BOYSWILLBEBOYS:GENDER,OVERCONFIDENCE ...

1 BOYSWILLBE BOYS:GENDER,OVERCONFIDENCE,ANDCOMMONSTOC KINVESTMENT*BRADM. BARBERANDTERRANCEODEANT heoreticalmodelspredictthat overcon this predictionby partitioninginvestorson ,in areassuchas nance,menare moreovercon ,theorypredictsthatmenwill over35,000householdsfroma largediscountbrokerage,we analyzethe commonstockinvestmentsof menand documentthatmentrade45 s net returnsby yearasopposedto s not whata mandon t knowthat makeshim a fool, but whathe doesknowthatain t ,nineteenthcenturyAmericanhumoristIt is dif cult to reconcilethe volumeof tradingobservedinequitymarketswiththe tradingneedsof ,rebalancetheirportfolios,andtradeto superiorinfor-mationmaytradespeculativel y,althoughrationalspeculativetraderswill generallynot chooseto isunlikelythat rationaltradingneedsaccountfor a turnoverrate of76 percenton the NewYorkStockExchangein believethereis a simpleand powerfulexplanationforhighlevelsof tradingon nancialmarkets:overcon overcon dentabouttheirabilities,theirknowl-edge, and [1998]showsthatover-con dentinvestors whobelievethatthe precisionof theirknowledgeaboutthe valueof a securityis greaterthanit actually* We are gratefulto the discountbrokerage rm thatprovidedus withthedatafor this studyand gratefulto PaulThomas,DavidMoore,PaineWebber,andthe GallupOrganizationfor appreciatethe commentsof DianeDel Guercio,DavidHirshleifer,AndrewKarolyi,T imothyLoughran,EdwardOptonJr.

2 , SylvesterSchieber,AndreiShleifer,MarthaS tarr-McCluer,RichardThaler,LuisViceira,a ndparticipantsat the Universityof Alberta,ArizonaStateUniversity,INSEAD,th e LondonBusinessSchool,the UniversityofMichigan,the Universityof Vienna,the Instituteon Psychologyand Markets,theConferenceon HouseholdPortfolioDecision-makingand AssetHoldingsat theUniversityof Pennsylvania,and the our reachedat (530)752-5332orodean] ;BradBarbercanbe reachedat (530)752-0512orbmbarber] NYSEFactBookfor the Year1999. 2001by the Presidentand Fellowsof HarvardCollegeand the QuarterlyJournalof Economics,February2001261is trademorethanrationalinvestorsand thatdoingso denceleadsto greatertradingand to directtest of whetherovercon dencecontributesto exces-sive markettradingis to separateinvestorsinto thosemoreandthoseless proneto overcon thentest whethermoreovercon denceleadsto moretradingand to test is the primarycontributionof this nd that in areassuchas nancemenare moreovercon overcon denceyieldstwo predictions.

3 Menwill trademorethanwomen,and theperformanceof menwill be hurtmoreby excessivetradingthanthe performanceof test thesehypotheses,we partitiona dataset of positionand tradingrecordsfor over35,000house-holdsat a largediscountbrokerage rm into accountsopenedbymenand accountsopenedby predic-tionsof the overcon dencemodels,we nd thatthe averageturnoverrateof commonstocksfor menis nearlyone and a halftimesthatfor womenreducetheirnet returnsthroughtrading,mendo so by yearthando differencesin turnoverand returnperformanceare evenmorepronouncedbetweensinglemenand percentmorethansinglewomentherebyreducin gtheirreturnsby thispaperis organizedas test of overcon dencein SectionI. We discussourdataand empiricalmethodsin SectionII. Ourmainresultsarepresentedin SectionIII. We discusscompetingexplanationsfor our resultsin SectionIV A Overcon denceand Tradingon FinancialMarketsStudiesof the calibrationof subjectiveprobabilities nd thatpeopletendto overestimatethe precisionof theirknowledge[AlpertandRaiffa1982;Fisch hoff,Slovic,andLichtenstein1977];see Lichtenstein,Fischhoff,and Phillips[1982]for a re-viewof the dencehas been262 QUARTERLYJOURNALOF ECONOMICS observedin manyprofessional [Os-kamp1965],physiciansand nurses[Christensen-SzalanskiandBushyhead 1981;Baumann,Deber,and Thompson1991],invest-mentbankers[Stae l von Holstein1972],engineers[Kidd1970],entrep reneurs[Cooper,Woo,andDunkelberg1988],la wyers[Wagenaarand Keren1986],negotiators[Nealeand Bazerman1990],and managers[Russoand Schoemaker1992]haveall beenobservedto exhibitovercon dencein theirjudgments.

