ContentslistsavailableatScienceDirect
Journal
of
Economic
Behavior
&
Organization
jo u r n al ho me p ag e :ww w . e l s e v i e r . c o m / l o c a t e / j e b o
Do
markets
reveal
preferences
or
shape
them?
夽
Andrea
Isoni
a,∗,
Peter
Brooks
b,
Graham
Loomes
a,
Robert
Sugden
caBehaviouralScienceGroup,WarwickBusinessSchool,CoventryCV47AL,UK bBarclaysWealth,Singapore
cSchoolofEconomicsandCentreforBehaviouralandExperimentalSocialScience,UniversityofEastAnglia,NorwichNR47TJUK
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received9June2014
Receivedinrevisedform28October2015 Accepted17November2015
Availableonline27November2015
JELclassification: C81 C91 D44 Keywords: Shapingeffects Marketdiscipline Repeatedmarkets Pricesensitivity Preferenceimprecision
a
b
s
t
r
a
c
t
Wecontrastthepropositionthatmarketsrevealindependentlyexistingpreferenceswith thealternativepossibilitythattheymayhelptoshapethem.Usingdemand-revealing experimentalmarketinstitutions,weseparatetheshapingeffectsofpricecuesfromthe effectsofmarketdiscipline.Wefindthatindividualvaluationsandprevailingpricesare systematicallyaffectedbybothexogenousmanipulationsofpriceexpectationsand endoge-nousbutdivergentpricefeedback.Theseeffectspersisttovaryingdegrees,inspiteoffurther marketexperience.Insomecircumstances,marketexperiencemayactuallyconsolidate them.Wediscusspossibleexplanationsfortheseeffectsofuninformativepricecueson revealedpreferences.
©2015TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCC BYlicense(http://creativecommons.org/licenses/by/4.0/).
1. Introduction
Thereisnowabundantevidencethat,contrarytothestandardassumptionsofeconomicmodels,thepreferencesof economicagentsareoftenaffectedbysalientbutinformationallyirrelevantcues,specifictotheparticularcontextsinwhich thosepreferencesarerevealed(SlovicandLichtenstein,1968;Bohmetal.,1997;Arielyetal.,2003;Mazaretal.,2013; Sugdenetal.,2013;Maniadisetal.,2014).
Inthispaper,wefocusonaspecificformofcontext-dependence:thesystematiceffectthatobservationsof,orexpectations about,actualmarketpricesexertonthevaluationsofeconomicagents,evenwhenthosepriceshavenousefulinformational contentfortheformationofprivatevalues(seeKnetschetal.,2001;Loomesetal.,2003;Tufano,2009).Thiseffectposes achallengetothelong-standingviewineconomicsthatmarketsareinstitutionsthatsimplyalloweconomicagentsto revealtheirpre-existingandmarket-independentpreferences:itsuggestsbycontrastthatmarketsmay,toaconsiderable extent,shapepreferences.Inthetraditionalview,themarketisliabletopenalisetraderswhomisreporttheirpreferences
夽 ThisresearchwascarriedoutaspartoftheProgrammeinEnvironmentalDecisionMaking,organisedthroughtheCentreforSocialandEconomic ResearchontheGlobalEnvironmentattheUniversityofEastAnglia,andsupportedbytheEconomicandSocialResearchCounciloftheUK(awardnos. M-535-25-5117andRES-051-27-0146).
∗ Correspondingauthor.Tel.:+4402476150370.
E-mail addresses: [email protected] (A. Isoni), [email protected] (P. Brooks), [email protected] (G. Loomes), [email protected](R.Sugden).
http://dx.doi.org/10.1016/j.jebo.2015.11.006
0167-2681/© 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
andtherebyencouragesthemtocorrecttheirbehaviour(e.g.Binmore,1999;List,2003,2004).Suchmarketdiscipline,itis argued,helpstorefinetherevelationofone’sunderlyingpreferenceswhenthesearenotimmediatelyaccessiblebutemerge astheresultofaprocessof‘discovery’(Plott,1996).
Shapingandmarketdisciplinearebothmechanismsbywhichpricesgeneratedinonetradingperiodcanaffectbehaviour insubsequentperiods,andtheeffectsofbothcanbecumulative.However,whilemarketdisciplineissupposedtopull revealedpreferences towardstheirunderlyingmarket-independent value,shaping mayarbitrarilypull them towards irrelevantcues.Inthenormaloperationofmarkets,discriminatingbetweentheseopposingforcesmayproveextremely challenging,astheycanresultinobservationallyequivalentbehaviour.Takingupthischallengeisthemaincontributionof ourpaper.Wepresentanexperimentwhichusestwomanipulationsaimedatidentifyingtheextentofshapingeffectsin theabsenceofmarketdiscipline,andthepowerofmarketdisciplinetoerodetheeffectsofshaping.
Inourexperimentalsetup,pricecuesaredevoidofanyrelevantinformationalcontent:peopletradeprivatelyowned andindividuallyconsumedgoods,forwhichvaluationscannotbeobjectivelyaffiliated(e.g.ListandShogren,1999).Since thesegoodsonlyexistinourexperimentandareconsumedinsidethelab,pricecuescannotconveyanyinformationabout alternativetradingopportunities(Harrisonetal.,2004).
Weuseanexogenousandnon-informativemanipulationtoinfluencestatedpriceexpectations,whichactascuesfor valuations.Noticethattheseareexpectationsaboutactualmarketprices,buthavenotbeeninfluencedbyanyprevious marketbehaviour.Soapureshapingeffect–thatis,aneffectthatcannotbeattributedtomarketdiscipline–canbeobserved, ifitoccurs,inthefirstroundoftrading.Then,subsequentroundsallowustoobservetheeffectsofmarketdiscipline.
Wealsouseamoreinnovativemanipulationthatallowsusto‘switchoff’marketdisciplineforidentifiablesubsetsof traders:traderswhorepeatedlytradeatpricesthatareclearlyveryadvantageous,andtraderswhorepeatedlydonottrade becausetheyfaceverydisadvantageousprices.Byrepeatingthemarketintheabsenceofmarketdiscipline,wecanobserve thecumulativeshapingeffectsofendogenouspricefeedback.Byswitchingmarketdisciplinebackonforthesametraders, wecanobserveitspowertoerodetheseshapingeffectswhich,giventheircumulativenature,couldbequitelarge.
Topreviewourresultswefindthat,intheabsenceofmarketdiscipline,bothmanipulationsinduceextensiveshaping effects.Whentheseshapingeffectsareexposedtothefullforcesofmarketdiscipline,theyaresometimesweakened,but arenoteliminated.
Ultimately,shapingmayariseforanumberofreasons.Oneextremepossibilityisthatagentsdonothavepreferences whentheygettothemarket,andconstructthemusingarbitrarycuesasanchors(e.g.Slovic,1995;LichtensteinandSlovic, 2006;Arielyetal.,2006;Stewartetal.,2006).Ortheymayreacttopriceinformationbecausetheyusepricesasreference points(e.g.Thaler,1985;Putler,1992;Isoni,2011;WeaverandFrederick,2012;Bordaloetal.,2012).Ourresultssuggestthat thepropositionthatpreferencesarecompletelymalleableistooextreme.Rather,theyarecompatiblewiththeexistenceof underlyingpreferencesofsomeform,butsuggestthatthesearesusceptibletoextraneousinfluencesthatmayhavesome longtermresidualeffect,inspiteofthediscipliningforcesoperatinginmarkets.
Thepaperisorganisedasfollows.InSection2,wedescribeourmarketinstitution–amedianpricesellingauction– andderivetheinstitution-specifichypothesesaboutshapingandmarketdisciplinewhichourexperimentswillexamine.In Section3,wedescribeourbroadexperimentaldesign.InSection4,wepresentourfirstmanipulation,designedtotestfor shapingeffectsobtainedthroughexogenouslyinfluencedpriceexpectations.Section5focusesonoursecondmanipulation, inwhichwestudytheroleofendogenouspricefeedback.SomeissuesraisedbyourresultsarediscussedinSection6.In Section7weoffersomeconcludingremarks.
2. Shapingandmarketdisciplineinrepeatedsellingauctions
Allofourtreatmentsemployrepeatedmedian-pricesellingauctionsformonetarylotteries.Ineachauctionround,each traderisendowedwithalotteryandisaskedtoconsiderasetofdiscreteamountsofmoneyspanningagivenrangeandto say,foreachamount,whetherornottheywouldacceptitinexchangeforthelottery.Theelicitationprocedure(described indetailinSection3)isconstrainedtoimposeconsistency:anyonewhoreportswillingness(unwillingness)toacceptsome amountxmustalsoreportwillingness(unwillingness)toacceptanyhigher(lower)amount.Thesmallestamountofmoney thetraderwouldacceptistheirwillingness-to-acceptvaluation(WTA);thelargestamountofmoneytheywouldnotacceptis theirnot-willing-to-acceptvaluation(NWTA).(Implicitly,thetraderisreportingthattheyvaluethelotteryatleastasmuch asNWTAbutnolessthanWTA.)ThemedianNWTAisidentifiedandannouncedasthemarketpriceforthatround.Traders whoseNWTAsarestrictlylessthantheannouncedprice(i.e.whohavereportedwillingnesstoacceptthatprice)selltheir lotteriesbacktotheexperimenteratthatprice;andifthisroundoftradingisrandomlyselectedattheendoftheexperiment tobethebasisofpayment,theyarethenpaidthatamount.Theothertraderskeeptheirlotteries:ifthisroundisselected, theyplayoutthelotteryandarepaidaccordingly.Sinceweareinterestedintheshapingeffectofpricecues,thereisno resolutionoflotteriesuntilafterthelastroundandthereisnoinformationaboutothertraders’responsesexceptintheform ofthemedianNWTAwhichconstitutesthemarketprice.
