitoflandorinvestmentonenter-
prises1and2respectively.Similarly,thevarianceoffarmincomefor
twoenterprisesis
2 2 2
(6.3)

2
= h
1
+
(
1 h
)
2
+ 2 1 h
)
, h
(

2 12
2
where
1
2
isthevarianceofenterprise,1incomeand

2
isthevari-
ance of enterprise 2 income, and
12
is the covariance of the two
incomes. Heady found that if a typical Iowa corn farm diversified by
halving corn acreage and correspondingly increasing hay production,
varianceofincomecouldbereducedsubstantiallywithoutasignificant
reductionofexpectedincome.Thisisbecausethecorrelationbetween
hay and corn income is relatively small, 0.45. Thus, large random
drawsinonecropsincomeareoftenoffsetwithlowrandomdrawsin
the other crops income. This incentive becomes most pronounced for
enterprises whose outcomes tend to be independent or are negatively
correlatedwithsimilarmeans.Whenexpectedreturnsareverydiffer-
ent in the two enterprises, then specialization of production becomes
morelikely.
Subsequent writers added the possibility of renting in or out land
(Johnson1967).Inthiscase,thesquarerootofthevarianceorthestan-
RiskandAgriculture 133
darddeviationofincomecanbelinearlyreducedbychoosingtopro-
duce fewer (or less) risky enterprises and engaging in more safe
activities.Examplesofthelatterarecashleaseoflandandinvestingin
arisk-freeassetinthecaseofcapital.Incaseswherethecashleaseor
risk-free rate exceeds the expected return from risky enterprises, then
specializationintherisk-freeassetispredictedunderriskaversion(see
Equation (6.2)). In cases where the expected risky return exceeds the
cashleaserate,firmscombinethetwoaccordingtotheirtastesforrisk
(risk aversion). A similar conclusion can be obtained for investment
capitalamongrisk-freeandriskyassets.
Many authors attempted to use quadratic programming or an
equivalentmathematicalprogrammingmodeltoidentifytherisk-effi-
cient(minimizingvarianceforagivenmean)setofenterprisechoices
for farms, regions, or countries. The main advantage is that quadratic
programming models of farms could integrate many production con-
straints on firm behavior. For example, perhaps machinery and labor
supplies were limited throughout the months of a growing season. In
addition, they could include many policy constraints or incentives,
such as land set-asides. However, the normative and positive content
(what farms should do and what they do) of these models is only as
good as the models themselves. Failure to reflect individual prefer-
ences, beliefs, or constraints will yield recommendations or insights
thatmaybeirrelevanttoadecisionmaker.Additionaleffortisneeded
to understand what decision makers actually do in their response to
risk.
Lin,Dean,andMoore(1974)attemptedtotestwhetherprogram-
mingmodelsincorporatingresponsetorisk(variance)werebetterthan
risk-neutralmodels.Usingelicitationtechniques,thepreferencefunc-
tions of a small set of farmers were estimated. Many of these prefer-
encefunctionsimpliedthatthemeanandvarianceofincomesshould
enter into farmers objective functions. Mathematical programming
models using these general objective functions were superior at pre-
dicting what farmers actually did when compared to models based
solelyonmaximizingexpectedfarmincome.Thoughbytodaysstan-
dards the techniques and evidence used to advance their argument
mightberatherunconvincing,itwasandisanimportantpaperinposi-
tiveeconomics,convincingmanythatriskwasfundamentaltounder-
standing behavior in agriculture. This paper confirmed empirically
134 Pope
what other researchers had suspected: risk aversion was superior to
riskneutralityforexplainingbehavior(e.g.,OfficerandHalter1968).
SpatialDiversification
Soon after Headys work, Emery Castle (1954) noted that area
diversificationwasalmostasimportantascropdiversification.Indeed,
spatialdiversificationapparentlyhasbeenasuccessfulstrategyforrisk
reduction since medieval times (McCloskey 1976). Formulae for the
varianceacrossfarmsalwaysinvolvethecovariance(seeEquation(3))
or correlation, which is the covariance divided by the product of the
square root of the variances. Spatial diversification becomes particu-
larly useful if the correlation across farms is sharply reduced as dis-
tancebetweentheplotsincreases.
Jensen(1961)arguedthatspatialdiversificationwasanimportant
managerial technique open to dryland farms in the Great Plains
because of idiosyncratic weather across areas. Thompson and Wilson
(1994)arguedthatoneoftheprimaryreasonsthatMexicanejidocom-
munal farmers resisted privatization of grazing land is that yields are
variable with highly idiosyncratic weather patterns. Farmers could
readilyreducethevarianceoftheiryieldsbyscatteringproductionspa-
tially.Ofcourse,spatialdiversificationhasacostintermsofexpected
return(increasedtravelcosts),butapparentlythebenefitsaresufficient
tomakeitviable.
Davisetal.(1997)foundthatthecorrelationbetweenyieldsofdif-
ferentpeachorchardsdecreased2.28percentforeachmileofsepara-
tion,whichcouldbeasignificantfactorinthepatternofoperation.It
should also be mentioned that larger farms often have a significant
advantage due to very subtle advantages in diversification. Many
farms, such as orchards, have different responses at different eleva-
tions.Thus,afarmcaninsomecasesgainasignificantreductioninthe
variance while having contiguous plots by diversifying by elevation.
However, more research is required to know how extensively spatial
diversificationtechniquesareused.
RiskandAgriculture 135
EconometricModelsofRiskResponse
Therewasanincreasingawarenessinthedecadeofthe1970sthat
theevidenceviaprogrammingmodelsofrisk-averseorrisk-responsive
behavior was not on sound statistical footing. In mathematical pro-
gramming models, parameters are usually estimated and treated as
exact.Hence,therewasnotareadilydeduciblemetrictodecidewhen
something like the null hypothesis of no aversion or risk response
couldberejected.Aswithotherfieldsofeconometrics,programming
modelsgavewaytothesearchforeconometricevidence.Thesemod-
els often made very simple assumptions about constraints, but the
resultsweremoreeasilyamenabletoinference.Theabilitytoincorpo-
ratemorecomplexconstraintsonbehaviorineconometricmodelsmay
implythatmoreoftheoldprogrammingconstraintswillfindtheirway
intoeconometricmodels(Andrews2001).
Foroverfourdecades,agriculturaleconomistshadbeenusingcom-
puterstoestimateshortrunsupplyfunctionsessentiallyoftheform
(6.4) A b
0
+ b
p
+ b z , =
1 2
where
p
is the expected price (or yield, or both) of the crop or live-
stock,Awouldbeacreageorsupply,z representsothervariables,and
theb
0
b
2
areconstantstobeestimatedeconometrically.Onepromi-
nent example of such an approach is the adaptive expectation model
discussedinundergraduateeconometricstexts.Thecoefficientofb
1
is
presumed to be positive and the larger the magnitude of b
1
, the more
elasticisthesupply.
Around1970,somearguedthatthisapproachwaslimitingbecause
it didnt capture risk response. Behrman (1968) incorporated risk
responseinagricultureashestudiedcropproductioninThailand.This
is a large and careful study. More importantly, a regression was esti-
matedofthefollowingform:
2
(6.5)
A b
0
+ b
p
+ b
2

p
+ b z , =
1 3
where
p
isanestimateofexpectedprice(yieldorboth),
2
isanesti-
p
matedofthevariance(orstandarddeviation)ofprice,yield,orboth,z
represents other variables (for example, the means and variances of
136 Pope
substitutesandcomplements),andtheb
0
b
3
areconstantstobeesti-
matedeconometrically.Behrmanfoundthatinapreponderanceofthe
cases, b
1
was estimated to be positive and statistically different from
zero, indicating that supply curves are upward-sloping in expected
price.Inamajorityofcases,b
2
wasestimatedtobenegativeandstatis-
ticallydifferentfromzero.Thiswasparticularlytrueforuplandcrops
thataresoldonthemarket,unlikerice,whichisoftenconsumedbythe
farmfamily.BehrmanconcludesthatTheestimatedresponsestothe
relative standard deviations do provide further support, however, for
thehypothesisthattheagriculturalsectorsinunderdevelopedcountries
respondnegativelytorisks(p.336).Sixyearslater,anotherworkon
risk response was very influential. Just (1974), using a Bayesian
approach,formalizedtheestimationandspecificationofthemeanand
variance of revenue, including complementary and substitute crops,
andestimatedanacreageresponsemodellikeEquation(6.4)forcoun-
ties in California. He concluded that there was convincing statistical
evidencethatb
2
isnegativeformanycrops.Thus,itappearedthatrisk
responsewasnotlimitedtodevelopingcountries.
Anumberofpapersduringthenextthreedecadessoughttodeter-
mine whether a model like Equation (6.5) captures something that
Equation(6.4)doesnot.Indeed,therehasbeenmountingevidenceof
risk-responsive behavior across many commodities, countries, and
aggregations.Table6.1summarizesasampleofthesestudies.Though
theelasticitiesmeasuringriskresponseareoftenlowinabsolutevalue
(column 5), they usually have the expected sign (negative), and risk
coefficientsarestatisticallysignificant(column3).Insomecases,the
response to the risk of competing crops can be captured (column 4).
Forexample,morecornacreagemaybeplantedwhentheriskinsoy-
beansincreases.Variousmeasuresofriskcanbeconstructed(column
6),andthisissuecontinuestobeamatterofresearchandcontroversy.
2
In many cases, both
p
and are estimated using weights of past
p
observations. This is called adaptive in the table. In this case, risk is
measured by a backward looking mechanism; surprises in the past
affecttheexpectationofthefuturevarianceofprice,yields,orrevenue
peracre.
Toillustrate,
p
mightbeaweightingofthepreviousthreeyearsof
prices with weights summing to one. Similarly,
2
p
is estimated by
weighting the last three years of squared deviations about
p
(Chavas
RiskandAgriculture 137
and Holt 1996). More sophisticated single-equation approaches use
long memory geometrically declining weights (Just 1974), ARCH/
GARCH with conditional or time varying variances, and/or rational
risk(AradhyulaandHolt1989;Myers1989;HoltandAradhyula1990,
1998; Holt and Moschini 1992). Rational risk implies that the mean
andvariancesimpliedbythemodelmatchthemarketdatagivenavail-
able information. One must build up a structural model of the supply
andthedemandsideofthemarkettoyieldexpectedpriceandthevari-
ance of price given available information. Then, the restrictions
implied by the rational expectation hypothesis must be imposed. One
of the most impressive but complex applications of rational risk is
foundinHoltandAradhyula(1998),whereacarefullyspecifiedmodel
of the broiler market is estimated. Risk-responsive behavior was evi-
dent.Morecomplicatedstillwouldbetoestimateacompletemodelof
production or supply and inputs demanded (factor demands), such as
chemicals,labor,land,andmachineryandproductsupplyusingratio-
nalexpectationsofthefirsttwomomentsofprice.Anumberofauthors
haveestimatedsuchmodelswithoutexplicitcomplicatedexpectational
schemesandfoundevidenceofrisk-responsivebehavior(Antle1987;
ChavasandHolt1996;LoveandBuccola1991;Saha,Shumway,and
Talpaz1994;Coyle1999).
In summary, the available econometric evidence suggests that
firms rebalance their production portfolios such that when the per-
ceivedrisksofanenterpriseincrease,farmssubstitutetowardlessrisky
enterprises. Taken as a whole, this evidence is very persuasive that
thesemodelscapturesomething.However,forsome,therearestillres-
ervations about the explanation of risk aversion for these risk effects.
Thatis,isitpossiblethatEquation(6.5)merelypicksupanonlinearity,
lags,oraggregationproblems(e.g.,Pope1981)?Partofthereasonfor
this skepticism is the very success of the approach. When Equation
(6.5)isappliedtohighlyaggregateddatawhereriskmeasuresaresub-
stantially compromised and/or in markets where reasonably good
futuresmarketsexist,itstillseemstoworkwell.Thequestionthenis
notoneofinsufficientevidence,butofinterpretationoftheevidence.
Table6.1 SelectedRiskResponseStudies
1
3
8

Significantnegative Significant
Dependent own crossrisk Ownriskelasticity Risk
Author variable riskcoefficient
a
coefficient (shortrun)
b
measure
c
Behrman(1968) Rice,corn,casaba,A
d
Yes N.A.
e
Smallbut 3-yearstd.dev.
negative (moving)
Just(1974) Grain,cotton,A Yes Yes Notcalculated Adaptivelike
(infinite)
Lin(1977) Wheat,A Yes N.A. Notcalculated 3-yearstd.dev.
Estesetal.(1981) Potato,A Yes N.A. Notcalculated Adaptive
HurtandGarcia Sowfarrowing Yes Yes 0.47to0.56 Adaptive
Brorsen,Chavas,and Wheatmargins Yes N.A. Notcalculated Adaptive
Grant(1985) (f-m&m-r)
f
AradhyulaandHolt (1989) Broilers Yes Yes 0.045 GARCH
rational
HoltandAradhyula(1990) Broilers Yes 0.232,0.012, GARCH
0.046
ChavasandHolt(1996) Cornandsoybean,A Yes No Notreported Adaptive
LoveandBuccola(1991) Cornandsoybean Yes No N.A. Yes
system,A
PopeandJust(1991) Potatoandsugarbeet, A Yes Yes Notreported Adaptive
vonMassowand Whitebeans,corn Yes Yes 0.073to0.220 Adaptive
Weersink(1993) soybeans,wheat,A
Saha,Shumway,and Wheatsystem Yes N.A. N.A. N.A.
Talpaz(1994)
Holt(1994) Corn,A Yes Yes 0.018 Rational
Duffy,Shalishali,and Cotton,corn,and Yes No Notreported Adaptive
Kinnucan(1994) soybean,A
Krause,Lee,andKoo Wheat,A Yes N.A. 0.062to0.003 Adaptive
(1995)
KrauseandKoo(1996) Wheat,barley, Yes Yes 0.05to0.01 Adaptive
flaxseed,andoil
sunflower,A
TronstadandMcNeill Sowfarrowing Yes Yes 0.0013to0.164 Downside
(1989)
Bar-Shira,Just,and Cropsystem Yes Yes N.A. Adaptive
Zilberman(1997)
Coyle(1999) Cropsandlivestock Yes No Notreported Adaptive
system
a
Oftenasinglepaperincludesavarietyofspecifications.Yesmeansthatsomeoftheriskparametersweresignificant.
b
Often a single paper includes a variety of specifications. Thus, the elasticities reported are an attempt to convey approximate risk
response.
c
Adaptivehereisusedveryloosely.Itisintendedtoimplyaweightingschemewheretheweightssumtoone.Someadaptiveused
polynomiallagsratherthangeometricdeclining;someusesimplefixedweightingschemes.
d
A=cropacreageorsimilarspatialmeasure.
e
N.A.meansnotapplicable.
f
f-mandm-raremargins:respectively,farmtomillandmilltoretail.