4 (Forfur-therdiscussionsee Lichtenstein,Fischhoff,and Phillips[1982]and Yates[1990].)Overcon denceis greatestfor dif culttasks,for forecastswithlow predictability,and for undertakingslackingfast,clearfeedback[Fi schhoff,Slovic,and Lichtenstein1977;Lichtenstein,Fischhoff, andPhillips1982;Yates1990;Grif n andTversky1992].Selectingcommonstockstha twill outperformthe marketis a dif cult low;feedbackis ,stockselectionis the typeof taskfor whichpeopleare mostovercon [1998]developsmodelsin whichovercon dentinves-torsoverestimatethe precisionof theirknowledgeaboutthevalueof a probabilitythattheirpersonalassessmentso f the security s valueare moreaccuratethanthe assessmentsof ,overcon dentinvestorsbelievemorestronglyin theirownvaluations,andconcernthemselvesl ess aboutthe beliefsof es differencesof opinioncausetrad-ing [Varian1989;Harrisand Raviv1993].Rationalinvestorsonlytradeand onlypurchaseinformationwhendoingso increasestheirexpectedutility( ,Grossmanand Stiglitz[1980]).

5 Over-con dentinvestors,on the otherhand,lowertheirexpectedutil-ity by tradingtoo much;theyholdunrealisticbeliefsabouthowh ightheirreturnswill be and howpreciselythesecan be esti-mated;andtheyexpendtoo manyresources( ,timeandmoney)on investmentinformation[Odean1998].Overcon dent2. Othermodelsof overcon dentinvestorsincludeDe Long,Shleifer,Sum-mers,and Waldmann[1991],Benos[1998],Kyleand Wang[1997],Daniel,Hirsh-leifer,andSubram anyam[1998],GervaisandOdean[1998],andCab alle andSa kovics[1998].Kyleand Wangarguethatwhentraderscompetefor duopolypro ts, overcon denttradersmayreapgreaterpro ts. However,this predictionisbasedon severalassumptionsthatdo not applyto [1998]pointsout thatovercon dencemayresultfrominvestorsover-estimati ngthe precisionof theirprivatesignalsor, alternatively,overestimatingtheirabiliti esto BOYS investorsalsoholdriskierportfoliosthando rationalinvestorswiththe samedegreeof risk aversion[Odean1998].BarberandOdean[2000] andOdean[1999]testwhetherinvestorsdecrea setheirexpectedutilityby tradingtoo samedataanalyzedin this paper,Barberand Odeanshowthatafteraccountingfor tradingcosts, mostrealize,by far, the whatthe modelsofovercon differentdataset, Odean[1999]

6 Nds thatthe securitiesindividualinvestorsbuy controlsfor li-quiditydemands,tax-lossselling,rebala ncing,andchangesinriskaversion,investors timingof tradesis not onlyare investorstoo willingto act on toolittleinformation,but theyare too willingto act investorstradetoo muchandto ndingsare inconsistentwithrationalityand not easilyexplainedin the absenceof overcon ,overcon denceis neitherdirectlyobservednor manip-ulatedin yet sharpertestof the modelsthatincorporateovercon denceis to partitioninvestorsintothosemoreand thoseless proneto overcon modelspredictthatthe moreovercon dentinvestorswill trademoreand test thesepredictions,we partitionourdataon Genderand Overcon denceWhilebothmenand womenexhibitovercon dence,menaregenerallymoreovercon dentthanwomen[Lundeberg,Fox,andPunc ochar 1994].3 Genderdifferencesin overcon dencearehighlytaskdependent[Lundeberg,Fo x,and Punc ochar 1994].Deauxand Farris[1977]write Overall,menclaimmoreabilitythando women,but this differenceemergesmoststronglyon.

7 Masculinetask[s]. Severalstudiescon rm thatdifferencesincon denceare greatestfor tasksperceivedto be in the masculinedomain[DeauxandEmswiller1974;Le nney1977;BeyerandBowden1997].Menare inclinedto feel morecompetentthan3. WhileLichtensteinand Fishhoff[1981]do not nd genderdifferencesincalibrationof generalknowledge,Lundeberg,Fox,and Punc ochar [1994]arguethat this is becausegenderdifferencesin calibrationare strongestfor topicsin ECONOMICS womendo in nancialmatters[Prince1993].Indeed,casual ob-servationrevealsthatmenare disproportionatelyrepresentedinthe expect,therefore,thatmenwill gen-erallybe moreovercon dentabouttheirabilityto make ,Lenney[1977]reportsthatgenderdifference sin self-con dencedependon the lackof clearand unequivocaland immediatelyavail-able,womendo not ,whensuchfeedbackis absentor ambiguous,womenseemto haveloweropinionsof theirabilitiesand oftendo under-estimaterelativeto men. Feedbackin the stockmarketis the morereasonto expectmento be morecon dentthanwomenabouttheirabilityto Odean[1998]developa modelin whichinvestorovercon in this modelinfertheirownabilitiesfromtheirsucc essesand theirtendencyto taketoo muchcreditfortheirsuccesses,theybecomeov ercon [1977],Meehanand Overton[1986],and Beyer[1990] nd thatthe self-servingattributionbiasis greaterfor so menare likelyto becomemoreovercon previousstudymostlike our own is Lewellen,Lease,andSchlarbaum s [1977]analysisof surveyanswersand brokeragerecords(from1964through1970)