Attheendofeachauctionround,tradersaretoldthemarketpriceforthatroundandhencewhethertheyhavesoldor not.Accordingtotheshapinghypothesis,tradersareliabletorevisetheirNWTAsforthenextroundinthedirectionofthe observedprices.ThiscreatesatendencyforNWTAstobepulledinthedirectionofpricecues,otherthingsbeingequal.
Thistendencyisconceptuallydifferentfromvalueaffiliation.Affiliatedvaluesreflectapositivecorrelationbetween non-observedobjectivepropertiesofagood(e.g.theworthofacommon-valueresource),whileshapingreflectsacorrelation
betweensubjectiveattitudes(e.g.eachindividual’svaluationofaprivategood)originatingfromapublicsignal.Fora con-trolledtestoftheshapinghypothesis,weneedadesignthatisabletoscreenoutobjectiveaffiliation,whileallowingfor subjectivevaluestobecorrelated.Whetherornotshapingcanbereconciledwithstandardrationalityassumptionsisa legitimatequestion,butnotthefocusofourinvestigation,whichcentresonwhetherthephenomenonexistsornot,and howstrongitis.
Ourcontrolledtestmakesuseofseveralconvenientpropertiesofmonetarylotteries.Becauseattitudestoriskcanvary acrossindividuals,agivenlotteryislikelytohavedifferentprivatevaluesfordifferentindividuals.Sothemarketpricefor alotteryshouldprovidenoobjectiveinformationthatisrelevantforanindividual’sjudgementofitsprivatevaluetothem, eventhoughitreflectsotherpeople’sjudgementsaboutitsworth.Inaddition,sincethelotteriesweusehavenoimmediate substitutesoutsidethelabandare‘consumed’insidethelab(inthesensethatlotteryoutcomesareresolvedattheendof theexperiment),thevalueofalotterytoaparticipantcannotberationallyaffectedbyinferencesaboutexternaltrading opportunities.Afurtherfeatureoflotteries,whichweshallexploit,isthattheirmaximumandminimumprivatevaluesare givenbythehighestandlowestoutcomes.
Controllingforobjectivelyaffiliatedvaluesandoutsidetradingopportunitiesisnottheonlychallengeonefaceswhen testingforshapingeffects.Asweexplainbelow,ifotherfactorssuchasmarketdisciplinearenotcontrolledfor,observing thatvaluationsmovetowardspreviouslyobservedpricesisnotnecessarilyevidencethatpreferencesaremarket-dependent. Thefirstgoalofourexperimentistoseeifshapingeffectsoccurwhenmarketdisciplineisswitchedoff.
Themarketdisciplinehypothesispositsthatagentspossesspreferencesofsomeform,whichareindependentofthemarket. Forexpositionalconvenience,herewemaketheconventionalassumptionthateachtraderhasauniquetrueprivatevalue forthetradedobject.1
OurmarketinstitutionisavariantoftheVickreyauction(Vickrey,1961).Soiftradershavetruemarket-independent privatevalues,itisaweaklydominantstrategyforeachtradertoreportanNWTAconsistentwiththeirprivatevalue(Milgrom andWeber,1982).Ifthesetruevalueswerereadilyaccessibleandthedominantstrategywereeasilycomprehended,then traderswouldbehaveoptimallyeachtimetheytookpartintheauction,andrepetitionwouldhavenoeffectwhatsoever. However,evenifunderlyingpreferencessatisfystandardassumptions,truevaluesmaynotbereadilyaccessible,orthebest strategymaynotbeimmediatelyobvious,sothattheremaybecircumstancesinwhichtradersdeviatefromthisstrategy inonewayoranother.Suchdeviationsmayprovecostly,inwhichcasemarketdisciplinecancomeintoplay.
Toseethis,consideramedian-priceauctionwithntraders(wherenisanoddnumber),indexedbyi=1,...,n.Assumeeach traderhasatruevalue
v
i.Ifthetraderrespondsoptimallytotheelicitationprocedure,theirreportedWTAwillbegreaterthanorequalto
v
i,andtheirreportedNWTA,whichwedenoteai,willbelessthanorequaltov
i.However,theiractualresponsemaybedifferentbecauseofsome‘error’(whichmayberandom,ormayreflectsomesystematicbehaviouralanomaly). MarketdisciplineoperatesbyexposingthecasesinwhichdeviatingfromtheoptimalstrategyofreportingWTA≥
v
i≥airesultsinanundesiredoutcome.Thishappenswheneverai≥p>vior
v
i>p>ai.Inthefirstcase,thetraderover-asksandfailstosellatp,whichismorethantheirtruevaluation,thusforegoingapotentialgain.Inthesecondcase,theyunder-askand sellatp,whichislessthantheirtruevaluation,makingthemworseoff.Inbothofthesesituations,giventherealisedprice, theywouldhaveexperiencedabetteroutcomehadtheyfollowedtheoptimalstrategy.Toavoidincurringthesameloss again,theyshouldrevisetheiraiinthedirectionof
v
iinsubsequentrounds.2Noticethat,iftheydoso,theyalsoreviseaiinthedirectionofthemarketprice.Thatis,marketdisciplineproducesanempiricaltendencyanalogoustoashapingeffect, whichiswhy,whenmarketdisciplineisinoperation,itisdifficulttoisolateshapingeffects.
However,thisdiscipliningeffectdoesnotnecessarilycometobearoneverytraderwhodoesnotbehaveoptimally.There aresituationsinwhichtradersover-askorunder-askbuttheauctionstillhasthesameimplicationsforthemasiftheyhad followedtheoptimalstrategy.Thishappenswheneverp>ai>viorelse
v
i>ai≥p.Sincemarketdisciplineworksonlyontradersforwhomai≥p>vior
v
i>p>ai,thetraderswhoaresubjecttodisciplinearelikelytohavetruevaluesthatarerelativelyclosetothemarketprice.Ifthemarketisrepeated,changesinbehaviour willcontinuetobeinducedamongsuchmarginaltradersuntiltheyarenolongervulnerabletomarketdiscipline.Thisisthe caseifandonlyifp>
v
iholdsforalltradersiwhosell,andv
j≥pholdsforalltradersjwhodonotsell.OnthesimplifyingassumptionthatthemedianNWTAisreportedbyonlyonetrader,the(n−1)/2traderswiththelowestNWTAssell,andthe (n+1)/2traderswiththehighestNWTAsdonotsell.Thus,whennotraderissubjecttomarketdiscipline,themarketprice liesbetweenthe[(n−1)/2]thhighestandthe[(n+1)/2]thhighesttruevalues–justasconventionalmicroeconomictheory wouldpredictforamedian-pricemarket.Inotherwords,ifonedefinesthetruemarketpriceasthepricethatwouldprevail ifalltradersfollowedtheoptimalstrategy,marketdisciplineproducesatendencyforthemarketpricetoconvergetoitstrue value.Thisistheimplicationofthemarketdisciplinehypothesisthatwewilltest.
1AswewillexplaininSection6,marketdisciplinedoesnotrequirethatpreferencesarecharacterisedbysuchadegreeofprecision,butcanoperatein
ananalogouswayonimprecisepreferences,aslongastheseareindependentofthemarket.
2Here,wemakenofurtherassumptionaboutthespeedwithwhichtruevalues(ortheoptimalstrategyforreportingthese)arediscovered.Onanextreme
interpretation,onecouldthinkthatasinglecostlyerrorwouldbesufficientforaparticipanttodiscovertheirtruevalue.Alternatively,convergencecould beagradualprocess,inwhichNWTAsareslowlyupdatedinthedirectionsuggestedbytheerrortheparticipantmade(i.e.bydecreasingone’sNWTAin caseofover-askingandincreasingitintheoppositecase).
However,asnotedearlier,theremaybetraderswhosetruevaluesarefarenoughawayfromthepricesgeneratedby repeatedmarketsthattheirNWTAsneverfallbetweentheirtruevaluesandtherealisedprice.Suchtradersneverexperience thecoststhatgiverisetomarketdiscipline.Nevertheless,theyareexposedtotheshapingeffectsofpricefeedback.Iftraders withthesecharacteristicscanbeidentified,itispossibletotestfortheexistenceandstrengthofshapingeffectswhile screeningouttheeffectsofmarketdiscipline.Oneofourmanipulationsallowsustoswitchoffmarketdisciplineandexpose traderstoendogenouspricefeedbackcueswhich,ifshapingoccurs,willtendtopulltheirNWTAsawayfromtheirtrue valuesinaparticulardirection.Then,bysettinguprepeatedmarketsinwhichallparticipants’NWTAshavepreviouslybeen shapedinthesameway,wecanswitchmarketdisciplineonandtestwhetherpricesinthesemarketsconvergetowardsa hypothesisedtruemarketprice.
Inourexperiment,wewilltestthefollowinggeneralimplicationsofshapingandmarketdiscipline:
Hypothesis1(Pureshaping). Absentmarketdiscipline,NWTAsarepulledinthedirectionofuninformativepricecues.
Hypothesis2(Marketdiscipline). Withrepetition,pricesconvergetovaluesindependentofuninformativepricecues.
3. Experimentalprocedures
Theexperimentwasconductedin18sessionsattheUniversityofEastAngliaandwasimplementedusingtheZurich ToolboxforReadymadeEconomicExperiments(Fischbacher,2007).The204participantswererecruitedviaemailshots fromthegeneralstudentpopulation.
Ineachsessionparticipantstookpartinthreeauctions,eachrepeatedeighttimes,foratotalof24auctionperiods. Ineachsession,therewereeithertenortwelveparticipants,whowereassignedtotwotradinggroupsofeitherfiveor sevenparticipants(oursecondmanipulationrequiredustohaveauctiongroupsoftwodifferentsizes).Participantswere re-matchedtoformdifferenttradinggroupsbeforethestartofeachauction,butthecompositionoftradinggroupswaskept constantduringtheeightroundsofanauction.