SOURCE:Authorscalculationsfromselectedcitedreferences.
1
3
9

140 Pope
OtherFormsofDiversification
Anotherrisk-reducingactivityistodiversifyfamilylabor.Mishra
and Goodwin (1997) find a significant positive relationship between
thecoefficientofvariationoffarmincome(standarddeviationdivided
by the mean) and off-farm employment. Thus, when farm income is
more variable, one risk-reducing strategy is to apply more of ones
laborportfoliotosaferoff-farmincome-generatingactivities.Further,
farmoperatorswhoreceivelargepaymentsfromgovernmentfarmpro-
grams are less likely to supply off-farm labor. Both of these findings
areconsistentwithfarmersbalancingrisksinaportfoliogeneratedby
ownerlaborandownercapital.
Not only can a farmer self-insure through reducing labor alloca-
tions to risky endeavors, but capital can be allocated to safe invest-
ments as well. Mishra and Morehart (2001) calculate that off-farm
financialassetsin1995fortheUnitedStateswere18percentoftotal
assets for farm families. This is up from 14 percent in 1990. Thus,
farms are becoming more diversified outside of agriculture. One way
to view these data is that agriculture is more risky so farmers are
increasingly diversifying outside of agriculture in order to reduce the
risk of total wealth or income. Perhaps recent market events have
reversedthattrend.
DiversificationandFarmSize
Because the foundation of much agricultural policy in the United
States historically has hinged on the survival of the family farm, one
issueofconcernistherelationshipofscaleandrisk.Unliketheportfo-
lio approach, there may be substantial economies of scale in produc-
tion;thatis,asafarmproducesmoreofaparticularcrop,marginaland
average costs of production fall. Economies of scope may also arise,
meaningtherearecostadvantagestodiversification.Partofthereason
forsucheconomiesofscopeisthatthereareinputsthatareproductive
across products. For example, a tractor can be utilized to produce a
varietyofcrops,especiallywhentheydontcompeteatthesametime
forservices.Economiesofscopeimplythatexpectedutilitywithtwo
products is greater than expected utility when specialized. These can
come from the diversification motive discussed above or from cost
RiskandAgriculture 141
advantages from public-like inputs. Economies of scale promote spe-
cialization while economies of scope promote diversification. Thus,
therearethreerelevanteffects:diversificationinresponsetoriskwith
no scale effects, scale economies in enterprises, and economies of
scope due to cost advantages in jointly producing two or more prod-
ucts.
Attemptingtoempiricallyuntanglethesethreeeffectsisdifficult,
andthereisnotasatisfactoryconclusion.Econometricstudiesprovide
someevidencethatlargerfarmsaremorediversified,ceterisparibus.
Further, wealthier farms, ceteris paribus, have less diversification
(Pope1976;PopeandPrescott1980;DunnandWilliams2000;Zenger
and Schurle 1981). Pope (1976), using factor analysis for California
farms,foundevidencethatthereisacombinationofminimumefficient
scale and economies of scope due to spreading the services of fixed
inputsacrosstime.
However, looking broadly across this literature, one is struck by
the large volume of prescriptive literature on optimal diversification
andtherelativelysmallsetthatpositivelyexaminesbehavior.Though
there is little doubt that the principles of diversification are always
potentially important, exactly where they are used is still a matter of
some debate. For example, perhaps farm diversification across enter-
prisesduetoriskaversionisrelativelyunimportantinexplainingfarm
behavior.
Ifinitiallyanoptimalportfolioofactionsorinvestmentsischosen,
thenpolicythatreducestheriskinaparticularagriculturalcommodity
willseegreatersupplyofthatcommodity(andattheexpenseofoth-
ers).Thus,awell-meaningpolicyattemptingtoassistwheatfarmersin
thenorthernplainsbecauseofvariableprofitabilitymayhavetheunin-
tended consequence of increased production and greater demand for
help in the future. Further, it is apparent that behavioral and market
responsestoriskmaybediminishedinresponsetoapublicpolicythat
attempts to reduce risk. For example, diversification may fall if the
governmentprovidesasafetynetforfarms.
142 Pope
RISKREDUCTIONANDINPUTUSE
Onecansurelyviewtheentireportfoliochoiceasoneofchoosing
inputs,suchaslandallocationsforcrops.However,oneaspectofinput
choiceappliestospecializedordiversifiedfarmsandaskswhatthedis-
tributional consequences are of input choice. To illustrate the issue,
supposethatafarmisspecializedintheproductionofcorn.Cornhasa
productionfunctionthatdependsonaninput,h.Bothitsmeanoutput
andthevarianceofoutputdependonh.Thatis,=(h);
2
=
2
(h).
Whatmightafarmerdoinchoosinghowmuchofthisinputtoapply?
Just as in diversification, there is a marginal benefit in that expected
outputwouldincrease.Forexample,iffertilizerisapplied,weexpectit
toraiseoutputortohaveapositiveexpectedmarginalproductgener-
ally.Atsomepoint,itisexpectedthatadditionalfertilizerwilldiminish
output(negativeexpectedmarginalproduct).Ifthefarmersonlycon-
cern were expected profit , the farmer would choose fertilizer such
thattheexpectedmarginalbenefitequaledthecost(price)perunitof
thefertilizer.Inwhichcase,economistssaythatafarmchoosesinputs
such that the expected marginal revenue product is equal to marginal
factor price. That is, the marginal benefit of input use is equal to its
marginalcost.
However, if the farmer is risk-averse, there is concern with how
increasing input allocations might alter the variance of profit. If the
decision maker is risk-averse, then increasing the variance of profit
will reduce utility. The important question is how each input contrib-
utestoexpectedprofitandthevarianceofprofitatthemargin.Wewill
callinputsthatreducethevarianceofprofitsatthemarginrisk-reduc-
ing,whileinputsthatincreasethevarianceofprofitsarerisk-increas-
ing.Ifaninputisrisk-reducingandthefarmisrisk-averse,therewill
beanadditionalmarginalbenefitfromusingmoreofitthanwouldbe
implied by maximizing expected profit. This is a self-insuring tech-
nique. Firms might have more machinery or labor than would seem
advisablebaseduponaveragemarginalproductbecauseusingmoreof
itreducesrisk.Farmsmightusemorepesticidesthanwouldseemprof-
itable on average because of its self-insuring capabilities. A little
reflection shows that irrigation may perform that function. Irrigation
oftenvirtuallylopsoffthelowertailofthedistributionofyields;nec-
RiskandAgriculture 143
essarily,itreducesthevariance(andraisestheaverage)ofcropyields.
However, land is likely risk-increasing under this definition: adding
more acres of corn increases the variance of profits (and expected
profit).Thepolicysignificanceofdescribingagriculturaltechnologyis
apparent.Ifagriculturaldecisionmakersbelievethattheenvironment
ismorerisky,thentheymayusemoreofinputsthatleadtodegradation
oftheenvironmentwhentheinputsarerisk-reducing.
Themotivationforansweringthequestionabouthowinputsaffect
thedistributionofoutputcomesfromtwosources.First,itisrelevant
toprescribingoptimalinputusetofarms.Itisparticularlyrelevantin
developingcountries.Ifamodernvarietyofacropischosen,thereare
often very large variations in output if there are modest variations in
inputslikefertilizer.Thisisoftenincontrasttonativevarieties,which
have a number of resistances to input variations. Secondly, there are
manyenvironmentalissuesregardingtheuseofmodernchemicals.
Roumasset(1976)consideredriceproductioninthePhilippinesin
19711972 and found that the green revolution was not as successful
as expected. Farmers often adopted miracle rice varieties, but they
did not use the recommended amount of nitrogenous fertilizer. It was
hypothesized that less than the recommended level of fertilizer was
used because of risk aversion. After estimating risk preference func-
tionsandtherandompropertiesoftechnology,Roumassetdiscovered
that risk neutrality was more consistent with observed behavior than
was risk aversion, contrary to the Officer and Halter (1968) and Lin,
Dean,andMoore(1974)conclusionthatriskaversionoftenexplained
behaviorbetterthanriskneutrality.
As mentioned earlier, most attention was focused on chemical
inputs.Theempiricalresultswereoftenmixed,butthereisnoreason
forinputstobehavesimilarlyacrosssoilqualities,climaticconditions,
and crops. Secondly, results vary because of many methodological
issuesassociatedwithfunctionalformsandestimationofhigherorder
moments. Regev, Gotsch, and Rieder (1997) found significant evi-
dence that fungicides are risk-increasing at low levels of rainfall, but
foundnoconclusiveevidenceofnitrogenbeingrisk-reducingorrisk-
increasing.HorowitzandLichtenberg(1993)foundevidencethatfer-
tilizerandpesticidesmayberisk-reducing.Mixedresultsonpesticides
are found in Carlson (1979), Horowitz and Lichtenberg (1994), and
Hurd (1994). There seems to be a growing consensus that there is no
144 Pope
evidencethatpesticidesarerisk-reducing.Thusfar,agriculturalecono-
mists are only beginning to build a consistent body of findings upon
which to infer a coherent set of stylized facts about risk-reducing/
increasinginputs(Antle1983;GriffithsandAnderson1982;Halland
Moffit1985;JustandPope1979).Themostclear-cutevidenceseems
to come from experimental plots commonly studied by agricultural
experimentstationsthroughouttheworld,buttherearequestionsabout
how these data apply to actual farming experience under less con-
trolledsituations.
CROPINSURANCE
As economists have thought about the new economics of uncer-
tainty, one of the early insights was that insurance markets rationally
couldnotexistunlesscoercionwasinvolvedorunlesstherewasfree
choice with significant risk aversion. Excluding coercion, a risk-neu-
tral person will maximize expected wealth and therefore will pay at
mosttheexpectedlossduetoactsofnature.Thatis,ifthereisa0.001
probability that fire will destroy a $200,000 building in a given year,
the largest insurance premium a risk-neutral individual would pay is
the expected loss, E(L), which equals $200. Insurance provision
involves marketing, adjusting, and other monitoring costs denoted by
c.LetthistotalcostofinsuranceprovisionbeC=c+E(L).Noinsur-
ancemarketcouldexistunlesspeoplearewillingtopayatleastCfor
insurance. The amount an individual is willing to pay beyond E(L)is
called the risk premium, . The risk premium is zero for risk-neutral
individuals and positive for those who are risk-averse. Thus, a risk-
averseindividualiswillingandabletopurchaseinsuranceiftheprovi-
sioncostsarelessthanthepremium,orc .Theleftsideoftheine-
qualityisthesupplyprice,andtherightsideisthedemandprice.No
insurancemarketcanexist(forc>0)withoutcompulsionunlessmar-
ketparticipantsarerisk-aversesuchthattheyarewillingtopayforthe
costs,c.Anyrisk-averseindividualwouldsurelypurchasefairinsur-
ancewheretheinsurancepremiumisequaltotheexpectedloss.This
isasimpleinitialinsightintoanecessaryconditionfortheexistenceof
aninsurancemarket.
RiskandAgriculture 145
A second insight comes from the notion of insurability. Insurers
generallyarethoughttohavelittleexposuretoriskiftheyhavealarge
number of independent contracts. In this case, the payouts (indemni-
ties)willthusberemarkablypredictable(lowvariance).Thisisevident
from the law of large numbers in probability. Using this indemnity
data,theinsuranceproductcanbereadilypricedandmostofthecom-
petitiveassumptionscanensue.Theseconditionsforinsurabilityhold
for life insurance and fire insurance. These conditions rarely hold for
actsofnaturetoagriculture(hailinsuranceisanexception),whichcan
oftenbecatastrophic.InsurersofactsofnatureintheMidwestwould
have a highly correlated portfolio if the insured losses were due to
drought.Thus,theliabilitiescouldbelargeoneyearandlowthenext,
implyingahighvarianceofthereturn.Thismaymeanthattheproba-
bility of ruin for an insurer would be substantial, leading to risk-
responsive behavior by insuring firms (Duncan and Myers 2000).
However,therearereinsurancemarketsandothermeanstotradeaway
someoftheriskinariskyundertaking.
It appears that no multiple-peril private crop insurance markets
haveemerged(e.g.,seeGlauberandCollins2002).Dueeithertoissues
of insurability or just plain old rent seeking, policy has focused in
recent decades on the provision of federally organized and provided
cropinsurance.Toillustratetheessenceoftheprogram,afarmmight
select the 0.75 option. When yields are 75 percent of approved pro-
gramyield,thistriggersapaymentfromthegovernment.
In1980,theCropInsuranceImprovementActwaspassedinthe
UnitedStates,allowingtheprivatesectortosellmultiple-perilfederal
crop insurance (MPCI) with a subsidized premium. Since that time,
fiveadditionalactshavebeenpassedtoextendandreformthefederal
program.Federalsubsidieshaverisentoaround$1.4billion.Liabilities
havegrownsevenfoldsince1980toaround$35billion,alsoshowing
thetremendousgrowthintheprogram.Duringthe1980sandmuchof
the 1990s, the ratio of indemnities/total premiums (ignoring the gov-
ernment subsidy) or loss ratio was greater than 1, indicating that the
programwasactuariallyunsound.Becauseofenormouspolicyinterest
in the program, significant amounts of intellectual effort, computer
time,andinkwerespentstudyingcropinsurance.
Crop insurance has also been a focus of international attention as
countries around the world study the viability of similar programs
146 Pope
(bothCanadaandJapanhaveprogramssimilartotheUnitedStates;see
the bibliography of Coble and Knight 2002). Because of the federal
subsidy,thisinsuranceismorethanfairtosomefarmers.Thus,even
arisk-neutralfarmermaystrictlyprefertheinsurance.Thus,risktools
arerelevant,butinsurancepurchaseisntprimafacieevidenceofrisk
aversion,asitwouldbeinalaissez-fairemarketforinsurance.
Therearemanypossiblereasonsfortheexcesslosses,includingthat
governmentmaywishtotransferwealthtoagriculture.However,using
thebestavailableactuarialmethods,therearegoodreasonstoexpectthe
programtofail.Atleastpartoftheansweriswellknowntoeconomists.