OneoftheauctionsimplementedourEXPmanipulation(shapingthroughpriceexpectations;seeSection4fordetails). Whetherthisauctioncamefirstorlastwascounterbalancedacrosssessions.Theremainingtwoauctionsimplementedour FBKmanipulation(shapingthroughpricefeedback),andwerealwaysruninsequence(seeSection5).3
Ineachroundofeachauction,eachparticipantsawalotterydisplayedontheirscreen.Lotteriesalwayshadexactlytwo possibleoutcomes;thesewerepositiveamountsofmoneyorprizes.Participantsweretoldtheyownedthelotterydisplayed ontheirscreenandhadtheopportunitytosellit.Theywerepresentedwithalistofpricesandaskedtostatewhetheror nottheywouldbewillingtoselltheirlotteryateachprice.Theprices(in
£
)alwaysrangedfromz+0.01toy,whereyand z,y>z,werethetwoprizes.4The0.01wasaddedtothelowerprizetoensurethatNWTAwasdefinedforparticipantswhoacceptedeveryprice,inwhichcasetheirNWTAwassetequaltoz.Forparticipantswhorejectedeveryprice,NWTAwasset equaltoy.Itwasexplainedtoparticipantswhyitwouldbeinconsistenttovaluethelotteryaboveyorbelowz.5Participants
weretoldthatiftheyagreedtosellfor
£
x,theyshouldalsobewillingtosellforanyamountgreaterthan£
x(i.e.theywere notallowedtoswitchbackandforthinthemultiple-pricelist).Onceentered,responsescouldbechangedbyrestartingthe wholeprocess,butthiswasonlypossibleonce.Attheendoftheprocess,thecomputerrecordedeachparticipant’sNWTA. AfteranNWTAwasrecordedforeachtraderinagroup,thecomputerdeterminedthemarketpricebyselectingthemedian oftheseamounts.Theinstructionsexplainedhowthemarketpricewasobtained,andthepossibletradingimplications. Beforethebeginningofthefirstauction,therewasapracticeroundforparticipantstofamiliarisethemselveswiththe elicitationprocedures.Theirunderstandingofthebasicruleswastestedbymeansofacomputerisedquestionnaire.Attheendofeachsessionthecomputerrandomlyselectedoneoftheauctionroundsthathadbeencompletedbythe participant.6Iftheparticipanthadsoldtheirlotteryinthatround,theyreceivedthemarketprice.Iftheyhadkeptthelottery,
theyplayeditoutandreceiveditsoutcome.Lotterieswereresolvedbydrawingoneofahundrednumbereddiscsfroma bag.Theoutcomeassociatedwitheachnumberwasclearlyindicatedbythelotterydisplay(seeonlineAppendixforan example).Iftheselectedauctionroundwasfromthefirstmanipulation,participantswerealsorewardedfortheiraccuracy inpredictingthemarketpriceintheothergroup(seeSection4.1).Sessionslastedanhourandfiveminutesonaverage,and theaverageearningwas
£
7.50.3 Thefulllistofsessions,outliningourmanipulationsandcounterbalancing,canbefoundintheAppendix,TableA.2.1.
4 Inordertoavoidhavingtopresentparticipantswithverylonglistsofprices,theelicitationprocedurehadanumberofsteps.Inthefirststep,the
pricesonofferstartedfrom£(z+0.01)andwentto£ycoveringallmultiplesof£1(foroneofourlotteries,wehadanextrastepatthebeginning,inwhich allpricesexceptthelowestweremultiplesof£10).Inthelaststep,the£1rangebetweenthelowestpricethattheindividualacceptedandthehighest pricesherejectedinthepreviousstepwasbrokendownin£0.10intervals,toarriveatanNWTAexpressedasamultipleof£0.10.Forsomeparticipants priceswentupfrom£(z+0.01)to£y,forotherstheywentdownfrom£yto£(z+0.01).Somescreenshotsofthiselicitationprocedurecanbefoundinthe Appendix.
5 AsweexplaininSection5.1,thisaspectoftheelicitationprocedurewascrucialforoursecondmanipulation.Gneezyetal.(2006)reportevidencethat
someparticipantsvaluelotterieslessthantheirworstpossibleoutcome,afindingthattheylabeltheuncertaintyeffect.However,oneoftheconditions forthiseffecttooccuristhattheprizesofthelotterytobevaluedarenotdirectlyexpressedinmonetaryterms(ibid.p.1290),whichisnottrueinour experiment.
6 Sinceoursecondmanipulationrequiredthatsomeparticipantsdidnottakepartinoneoftheauctions,someparticipantscompleted24roundsand
Table1
RangesforpriceexpectationcuesinEXPmanipulation.
HighCue LowCue
£20ormore £1.00–£2.00
£19.90–£15.00 £2.10–£3.00
£14.90–£10.00 £3.10–£4.00
£9.90–£5.00 £4.10–£5.00
£4.90–£1.00 £5.10ormore
InR1andR8,participantsstatedtheirpricepredictionfortheothergroupin theirsessionbyselectingoneoftherelevantfiverangesandthenenteringa valuefallingwithinthatrange.
4. EXPmanipulation:shapingthroughpriceexpectations
4.1. Design
Our firstmanipulationwasdesigned toinvestigatewhetherNWTAscanbeshapedbydivergentpriceexpectations
(Hypothesis1),andwhetheranyresultingshapingeffectsareremovedbymarketdiscipline(Hypothesis2).Thesetreatments involvedarepeatedauction,inwhicheverytraderwasendowedwiththesamelottery.Inordertoallowroomforsizeable shapingeffects,wechosealotterywithafairlyhighvariance.LetK=(
£
z,p;£
y,1−p)denoteagenericbinarylotteryoffering£
zwithprobabilityp,and£
ywithprobability1−p.Thelotteryusedinthispartoftheexperimentwas: E=(£
1,0.95;£
50,0.05).Werandomlydividedparticipantsintotwogroups,whichwelabeltheHighCueandLowCuemarkets.Eachmarket consistedofeightauctionroundsconductedasexplainedinSection3.Wetoldparticipantsthatthereweretwotrading groupsintheroom,andthateveryonehadthesamelottery.Beforethefirstauctionround,weaskedthemtopredictwhat thepricewouldbeintheothertradinggroup,byselectingarangefromalistandthenreportingapointestimatewithinthe chosenrange.Ifoneoftheseauctionroundswasselectedtobepaidattheendoftheexperiment,theywererewardedfor theaccuracyoftheirpredictions,usinganabsolutelossfunctionwithmaximumrewardof
£
2.(Becausepredictionswere madeabouttheothergroup,therewasnoincentiveforparticipantstomisreportvaluationsinanattempttomaketheir predictionscometrue.)Thesameprocedurewasrepeatedbeforethelastauctionroundtoseeifparticipants’expectations adjustedtothepricefeedbacktheyreceived.TheHighandLowCuegroupsdifferedintermsoftherangesofpricestheywerepresentedwith.Theseareshownin Table1.
BothlistsinTable1coverthewholerangeofvaluesbetweenthetwoprizesoflotteryE,buttheydifferintwoimportant respects.First,themiddlerangecontainsmuchhighervaluesfortheHighCuemarketthanfortheLowCuemarket.Second, theHighCuerangesarepresentedindecreasingorder,whiletheLowCuerangesarepresentedinincreasingorder.We expectedbothoftheseaspectstoinducehigherpriceestimatesintheHighCueconditionthanintheLowCuecondition,on thegroundsthatthefirstandthemiddlerangesaremoresalientthantheothers(theseeffectsareknowninthepsychology literatureas‘rangefrequencyeffects’;seeParducci,1965;ParducciandWeddell,1986).7
Ifparticipantshavewell-formedtruevalues,suchmanipulationsshouldhavenoeffect;butiftheyarepronetoshaping effects,thefirst-roundNWTAsintheHighCuemarketscouldbesubstantiallylargerthanthefirst-roundNWTAsintheLow Cuemarket.Sincenofeedbackisprovideduntiltheendofthefirstround,theinitialNWTAsmaybesubjecttoshaping,but cannothavebeenaffectedbymarketdiscipline.Inthissetting,Hypothesis1impliesthatNWTAswillbehigherintheHigh CuegroupthanintheLowCuegroup.
Astheauctionisrepeatedwithfeedback,marketdisciplineisswitchedonandwecaninvestigateitseffects.Because tradershavebeenrandomisedbetweenthetwotreatments,anyunderlyingtruevalues–includingthetruevalueofthe marketprice–shouldhavesimilardistributionsinthetwosubsamples.Therefore,astheauctionisrepeated,Hypothesis2 impliesthatmarketdisciplineshouldinduceconvergenceofthepricesprevailinginthetwomarkets.
4.2. Results
Table2reportsthedataoftheHighCueandLowCuemarketsofourEXPmanipulation.Forbothtreatments,thetable reportsthenumberofobservations,mean,medianandstandarddeviationofthepredictedmarketprices,theactualNWTAs andtheactualmarketpricesinrounds1(R1)through8(R8).
WewilldiscussthedatawithreferencetoFig.1,whichusesthefollowingconventions.Solidmarkersjoinedbysolid linesrefertotheHighCuegroup,whileemptymarkersjoinedbydashedlinesrefertotheLowCuegroup.MeanNWTAsare
7Thismanipulationbearssomesimilaritytoanchoringmanipulations(e.g.Arielyetal.,2003),butthefinalpricepredictionis,effectively,aself-generated
Table2
ResultsofEXPmanipulation:shapingthroughpriceexpectations.NWTAvaluationsandmarketpricesforlotteryE=(£1,0.95;£50,0.05).
NWTA Marketprices
Obs. Mean Median St.dev. Obs. Mean Median St.dev.