Whenfarmershavemoreinformationthanthosesettingtherates(asym-
metric information), moral hazard and adverse selection may occur.
Thesewillbeexploredconceptuallyfirstandthenempirically.
AdverseSelection
Supposethatratemakershaveaccesstoaverageactuarialdataand
setwhatisknownasapooledrate,thatratewheretheaveragelossratio
is1.Supposealsothatthereisheterogeneity;thatis,somefarmshavea
highprobabilityoflossbelowtheinsuredlevel,whileotherfarmshave
alowprobabilityofloss.Farmsthataregoodrisks(lowprobabilityof
a payout) in that the probability of yields falling below the threshold
valueislow,willfindthepriceofinsurancetoohighbecauseitisbased
upontheaveragefarmsprobability.Theywillnotpurchasetheinsur-
ance.Farmsthatarepoorriskswillfindtheaveragerateattractiveand
willself-selectintotheinsuranceprogram.Thus,risksthatareadverse
to the long-run viability of the program select in and low risks select
out.Thisimpliesthatthegovernmentwilllosemoneyandmaywishto
raiserates.Ifratesareraised,someofthegood-riskfarmswillexitthe
program.Again,theprogramwilllosemoney.
Theincentivestopurchaseinsurancearenowthreefoldunderrisk-
aversion: 1) the incentive to participate based upon an increase in
expectedprofitduetothesubsidy,2)anincentiveduetoriskaversion
(reducedrisk),and3)anincentivedueadverseselection.Onewayto
calculatethethreeeffectsisasfollows:thefirstcalculatestheincrease
inexpectedprofitfrombeinginsured,thesecondcalculatesthediffer-
ence in the risk premium due to being insured, and the third follows
fromtheincreaseinexpectedindemnityduetoadverseselection.Each
RiskandAgriculture 147
oftheseprovidesincentivetoparticipate.Thegreaterthesubsidy,the
largerthefirstincentive.Greaterriskaversionimpliesgreaterincentive
topurchaseinsurance.Finally,fortheadverseselectioneffect,higher-
riskfirmswillbenefitbecauseoflargerexpectedindemnitiesthantyp-
ical. The third incentive will imply that the expected loss to insurers
increaseswithparticipationininsurancebyhigh-riskproducers.
Adverseselectionneednotbeaproblemiftheinsuranceprovider
canmonitororknowthenatureoftheheterogeneousfirm.Experience
ratingisanexampleoftryingtoadjustpremiumsforthetypeoffirm
demandinginsurance.
MoralHazard
Moral hazard implies another type of asymmetric information.
Here, knowledge of the insureds actions is hidden from the insurer
when comparing pre- and post-insurance behavior. The most extreme
formofhiddenactionisarson,butmoresubtlebehaviorsinvolvetak-
ing inappropriate care or effort. Antitheft devices might not be pur-
chasedifacarisfullyinsured.Regardinghealthinsurance,aninsured
personmightseeadoctormoreoftenthanifuninsured.ForMCPI,the
opportunities to change behavior if insured are many. Thus, a fourth
incentive to purchase insurance relates to moral hazard: fewer inputs
maybeappliedwheninsured.Thiswillsavecostsandwillincreasethe
probability of collecting indemnity payments. Again, if the provider
can monitor behavior and pay indemnities according to deviations
from best practice, moral hazard need not be an issue. Monitoring is
expensiveanddifficulttodo,exceptforobviousbehaviors.
Many policy proposals have tried to deal with the moral hazard
problem. One such program makes payments based upon area yields
ratherthanindividualyields.Inthiscase,adverseselectionandmoral
hazardarevirtuallyeliminated.However,theamountofinsurancethat
anindividualreceivesisdependentonhowthefarmoutcomesarecor-
related with the area outcomes. If a farm risk is largely idiosyncratic,
thentheareawideinsurancewillprovidelittlebenefittothefarm.
148 Pope
EmpiricalResults
Consider first the demand for insurance. Empirical work on crop
insurance demand has used simulation methods assuming particular
characteristicsandriskpreferencesofthefarm(e.g.,KramerandPope
1982;MappandJeter1988)oreconometrictechniques(e.g.,Gardner
and Kramer 1986; Goodwin 1993; Barnett and Skees 1995; Richards
2000; Vandeveer and Loehman 1994; Coble et al. 1996). A central
question is how does the demand for insurance respond to various
characteristics of the farm and the contract and insurance premiums?
These studies find that the demand for crop insurance is very price
(premium)inelasticdespitewidevariationincrops,regions,subsidies,
andinthenatureoftheprogram(contract).The1998adhocdisaster
relief bill provided for an additional 30 percent of subsidies for pre-
miumsubsidy.Studyingthischange,CobleandBarnett(1999)findthe
price elasticity of demand to be approximately 0.65 in terms of acres
insured.Thatis,a1percentdecreaseinpremiumswouldincreaseacres
insuredby0.65percent.
Empirical work on moral hazard and adverse selection is much
moredifficultthanmeasuringinsurancedemandelasticities.Anumber
ofstudiesfindsubstantialscopeforordirectevidenceofadverseselec-
tion (Goodwin 1994; Ker and McGowan 2000; Luo, Skees, and
Marchant1994;Just,Calvin,andQuiggin1999).Adverseselectionisa
large problem in the program for at least three reasons. First, farmers
can choose to participate knowing early spring soil moisture and
weatherforecasts.Forexample,soilmoistureatenrollmentandlong-
run weather forecasts can be beneficial. Using El Nio/La Nia
weather patterns can exacerbate the adverse selection problem for
insurers (Ker and McGowan 2000). This implies that farmers often
havemoreinformationthanratemakers.Second,thereisgreathetero-
geneity, and farmers may choose to insure particular parcels of their
land.Third,theU.S.programismarkedbyproceduresthatimplylarge
difficulties. For example, a farm without an approved yield history
could use the county average. If a farms yields were substantially
lowerthanthisaverage,therewouldbealargeindemnitypaidandrel-
ativelysmallpremiumreceived,leadingtoprogramlosses.
Regarding moral hazard, Horowitz and Lichtenberg (1993) esti-
mated chemical use for Midwest corn producers. They estimated that
RiskandAgriculture 149
insuranceparticipationinMPCIledtoincreaseduseofnitrogen,herbi-
cides, and pesticides. Smith and Goodwin (1996) examined Kansas
dryland wheat production and obtained opposite results. Firms pur-
chasinginsurancesignificantlyreducedtotalchemicalinput.Babcock
and Hennessy (1996) argued that using reasonable measures of risk
aversion and estimates of technology, insurance implied very modest
reductions in fertilizer usage. Coble et al. (1996) found evidence of
increased yield shortfalls for those insured. Taken as a whole, these
resultssuggestthatmoralhazardisapotentiallyseriousproblem.
Thereisalsoresearchthatsubstantiatesthattheuninsuredbehave
differentlyfromtheinsured,butthatdoesnotattributethistoaparticu-
lar explanation. Quiggin, Karagiannis, and Stanton (1993) examined
typicalrevenueandinputshareequationsandnotedthatrevenuewas
statisticallylessforinsuredfarms.Nocorrespondingsignificantresults
werefoundforinputs.Theimpactsoncropsgrown(likelymoralhaz-
ard)aresubstantialandlikelyclearer.Glauber(1999)estimatesthata
revenue insurance program for North Dakota durum wheat producers
led to a 25 percent increase in production. Wu (1999) estimates that
cropinsuranceforcorncausescornacreagetoincrease.Keaton,Skees,
andLong(1999)estimatethata10percentincreaseincropinsurance
participation increased an increased planted area of 6 major crops of
5.9millionacres.Thisisanunusuallylargeresponseandlikelyoveres-
timates the response to crop insurance alone (Glauber and Collins
2002). Goodwin and Vandeveer (2000) estimate a 2.23.3 percent
increaseincornandsoybeanacreageplanted.Orden(2001)estimates
thatthatwouldincreaseproductionby0.284.1percent.Finally,ifpro-
duction increases, price must fall. Babcock and Hart (2000) conclude
thattheeliminationofcropinsurancesubsidiesforcornwouldincrease
priceby$0.02$0.16perbushel.
To summarize, it seems that there is every expectation to believe
thatadverseselectionandmoralhazardwillbeaproblemintheMPCI.
The dates allowed for enrollment, the fact that separate fields can be
enrolled,andthedifficultiesofmonitoringcomplexbehaviorallcon-
tribute to these possibilities. Though the empirical research is not as
broadanduniformasdesired,theavailableevidencesuggeststhatthe
two economic problems identified with provision of insurance under
asymmetricinformationarealiveandwellinMPCI(CobleandKnight
2002).
150 Pope
As this section concludes, a question arises: Should the govern-
ment be insuring yields in the first place? First, if the elasticity of
demandisunity,yieldsmayvaryconsiderablybuttotalrevenue(price
times quantity) is fixed. This suggests that if policy wants to provide
somesafetynetforfarmsratherthantransferwealthtothem,thenrev-
enueinsurancemaybeapreferredpolicytoyieldinsurance.Second,it
is far from clear that there is a strong demand for agricultural crop
insurance.Thisistobedistinguishedfromastrongdemandforasub-
sidyortransfertofarmers.
HEDGING/FORWARDMARKETS
As discussed in virtually every textbook on economic theory or
practice, hedging can reduce exposure to risk. Examples abound of
markets for risk. Many commodities are listed on the Chicago Board
of Trade, and a number of instruments are relevant. Though using
futures markets is available, there is no reason to suspect that this is
the efficient mechanism to trade risk. Often the efficient mechanism
for a farm to shed risk is a forward contract. A forward contract is
merelyacontractatanegotiatedpricetodayfordeliveryinthefuture.
A futures market is an organized forward market specifying delivery
at a particular date, quantity, and grade of the commodity at a speci-
fiedplace(e.g.,Chicago).However,thebasicadvantagesandrisksof
farm hedging can be told equivalently with either a forward or a
futuresmarket.Ishallusethelatterbecauseitiscommonlydiscussed
inmosttextsinmicroeconomics.
Afarmerplantscorninthespringandknowsthatthefuturesprice
is$3.00forSeptembercorn.Thisisthecurrentpriceforfuturedeliv-
ery of corn in September. If the futures price converges to the actual
priceofcornonthespotmarket,thenwhenfallcomes,bothpriceswill
beequal.Thesepricesmightbeequalto$4.00or$2.00.Theyareran-
domwhenviewedfromthepointofviewofthefarmerinthespring.
RiskandAgriculture 151
Now consider the following three transactions for a bushel of corn
assumingthefallpriceis$x:
Spot Futures
Sell(spring) $3.00
Sell(fall) $x
Buy(fall) $x
Summing yields +$3.00. The farmer, by placing the hedge, has
received a certain $3.00 for corn rather than the random price $x. If
risk-averseandifthefuturesmarketisfairorunbiased(expectedspot
isthefuturesprice),farmerswouldsurelyprefertousethefuturesmar-
ket. The farmer could forget risk aversion and use the certain $3.00
pricesignaltodecidehowmuchacreagetoplantincorn.Summarizing
andgeneralizingtheaboveexample,farmprofitusingthefuturesmar-
ketcanbedescribedequivalentlyas:
(6.6)profit=totalrevenueinthespotmarketcostsofproduction+
(futurespricespotprice)(quantityofoutputhedged),
orequivalently,
(6.7)profit=(unhedgedoutput)(spotprice)costsofproduction+
(hedgedoutput)(futuresprice),
whenthefuturespriceconvergestothespotpriceatanypointintime.
The example and concepts discussed above bring about four
importantissues.First,theseparationresultofproductionfromhedg-
ingdoesnotextendtotheamounthedged;itdependsonthemagnitude
of risk aversion. However, when production itself is uncertain, the
farmer does not know how much of her crop is hedged by a given
quantity sold forward in the spring. Second, there is basis risk where
basis is the difference between the spot price and the futures price at
anypointintimeandintheplacewhereproductiontakesplace.Third,
how should a hedge change over time in reaction to new information
and what are the time series properties of prices? Dynamic or rolling
hedgesareanimportantissue.Finally,theremaybesubstantialtrans-
152 Pope
actionscostsinusingthefuturesmarket(fees/margincallsetc.).These
issuesarereviewedinsequence.
TheHedge
Whenproductioniscertain,whenthereisnobasisrisk,andwhen
firms are risk-averse, one of the first observations involving the opti-
mal hedge is that it will be less (greater) than output produced as the
expected spot price is greater (less) than the futures price. This states
that in order for a risk-averse firm to be rewarded for risk taking by
sellingmorefuturescontractsthanithasoutput(speculating),itmust
betruethatthefuturespriceisgreaterthanexpectedspotprice.When
thefuturesorforwardpricepredictsunbiasedlythespotprice,output
willbecompletelyhedgedbecausethereisnoincentivetospeculatein
either the cash or futures markets. Thus, a key question is whether
futuresorforwardpricesunbiasedlypredictspotprices.Theavailable
evidenceismixed.However,acrossmanycommoditiesandcountries,
my reading of the evidence suggests that when spot prices are not
longer than 36 months out, futures prices are unbiased estimates of
futurespotprices.
ProductionUncertainty
Whenproductionisuncertain,thecorrelationbetweenproduction
andpriceuncertaintyiscrucialtoanyanalysis.Forafarmerproducing
in the corn-belt, this correlation is likely significantly negative. To
illustratewhythiscovariancematters,consideracommondescription
of technology where production shocks enter production multiplica-
tively.Whenexpectedproductionisexpanded,themarginalbenefitin
termsofexpectedprofitisexpectedpriceplusthecovariancebetween
theproductionshockandprice.Becausewepresumethiscovarianceis
negative,firmswillproducelessoutputbecausethemoreoutputpro-
duced,thegreaterthereductioninprofitonaverage.Further,increas-
ing the scale of production will increase the variance because the
varianceofprofitisproportionaltothescalesquared.Nowweaskhow
thepossibilityofaforwardorfuturescontractaffectshedgingandpro-
ductionchoice.
RiskandAgriculture 153
Many authors have used mean-variance notions to calculate the
optimalhedgewhenbothproductionandthepriceofoutputareuncer-
tain.Becausepriceandproductiontendtobenegativelycorrelated,the
optimal hedge under risk aversion is generally found to be less than
expectedfarmproduction.
Dynamics
Theoptimaldynamichedgedependscruciallyontheevolutionof
prices(time-seriesproperties)andwhethertheyareunbiased.Ifprices
areunbiasedandproductioniscertain,thenreasonablysimpledynam-
icsareimpliedintheoptimalhedge(orratioofhedgetoproduction)in
most cases (Myers and Hanson 1996). When production is uncertain,
thenstrongassumptionsarerequiredinordertomakemuchheadway
onsolvingtheproblem(seereferencesinMyersandHanson).
UseofFuturesandForwardMarkets
MoschiniandHennessy(2001),citingareportfromtheU.S.Gen-
eralAccountingOffice(1999),statethattheavailableevidenceisthat
farmersusefuturesmarketssomebutuseforwardmarketsfrequently.
For farmers with sales exceeding $100,000, forward contracts were
usedby55percentoffarmsandfuturescontractsoroptionswereused
by32percentoffarms.Patrick,Musser,andEckman(1998)surveyed
large, well-educated, progressive Indiana farmers over a three-year
period on their use of forward and futures markets. Those who used
some form of forward contracts exceeded 75 percent. Use of futures
marketstohedgewaslimitedtolessthan25percentforcornandsoy-
beans and usually was less than 15 percent. My interpretation of the
generaltoneofmuchagriculturalextensionworkseemstobe:futures
areariskreductiontoolthathasbeenunder-exploited.However,itis
a very costly and imperfect mechanism for trading risk compared to
forward contractsparticularly where a large purchasing entity can
usethefuturesmarkettolockinpriceandthenextendforwardcon-
tractstofarmers.
154 Pope
FOODSAFETY
Consumerconfidenceaboutfoodsafetyhasfallenprecipitouslyin
recent times (Kramer 1990). This is likely due to highly publicized
occurrences in the 1980s and 1990s. The Alar and Chilean grapefruit
scaresareexamplesofconcernaboutchemicalresiduesonproduce.In
1993, an Escherichia coli outbreak in several fast-food restaurants
sickenedhundredsofpeopleandresultedinfourdeaths.Inthesummer
of1997,therewasamuch-publicizedcasewhere25millionpoundsof
hamburger were produced with suspected E. coli contamination. The
Centers for Disease Control and Prevention (CDC) estimate that
between6.5and33millionpeopleintheUnitedStatesbecomeilleach
yearfromfood-bornepathogens,andthatupto9,000die(Buzbyetal.
1998).Ofthesecasesofillness,morethan4,000deathsmaybeassoci-
ated with meat and poultry products. In addition, chemical residues
fromfertilizer,herbicides,andpesticidesmayposelong-termrisksto
thepublic.
Safety policy is concerned with the delivery of existing foods
within some level of confidence that it is safe. It also extends to new
foodstuffssuchasgeneticallymodifiedorganisms.Weexpectthatthe
usual marginal benefitmarginal cost calculations inform decision
making: absent externalities, the optimal level of care or safety is
where the marginal private benefit equals marginal private costs. The
marginal benefit could be modeled with expected utility or a mean
variance utility and the willingness to pay for each additional unit
increaseinsafety.Apparently,however,therearesignificantexternali-
ties to other firms and consumers if a firm chooses a low level of
safety. Thus, because the optimal level of safety is where marginal
socialbenefitequalsmarginalsocialcost,privateincentivesasembod-
iedinsupplyanddemandmaynotleadtothesocialoptimum.Contrary
to the rhetoric often heard, this optimum will most often allow for
somecontamination/risk.
Though measurement of each of these entities is not easy to do
well,therehavebeennumerousattemptstoshedlightonthecostsand
benefitsofapolicyproposal.Thecostsarerelativelyeasytoconceptu-
alizeandcalculate.Thesearetheadditionalcoststofirmswhensafety
isefficientlyincreased.ForarecentpolicychangebytheFoodSafety
RiskandAgriculture 155
and Inspection Service (FSIS) called Hazard Analysis and Critical
Control Points (HACCP), the costs are estimated to be at least $100
million annually. Antle (2000), studying meat-processing plants,
arguesthattheseestimatedcostsaremuchtoosmallduetothelossin
productiveefficiencyinvolvedincomplyingwiththeHACCPregula-
tions.Anyattempttomeasurethecostsofaregulationmustcountboth
thedirectcostsoftheprogramandtheindirectcostsduetothelossin
productiveefficiency.
Many estimates of the benefits have been much larger: $3.7 bil-
lion$19.1billion,dependingonquantityandtypeofpathogensame-
lioratedandassumptionsaboutthevalueoflife.Cuttingsomecorners,
theconceptualnotionofwillingnesstopay(WTP)forfoodsafetycan
beillustratedusingEquation(6.2)foraconsumer.Letaconsumerbe
givenachoicebetweentwoprobabilitydistributions.Thecurrentdis-
tribution possessed has a mean of $100 and a variance of a 500. The
seconddistributionhasameanof$96andavarianceof400.TheWTP
isthevaluethatequatesthefollowingutilities:
U(100,500)=U(96WTP,400).
(Amorerealisticdepictionwouldembodynottwomeansorvariances
but two probability distributions of contamination.) It is the purchase
or demand price for the second probability distribution given that the
individualpossessesthefirstone.Ingeneral,itcanbepositiveornega-
tive. The Food Safety and Inspection Service estimates of the yearly
publicbenefitsusingthecostofillnessmethod(discussedbelow)are
$990 million$3.7 billion. This wide range of numbers immediately
suggeststhedifficultyofmeasuringconsumerbenefitsfortheUnited
States.
Thefourmethodsusedtoestimatebenefitsare:1)askpeopleina
survey(contingentvaluation,orCVmethod)howtheywouldvaluean
increaseinsafety,2)useexperimentalauctionstotrytoevaluatecon-
sumerswillingnesstopayforimprovedsafety(experimentalmethod),
3)usecostofillnessorliabilityasmeasureofconsumerbenefits,and
4)directeconometricestimationoftheshiftindemandfunctionscon-
trollingforotherfactors(Caswell1998;Buzbyetal.1998).
Afewintroductorycommentswillserveasbackground.Attempts
to measure econometrically the effects of food safety on consumer
156 Pope
demand are fraught with measurement problems and often cannot
applytoaprospectiveprogram.Ifonehadmeasuresoffoodcharacter-
istics, including safety attributes, then a regression of price on food
characteristics (hedonic regression) could yield the WTP for safety
changes. For example, as safety varied, the economist could measure
theeffectonprice.Thismarginaleffectonpricecouldbeusedtoinfer
WTP. However, one seldom has such data. Yet, it may be possible to
measuretheimpactofinformationorsafetyondemand.Further,some
economists make a distinction between safety claims by a manufac-
turerandscientificsupportableclaims.Thatis,ifamanufacturerlabels
eggswithaparticularlylowprobabilityofSalmonella,andcharges$x
more for them, is that the correct measure of the social value of
improvedsafetyirrespectiveofscientificevidenceofefficacy?
ApplicationsoftheMethods
Inthefirstmethod,surveyselicitaresponsetoahypotheticalenvi-
ronment. For example, one might propose a baseline probability of
foodpoisoningandseverityandasktherespondentwhattheywouldbe
willingtopayforaparticularscenarioofrisk/severityreduction.This
is conceptually the most direct and appealing method, although there
may not be sufficient incentives and context for respondents to be
truthful. The second method need not rely on hypothetical scenarios,
buttheexperimentmaynotberepresentativeofactualdecisionmaking
by the population at large (a sampling problem), or the experiment
itselfmaynotrepresentthecomplexityoftheenvironmentandchoice.
The third method often is not necessarily linked to WTP or social
value.Forexample,thecostofillnessmaynotincludepainandsuffer-
ingandmaymissthelong-termconsequencesofillnessongrowthand
development.Liabilitymaybeabettermeasure,butitisnotveryhelp-
fulforaprospectiveevaluationofapolicy.Eitherthecostofillnessor
producer liability likely underreports the WTP for improved food
safety.
Theempiricalfindingsareinterestingbutoftendonotyetyielda
precise and consistent pattern (Shogren et al. 1999). Buzby et al.
(1998)discussthefollowingCVexperiment.StoreAisaconventional
U.S.groceryoutlet,butstoreBeliminatesorreduces,throughtesting,
the amount of pesticide residues on fresh produce. Store A is called
RiskandAgriculture 157
pesticide-free and store B is set to government residue standards.
Demographicvariablesandariskindexthattherespondentestimates
are included in the regressions. The only demographic variable that
wasstatisticallysignificantwasgender:womenaremorelikelytoshop
at store B and have a higher WTP. As expected, those who estimated
therisksfromresiduesasbeinghighwerealsomorelikelytoshopat
storeBandhaveahigherWTP.ThemedianweeklyWTPsforagov-
ernmentstandardstoreandapesticide-freestorewere$5.31perweek
and $5.88, respectively. Buzby, Ready, and Skees (1995) used CV to
measurethecostsandbenefitsofeliminatingapost-harvestchemical
sodium ortho-phenylphenate (SOPP) from use on Florida grapefruit
designatedforthefreshmarkets.Sodiumortho-phenylphenateisafun-
gicide that reduces molds and rots but is perceived by consumers to
havehealthrisks.Aftercalculatingthecosts(lostfruit)totheindustry
from the ban, CV is used to calculate the WTP. Average WTP was
between $0.19 and $0.28, depending on what one assumed about the
WTP to nonrespondents. On average, respondents are willing to pay
about 38 percent more for SOPP-free grapefruit. Regression analysis
found no significant evidence that household size, race, or gender
affected WTP. More affluent and older people were found to have a
lowerWTP.vanRavenswaayandHoehn(1991)foundthatconsumers
were willing to pay about 17 percent of the current purchase price to
avoidAlarinfreshapples.
In a typical experimental market, participants are given a choice
betweenachickensandwichwiththeusualchanceofcontaminationby
Salmonella(probabilityofcontaminationmaynotbespecified)ifpur-
chasedatalocaloutletandasandwichthatisscreenedandisreported
to have 1/1,000,000 chance of contamination. Bids are in increments
over the price of the sandwich with the usual risk of contamination.
SimilarexperimentshavebeendoneinArkansas,Massachusetts,Iowa,
and California. Incentives are put in place to obtain relevant bids. In
Arkansas and Massachusetts, average bids often exceeded $1, but in
Iowa,California,andKansastheaveragewasapproximately$0.55or
lessforagivenrunoftheexperiment.Itisunclearhowoneextrapo-
latesthistoacountrywidecost/benefitcalculationwhichincludesnon-
studentparticipants.
The essence of the methodological difficulty involving eliciting
WTP is found in the excellent experimental study of Shogren et al.
158 Pope
(1999). They designed an experiment where subjects chose irradiated
or nonirradiated chicken breasts. The price of nonirradiated chicken
breasts was held constant at $2.88/lb. The price of the irradiated
chicken breasts varied from a 10 percent discount ($2.59/lb.) to a 20
percent premium ($3.45/lb.). The first experiment involved actual
retailmarkettrialswithclearlabelingandprominentdisplayofUSDA
summarydataonfoodirradiation.Inthesecondexperiment,anexperi-
mental auction was conducted. A budget of $30 was offered, and the
participants were asked to spend approximately $5.00 and keep the
rest.Briefly,afterprovidingeachparticipantwiththeUSDAsummary
data on irradiation, each participant responded with their preferred
choice.Thefinalexperimentwasarandomsampleof400households
where the survey requested information on purchase behavior given
thesamechoicesasintheretailandexperimentalmarkets.Inthelatter
case,amuchmorerichsetofattitudinal,experience,anddemographic
datawereavailabletotheresearchers.
There was general agreement among all three approaches in that
the demand for irradiated chicken is downward sloping. However,
informing market participants with the best available scientific infor-
mation(whichisgenerallysupportivetohigherhealthandsafetywith
irradiation) led a significant percentage of customers to demand a 10
percent discount on irradiated chicken. Further, in this category
(requiringa10percentdiscount),therewasareasonablylarge(greater
than33percent)differenceamongthethreemethodsinthepercentage
thatwouldpurchasethetreatedchicken.Thenatureandexplanationof
these anomalies are part of an ongoing debate (e.g., Bockstael 1999).
When it comes to the value of human life and safety, there are many
methodologicalandpolicyissues(Hooker,Nayga,andSiebert1999).
The concluding question arising from these experiments is: Based
upon available information, is consumer sovereignty to be respected
even if tastes and preferences conflict with accepted scientific evi-
dence?
Thelastmethodisbaseduponsecondarydata.Henneberry,Piew-
thongngam,andQiang(1999)triedtomeasureariskinformationvari-
ableandplaceditinasystemofdemandsfor14majorfreshproduce
categories.Theriskinformationvariablewasseldomstatisticallysig-
nificant but suggested an average percentage elasticity of 0.050.07
percentduetoamarginaldecreaseinriskinformation.Forexample,a
RiskandAgriculture 159
1percentincreaseintheriskinformationindexreducedcrucifers,car-
rots,andfoliageconsumptionbyanaverageof0.07percent.However,
ifhigh-frequencydataareusedwithaspecificrisk,itappearsthatone
can establish through event studies the impact of contamination on
prices.Forexample,forspecificUSDAE.coliO157:H7recalls,McK-
enzie and M.R. Thomsen (2001) established that prices (using daily
prices) for boneless beef react significantly to the recalls. This is the
most likely category affected by the bacteria. However, no such rela-
tionshipcanbeestablishedforthemoreaggregativecategoriesoflive
cattleandboxedbeefprices.
CONCLUSION
Itisadauntingtasktotrytosummarizethecontentofriskresearch
inagriculture.Largeareasofagriculturaleconomicresearchhavebeen
neglected:adoptionoftechnology,storage,gradingandstandards,con-
tracting, environmental risks, finance, and others. Risk research per-
vades agricultural economic research because risk is pervasive in
agriculture.Biologicalandphysicalprocesses(suchasweather)areso
complexthatriskisoftentreatedasendemic.Thisisnottheonlyway
to view research. Perhaps more investment should be made to under-
stand these biological processes so that deterministic methods can be
coherently employed. My conclusion is that risk research in agricul-
tural economics has been a very fruitful intellectual endeavor. How-
ever,asislikelyapparentthroughoutthischapter,Iamnotsurethatthe
profession has invested sufficient attention to carefully measuring
behavior. Normative prescriptions to government or individuals are
likelytomisleadifthereisnofirmgroundinginbehavioralsocialsci-
enceknowledge.Tobesure,therearesomerisk-relatedstylizedfacts
such as the econometric response of enterprise choices to changing
risk.However,thereismuchmoreworktobedoneinordertounder-
standwhethermanycurrentinterpretationsofresearchresultsbasedon
aggregatedatarestsonfirmmicro-foundations.