ELowCuepredictionsR1 102 6.72 4.05 7.23
ELowCuepredictionsR8 102 4.94 4.35 3.18
ELowCueR1 102 10.21 5.40 10.46 18 6.90 6.30 3.26 ELowCueR2 102 8.56 5.35 8.54 18 7.37 5.10 5.68 ELowCueR3 102 7.61 4.90 7.21 18 6.43 4.65 3.95 ELowCueR4 102 8.61 4.80 10.50 18 6.42 4.70 4.65 ELowCueR5 102 7.64 4.40 9.62 18 5.03 4.45 2.08 ELowCueR6 102 6.85 4.30 7.62 18 4.74 4.25 2.07 ELowCueR7 102 7.20 4.20 9.47 18 4.61 4.30 1.92 ELowCueR8 102 6.73 4.20 7.92 18 4.38 4.20 1.98
EHighCuepredictionsR1 102 12.11 12.25 7.99
EHighCuepredictionsR8 102 7.51 6.05 4.82
EHighCueR1 102 12.11 10.05 9.97 18 10.19 10.00 4.17 EHighCueR2 102 9.64 8.90 6.95 18 8.37 7.90 3.10 EHighCueR3 102 9.36 7.40 8.28 18 8.68 8.00 3.91 EHighCueR4 102 8.04 7.30 5.04 18 8.19 7.40 3.61 EHighCueR5 102 8.13 6.40 5.99 18 7.48 7.10 3.14 EHighCueR6 102 8.28 6.20 7.57 18 6.77 6.50 2.86 EHighCueR7 102 8.29 5.40 8.64 18 6.06 5.20 2.51 EHighCueR8 102 8.08 5.15 8.96 18 6.33 5.70 2.71
102participants(18groupsof5or7)wereassignedtotheLowCuetreatment,102(18groupsof5or7)totheHighCuetreatment.Eachgrouptradedfor8 rounds,withfeedbackaboutmarketpriceandtradingoutcomeaftereachround.PricepredictionswereelicitedbeforeR1andbeforeR8usingeitherLow orHighcues. 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 0 1 2 3 4 5 6 7 8 V al ues (£ ) Auction Round Exp. High Ask High Price High Exp. Low Ask Low Price Low NWTA High NWTA Low
Fig.1. ResultsoftheEXPmanipulation.EvolutionofNWTAsandpricesintheHighCueandLowCuemarkets.Note:HighCuedenotedbysolidmarkers joinedbysolidlines.LowCuedenotedbyemptymarkersjoinedbydashedlines;squaresaremeanNWTAs;trianglesaremeanactualmarketprices;solid (empty)diamondsaremeanpricepredictionsforHigh(Low)Cue.
presentedbysquares,whilemeanactualmarketpricesareshownbytriangles.Themeanpricepredictionsarerepresented
bythediamonds.
Thefirstissueistoestablishwhetherourmanipulationhasaffectedpriceexpectations.Thisisaprerequisitefortesting
theimplicationsofHypothesis1.InR1,themean(median)expectationis
£
12.11(£
12.25)intheHighCuemarketand£
6.72(
£
4.05)intheLowCuemarket;thisdifferenceisstronglysignificantinaone-tailMann–Whitneytest(Z=−6.44,p<0.001). Giventhatwehavebeensuccessfulatmanipulatingparticipants’priceexpectations,wecantesttheimplicationsof Hypothesis1bycomparingNWTAsinR1betweenthetwogroups,whenmarketdisciplinecannot operate.Themean (median)NWTAinR1is£
12.11(£
10.05)intheHighCuemarketand£
10.21(£
5.40)intheLowCuemarket.8ThetwoNWTA8 Somereadersmayfinditsurprisingthatourparticipants’NWTAsforalotterywhoseexpectedvalueisjust£3.45weresohigh.However,lotteryEhas
thecharacteristicsofa‘$-bet’(i.e.alotteryofferingsmallchanceofwinningalargeprize).Thepreferencereversalliteratureprovidesrobustevidencethat suchlotteriesareoftenovervaluedrelativetotheirexpectedvalues(seeSeidl,2002).
Table3
EvolutionofNWTAvaluationsinLowCueandHighCuemarketsforsellersandnon-sellers.
Auctionround t NWTAtifsoldatt−1 NWTAtifnotsoldatt−1
<0 0 >0 Mean <0 0 >0 Mean LowCue R1 1 – – – – – – R2 2 3 7 31 2.85 48 7 6 −4.66 R3 3 7 11 24 1.03 38 13 9 −2.34 R4 4 6 12 21 2.42 43 9 11 0.12 R5 5 11 16 13 1.23 37 11 14 −2.38 R6 6 10 16 13 0.41 46 9 8 −1.55 R7 7 12 16 11 1.18 31 22 10 −0.16 R8 8 15 14 12 1.14 28 22 11 −1.55 HighCue R1 1 – – – – – – R2 2 3 11 26 3.29 53 3 6 −6.20 R3 3 5 11 23 3.97 48 6 9 −2.91 R4 4 5 14 23 1.40 42 9 9 −3.22 R5 5 6 17 18 1.42 44 8 9 −0.80 R6 6 9 20 9 0.77 43 13 8 −0.22 R7 7 11 14 17 1.63 40 12 8 −1.11 R8 8 7 12 20 1.12 33 16 14 −1.03
ChangeinNWTAfromroundtoround,NWTAt=NWTAt−NWTAt−1,forparticipantswhosoldandparticipantswhodidnotsellinthepreviousround.For
eachroundoftheauction,thenumbersofsellersandnon-sellerswhodecreased,keptconstantandincreasedtheirNWTAvaluationisreportedtogether withthemeanNWTAt.
distributionsdiffersignificantlyinaone-tailMann–Whitneytest(Z=−2.17,p<0.05).Thesefindingsprovidesupportforthe
implicationsofHypothesis1,andconstituteourResult1a.
Result1a–Shapingthroughpriceexpectations.Intheabsenceofmarketdiscipline,NWTAsarepulledinthedirection ofuninformativepriceexpectationcues.
Wecannowaskhowfartheshapingeffectsofpriceexpectationcuespersistinthefaceofmarketdiscipline.According toHypothesis2,marketdisciplineshouldleadmarketpricestoconvergetovaluesthatareindependentofuninformative pricecues.Sinceparticipantswererandomlyallocatedtothetwotreatments,thismeansthat,withrepetition,market pricesshouldconvergetothesamevalueintheHighandLowCuegroups.Forthistesttobemeaningful,weneedour EXPmanipulationtohaveproducedadifferenceinmarketprices.InR1,themean(median)marketpriceis
£
10.19(£
10) intheHighCuegroupand£
6.90(£
6.30)intheLowCuegroup,adifferencethatisstatisticallysignificantinaone-tail Mann–Whitneytest(Z=−2.36,p<0.05).InordertotesttheimplicationsofHypothesis2,wecomparemarketpricesinthelastroundoftheauction,whenmarket disciplinehashadthebestopportunitytoexertitseffects.InR8,themean(median)priceis
£
6.33(£
5.70)intheHigh Cuegroupand£
4.38(£
4.20)intheLowCuegroup.Althoughtheabsolutedifferencebetweenthemean(median)prices observedinthetwogroupsissmallerinR8thaninR1,thedistributionsofmarketpricesstilldiffersignificantlybetween thetwogroupsinaone-tailMann–Whitneytest(Z=−2.66,p<0.01).9Therefore,wedonotfindsupportfortheimplicationsofHypothesis2.10ThisconstitutesourResult2a.
Result2a–Persistenceofshapingthroughpriceexpectations.Aftereightauctionrounds,pricesthathavebeenshaped byuninformativepriceexpectationcuesarestillsignificantlyaffectedbythosecues.
TheevolutionofNWTAvaluationsovertheeightauctionroundsissummarisedinTable3.Forrounds2through8ofthe LowCueandHighCuemarkets,thetablereportsthenumbersofparticipantswhodecreased,keptunalteredorincreased theirNWTA,dependingonwhethertheysoldorkepttheirlotteryinthepreviousround.Forbothsellersandnon-sellers, thetablealsoshowsthemeanchangeinNWTA,NWTAt=NWTAt−NWTAt−1,fromroundtoround.
9Participants’priceexpectations,whichweelicitedagainbeforeR8,areconsistentwiththispersistentdifference.Themean(median)R8predictionis
£7.51(£6.05)intheHighCuegroupand£4.94(£4.35)intheLowCuegroup.Whilethetrendsinthedistributionsofmarketpricesareclearlyvisiblein Fig.1,itispossiblethattheelicitationofpriceexpectationsbeforeR8hadanamplifyingeffectonNWTA(andprice)differences.Ifwetesttheimplications ofHypothesis2usingtheR7pricesinstead,westillfindsignificantdifferences(Z=–2.06,p<0.05inaone-tailMann–Whitneytest).
10ThetheoreticalanalysisofSection2,whichformsthebasisforHypothesis2,presupposesthatcostlyerrorsareentirelyremoved.Thisraisesthequestion
ofwhethereightauctionroundsaresufficientforpricestohavestabilised,aquestionthatwehavetriedtoanswerbyestimatingasimpleasymptotic functionforthemarketprice:pt=p*+(1/t)bwhereptisthemarketpriceinroundt,p*istheasymptoticvalueofthemarketprice,andbisaparameter
relatedtothespeedofconvergence.Apartfromitseaseofinterpretation,anadvantageofthismodelisthatitcanbeeasilyestimatedwithOLS.Thekey parameterisp*.BycomparingitsvalueintheHighCueandtheLowCuetreatments,wecancheckwhetherpricesaretendingtowardsdifferentvalues.The estimatedvaluesofp*are£6.35intheHighCuegroupand£4.75intheLowCuegroup.Thesevaluesaresignificantlydifferentfromeachother(p<0.1), indicatingthatpricesappeartobetendingtowardsdifferentvaluesinthetwogroups.SeeTableA.3.1intheAppendixfordetails.
Table3revealsastrikingregularity.InboththeHighandLowCuegroups,participantshaveaprevalenttendencyto increasetheirNWTAiftheysoldinthepreviousroundandanoppositetendencytodecreasetheirNWTAiftheydidnot sell.11Theeffectweakensovertime,asmoreparticipantskeeptheirNWTAsunaltered.Thisconstitutesourfirstobservation.
Observation1–NWTAupdatinginHighandLowCuemarkets.InHighandLowCuemarkets,sellerstendtoincrease theirNWTAandnon-sellerstendtodecreasetheirNWTA.
Observation1showsthatbothsellersandnon-sellersrevisetheirNWTAsinthedirectionofthemarketprice.Butbecause wehavenoinformationabouttheirtruevalues,wecannottellwhetherthisisaneffectofshaping,marketdisciplineorboth. Thisclearlyillustrateswhy,ifoneisinterestedinthecumulativeeffectofshaping,itisusefultoshutoffmarketdiscipline. ThisistheobjectiveofourFBKmanipulation.
5. FBKmanipulation:shapingthroughpricefeedback
5.1. Design
OursecondmanipulationisdesignedtostudytheeffectofpricefeedbackcuesonNWTAs.Unlikethepriceexpectation cuesthatweemployedinourEXPmanipulation,thesecuesareendogenous,astheyaregeneratedbythenormaloperation ofthemarket.
Werandomlydividedparticipantsbetweentwotreatments,thecontroltreatmentandtheshapingtreatment.Thecontrol treatmentwasruninsessionswithtenparticipants.Eachparticipanttookpartintwoseriesofrepeatedauctionsbelonging totheFBKmanipulation,oneforeachoftwodifferentlotteries,alow-valuelotteryLandahigh-valuelotteryH(seebelow forthespecificsoftheselotteries).Ineachrepeatedauction,therewereeightrounds.Theorderofthetwolotterieswas counterbalancedbetweensessions.Inthistreatment,alltraderstakingpartinthesameauctionwereendowedwiththe samelottery,sothatthepricefeedbackreflectedthemedianNWTAforthatlotteryintherelevantround.Eachofthese marketshadexactlyfivetraders.
Bycontrast,theshapingtreatmentwasruninsessionswithtwelveparticipants.Asinthecontroltreatment,therewere twoseriesofrepeatedauctionsinvolvingeightroundsperseries.Thefirstseriesineachsessionwasdesignedtocreate conditionsunderwhichthecumulativeeffectsofshapingcouldbeobservedintheabsenceofmarketdiscipline,hence allowingustotesttheimplicationsofHypothesis1.Thesecondseriesofeightroundswasdesignedtoexaminetheimpact ofmarketdisciplineonanysucheffects,asatestofHypothesis2.
Ineachsessionoftheshapingtreatment,thefirstseriesconsistedofmixedmarkets,inwhichdifferenttraderswere endowedwithdifferentlotteries.Therewasonemarketwithseventraders–threeendowedwithL,threewithH,andone withamedium-valuelotteryM–andanothermarketwithfivetraders–twoendowedwithL,twowithH,andonewithM. ThisarrangementallowedustodropthetwotraderswhohadbeenendowedwithMfromthesecondseries12,andreassign
thetenparticipantswhohadpreviouslybeenendowedwithLorHtotwohomogenousmarkets.Thatis,thefivetraders inasessionwhohadbeenendowedwithLinthefirstseriescontinuedtotradeL,butnowaspartofonemarketwhere allfivetraderswereendowedwithL,whiletheothermarketconsistedofthefivetraderswhohadbeen,andcontinuedto be,endowedwithH.Sincebothhomogeneousandcontrolmarketsconsistedoftradersendowedwiththesamelottery(L insomemarkets,Hinothers),wecanjudgethepersistenceofshapingeffectsbycomparingthepatternsofpricesinour second-serieshomogeneousmarketswiththoseinthecorrespondingcontrolmarkets.
ThekeyinnovationoftheFBKmanipulationistheabilitytoshapeNWTAsthroughthefirst-seriesmixedmarkets.We nowexplainthismanipulationinmoredetail.Weusedthefollowingbinarylotteries:
L=(
£
2,0.8;£
9,0.2), M=(£
7,0.5;£
9,0.5), H=(£
7,0.2;£
14,0.8).Sinceeachtrader’sNWTAwasconstrainedtobenogreaterthanthehigherlotteryprizeandnolessthanthelower(see Section3),thedistributionofendowmentsinthemixedmarketsensuredthatthemarketpriceinthesemarketswasbound tolieinthe(
£
7,£
9)interval.13Inthesemarkets,tradersweregivennoinformationaboutthelotteriesownedbyothertraders.Whenweexplainedhow participantsshouldrespondtoquestionsabouttheirwillingnesstoacceptstatedprices,westressedthattheseresponses
11 Inbothgroups,beingasellerinthepreviousroundhasahighlysignificantpositiveeffectonNWTA
tinarandom-effectspanelregression.SeeTable
A.3.2intheAppendixfordetails.
12 Theseparticipants,whosepresencewasrequiredforourmanipulationofthepricefeedback(seebelow),wereinsteadaskedtocompleteanunrelated
questionnaire.
13 BecauseoftheconstraintonindividualNWTAs,onlyLtraderscouldreportNWTAsbelow£7,andonlyHtraderscouldreportNWTAsabove£9.The
presenceofoneMtraderineachgroupensuredthatfewerthanhalfofthetradersinthemarketwereendowedwithL,andfewerthanhalfwithH,sothe medianaskwasboundtobeinthe(£7,£9)interval,therangeofadmissibleasksforMtraders.
werestrictlypersonalandshouldsimplyreflecttheirownevaluationofthelotterytheywereendowedwith.Underthese circumstances,fromaconventionalpointofview,itshouldbeirrelevantwhatotherpeopledoandwhattheactualmarket priceturnsouttobe.14
Akeyfeatureofourmixedmarketdesignisthat,sinceNWTAsinthe(
£
7,£
9)intervalwereadmissibleforalllotteries, thepricefeedbackwascredibleforalltraders.However,theexpectedvalueofL(£
3.40)wasmuchlowerthanthelowest possibleprice(£
7),whiletheexpectedvalueofH(£
12.60)wasmuchlargerthanthehighestpossibleprice(£
9).Thismakes thepricefeedbackfairlyhighforLtraders,andfairlylowforHtraders,creatingthepotentialforsizeableshapingeffects.BecausetheexpectedvaluesofLandHliewelloutsidetherangeofpossibleprices,mostLandHtraderswereunlikelyto besubjecttomarketdiscipline,eveniftherewerestrongshapingeffects.ConsiderLtradersundertheassumptionthattrue valuesexistandareunique(thecaseofHtradersisamirrorimage).SincethesetraderswerereportingNWTAsforlotteries thatwerefarinferiortothoseheldbythemajorityoftradersinthemarket,itisnaturaltoexpectthatinthefirstauction roundthepricewouldbemuchhigherthanthetruevaluesandNWTAsofalmostallLtraders.Ifthiswasthecase,Ltraders wouldsellatsuchanadvantageouspricethatmarketdisciplinewouldnotbite.Nevertheless,thehighpricewouldstillhave thepotentialtoexertanupwardshapingeffectontheirNWTAs.Becauseofitscumulativenature,oversuccessiveauction roundsthisshapingeffectcouldbequitestrongandyetstillnotraiseNWTAstoalevelatwhichanyLtraderswouldfail tosell(andhencebecomesubjecttomarketdiscipline).Thisfeatureofourdesignallowsus,ineffect,toswitchoffmarket disciplineandtestwhethervaluationscanbeshapedbyendogenouspricecues.15Therefore,Hypothesis1impliesthat,after
repeatedexposuretomarketpricefeedbackcues,NWTAsforeachoflotteryLandHshoulddifferbetweenmixedmarkets andthecorrespondingcontrolmarkets.
Marketdisciplinewasswitchedonwhentraderswerereassignedtohomogeneousmarkets.SupposethatfiveLtraders whoseNWTAshaveallbeenshapedupwardsinmixedmarketsnowtakepartinthesamehomogeneousmarket,andthat eachoftheminitiallyreportsthesameNWTAasinthelastroundofthemixedmarket.Then,thetradersreportingthe highestoftheseNWTAswillnolongersell.Forthesetraders,ai≥p>vi:theyover-askandfailtosellatanadvantageous
price.Accordingtothemarketdisciplinehypothesis,theyshouldreactbyreducingtheirNWTAsinthenextround.This tendstoproduceafallinthepriceinthenextround.Again,thetraderswhonowstatethehighestNWTAsfailtosell; ifthesetradersarestillover-asking,marketdisciplineleadsthemtolowertheirNWTAsinthefollowinground,andthe pricefallsagain;andsoon,untilanyover-askingiseliminatedinthemarginaltraders.Amirror-imageargumentappliesto homogeneousHmarkets.
AsexplainedinSection2,ifagentshadpreciseandmarket-independenttruevalues,thisprocessofadjustmentwould cometoanendwhenthepricereacheditstruevalue,whichisfullydeterminedbythedistributionoftraders’truevalues. Sinceparticipantswererandomisedbetweenhomogeneousandcontrolmarkets,thedistributionsoftraders’truevalues shouldnotdiffersystematicallybetweenthesemarkets.Thus,Hypothesis2impliesthat,withrepetition,pricesin homo-geneousmarketsshouldconvergetovaluesthatarenotsignificantlydifferentfromthoseobservedinthecorresponding controlmarkets.
5.2. Results
Table4reportsthedatawegatheredinthemixed,homogeneousandcontrolmarketsfortheLlottery(inthetoppanel) andtheHlottery(inthebottompanel).Eachrowreportsthenumberofobservations,mean,medianandstandarddeviation ofboththeNWTAsandmarketpricesforeithertheHorLlottery.