160 Pope
References
Allen,D.W.,andD.L.Lueck.1998.TheNatureoftheFarm.JournalofLaw
andEconomics41(2):343386.
AndrewsD.W.K.2001.TestingWhenaParameterisontheBoundaryofthe
MaintainedHypothesis.Econometrica69(3):683734.
Antle,J.M.1983.TestingtheStochasticStructureofProduction:AFlexible
MomentbasedApproach.JournalofBusinessandEconomicStatistics
1(3):192201.
.1987.EconometricEstimationofProducersRiskAttitudes.Amer-
icanJournalofAgriculturalEconomics69(3):509522.
.2000.NoSuchThingasaFreeSafeLunch:TheCostofFoodSafety
RegulationintheMeatIndustry.AmericanJournalofAgriculturalEco-
nomics82(2):310322.
Aradhyula,S.,andM.Holt.1989.RiskBehaviorandRationalExpectations
intheU.S.BroilerMarket.AmericanJournalofAgriculturalEconomics
71(4):892902.
Babcock,B.A.,andD.A.Hennessy.1996.InputDemandUnderYieldand
RevenueInsurance.AmericanJournalofAgriculturalEconomics78(2):
416427.
Babcock,B.,andC.Hart.2000.ASecondLookatSubsidiesandSupply.
IowaAgriculturalReview6(4):3.
Barnett,B.J.,andJ.R.Skees.1995.RegionandCropSpecificModelsofthe
DemandforFederalMultiplePerilCropInsurance.JournalofInsurance
Issues19(2):4765.
Bar-Shira,Z.,Just,R.E.,andD.Zilberman.1997.EstimationofFramers
RiskAttitude:AnEconometricApproach.AgriculturalEconomics17:
(23):211222.
Behrman,J.R.1968.SupplyResponseinUnderdevelopedAgriculture:ACase
StudyofFourMajorCropsinThailand,193763.Amsterdam:North-Hol-
landPublishingCo.
Bockstael,N.1999.ValuingtheBenefitsofMicrobialFoodSafetyRisk
Reduction:Discussion.AmericanJournalofAgriculturalEconomics
81(5):12001204.
Brorsen,W.B.,Chavas,J-P,andW.R.Grant.1985.MarketingMarginsand
PriceUncertainty:TheCaseoftheU.S.WheatMarket.AmericanJour-
nalofAgriculturalEconomics67(3):521528.
Buzby,J.C.,R.C.Ready,andJ.Skees.1995.ContingentValuationinFood
PolicyAnalysis:ACaseStudyofaPesticide-ResidueRiskReduction.
JournalofAgriculturalandAppliedEconomics27(2):613625.
RiskandAgriculture 161
Buzby,J.C.,J.A.Fox,R.C.Ready,andS.R.Crutchfield.1998.Measuring
ConsumerBenefitsofFoodSafetyRiskReduction.JournalofAgricul-
turalandAppliedEconomics30(1):6982.
Carlson,G.A.1979.Insurance,Information,andOrganizationalOptionsin
PestManagement.Ann.Rev.Phytopath.17:149161.
Castle,E.1954.FlexibilityandDiversificationasaMeasureofMeeting
PriceandYieldUncertaintyinWesternKansas.JFE36:273284.
Caswell,J.A.,ed.1995.ValuingFoodSafetyandNutrition.Boulder:West-
viewPress.
.1998.ValuingtheBenefitsandCostsofImprovedFoodSafetyand
Nutrition.TheAustralianJournalofAgriculturalandResourceEconom-
ics42(4):409424.
Chavas,J.P.,andM.T.Holt.1996.EconomicBehaviorunderUncertainty:A
JointAnalysisofRiskPreferencesandTechnology.ReviewofEconomics
andStatistics78(2):329335.
Coble,K.H.,andB.J.Barnett.1999.PreliminaryEvidenceofCropInsurance
DemandElasticityfromthe1999ResponsetoAdditionalSubsidies.
DepartmentofAgriculturalEconomics,MississippiStateUniversity.
Availableat:<http://www.agecon.msstate.edu/risk/documents>.
Coble,K.H.,andT.O.Knight.2002.CropInsuranceasaToolforPriceand
YieldRiskManagement.InAComprehensiveAssessmentoftheRoleof
RiskinU.S.Agriculture,R.E.JustandR.D.Pope,eds.Boston:Kluwer
AcademicPublisher,pp.445468.
Coble,K.H.,T.O.Knight,R.D.Pope,andJ.Williams.1996.Modeling
Farm-LevelCropInsuranceDemandwithPanelData.AmericanJournal
ofAgriculturalEconomics78(2):439447.
Coyle,B.1999.RiskAversionandYieldUncertaintyinDualityModelsof
Production:AMean-VarianceApproach.AmericanJournalofAgricul-
turalEconomics81(3):553567.
Davis,S.B.,T.J.Price,M.Wetzstein,andM.W.Rieger.1997.Reducing
YieldVariationinPeachOrchardsbyGeographicScattering.American
JournalofAgriculturalEconomics79(4):11191126.
Duffy,P.A.,Shalishali,K.andH.W.Kinnucan.1994.AcreageResponse
underFarmProgramsforMajorSoutheasternFieldCrops.Journalof
AgriculturalandAppliedEconomics26(2):367378.
Duncan,J.andR.J.Myers.2000.CropInsuranceunderCatastrophicRisk.
AmericanJournalofAgriculturalEconomics82(4):842855.
DunnJ.W.,andJ.R.Williams.2000.FarmCharacteristicsthatInfluenceNet
FarmIncomeVariabilityandLossesPaperpresentedattheWesternAgri-
culturalEconomicsAssociationAnnualMeetings,Vancouver,British
Columbia,June29July1.
162 Pope
Estes,E.A.,Blakeslee,L.L.,andR.C.Mittelhammer.1981.OnVariancesof
ConditionalLinearLeasto-SquaresSearchParameterEstimates.Ameri-
canJournalofAgriculturalEconomics(February):141145.
Gardner,B.L.,andR.A.Kramer.1986.ExperiencewithCropInsurancePro-
gramsintheUnitedStates.InCropInsuranceforAgriculturalDevelop-
ment:IssuesandExperience,P.Hazell,C.PomaredaandA.Valdes,eds.
Baltimore:JohnHopkinsPress,pp.195222.
Glauber,J.W.1999.DeclarationintheU.S.DistrictCourtfortheDistrictof
NorthDakota,SoutheasternDivision.TestimonygiveninPaulWileyet
al.DanielGlickman,DivilNo.A3-99-32(May18).
GlauberJ.W.,andK.J.Collins.2002.RiskManagementandtheRoleofthe
FederalGovernment.InAComprehensiveAssessmentoftheRoleofRisk
inU.S.Agriculture,R.E.JustandR.D.Pope,eds.Boston:KluwerAca-
demicPublisher,pp.469488.
Goodwin,B.K.1993.AnEmpiricalAnalysisoftheDemandforCropInsur-
ance.AmericanJournalofAgriculturalEconomics75(2):425434.
.1994.PremiumRateDeterminationintheFederalCropInsurance
Program:WhatDoAveragesHavetoSayAboutRisk?JournalofAgri-
culturalandResourceEconomics19(2):382395.
Goodwin,B.K.,andM.Vandeveer.2000.AnEmpiricalAnalysisofAcreage
DistortionsandParticipationintheFederalCropInsuranceProgram.
PaperpresentedattheEconomicResearchServiceworkshop,CropInsur-
ance,LandUse,andtheEnvironment,heldinWashington,DC,Septem-
ber1220.
GriffithsW.E.,andJ.R.Anderson.1982.UsingTime-SeriesandCross-Sec-
tionDatatoEstimateaProductionFunctionwithPositiveandNegative
MarginalRisks.JournaloftheAmericanStatisticalAssociation77(Sep-
tember):529536.
Hall,D.C.,andL.J.Moffitt.1985.ApplicationoftheEconomicThreshold
forInterseasonalPestControl.WesternJournalofAgriculturalEconom-
ics10(December):223229.
Heady,E.O.1952.DiversificationinResourceAllocationandMinimization
ofIncomeVariability.JournalofFarmEconomics34:482496.
Henneberry,S.R.,K.Piewthongngam,andH.Qiang.1999.ConsumerFood
SafetyConcernsandFreshProduceConsumption.JournalofAgricultural
andResourceEconomics24:98113.
Holt,M.T.1994.Price-BandStabilizationProgramsandRisk:AnApplica-
tiontotheU.S.CornMarket.JournalofAgriculturalandResourceEco-
nomics19(2):239254.
RiskandAgriculture 163
Holt,M.T.,andS.V.Aradhyula.1990.PriceRiskinSupplyEquations:An
ApplicationofGARCHTime-SeriesModelstotheU.S.BroilerMarket.
SouthernEconomicJournal57(1):230242.
.1998.EndogenousRiskinRational-ExpectationsCommodityMod-
els:AMultivariateGeneralizedARCH-MApproach.JournalofEmpiri-
calFinance5:99129.
Holt,M.T.,andG.Moschini.1992.AlternativeMeasuresofRiskinCom-
monditySupplyModels:AnAnalysisofSowFarrowingDecisionsinthe
UnitedStates.JournalofAgriculturalandResourceEconomics17(1):1
12.
Hooker,N.H.,R.M.Nayga,Jr.,andJ.Siebert.1999.PreservingandCommu-
nicatingFoodSafetyGains.AmericanJournalofAgriculturalEconomics
81(5):11021106.
Horowitz,J.K.,andE.Lichtenberg.1993.Insurance,MoralHazard,and
ChemicalUseinAgriculture.AmericanJournalofAgriculturalEconom-
ics75(5):926935.
.1994.Risk-ReducingandRisk-IncreasingEffectsofPesticides.
JournalofAgriculturalEconomics45(1):8289.
Hurd,B.H.1994.YieldResponseandProductionRisk:AnAnalysisofInte-
gratedPestManagementinCotton.JournalofAgriculturalandResource
Economics19(2):313326.
Hurt,C.,andP.Carcia.1982.TheImpactofPriceRiskonSowFarrowing,
196778.AmericanJournalofAgricultureEconomics64(3):565568.
Jensen,C.W.1961.SpatialDiversificationoftheDrylandFarmUnit.Man-
agementStrategiesinGreatPlainsFarming,NebraskaAgriculturalExper-
imentStation,MP7:9195.
Johnson,S.R.1967.ReexaminationoftheFarmDiversificationProblem.
JournalofFarmEconomics49:610621.
Just,R.E.1974.AnInvestigationoftheImportanceofRiskinFarmers
Decisions.AmericanJournalofAgriculturalEconomics56:1425.
Just,R.E.,L.Calvin,andJ.Quiggin.1999.AdverseSelectioninU.S.Crop
Insurance:ActuarialandAsymmetricInformationIncentives.American
JournalofAgriculturalEconomics80(4):834849.
Just,R.E.,andR.D.Pope.1979.ProductionFunctionEstimationandRelated
RiskConsiderations.AmericanJournalofAgriculturalEconomics61:
276283.
.2001.TheAgriculturalProducer:TheoryandStatisticalMeasure-
ment.InHandbookofAgriculturalEconomics,B.L.GardnerandG.C.
Rausser,eds.Amsterdam:Elsevier,pp.629741.
Just,R.E.,andR.D.Pope,eds.2002.AComprehensiveAssessmentoftheRole
ofRiskinU.S.Agriculture.Boston:KluwerAcademicPublishers.
164 Pope
Keaton,K.,J.Skees,andJ.Long.1999.ThePotentialInfluenceofRisk
ManagementProgramsonCroppingDecisions.Paperpresentedatthe
annualmeetingofTheAmericanAgriculturalEconomicsAssociation,
heldinNashville,TN,August811.
Ker,A.P.,andP.McGowan.2000.Weather-BasedAdverseSelectionandthe
U.S.CropInsuranceProgram:ThePrivateInsuranceCompanyPerspec-
tive.JournalofAgriculturalandResourceEconomics25(2):386410.
Kramer,C.S.1990.FoodSafety:TheConsumerSideoftheEnvironmental
Issue.SouthernJournalofAgriculturalEconomics22(1):3340.
Kramer,R.,andR.D.Pope.1982.CropInsuranceforManagingRisks.The
JournaloftheAmericanSocietyofFarmManagersandRuralAppraisers
46(1):3440.
Krause,M.A.,andW.W.Koo.1996.AcreageResponsestoExpectedReve-
nuesandPriceRiskforMinorOilseedsandProgramCropsintheNorthern
Plains.JournalofAgriculturalandResourceEconomics21(2):309324.
Krause,M.A.,Lee,J-H,andW.W.Koo.1995.ProgramandNonprogram
WheatAcreageResponsestoPricesandRisk.JournalofAgriculturaland
ResourceEconomics20(1):96107.
Lin,W.1977.MeasuringAggregateSupplyResponseUnderInstability.
AmericanJournalofAgriculturalEconomics(December):903907.
Lin,W.,G.W.Dean,andC.V.Moore.1974.AnEmpiricalTestofUtilityvs.
ProfitMaximuminAgriculturalProduction.AmericanJournalofAgri-
culturalEconomics56:497508.
Love,A.,andS.Buccola.1991.JointRiskPreference-TechnologyEstima-
tioninaPrimalSystem.AmericanJournalofAgriculturalEconomics73:
765774.
Luo,H.,J.R.Skees,andM.A.Marchant.1994.WeatherInformationandthe
PotentialforInter-temporalAdverseSelection.ReviewofAgricultural
Economics16:441451.
Mapp,H.P.,andK.L.Jeter.1988.PotentialImpactofParticipationinCom-
modityProgramsandMultiplePerilCropInsuranceonaSouthwestOkla-
homaFarm.InMultiplePerilCropInsurance:ASCollectionofEmpirical
Studies.SouthernCooperativeSeriesBulletinNo.34,Washington,DC.
Marshall,A.[1890]1920.PrinciplesofEconomics.8thed.London:Mac-
millanandCo.,Ltd.
McCloskey,D.1976.TheOpenFieldsasBehaviorTowardsRisk.Research
inEconomicHistory:AnAnnualCompilationofResearch.Vol.1.Paul
Uselding,ed.Greenwich:JAIPress.
McKenzie,A.M.,andM.R.Thomsen.2001.TheEffectofE.ColiO157:H7
onBeefPrices.JournalofAgriculturalandResourceEconomics26(2):
431444.