ThekeyresultswillbepresentedwiththeaidofFig.2,whichshowstheevolutionofNWTAsandpricesinmixedand controlmarkets(panelsaandb)andinhomogeneousandcontrolmarkets(panelscandd).Ineachpanel,squaresdenote meanNWTAandtrianglesdenotemeanmarketpricesintherelevantmarket.Dataformixedandhomogeneousmarkets areshownusingsolidlines(rememberthatthesameparticipantstookpartinmixedmarketsfollowedbyhomogeneous markets),whiledashedlinesrefertocontrolmarkets.
Westartbylookingatmixedmarkets.Oneofourobjectiveswiththesemarketswastoprovidepricefeedbackwithout thepotentialforthisfeedbacktotriggermarketdiscipline.Inordertohaveasenseofwhetherweachievedthis,wecan lookatthefrequencieswithwhichLtraders(whowereexpectedtotrade)andHtraders(whowereexpectedtokeeptheir lotteries)didactuallysell.Overthefirstsevenroundsofthemixedmarkets(wedonotconsiderthelastroundbecause feedbackreceivedinthatroundcouldhavenoeffectonobservedNWTAsorprices),68%oftheLtraderssoldineveryround, and93%soldatleastsixtimes.88%oftheHtradersneversold,and97%soldatmostonce.16Thus,althoughasmallnumber
ofexceptionallyhighLNWTAsandanevensmallernumberofexceptionallylowHNWTAsprobablyweresubjecttomarket
14Wehavenowaytotellwhetherornotparticipantsthoughtothertradersintheirgroupweretradingthesamelottery.Eachindividual’sdecisionwas
framedasadecisionastowhetherornottheywantedtoselltheirlotteryatanygivenprice,andincentivisedaccordingly.Thisisallweneedforourtestsof theshapinghypothesis.Ourassignmentoflotteriestoparticipantsisanalogoustotheinducedvaluetechniqueadoptedinexperimentaleconomicssince Chamberlin(1948)andSmith(1962)tostudymarketbehaviour.Weuselotteriesinsteadoftokenstomakelargeshapingeffectspossible,andemploya median-priceinsteadofadouble-oralauction,butparticipantsonlytradeifthepriceisacceptabletothem.
15TheauxiliaryhypothesisthatLandHtradersarerarelysubjecttomarketdisciplinecanbeverifiedbyconsideringthefrequencieswithwhichLtraders
failtosellandHtraderssell(seeSection5.2below).
Table4
ResultsofFBKmanipulation:shapingthroughpriceexpectations.
NWTA Marketprices
Obs. Mean Median St.dev. Obs. Mean Median St.dev.
NWTAvaluationsandmarketpricesforlotteryL=(£2,0.8;£9,0.2)a
LcontrolR1 60 4.25 4.00 1.10 12 4.15 3.95 0.72 LcontrolR2 60 4.20 4.00 1.16 12 4.04 4.00 0.45 LcontrolR3 60 4.21 3.90 1.38 12 3.85 3.80 0.44 LcontrolR4 60 4.09 3.70 1.37 12 3.78 3.55 0.58 LcontrolR5 60 3.94 3.80 1.15 12 3.78 3.80 0.53 LcontrolR6 60 3.99 3.80 1.14 12 3.80 3.85 0.54 LcontrolR7 60 3.96 3.70 1.20 12 3.73 3.65 0.56 LcontrolR8 60 3.83 3.70 0.93 12 3.68 3.55 0.55 LmixedR1 60 4.64 4.40 1.62 24 7.93 7.90 0.55 LmixedR2 60 4.93 4.70 1.54 24 7.74 7.70 0.53 LmixedR3 60 5.04 4.95 1.51 24 7.60 7.50 0.59 LmixedR4 60 5.53 5.45 1.64 24 7.65 7.60 0.62 LmixedR5 60 5.46 5.45 1.52 24 7.61 7.80 0.55 LmixedR6 60 5.67 5.80 1.64 24 7.64 7.70 0.55 LmixedR7 60 5.72 5.90 1.71 24 7.67 7.65 0.58 LmixedR8 60 5.67 6.45 1.82 24 7.57 7.50 0.50 LhomogeneousR1 60 5.35 5.40 1.49 12 5.53 5.40 0.88 LhomogeneousR2 60 4.94 4.85 1.50 12 4.93 4.85 0.92 LhomogeneousR3 60 4.52 4.40 1.48 12 4.38 4.20 0.87 LhomogeneousR4 60 4.81 4.40 1.75 12 4.87 4.75 1.18 LhomogeneousR5 60 4.65 4.40 1.59 12 4.62 4.35 1.19 LhomogeneousR6 60 4.57 4.40 1.56 12 4.43 4.00 1.13 LhomogeneousR7 60 4.40 4.10 1.47 12 4.23 4.00 0.98 LhomogeneousR8 60 4.60 4.25 1.72 12 4.42 4.20 1.15
NWTA Marketprices
Obs. Mean Median St.dev. Obs. Mean Median St.dev.
NWTAvaluationsandmarketpricesforlotteryH=(£7,0.2;£14,0.8)b
HcontrolR1 60 11.17 11.45 1.66 12 11.46 11.45 0.71 HcontrolR2 60 11.09 11.35 1.36 12 10.97 10.90 0.88 HcontrolR3 60 11.10 10.95 1.34 12 10.93 11.05 1.00 HcontrolR4 60 11.08 11.40 1.39 12 10.93 11.50 1.18 HcontrolR5 60 11.21 11.50 1.45 12 11.02 11.45 1.26 HcontrolR6 60 11.19 11.40 1.41 12 10.98 11.30 1.09 HcontrolR7 60 11.14 11.30 1.34 12 10.98 11.10 1.13 HcontrolR8 60 10.97 11.20 1.32 12 10.93 11.30 1.11 HmixedR1 60 10.91 11.10 1.66 24 7.93 7.90 0.55 HmixedR2 60 10.91 11.25 1.66 24 7.74 7.70 0.53 HmixedR3 60 10.26 10.45 1.97 24 7.60 7.50 0.59 HmixedR4 60 9.95 10.10 1.99 24 7.65 7.60 0.62 HmixedR5 60 9.93 10.15 2.07 24 7.61 7.80 0.55 HmixedR6 60 10.05 10.40 2.07 24 7.64 7.70 0.55 HmixedR7 60 9.86 9.95 2.00 24 7.67 7.65 0.58 HmixedR8 60 10.09 10.50 2.06 24 7.57 7.50 0.50 HhomogeneousR1 60 10.26 10.40 1.81 12 10.55 10.35 1.05 HhomogeneousR2 60 10.55 10.50 1.54 12 10.51 10.50 0.90 HhomogeneousR3 60 10.51 10.50 1.42 12 10.41 10.50 0.98 HhomogeneousR4 60 10.60 10.70 1.47 12 10.60 10.70 1.02 HhomogeneousR5 60 10.61 10.55 1.37 12 10.55 10.60 0.97 HhomogeneousR6 60 10.48 10.50 1.47 12 10.48 10.45 1.01 HhomogeneousR7 60 10.47 10.50 1.51 12 10.40 10.40 0.95 HhomogeneousR8 60 10.56 10.55 1.28 12 10.33 10.45 0.94
a60participants(12groupsof5)wereassignedtoacontroltreatmentinwhicheachtradersubmittedanNWTAtosellthesamelottery(L).Eachgroup
tradedfor8rounds,withfeedbackaboutmarketpriceandtradingoutcomeaftereachround.60participantsweregiventheLlotteryandtradedinmixed groups(of5or7)inwhichsometradershaddifferentlotteries.Eachgrouptradedfor8rounds,withfeedbackaboutmarketprice(constrainedbetween£7 and£9)andtradingoutcomeaftereachround.Thesameparticipantswerethenre-assignedtotradeinhomogeneousgroupsof5withotherparticipants endowedwiththesamelottery.
b 60participants(12groupsof5)wereassignedtoacontroltreatmentinwhicheachtradersubmittedanNWTAtosellthesamelottery(H).Eachgroup
tradedfor8rounds,withfeedbackaboutmarketpriceandtradingoutcomeaftereachround.60participantsweregiventheHlotteryandtradedinmixed groups(of5or7)inwhichsometradershaddifferentlotteries.Eachgrouptradedfor8rounds,withfeedbackaboutmarketprice(constrainedbetween£7 and£9)andtradingoutcomeaftereachround.Thesameparticipantswerethenre-assignedtotradeinhomogeneousgroupsof5withotherparticipants endowedwiththesamelottery.
a) L lottery: mixed vs. control c) L lottery: homogeneous vs. control
b) H lottery: mixed vs. control d) H lottery: homogeneous vs. control
2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 1 2 3 4 5 6 7 8 Price MIX L MIX L CTRL Price L CTRL 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 1 2 3 4 5 6 7 8 Price L HOM L HOM L CTRL Price L CTRL 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 1 2 3 4 5 6 7 8
Price MIX H MIX
H CTRL Price H CTRL 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 1 2 3 4 5 6 7 8
Price H HOM H HOM
H CTRL Price H CTRL
Fig.2. ResultsoftheFBKmanipulation.EvolutionofNWTAsandpricesinthemixed,homogeneousandcontrolmarkets.Note:Ineachpanel,dashedlines depicttherelevantcontrolmarket,andsolidlineseitherthecorrespondingmixed(panelsaandb)orhomogeneousmarket(panelscandd).Trianglesare meanpricesandsquaresaremeanNWTAs.
discipline,itseemsclearthatmarketdisciplinecouldhavehadverylittleeffectonanygeneraltrendsintheNWTAsreported byHandLtradersinmixedmarkets.Suchtrendscansafelybetreatedasisolatingtheeffectsofshaping.
Byconstruction,theaveragemarketpriceisconstrainedtobebetween
£
7and£
9,asindicatedbythehorizontallinesin panelsaandbofFig.2.Throughouttheeightroundsthemean(median)pricefellslightlyfrom£
7.93(£
7.90)inR1to£
7.57 (£
7.50)inR8.AspricesforL,thesefiguresarequitehigh;aspricesforH,theyareratherlow.Didthispricefeedbackhave anyeffectonNWTAs?Theanswerisaclearyes.InthetoppanelofTable4andpanelaofFig.2,wecanseethatthemean(median)NWTAfortheLlotteryrosefrom
£
4.64 (£
4.40)inR1to£
5.67(£
6.45)inR8.17Thestrengthoftheseshapingeffectsisconfirmedbythetotallydifferentpatternobservedincontrolmarkets,inwhichNWTAsdeclinedoverrepetitions.Asexpected,becauseparticipantswererandomly assignedtotheshapingandcontroltreatments,inR1NWTAvaluesinmixedmarketsdonotdiffersignificantlyfromthose inthecontrolmarkets(two-tailMann–Whitneytest:Z=−1.32,p=0.1871).However,thedifferencebetweenthetwogroups isstronglysignificantinR8(one-tailMann–Whitneytest:Z=−5.39,p<0.001).
Markedshapingeffects,althoughnotasstrongasforL,alsoshowupforlotteryH.Themean(median)NWTAwas
£
10.91(£
11.10)intheinitialround,droppingto£
9.91(£
10.45)inR8.18Inthecorrespondingcontrolmarket,NWTAsvariedverylittleoverthecourseoftheeightrounds,startingat
£
11.17(£
11.45)andendingat£
10.97(£
11.20).The between-subjectcomparisonswiththecontrolmarketconfirmthattheseshapingeffectsweresizeable.NWTAswerenotsignificantly17Infact,asignificantincreaseinNWTAsisalreadydetectableinR2(Mann–Whitneytest:Z=−3.11,p<0.001). 18ThisdecreasewasalreadysignificantinR2(Mann–Whitneytest:Z=2.26,p<0.05).
Table5
EvolutionofNWTAvaluationsinmixedmarketsforsellersandnon-sellers.
Auctionround t NWTAtifsoldatt−1 NWTAtifnotsoldatt−1
<0 0 >0 Mean <0 0 >0 Mean L=(£2,0.8;£9,0.2) R1 1 – – – – – – – – R2 2 8 13 35 0.49 3 1 0 −2.43 R3 3 10 18 28 0.22 4 0 0 −1.43 R4 4 10 18 30 0.51 2 0 0 −0.10 R5 5 11 17 24 0.11 5 1 2 −1.26 R6 6 11 22 24 0.23 3 0 0 −0.13 R7 7 14 22 18 0.15 4 0 2 −0.83 R8 8 11 22 20 0.11 7 0 0 −1.30 H=(£7,0.2;£14,0.8) R1 1 – – – – – – R2 2 0 1 2 0.20 35 15 7 −0.69 R3 3 0 1 1 0.20 31 13 14 −0.32 R4 4 0 0 1 0.30 23 23 13 −0.03 R5 5 0 0 0 (n/a) 14 30 16 0.12 R6 6 1 0 0 −0.20 17 31 11 −0.18 R7 7 0 0 2 2.35 17 28 13 0.16 R8 8 0 2 1 0.27 19 27 11 −0.21
ChangeinNWTAfromroundtoround,NWTAt=NWTAt−NWTAt−1,forparticipantswhosoldandparticipantswhodidnotsellinthepreviousround.For
eachroundoftheauction,thenumbersofsellersandnon-sellerswhodecreased,keptconstantandincreasedtheirNWTAvaluationisreportedtogether withthemeanNWTAt.
differentinR1(two-tailMann–Whitneytest:Z=0.87,p=0.3846),butdifferedstronglyaftereightauctionrounds(one-tail
Mann–Whitneytest:Z=2.78,p<0.01).19
Takentogether,thesefindingsprovidestrongsupportfortheimplicationsofHypothesis1.Wecansummarisethemas Result1b:
Result1b–Shapingthroughpricefeedback.Intheabsenceofmarketdiscipline,NWTAsarepulledinthedirectionof uninformativepricefeedbackcues.
Ourdataseemtoindicatethatshapingeffectsarestrongerwhenfeedbackisrelativelyhigh(forlotteryL)thanwhenit islow(forlotteryH).Thisformsoursecondobservation.
Observation2–Asymmetryin thestrength ofshaping.Whenshapingisgenerated throughuninformativeprice
feedbackcues,itseffectsarestrongerwhenthefeedbackisrelativelyhighthanwhenitisrelativelylow.
BeforeturningtotheimplicationsofHypothesis2,itisinterestingtoasktowhatextentthisaggregatepicturereflects generaltendenciesratherthanbeingdrivenbyasmallnumberofparticularlymalleableindividualvaluations.Table5reports informationontheround-by-roundevolutionofNWTAsinmixedmarketsforLandH,separatelyforparticipantswhosold inthepreviousroundandparticipantswhokepttheirlotteries.Thestructureofthetableisentirelyanalogoustothatof Table3.20
Aswediscussedabove,becauseofourmanipulationofthepricefeedback,thereareveryfewLtraderswhodonotsell andveryfewHtraderswhodo.AmongsttheLsellers,thegreatmajorityrespondsbyincreasingtheirNWTAs,whileH non-sellershaveapreponderantpropensitytodecreasetheirs.21AsinTable3,theseeffectsslowdownslightlyovertime.
Thisformsourthirdobservation.
Observation3–Shapingattheindividuallevel.UninformativepricefeedbackcuesaffecttheNWTAsofthemajority ofparticipantsinthedirectionofobservedprices.
WenowturntothehomogeneousmarketsandourtestsoftheimplicationsofHypothesis2.TheevolutionofNWTAs andpricesinthehomogeneousandcontrolmarketsisdisplayedontheright-handsideofFig.2.Inhomogeneousmarkets, theshapingeffectsgeneratedthroughpricefeedbackcuesareexposedtotheeffectofmarketdiscipline.
Thefirstimportantaspecthighlightedbytheright-handsideofFig.2isthatforbothlotteriesthereisare-starteffect thatpartlyoffsetstheshapingeffectsgeneratedinmixedmarkets.NWTAsforlotteryLdropfromtheirR8levelinmixed markets,whileNWTAsforHincrease.Thatis,Ltraders(whoexperiencedconsistentlyhighpricesinamixedtradinggroup andrespondedbygraduallyincreasingtheirNWTAs)loweredtheirNWTAsonjoininganewtradinggroup,priortoanyprice feedback;thereisamirror-imageeffectforHtraders.Theseeffectsaresuggestiveofsomeformofunderlyingtruevalues.
19 AsexplainedinSection3,insomesessionsmixedmarketscamefirst,whileinotherstheyfollowedthemarketinvolvinglotteryE,inwhichalltraders
tradedthesamelottery.Itisreasonabletoaskwhethershapingeffectsarestrongerinthelattercase.Wefoundnoordereffectsofthiskind.Forboth lotteriesandinbothR1andR8,thedistributionsofNWTAswhenmixedmarketsprecededthemarketforlotteryEarenotstatisticallydifferentfromthe distributionswhentheyfollowedit.
20 ThecorrespondinginformationforcontrolmarketsisreportedintheAppendix,TableA.3.4.
21 Forbothlotteries,beingasellerinthepreviousroundhasasignificantpositiveeffect(atthe1%levelforLand10%levelforH)onNWTA
tina
Table6
EvolutionofNWTAvaluationsinhomogeneousmarketsforsellersandnon-sellers.
Auctionround t NWTAtifsoldatt−1 NWTAtifnotsoldatt−1
<0 0 >0 Mean <0 0 >0 Mean L=(£2,0.8;£9,0.2) R1 1 – – – – – – – – R2 2 3 8 13 0.40 29 3 4 −0.94 R3 3 8 5 11 0.15 25 5 6 −0.81 R4 4 7 7 7 0.60 20 7 12 0.13 R5 5 3 7 14 0.52 21 7 8 −0.63 R6 6 4 5 13 0.54 20 14 4 −0.43 R7 7 1 11 10 0.26 17 14 7 −0.43 R8 8 5 10 9 0.57 13 14 9 −0.04 H=(£7,0.2;£14,0.8) R1 1 – – – – – – R2 2 0 4 17 0.99 17 10 12 −0.08 R3 3 1 5 17 0.51 16 15 6 −0.39 R4 4 3 9 11 0.18 12 14 11 0.04 R5 5 3 7 13 0.34 16 17 4 −0.20 R6 6 4 7 13 0.17 17 17 2 −0.34 R7 7 2 13 7 0.26 14 20 4 −0.16 R8 8 2 11 9 0.45 11 21 6 −0.13
ChangeinNWTAfromroundtoround,NWTAt=NWTAt−NWTAt−1,forparticipantswhosoldandparticipantswhodidnotsellinthepreviousround.For
eachroundoftheauction,thenumbersofsellersandnon-sellerswhodecreased,keptconstantandincreasedtheirNWTAvaluationisreportedtogether withthemeanNWTAt.
TotesttheimplicationsofHypothesis2,welookatmarketprices.ForbothLandH,theR1pricesinhomogeneousmarkets
weresignificantlydifferentfromthoseofthecorrespondingcontrolmarkets(one-tailMann–Whitneytests:forL,Z=−4.85, p<0.001;forH,Z=4.61,p<0.001),showingthecontinuingeffectsofthepricefeedbackcuesthattradersreceivedinthe mixedmarkets,despitethere-starteffect.Overtheeightroundsofthehomogeneousmarkets,thepricesforLdecreased, whilethoseforHremainedmoreorlessstationary.Butattheendofthelastauctionround,priceswerestillsignificantly higherthaninthecontrolgroupforL(one-tailMann–Whitneytest:Z=−1.70,p<0.05),andstillsignificantlylowerfor H(one-tailMann–Whitneytest:Z=1.68,p<0.05),indicatingthepersistenceofshapingeffectsonprices.AsforourEXP manipulation,wedonotfindsupportfortheimplicationsofHypothesis2.22OurfindingsaresummarisedinResult2b.