RiskandAgriculture 165
Meyer,J.1987.Two-MomentDecisionModelsandExpectedUtilityMaxi-
mization.AmericanEconomicReview77(3):421430.
Mishra,A.K.,andB.K.Goodwin.1997.FarmIncomeVariabilityandthe
SupplyofOff-FarmLabor.AmericanJournalofAgriculturalEconomics
79(3):880887.
Mishra,A.K.,andM.J.Morehart.2001.Off-FarmInvestmentofFarm
Households:ALogitAnalysis.AgriculturalFinanceReview61:87101.
MoschiniG.,andD.A.Hennessy.2001.Uncertainty,RiskAversion,andRisk
ManagementforAgriculturalProducers.InHandbookofAgricultural
Economics,Vol.1A,B.L.GardnerandG.C.Rausser,eds.Amsterdam:
North-Holland,pp.87153.
Myers,R.J.1989.EconometricTestingforRiskAverseBehaviourinAgri-
culture.AppliedEconomics21(4):541552.
Myers,R.J.andS.D.Hanson.1996.OptimalDynamicHedginginUnbiased
FuturesMarkets.AmericanJournalofAgriculturalEconomics78(1):13
20.
Nelson,R.G.,andD.A.Bessler.1989.SubjectiveProbabilitiesElicited
UnderProperandImproperScoringRules:ALaboratoryTestofPredicted
Responses.AmericanJournalofAgriculturalEconomics71(2):363369.
Officer,R.R.,andA.N.Halter.1968.UtilityAnalysisinaPracticalSetting.
AmericanJournalofAgriculturalEconomics50:257277.
Orden,D.2001.ShouldThereBeaFederalIncomeSafetyNet?Paperpre-
sentedattheAgriculturalOutlookForum2001,heldinWashington,DC,
February22.
Patrick,G.F.,W.N.Musser,andD.T.Eckman.1998.ForwardMarketing
PracticesandattitudesofLarge-ScaleMidwesternGrainProducers.
ReviewofAgriculturalEconomics20:3853.
Pope,R.D.1976.AnAnalysisofFactorsAffectingFarmDiversification:
WithSpecialReferencetoSanJoaquinValleyCropFarms.Unpublished
Ph.D.Dissertation.Berkeley:DepartmentofAgriculturalEconomics,Uni-
versityofCalifornia.
Pope,R.,andR.Prescott.1981.SupplyResponseandtheDispersionof
Expectations.AmericanJournalofAgriculturalEconomics63(February):
161163.
.1980.Diversification,inRelationtoFarmSizeandOtherSocioeco-
nomicCharacteristics.AmericanJournalofAgriculturalEconomics
62(August):554559.
Pope,R.,andR.Just.1991.OnTestingtheStructureofRiskPreferencesin
AgriculturalSupplyAnalysis.AmericanJournalofAgriculturalEconom-
ics73(3):743748.
166 Pope
Quiggin,J.,G.KaragiannisandJ.Stanton.1993.CropInsuranceandCrop
Production:AnEmpiricalStudyofMoralHazardandAdverseSelection.
AustralianJournalofAgriculturalEconomics37(2):95113.
Regev,U.,N.Gotsch,andP.Rieder.1997.AreFungicides,Nitrogenand
PlantGrowthRegulatorsRisk-Reducing?EmpiricalEvidenceFromSwiss
WheatProduction.JournalofAgriculturalEconomics48(2):167178.
Richards,T.J.2000.ATwo-StageModelfortheDemandforSpecialtyCrop
Insurance.JournalofAgriculturalandResourceEconomics25(1):177
194.
Robison,L.J.,andP.J.Barry.1987.TheCompetitiveFirmsResponsetoRisk.
NewYork:Macmillan.
Roumasset,J.A.1976.RiceandRisk.Amsterdam:North-HollandPublishing
Company.
Saha,A.,C.R.Shumway,andH.Talpaz.1994.JointEstimationofRiskPref-
erenceStructureandTechnologyUsingtheExpo-PowerUtility.American
JournalofAgriculturalEconomics76(2):173184.
Samuelson,P.A.1967.AGeneralProofthatDiversificationPays.Journal
ofFinancialandQuantitativeAnalysis2(1):113.
Schultz,T.W.1980.NobelLecture:TheEconomicsofBeingPoor.Journal
ofPoliticalEconomy88(4):639651.
Shogren,J.F.,J.A.Fox,D.J.Hayes,andJ.Roosen.1999.ObservedChoices
forFoodSafetyinRetail,Survey,andAuctionMarkets.AmericanJour-
nalofAgriculturalEconomics81(5):11921199.
Smith,V.H.,andB.K.Goodwin.1996.CropInsurance,MoralHazard,and
AgriculturalChemicalUse.AmericanJournalofAgriculturalEconomics
78(2):428438.
Thompson,G.D.,andP.N.Wilson.1994.CommonPropertyasanInstitu-
tionalResponsetoEnvironmentalVariablility.ContemporyEconomic
Policy12(3):1021.
Tronstad,R.,andT.J.McNeill.1989.AsymmetricPriceRisk:AnEconomet-
ricAnalysisofAggregateSowFarrowings.AmericanJournalofAgricul-
turalEconomics71(3):630637.
U.S.GeneralAccountingOffice(USGAO).1999.AgricultureinTransition
FarmersUseofRiskManagementStrategies.GAO/RCED-99-90.Wash-
ington,DC:U.S.GeneralAccountingOffice.
vanRavenswaay,E.O.,andJ.P.Hoehn.1991.TheImpactofHealthRiskon
FoodDemand:ACaseStudyofAlarandApples.InEconomicsofFood
Safety,J.A.Caswell,ed.NewYork:ElsevierSciencePublishingCo.,Inc.,
pp.155174.
RiskandAgriculture 167
Vandeveer,M.L.,andE.T.Loehman.1994.FarmerResponsetoModified
CropInsurance:ACaseofCorninIndiana.AmericanJournalofAgricul-
turalEconomics76(1):128140.
vonMassow,M.,andA.Weersink.1993.AcreageResponsetoGovernment
StabilizationProgramsinOntario.CanadianJournalofAgricultureEco-
nomics41:1326.
Wu,J.1999.CropInsurance,AcreageDecisions,andNonpoint-SourcePol-
lution.AmericanJournalofAgriculturalEconomics81(2):305320.
YoungD.,W.Lin,R.D.Pope,L.Robison,andR.Selley.1979.RiskPrefer-
encesofAgriculturalProducers:TheirMeasurementandUse.Paperpre-
sentedatseminarofWesternRegionalResearchProject,heldinSan
Francisco,January2526.
Zenger,S.,andB.W.Schurle.1981.TheImpactofDiversificationonRiskat
theFarmLevel.Abstract.AmericanJournalofAgriculturalEconomics
63:1043.
Index
The italic letters f, n, and t following a page number indicate that the subject
informationoftheheadingiswithinafigure,note,ortable,respectively,onthatpage.
Actuarialmodels,riskreductionand,
113114,123n5,145146
Adults,ashealthyuninsuredpersons,
103,107t
healthofnonelderly,109
incomeofnonelderly,105,106108,
107f
laborforcestatusofnonelderly,108,
109
Adverseselectionproblem
cropinsuranceand,146147
empiricalworkon,148,149
healthinsuranceand,67,103,110,
112,117
healthinsuranceavoidance
mechanismsfor,113,114115,
116
winnerscurseas,67,68,6970,
7475,89
Agriculturemarkets,2,78,127128,
130167
cropinsurancein,144150
diversificationasriskresponsein,
130141
economicsof,127130
foodsafetyand,154159
hedgingin,150153
nondiversificationriskreductionin,
142153
riskresponseandmathematical
modelsin,135136
riskresponsestudiesin,136137,
138t139t
SeealsoProducemarkets
Airwaverightssale,69,95,98n19
Alaska,nongrouphealthinsurance
carriersin,111
AllaisParadox,3,2324,29,30
Almost-objectiveuncertainty
beliefsandbettingpreferencesover,
4446
structureand,3537,4246
Apples,chemicalresidueon,154,157
Arkansas,SalmonellaandWTP
experimentsin,157
Artauctions,commonandprivatevalues
in,66
Asbestosexposure,cleanupand
litigationdueto,3,1112,15
AtlanticCity,NewJersey,gamingin,52
casinos,54,63n4
destinationresortcasinos,4,59
Attitudes
riskand,2,10,25
towardsgaming,5456,58,61,
6263
Auctionmarkets,65101
ascending-price,andpublic
information,9596
commonvalue,66,94,96n2,97n11
English,6,66,8288,94
first-price(sealed-bid),6,6566,
6970,82,8488
insiderinformationand,65,8283,
8893,90t,92t,94
outsiderbidsin,8893,90t,92t,
97n14
privatevalue,65,66
riskon,2,56,6566
second-price,81,94
winnerscursein,56,6682,86,89,
91,9496
Australia,gamblingin,55,61,64n4
Automobileinsurance,1314,16n3,147
169
12
170 Index
Babcock&Wilcox(firm),bankruptcy,
Bacterialcontamination,foodpoisoning
and,154,157,159
Bankruptcies
asbestosexposureand,12
auctionexperimentsand,71,
72t73t,80,94
Barley,inriskresponsestudies,139t1
Baseball,winnerscurseclaimsand,67
Bayesianapproach,toriskresponse,136
Behavioralsocialscience,needfor,159
Benefits
consumersurplus,5,61
differential,inhealthinsurance
markets,113,114115,116
fringe,andcompetitive
compensation,109
SeealsoCostbenefitanalysis
Bettingpreferencesoveralmost-
objectiveuncertainty,4446
Biddersinauctionexperiments
experienced,andEnglishauctions,
82,83,8688,87t
inexperienced,7174,72t173t,
8486,85t,8991,90t,94,97n13
moderatelyexperienced,7478,76t
77t
outsidersas,8893,90t,92t
studentsas,6970,96n5
super-experienced,andasymmetric
information,9193,92t,93t
unlimitedliabilityfor,8081
Bidding
aggressive,andlimitedliability,
7881,96n6,97nn89
Nashequilibrium,79,82,84,86,88,
94,96n797n7
RNNEin,experiments,7179,
7273,7677,8183
SeealsoSealed-bidauctions
Bingomarketshare,54
Blacklungdisease,occupationalrisk
and,1112
Bookpublicationrights,winnerscurse
claimsand,67
Broilers,inriskresponsestudies,138t
California
agriculturaleconomicsin,136,141,
157
state-authorizedtribalcasinosin,53,
58,63n2
Canada
casinosuccessin,andeffectonNew
YorkState,54
cropinsurancein,146
effecton,ofU.S.catastrophe,9
Cancerasoccupationalrisk,12
Carteltheory,winnerscurseand,67
Casabamelons,inriskresponsestudies,
138t
Casinomarkets,2,45,5164
categorieswithin,45,5960
costbenefitanalysisof,5,5661,
6263
economicperspectiveson,4,5164
socialperspectiveson,5,51,52,
5456,5863
Catastrophes
effectsof,9,10
planningfor,3,10,1415
reliefafter,121,130
CDC.SeeCentersforDiseaseControl
andPrevention
CentersforDiseaseControland
Prevention(CDC),asdatasource,
154
CentralLimitTheorem,asprobability
tool,35
Certainty
inchoicetheory,22
inprivatevalueauctions,65
inproductionsituations,152
SeealsoUncertainty
Charitablegambling,54
ChebyshevsInequality,asprobability
tool,35
ChicagoBoardofTrade,150
Chickenproducts,138t,156
irradiatedvs.nonirradiated,157158
Children,asuninsuredpersons,105,
106,107t,109
Chileangrapefruit,154
Choice
inputuseand,142144
objectiveuncertaintyand,34,2225
riskand,12,128
scientificconsiderationsand,1922
statesofnaturein,4,18,2932,37
subjectiveuncertaintyand,4,2529
Churchill,Winston,andrisk
identification,15n1
Cleanupcosts,liabilityfor,11,
15n216n2
Colorado,state-authorizedgamingin,
52,60,63n1
Combinatorics,asprobabilitytool,22,
35
Commercialgamingcontroversies,
5356,62,63
Commoditybundles
certaintyand,22
uncertaintyand,2021,23
Commoditymarkets,hedgingand,
150153
Commonvalueauctions,66,81,94,
96n2,97n11
Compensation,2,109,110
Competition
cropsin,136,140
demandand,59,104,109,129
formof,shapedbyinformation
asymmetry,113116
healthinsurancecompaniesin,
110116
limited,andmonopolisticstructures,
52,57,64n5,115
programsin,119
ComprehensiveEnvironmental
Response,Compensationand
LiabilityAct,11
Index 171
Connecticut,state-authorizedcasinosin,
53,54,59
Consumerbehavior
benefitsvs.costsin,154156
spendingchoicesas,2,57,61,64n5,
155159
(SeealsoChoice)
strategicinformationand,65
Consumerdemand
inagriculturemarkets,148150,
158159
forgamblingandgamingactivities,
59,62
Consumersurplus
attitudetowardvicesand,58,62
benefitsof,5,61
definition,57
economicrentsand,57,64n5
Contingentvaluation(CV)method,
benefitsestimationwith,155,
156157
Control,asriskmanagementstep,10,
1213
Conveniencegambling.See
Neighborhoodgaming
Corn
incomefrom,130132,149
Midwesternsitesfor,148149,153
moralhazardand,148149
priceof,andfuturesmarket,150151
riskresponsestudieswith,138t139t
Corporateagriculture,129130
Corporatetakeovers,winnerscurse
claimsand,67
Costbenefitanalysis
benefitestimationmeasures,
154156
casinomarketsand,5,5661,6263
decisionstrategiesand,1213,15
foodsafetyapplications,156159
Cotton,inriskresponsestudies,
138t139t
CPS.SeeCurrentPopulationSurvey
149
172 Index
Cropinsurance,7,144150
adverseselectionin,146147
empiricalresultson,demand,
148150
moralhazardin,147
MPCIwithfederalsubsidyas,145,
policydebateson,130
CropInsuranceImprovementAct,145
Crops,inriskresponsestudies,
138t1139t1
CurrentPopulationSurvey(CPS),as
datasource,106108,107f,107t
CVmethod.SeeContingentvaluation
(CV)method
Dataanalysis,byUrbanInstitute,108
Datasources
experimentalauction,7181,
72t73t,76t77t
fieldvs.experimentalauction,and
reliability,6869
U.S.agriculture,130,153
U.S.demographic,105,106108,
107f,107t,109
U.S.medical,118,154
Deathspiral,definition,117
Decisionmakers.SeeConsumer
behavior
Decisionscience
constructsformodern,1718,32
expectedutilitytheoryin,30
modernprobabilitytheoryand,22
observabilityin,1920,3234
riskaversionand,6,10
Delaware
nongrouphealthinsurancecarriers
in,111
racinosin,53
Demand
agricultureand,129,148150
competitionand,59,104,109,129
shiftin,andbenefitsestimation,155,
158159
SeealsoConsumerdemand
Demographics
genderandWTP,157
U.S.datasources,105,106108,
107f,107t,109
Destinationresortcasinos,4,59
Detroit,Michigan
effectsofcatastrophein,9
state-authorizedgamingin,52,59,
60,63n1
Diversification
financialinvestments,2,140
inputusechoicesand,142144
labor,on-andoff-site,140
reducingriskwith,2,130141
sizeandspace,134,140141
E.coli. SeeEscherichiacoli
Economicrents
auctionsellersrevenueas,93
captureof,58,60,61,62
consumersurplusand,57,6061,
64n5
definition,57
Economics
agricultural,asasubfield,127130,
135137,138t139t,150153