Result2b–Persistenceofshapingthroughpricefeedback.Aftereightauctionrounds,pricesthathavebeenshaped byuninformativepricefeedbackcuesarestillsignificantlyaffectedbythosecues.
Table6summarisestheevolutionofNWTAsovertheeightauctionroundsofthehomogeneousmarketsfortheLandH lotteries.ItsstructureisanalogoustothatofTables3and5.Soaretheresults.Forbothlotteries,thereisaprevalenttendency forthosewhohavesoldinroundttoincreasetheirNWTAinroundt+1,andanoppositetendencyforthosewhohavenot soldtoreducetheirstatedvaluations,withthenumberofparticipantswhokepttheirNWTAunalteredslowlyincreasing overtheeightauctionrounds.23Thisformsthebasisforourfourthobservation.
Observation4–NWTAupdatinginhomogeneousmarkets.Inthehomogeneousmarkets,sellerstendtoincreasetheir NWTAandnon-sellerstendtodecreasetheirNWTA.
ThebehaviourofLsellersandHnon-sellersisofparticularinterestbecauseitisconsistentwithshapingbuthardto reconcilewithmarketdiscipline.Rememberthatallparticipantsenteredhomogeneousmarketsafterbeingexposedtothe mixedmarketsshapingmanipulation,whichpulledtheirNWTAsawayfromtheir(unobservable)truevalues.Giventhe magnitudeoftheseshapingeffects,itisreasonabletoinferthat,inthefinalroundsofthemixedmarkets,mostLtraders wereover-askingandsellingatveryadvantageousprices,i.e.
v
i<ai<p.However,whenthesetradersweresubsequentlyassignedtohomogeneousmarkets,themedianpricerulemeantthat(approximately)three-fifthsofthemwouldnolonger sell.Accordingtomarketdiscipline,thesetraders,forwhomnow
v
i<p<ai,shoulddecreasetheirNWTAs,whichiswhatweseeinTable6.ButeventhoseLtraderswhocontinuedtosellwerelikelytobeover-asking.Forthem,
v
i<ai<p,somarketdisciplinewouldnotbiteandtheyshouldnothavechangedtheirbehaviourinanysystematicway.Yet,aswealsoseein Table6,theyactuallyincreasedtheirNWTAsmorethantwiceasoftenastheyreducedthem–apatternthatsuggeststhey werecontinuingtobeinfluencedbymarketpricefeedback.
ThebehaviourofHnon-sellersshowsamirrorimageofthispattern.Inthefinalroundsofthemixedmarkets,mostH traderswerepresumablyfacingdisadvantageousprices,under-asking,butnotsellingandthereforenotsubjecttomarket
22SinceinboththehomogeneousandcontroltreatmentsvaluationsandpricesvarylittlefortheHlottery,wedonotneedtoworryaboutpriceconvergence
inrelationtothatlottery.ForL,wehaveestimatedasymptoticfunctionsofthesameformasforE(seefootnote12).Theresultingestimatesofp*inthe homogeneousmarket(£4.23)andinthecontrolmarket(£3.67)aresignificantlydifferentfromeachother(p<0.1).SeetheAppendixfordetails.
23Forbothlotteries,beingasellerinthepreviousroundhasasignificantlypositiveeffect(atthe1%level)onNWTA
tinarandom-effectspanelregression.
discipline(i.e.p≤ai<vi).Whentheywereassignedtohomogeneousmarkets,(approximately)two-fifthsofthemwouldsell
theirlotteries;marketdisciplinewouldthenleadthesetraderstoraisetheirNWTAs.Butmarketdisciplinewouldnotbite forthosewithp<ai<vi,whowouldcontinuenottosell.Andyet,asweseeinTable6,thesetraderssubsequentlyreduced
theirasksmorethantwiceasoftenastheyincreasedthem,onceagainsuggestingthatthemarketpricecontinuedtohave someshapinginfluence.
Observations1,3and4highlightapervasivetendencyforNWTAstomoveinthedirectionofthemarketprice.This tendencyiscompatiblewithanotherregularityinourdata:thatthevarianceofNWTAswithinauctiongroupsdecreases overrepeatedrounds,whileNWTAdifferencesbetweengroupsincreases.24
6. Discussion
Inthissection,wediscusssomeissuesraisedbyourresults.
Thefirstissueconcernstheexplanationofshapingeffects.Ourdatashowaninterestingasymmetrywhich,wereittobe replicated,mightthrowsomelightonthemechanism(s)thatliebehindshaping:shapingeffectsseemtobestrongerwhen NWTAsarebeingpulledupbyhighpricesthanwhentheyarebeingpulleddownbylowprices(seeobservation2).
Thisregularityisconsistentwiththetheoryof‘baddealaversion’(Isoni,2011;WeaverandFrederick,2012;forrelated ideas,seeThaler,1985;Putler,1992).Inthistheory,preferencesarereference-dependentinawaythatmakesindividuals’ valuationsofgoodsdependdirectlyonpriceexpectations.Specifically,individualsderiveutilityfrommaking‘gooddeals’ (buyingatpricesthatarelowerthanexpected,sellingatpricesthatarehigherthanexpected)anddisutilityfrommaking ‘baddeals’(theopposite).Inthecontextofasellingauction,atraderforwhomtheexpectedpriceishigherthantheir reference-independentvaluationofthegoodwillreportanNWTAhigherthanthatvaluationinordertoavoidparticularly baddeals;conversely,atraderforwhomtheexpectedpriceislowerthantheirreference-independentvaluationwillstate anNWTAlowerthanthatvaluationinordertoenjoyparticularlygooddeals.If,otherthingsbeingequal,thepainofbad dealsisstrongerthanthepleasureofgoodones,thetendencyforNWTAstobepulledupbyhighpriceexpectationswillbe strongerthanthetendencyforthemtobepulleddownbylowexpectations.
Analternativeexplanationofthisapparentasymmetryisthefollowing.WhenNWTAsarebeingpulledupbyahigh marketprice,thetraderswhosepreferencesarebeingshapedaretradingat,andbenefitingfrom,thatprice.Thesefeatures ofthepricemaycontributetoitspsychologicalsalienceandmayencouragefurtherover-asking,inthebeliefthatthiscould prolongaparticularlyfavourablestateofaffairs.Conversely,whenNWTAsarebeingpulleddownbyalowprice,thetraders whosepreferencesarebeingshapedarenottradingatthatprice,andthelowlevelofthepriceisasourceofdisbenefitto them.25
Amorefundamentalissueraisedbyourresultsisthatofexplainingwhyshapingeffectsarepartiallybutnotwhollyeroded bymarketdiscipline.Hadwefoundthatshapingeffectswerecompletelyresistanttomarketdiscipline,thatwouldhave beenconsistentwiththehypothesis,advocatedbysomepsychologists,thatpreferencesdonotexistpriortoelicitation,and areconstructedbyindividualsasandwhentheyareneeded(e.g.Payneetal.,1993,1999;Slovic,1995;Stewartetal.,2006). Butthatisnotwhatwefound.
Alternatively,supposewehadfoundthat,whenmarketdisciplinewasswitchedon,marketpricesconvergedtowards valuesthatwereindependentofourshapingmanipulations.Thatwouldhavebeenconsistentwithaninterpretationof Plott’s(1996)discoveredpreferencehypothesisinwhichthediscoverymechanismismarketdiscipline(Loomesetal.,2003). Accordingtothisaccount,psychologicalfactorsmaycauseirrelevantcuestohavesignificantimpactsonmarketbehaviour, solongasthatbehaviourdoesnotleadindividualstomakelosses;butwhenbehaviourissubjecttomarketdiscipline,true preferencesaregraduallydiscovered.Butagain,thatisnotwhatwefound.
Inbroad-brushterms,whatweseemtohavefoundisthefollowing.Whenourshapingmanipulationsinducelarge changesinNWTAs(asinthemixedmarketforlotteryL),theinitialeffectofintroducingmarketdisciplineisthatNWTAs moveinthedirectionthatoffsetsthisshaping(NWTAsforLdecreaseintheearlyroundsofthehomogeneousmarket). Indeed,adetectablemoveinthisdirectionisinducedmerelybytellingparticipantsthattheyhavejoinedanewtrading group(NWTAsforLarealreadylowerinR1ofthehomogeneousmarketthantheyareinR8ofthemixedmarket).These effectsmightindicatethatparticipantshavesomevaguesensethattheirshaping-inducedNWTAsdonotaltogetherreflect theirunderlyingpreferences.However,theeffectsofshapingarenoteliminated.Aftereightauctionroundsinwhichmarket disciplinehasbeenswitchedon,marketpricesstilldonotconvergetovaluesindependentoftheshapingmanipulation.The implicationisthatparticipantsarenotdiscoveringpreciseunderlyingvaluations.
24 TableA.3.5intheAppendixreportsthewithin-andbetween-groupvariancesofNWTAsforallorourtreatmentsinR1andR8.Infiveoutofsixcases,
thewithin-groupvariancefallsbetweenR1andR8.NWTAdistributionsarenotdifferentacrossauctiongroupsinmatchedtreatmentsinR1,asonewould expectgiventhatparticipantshavebeenassignedtotreatmentsatrandom.ButthenullhypothesesthatNWTAsarethesameacrossgroupscanberejected infiveoutofsixcasesinR8.AsimilarresultisfoundbyTufano(2009).
25 ThisideawassuggestedtousbyBenMcQuillin.Notethat,whileevidencesuggeststhatparticipantsinauctionsmayderiveutilityfromthefactthat
theyare‘winning’theauction(e.g.Dechenauxetal.,2014),joyofwinningseemsincompatiblewiththeasymmetrythatwefind(aswellasthegeneral trendsinvaluationupdatingshowninTables3,5and6)because,inasellingauctionlikeours,Ltradersshould,ifanything,decreasetheirNWTAsinstead ofincreasingthemastheyclearlydo.