casinomarketsand,4,5164
competitionanddemand,59,104,
109,129
consumersurplusandeconomic
rents,57,6061,64n5
exchangetheoryandauctionsin,65
marketefficiencies,104,109,
116117,120122,150
recessionsin,andcasinomarkets,4,
5354
scalevs.scopein,109,140141
SeealsoCostbenefitanalysis;
Markets;Monopolisticstructures
Electronicgamingdevices,54,55,62
neighborhoodsand,51,61,64n8
EllsbergParadox,violationsof
probabilisticsophisticationand,4,
2729
EmployeesRetirementandIncome
SecurityAct(ERISA),111
Employers
ashealthinsuranceproviders,103,
105,106,109,110
large,andstateoversightavoidance,
111
largevs.small,andgroupinsurance,
110,111112,116
small,withlow-wageworkers,120
England,asbestosexposurein,12
Englishauctions,6,66,8288,94
experiencedbiddersand,82,83,
8688,87t
first-pricesealedbidvs.,8388,85t,
87t,94
inexperiencedbiddersand,8486,
85t,97n13
ERISA.SeeEmployeesRetirementand
IncomeSecurityAct
Escherichiacoli,foodpoisoningfrom,
154,159
Euclideanstatespace,35,36,3940,42,
45
EuropeanUnion,marketsand
governmentpolicyin,129
Exchangetheory,auctionsrolein,65
Expectedutilitytheory
asdecision-makingmodel,30,128
economiesofscalevs.scopein,
140141
linearityinprobabilitiesand,25
mathematicalmodelfor,2324,
37nn23,131132,155
propertiesofpersonsunder,42,46
Experimentalauctions
benefitsestimationwith,155,156,
157158
fieldvs.,andreliability,6869
inexperiencedbiddersand,7174,
72t73t
moderatelyexperiencedbiddersand,
7478,76t77t
outsiderbiddersin,8893,90t,92t
Index 173
resultsof,andalternative
explanations,7881
studentbiddersin,6970,96n5
unlimitedliabilitiesin,8081
SeealsoBidding
Farmbills,riskincentiveproblemswith,
2
FarmBureauInsurance,1
FederalEmergencyManagement
Agency,121
FederalHousingAuthority(FHA),121
Federal-Mogul(firm),asbestosliability
of,12
Fertilizers
residuesof,andfoodsafety,154
riskreductionand,141142
useof,andMPCI,149
FHA.SeeFederalHousingAuthority
First-priceauctions
Englishvs.,withsealedbids,8388,
85t,87t
sealed-bidsin,6,6566,6970,82,
8488
uncontrolleduncertaintyexperiment
via,6970
Flaxseed,inriskresponsestudies,139t
Floridagrapefruit,WTPand,157
Foodsafety,154159
appliedcostbenefitanalysisin,
156159
consumerconfidencein,154
WTPand,155159
FoodSafetyandInspectionService,
154155
FordMotorCompany(firm)
asbestosexposureand,12
RiverRougecatastropheof,3,910,
13,1415
Forwardmarkets
contractdefinitionin,150
futuresmarketscomparedto,150,
153
153
174 Index
Freeagencymarkets,winnerscurse
claimsand,67
Fringebenefits,laborskillsand,109
Fungicides,154,156
Futuresmarkets
inagriculture,153
forwardmarketscomparedto,150,
hedgingin,150153
GAF(firm),bankruptcy,12
Gambling
acceptabilityof,industry,45,52,
5456
characteristicdimensionsof,6263
costbenefitframeworkfor,5661
highrisk,andcompensation,2
nationalresearchinvarious
countries,63n464n4
trendsinU.S.gaming,5154,63
SeealsoCasinomarkets;Internet
gambling;Pathologicalgamblers
Georgia-Pacific(firm),asbestos
exposureand,12
Governmentrevenues
auctionsalesand,65,6668,69,95,
98n19
commercialgamingand,5356,60,
64n8
competitionfor,119
economicrentsas,57,58,61,62
Governmentroles
inagriculture,145146,154155
ingamingregulations,51,52,5859,
64n5
inhealthinsurance,104105,115,
116122
ininsuranceregulations,111,114
nationaldebtand,65
publicpolicyintervention,5859,62,
64n6,104,121,122n1,129,
145146,154155
Grain
inriskresponsestudies,138t
Seealsospecifics,e.g.,Wheat
Grapefruit,chemicalresiduesand,154,
157
Greenrevolution,Philippines,142
HACCP.SeeHazardAnalysisand
CriticalControlPoints
Hawaii,slotmachinedebatesin,53
HazardAnalysisandCriticalControl
Points(HACCP),foodsafety,155
Healthinsurancecoverage
CPSdataon,110,122n2123n2
privatevs.quasi-publicprovidersfor,
110
stateregulationof,111
Healthinsurancemarkets,2,67,
103125,147
adverseselectionproblemin,67,
103,110,112,113
competitionamongcarriers(insurers)
in,110116
differentialbenefitsasseparation
mechanismin,113,114115,116
governmentrolesin,104105,111,
115,116122
interconnectedmarketsand
submarketsin,111112
uninsuredpersonsin,103,105110,
122,147
HealthInsurancePortabilityand
AccountabilityAct(HIPAA),
123n4
Healthstatus
chemicalresiduesand,154
ofuninsuredpersons,109
HealthyNewYork(program),120
Hedges(riskreductionmechanism),
150153
HenryJ.KaiserFamilyFoundation,
CommissiononMedicaidandthe
Uninsured,asdatasource,105,
109
Herbicideresidues,154
Highrisk
gamblingas,2
sharing,inhealthinsurance,67,
116125
stateregulationsand,115,123n6
SeealsoAdverseselectionproblem
HIPAA.SeeHealthInsurancePortability
andAccountabilityAct
Horses
gamingwith,5,53,60
ridingon,1
Idaho,nongrouphealthinsurance
carriersin,111
Identificationofrisk
aslosspreventionmeasure,13
asriskmanagementstep,1012,14,
15n1
Illinois
Chicago,andcommoditymarkets,
150
riverboatmonopoliesin,59
state-authorizedgamingin,52,63n1
Importsubstitution,urbancasinosand,
6061
Income
agriculturaleconomicsand,129,
130133,149
health-uninsurednonelderlypersons
by,103,106108,107f
low-,personsandHealthyNewYork,
120
IndianGamingRegulatoryAct,51,53
Indiana
commoditymarketsusedin,153
riverboatmonopoliesin,59
state-authorizedgamingin,52,63n1
Individual(nongroup)healthinsurance
carriersof,111,116,122
marketfailureand,103
riskreductionforcarriersof,113,
114,115,123nn45
standardizationofpoliciesby,and
government,119120
stateregulationsand,123n6
Index 175
vs.large-employergroupinsurance,
112
Information
asymmetryof,8893,113116,120,
145147,149
marketfailuresand,104,120
new,andhedges,151
private,signalsandestimates,66,88
public,andauctionsellersrevenue,
68,7475,78,94,9596,
96n797n7
risk,indexesandconsumerdemand,
157,158159
symmetryof,andauctions,8889,
91,93,93t
SeealsoInsiderinformation
Insiderinformation
auctionmarketsand,65,8283,
8893,90t,92t,94
definition,82
outsiderbidsvs.,8893,90t,92t,
97n14
Insurance
carriers(seeInsurers)
deathspiralin,117
federalgovernmentrolesin,121,145
purchaseincentivesfor,146147
reducingriskwith,2
shiftingriskwith,1,1314
SeealsoAutomobileinsurance;Crop
insurance;Healthinsurance
markets
Insurancepremiums
pricesubsidiesand,148,150
risk,definition,144
riskaversionand,144,147
Insurers
agriculture,andriskexposure,145
health,andcompetition,110116
health,andlowrisk,103,112
health,nongroup,111,116,122
health,small-employergroup,111,
112,122
housing,121
176 Index
Internationalissues
agriculturalproduction,128
federalcropinsurance,145146
governmentleaseauctions,9596,
98n19
Internetgamblingcontroversy,
5556,61,63
Internetproviders
auctionsby,65
gamblingand,5556,61,63
Iowa
racinosin,53
SalmonellaandWTPexperimentsin,
157
state-authorizedgamingin,52,63n1
Irradiatedchickenproducts,vs.
nonirradiated,157158
Japan,cropinsurancein,146
Jobtraining,asriskcontrol,13
Johns-Manville(firm),bankruptcy,12,
15
Johnson&Johnson(firm),loss
reductionstrategyof,15
Jointandseveralliability,definition,
1516n2
KaiserCommissiononMedicaidandthe
Uninsured.SeeunderHenryJ.
KaiserFamilyFoundation
Kansas
SalmonellaandWTPexperimentsin,
157
wheatandchemicalinputsin,149
KansasCity,Missouri
effectofout-of-statecatastropheon,
9
urbancasinosin,60
Kentucky,slotmachinedebatesin,53
Laborforce
farm,andsitediversification,140
healthuninsuredpersonsin,108,109
skillsof,andcompensation,110
Land-grantcolleges,originof,127
Large-employergroup(health
insurance),small-employergroup
insurancevs.,110,111112,116
LasVegas,Nevada
casinocompetitionin,59
destinationresortcasinos,4,59
LawofLargeNumbers,asprobability
tool,35
Liabilities
benefitsestimationwith,155,156
environmentand,1112,15n216n2
ofequineprofessionals,1
federal,andMPCI,145
limited,forlossesandwinners
curse,7881,97n8
unlimited,forexperimentalauction
bidders,8081
Litigation
asbestosproductionand,12
deeppocketsand,11
Livestock,inriskresponsestudies,139t
Lossreduction
criticalstrategyfor,1415
measureof,13
Lotteries
marketshareof,54
modernprobabilitytheoryand,3,
2224
statelegislationfor,52
videoterminalsfor,51
Louisianagaming
backlashto,61
casinomonopoliesinNewOrleans,
59
racinosfor,53
revocationofcommercial,5455
riverboatmonopoliesin,59
stateauthorizationfor,52,63n1
Lowrisk
healthinsuranceand,67,112,
114115
healthinsurersand,103,112
Lungcancer,asoccupationalrisk,12
Marketfailures
adverseselectionand,103,104,116
informationasymmetryand,120
Marketshareingaming,54
Markets
commodity,andhedging,150153
efficienciesin,104,109,116117,
120122,150
EuropeanUnion,andgovernment
policy,129
forward,comparedtofutures
markets,150,153
futures,andhedging,150153
impactofriskon,2,48
SeealsoAgriculturemarkets;
Auctionmarkets;Casinomarkets;
Healthinsurancemarkets;
Producemarkets
Marx,Groucho,winnerscurseand,
96n3
Maryland,slotmachinedebatesin,53
Massachusetts,SalmonellaandWTP
experimentsin,157
Mathematicalmodels
expectedutilitytheory,2324,
37nn23,131133,155
riskresponseinagriculturemarkets,
135136
subjectiveuncertaintyandchoice,
2526,26t
vonNeumannMorgensternutility
function,23
Meatproducts,food-bornepathogensin,
155,158
E.coli,154,159
Salmonella,156,157
Medicaid
asoneofmultipleinsurers,123n2
uninsuredpersonsand,104,105,106
Medicalbills
datasourcesfor,118,123n7
asdebtsofuninsuredpersons,105,
109,122
high,andreinsurance,117119
Index 177
MedicalExpenditurePanelSurvey,as
datasource,118
Medicalunderwriting
adverseselectionavoidancewith,
113,123n3
principlesof,114
Medicare,asoneofmultipleinsurers,
123n2
Mesothelioma,asoccupationalrisk,12
Michigan
Detroit,effectsofcatastrophein,9
Detroit,state-authorizedgamingin,
52,59,60,63n1
tribalcasinosin,53
MichiganEquineActivityLiabilityAct,
1
Mineralrightsauctions,government
leasesand,66
Minnesotagaming,53,54
Mississippigaming
Biloxi,destinationresortcasinos,4,
59
competitionin,59
stateauthorizationfor,52,63n1
Missouri
KansasCity,effectofout-of-state
catastropheon,9
state-authorizedgamingin,52,63n1
urbancasinosin,60
Mitigationofrisk
poolingas,110
asriskmanagementstep,10,1213
SeealsoRiskreductionmechanisms
Mobilephonerights,saleof,98n19
Monopolisticstructures
adverseselectionavoidancewith,
115
authorizedgamingwith,52,57
Montana,neighborhoodgamingin,60
Moralhazard
cropinsuranceand,78,147
empiricalworkon,148149
riskshiftingas,1314
vicesas,58
178 Index
Mortgageinsurance,federalgovernment
and,121
MPCI.SeeMultiple-PerilCrop
Insurance
Multiple-PerilCropInsurance(MPCI)
chemicalusageand,149
federalsubsidyfor,145
informationasymmetryand,149
Nashequilibriumbidding,79,82,84,
86,88,94
SeealsoRiskneutralNash
equilibrium
NationalScienceFoundation,research
supportfrom,37,96
NativeAmericanenterprises.SeeTribal
casinos
Nature.SeeStatesofNature
Neighborhoodgaming
costbenefitof,61
debatesaboutslotmachinesin,
5354
electronicdevicesin,51,60
Nevadagaming,51,52
destinationresortcasinosin,4,59
lawandlegislationforcasinosin,4,
52,63n4
neighborhoodsand,60
NewHampshiregaming,52,54
NewJersey
AtlanticCity,gamingin,4,52,54,
59,63n4
individualhealthinsurance
regulationsin,123n6
NewMexico,racinosin,53
NewYorkState
gamingauthorizationsby,53,54
healthinsurancesubsidiesby,120
nongrouphealthinsurancein,111,
123n6
Nongrouphealthinsurance.See
Individual(nongroup)health
insurance
Normative(prescriptive)applications
behavioralsocialscienceand,159
statesofnatureand,3032
NorthDakota,durumwheatandinsured
incomein,149
Objectiveuncertaintyandchoice,34,
21,2225
expectedutilityhypothesisin,3,23,
30
probabilityin,22,35
propertiesofobjectiveevents,41
violationofexpectedutilityin,3,
2324
Observability,inriskanalysisdecision-
making,1920,3234
Occupationalhealthhazards,1112,15
OCS.SeeOutercontinentalshelf
Ohio,urbancasinosinCincinnati,60
Oilleaseauctions
governmentrolein,66
winnerscurseand,6668,95,
97n1798n17
Oilsunflower,inriskresponsestudies,
139t
Ontario,Canada,9,54
Oregon,neighborhoodgamingin,60
Outercontinentalshelf(OCS),
governmentoilleasesfor,6667
Outsiderbidsinauctions,insider
informationvs.,8893,90t,92t,
97n14
OwensCorning(firm),bankruptcy,12
Pathologicalgamblers,5,55
protectionfor,58
socialcostsand,58,61,62,63
Peachorchards,spatialdiversification
andyield,134
Pennsylvania,slotmachinedebatesin,
54
Pesticides
residuesof,andfoodsafety,154,
156157
riskreductionand,141,142143
usageof,andMPCI,149
PetersburgGame,descriptivevs.
prescriptivesolutionsto,30
Potato,inriskresponsestudies,138t
Poultryproducts,138t,154,156,
157158
Preferencefunction
formsof,23,3940
lotteriesand,3,22,2324
riskaversionand,37n2
Pricesupports
cropsand,141,149
insurancepremiumsand,148,150
Privatevalueauctions,65,66
Probabilisticsophisticationhypothesis,4
non-probabilisticsophistication
modelsvs.,3940
propertiesofpersonsunder,4142,
4546
insubjectiveuncertaintyandchoice,
2627
violationsof,insubjective
uncertaintyandchoice,2729
Probabilitytheory
modern,founders,3,22
toolsof,22,35
Producemarkets
chemicalresiduein,154,156157
fruitin,134,138t,154,157
riskinformationindexesin,158159
vegetablesin,159
Production
agriculturaleconomicsand,128130,
152153
asbestos,andlitigation,12
catastrophesin,910,11,1415
gamblingservicesas,57
Profits
auction,andinexperiencedbidders,
71,72t73t,85t,94
Index 179
auction,andmoderatelyexperienced
bidders,7478,76t77t,87t
calculationfor,incommonvalue
auctions,97n11
Englishauctionsand,8486,85t,87t,
97n13
insiderinformationand,8993,90t,
92t,93t,97nn1416
negativeaverage,andhighauction
bids,71,74,81,86,94
OCSleasesand,6768
variable,andcrops,141143,
151152
Publicpolicy
agricultureand,129,130
gamblingand,5859,62,64n6,104,
122n1
hazardanalysisand,1112,15,
154155
healthinsuranceand,104,109110,
122n1
Quantification
aslossreductionmeasure,13
asriskmanagementstep,1012,14,
15n1
Racetracks
horsesand,5,53,60
pari-mutuelwageringmarketshare
at,54
Racinos
definition,5,51
slotmachinedebatesabout,5354
statedevelopmentof,53,54
urbanlocationsfor,60
RaybestosManhattan(firm),
bankruptcy,12
Realestateauctions,winnerscurse
claimsand,67
Recession(economics),casinomarkets
and,4,5354
180 Index
Reinsurance
federalgovernmentand,121122
highmedicalbillsand,117119
Reliability
data,inscientificsupport,68
measurementof,97n12
RhodeIslandgaming,53,54
Rice
Asianstudiesof,135136,142143
inriskresponsestudies,138t
Risk
auctionsand,6566
basicconcepts,3,916
definitions,1,10,137
impactof,onmarkets,2,48
informationindexesof,158159
reductionof,110,142159
SeealsoHighrisk;Lowrisk
Riskanalysistheories
classificationandmeasurement
issuesin,2022,34
objectiveuncertaintyandchoice,
34,2225
observabilityissuesin,1920,3234
scientificconsiderationsandchoice,
1922
statesofnaturein,4,18,2932,37
structureissuesin,3537
subjectiveuncertaintyandchoice,4,
2529
Riskaversion
decisionmakersand,6,10,142,143
definition,2,10
expectedutilityfunctionand,23,
37n2,132,155
futuresmarketsand,151152
incentivesand,130134,146147
insurancepremiumfor,144,147
Riskincentiveproblems
agricultureand,2,146147
healthinsuranceand,117118
trade-offsin,14
Riskinformationindexes
grocerystoresand,157
producemarketsand,158159
Riskmanagement,24,8
Riskmanagementconcepts,3,916
catastropheplanning,10,1415
financing,10,1314
identificationandquantification,
1012,15n1
mitigationandcontrol,10,1213
Riskneutralmodels
farmapplicationsof,133134,142
insurancepremiumfor,144
RiskneutralNashequilibrium(RNNE)
auctionexperimentswith,7179,
72t73t,76t77t,8182
Englishauctionsexperimentswith,
83
Riskpreferences
agriculturaleconomicsand,127128
healthinsurancecarriersand,103,
112,116
nonexpectedutilitymodelsof,4,
2425
violationsof,asAllaisParadox,3,
2324
Riskreductionmechanisms
costof,andgovernmentintervention,
116122,145
cropinsurance,144150
diversification,130141
foodsafety,154159
forwardandfuturesmarkets,
150153
healthinsurance,113,114115,116
inputchoice,142144
Riskresearch
agriculturaleconomicsand,136137,
138t139t,159
foundationsupportfor,37,96,122
gamblinginvariouscountries,
63n464n4
RiverRougepowerplant,explosion,3,
910,13,1415
Riverboatgambling,state
authorizations,5253,58,59
Index 181
RNNE.SeeRiskneutralNash
equilibrium
RobertWoodJohnsonFoundation,
researchsupportfrom,122
Ruralcasinos,5,5960
RussellSageFoundation,research
supportfrom,96
St.PetersburgParadox,solutionto,30
Salmonellasp.,foodpoisoningfrom,
156,157
Savage,Leonard
choicetheoryand,4,2526,27
Savagesomelet,26,26t,3031
statesofnatureand,18
SCHIP.SeeStateChildrensHealth
InsuranceProgram
Science
behavioralsocial,andneedfor,159
characteristicsof,18
datainsupportof,68
theoryofchoiceand,1922
Sealed-bidauctions
controlleduncertaintyexperiments,
7081
first-priceauctionsas,6,6566,
6970,82,8488
inexperiencedbiddersin,
experiments,7174,72t73t
limitedliabilityforlossesin,
experiments,7881
moderatelyexperiencedbiddersin,
experiments,7478,76t77t
second-priceauctionsas,81,94
winnerscursein,9596
SloanFoundation,researchsupport
from,96
Slotmachines
debatesabout,5354
racinoswith,5,53
Small-employergroup(health
insurance)
carriersand,111,112,122
marketfailureand,103
riskreductionforcarriersof,
113114,115116,123n5
standardizationofpoliciesby,and
government,119120
vs.large-employergroupinsurance,
110,111112,116
Sobczynki,DanielS.,onrisk
management,12,15
Socialconcerns
casinomarketsand,5,51,52,5456,
5863
foodsafety,154,156
gamblingcosts,5,56,58,60,61,
6263
Internetgamblingand,55,61
Sodiumortho-phenylphenate(SOPP),
WTPand
eliminationof,157
SOPP.SeeSodiumortho-phenylphenate
SouthCarolinagaming
backlashagainst,61
revocationof,54
SouthDakotagaming,52,60,63n1
Sowfarrowing,inriskresponsestudies,
138t139t
Soybean
acreagein,andinsuredincome,149
commoditymarketsfor,inIndiana,
153
inriskresponsestudies,138t139t
StateChildren'sHealthInsurance
Program(SCHIP),uninsured
personsand,104,105
Stateregulations
avoidanceof,bylargeemployers,
111
healthinsuranceand,114,115,
123n6
Staterevenues
collectedfromgamingdevicesvs.
racingwagers,53
taxavoidancebylargeemployers,
111
182 Index
Statesofnature
alternative,comparedtopermitted
gambling,5759
descriptiveapplicationsand,26t,29
normative(prescriptive)applications
and,3032,31t
observabilityissuesin,1920,3234
inriskanalysis,4,18,2932,37
Subjectiveprobability
agriculturaleconomicsand,127128
densityfunction,43f1
Subjectiveuncertaintyandchoice,4,18,
2122,2529
inmathematicalmodels,2526,26t
probabilisticsophistication
hypothesisin,4,2627
probabilisticsophisticationviolations
in,2729
Suburbancasinos.SeeUrbancasinos
Sugarbeet,inriskresponsestudies,138t
Texas,nongrouphealthinsurance
carriersin,111
Thailand,ricecropin,135136
3M(firm),asbestosexposureand,12
T&N(firm),asbestosexposureand,12
Trade-offs
auctionsand,66,96n1
automobileinsuranceand,1314,
16n3
Tribalcasinos,5,5960,63n2
authorizedbystates,51,53,54,58
most,asrural,5,5960
Tylenolcapsules,taintingof,3,15
Unarco(firm),bankruptcy,12
Uncertainty
almost-objective,andstructure,
3537,4346
normativeapplicationcriterionand,
31,31t
objective,andchoice,34,18,21,
2225
privatevalueauctionsand,65
inproduction,144,152153
asqualitativevs.quantitative
variables,2122,34
subjective,andchoice,4,18,2122,
2529
Unemployment,healthinsurance
coverageand,106,108
Uninsuredpersons(cropinsurance),
147,149
Uninsuredpersons(healthinsurance),
103,105110
age,103,105,106,107t
datasourcesfor,105,106
ethnicity,105
healthstatus,109
income(nonelderly),103,106108,
107f,122
laborforcestatus,108
policydilemmaposedby,109110
UnitedKingdom
Britain,mobilephonelicenses,95,
98n19
DepartmentforCultureMediaand
Sport,GamblingReviewReport,
63n464n4
England,asbestosexposurein,12
UnitedStates
agenciesforfoodsafety,154155,
158,159
agriculturalproductionin,129130
DepartmentofAgriculture(USDA),
158,159
FederalCommunications
Commission,airwavelicenses,
98n19
GeneralAccountingOffice,asdata
source,153
NationalGamblingImpactStudy
Commission,FinalReport,63n4
Seealsocomponentstates
U.S.lawandlegislation
CropInsuranceImprovementAct,
145
ERISA,111
farmbills,2
HIPAA,123n4
IndianGamingRegulatoryAct,51,
53
Superfundhazardouscleanup,11
SeealsoGovernmentroles
Urbancasinos,45,60
importsubstitutionbehaviorand,
6061
UrbanInstitute,dataanalysisby,108
VA. SeeVeteransAdministration
Variables
dependentvs.independent,1920,
21,29
qualitativevs.quantitative,2122,34
VeteransAdministration(VA),
mortgagesguaranteedby,121
Vickreyauction,66
Videogaming
lotteriesauthorizationfor,51
neighborhoodswith,60
pokermachineauthorizationand,51,
5455
Virginia,effectofout-of-state
catastrophesin,9
VonNeumannMorgensternutility
function,3,23
Wagering
pari-mutuel,andmarketshare,54
staterevenuesfromrace,vs.gaming
devices,53
WashingtonState,individualhealth
insuranceregulationsin,123n6
Weather,effectsonagriculture,129,
134,148,159
WestVirginia,racinosin,53
Wheat
prairiestatesand,149
pricesupportpolicyfor,141
inriskresponsestudies,138t139t
Whitebeans,inriskresponsestudies,
138t
Willingnesstopay(WTP),foodsafety
and,155159
Index 183
Winnerscurse
attributionsof,6667,96n3
auctionmarketsand,2,56,6682,
91,9496
definitions,2,67,89
demonstrationexperimentsof,6978
limitedliabilityforlossesand,7881
real-worldclaimsof,6668,95,
97n1798n17
Wisconsin,slotmachinedebatesin,54
W.R.Grace(firm),bankruptcy,12
WTP.SeeWillingnesstopay
Youngadults,ashealthuninsured
persons,106,107t
healthstatus,109
AbouttheInstitute
The W.E. Upjohn Institute for Employment Research is a nonprofit
researchorganizationdevotedtofindingandpromotingsolutionstoemploy-
ment-relatedproblemsatthenational,state,andlocallevels.Itisanactivity
of the W.E. Upjohn Unemployment Trustee Corporation, which was estab-
lished in 1932 to administer a fund set aside by the late Dr. W.E. Upjohn,
founder of The Upjohn Company, to seek ways to counteract the loss of
employmentincomeduringeconomicdownturns.
TheInstituteisfundedlargelybyincomefromtheW.E.UpjohnUnem-
ploymentTrust,supplementedbyoutsidegrants,contracts,andsalesofpubli-
cations. Activities of the Institute comprise the following elements: 1) a
research program conducted by a resident staff of professional social scien-
tists; 2) a competitive grant program, which expands and complements the
internal research program by providing financial support to researchers out-
sidetheInstitute;3)apublicationsprogram,whichprovidesthemajorvehicle
fordisseminatingtheresearchofstaffandgrantees,aswellasotherselected
works in the field; and 4) an Employment Management Services division,
which manages most of the publicly funded employment and training pro-
gramsinthelocalarea.
The broad objectives of the Institutes research, grant, and publication
programsareto1)promotescholarshipandexperimentationonissuesofpub-
lic and private employment and unemployment policy, and 2) make knowl-
edge and scholarship relevant and useful to policymakers in their pursuit of
solutionstoemploymentandunemploymentproblems.
Currentareasofconcentrationfortheseprogramsincludecauses,conse-
quences, and measures to alleviate unemployment; social insurance and
income maintenance programs; compensation; workforce quality; work
arrangements; family labor issues; labor-management relations; and regional
economicdevelopmentandlocallabormarkets.
185