8 Of 972 ,Lease,and Schlarbaum s reportthatmenspendmoretimeand moneyon securityanalysis,rely less on theirbrokers,makemoretransactions,believ ethatreturnsare morehighlypredictable, all theseways,menbehavemorelike overcon dentinvestorsthando moreovercon dentinves-torsthanwomencomesfromsurveysc onductedby the GallupOrganizationfor survey f-teentimesbetweenJune1998and additionto otherquestions,respondentswereasked Whatoverallrateof returndo youexpectto get on yourportfolioin the NEXT twelvemonths? and Thinkingaboutthe stockmarketmoregenerally,whatoverallrate of returndo you thinkthe stockmarketwill265 BOYSWILLBE BOYS provideinvestorsduringthe comingtwelvemonths? On average,bothmenand womenexpectedtheirown portfoliosto outperformthe ,menexpectedto outperformby a greatermargin( percent)thandid women( percent).The differencein the averageanticipatedoutperformanceof menand womenisstatisticallysigni cant(t=3. 3).4In summary,we havea naturalexperimentto (almost)di-rectlytest theoreticalmodelsof investorovercon ratio-nal investoronlytradesif the expectedgainexceedsthe overcon dentinvestoroverestimatesthe precisionof his informationand therebythe expectedgainsof true expectednet gainis moreovercon dentthanwomen,this givesus two test-ablehypotheses:H1: : By tradingmore, is thesetwo hypothesesthatare the focusof our DemographicDataOur mainresultsfocuson the commonstockinvestmentsof37,664households for whichwe are ableto identifythe genderofthe personwhoopenedthe household s rst compiledfromtwo is informationfroma largediscountbrokerage rm on the investmentsof 78,000householdsfor the sixyearsendingin this period,we haveend-of-monthpositionstatementsand tradesthatallowus to reasonablyestimatemonthlyreturnsfromFebr uary1991throughJanuary4.

9 Somerespondentsansweredthat theyexpectedmarketreturnsas highas900. We suspectthat theserespondentswerethinkingof ,we havedroppedfromour calculationsre-spondentswho gaveanswersof morethan100 to this , alternatively,we Windsorizeanswersover900 at 100 thereis no signi cantchangein Overcon dencemodelsalso implythatmoreovercon SectionIII we ,genderdifferencesin portfolioriskmaybe due to differencesin risktoleranceratherthan(or in additionto)differencesin overcon dataset includesall accountsopenedby the 78,000house-holdsat this discountbrokerage rm. Sampledhouseholdswererequiredto havean openaccountwiththe discountbrokerage the accountsin our analysiswereopenedpriorto 1987,whilehalf average,menopenedtheir rst accountat this yearsbeforethe beginningof our sampleperiod, sampleperiod,men saccountsheldcommonstocksfor 58 monthson averageand wom-en s for 59 mediannumberof monthsmenheldcom-monstocksis 70. For womenit is this research,we focuson the commonstockinvestmentsof excludeinvestmentsin mutualfunds(bothopen-and closed-end),Americandepositoryreceipts(A DRs),war-rants,and the 78,000sampledhouseholds,66,465hadpositio nsin commonstocksduringat leastone month;the re-mainingaccountseitherheldcashor investmentsin averagehouseholdhad approxi-matelytwo accounts,and roughly60 percentof the marketvaluein theseaccountswas heldin milliontradesin all securitiesduringour sampleperiod;commonstocksaccountedfor slightlymorethan60 per-centof all averagehouseholdheldfourstocksworth$47,0 00duringour sampleperiod,althougheachof these guresis stocksworth$16, aggregate,thesehouseholdsheldmorethan$ commonstocksin secondarydataset is demographicinformationcompiledbyInfobase Inc.

10 (as of June8, 1997)and providedto us by the broker-age genderof the personwho openeda household s rst accountfor 37,664households,of which29,659(79 percent)had accountsopenedby men and 8,005(21 percent)hadaccountsopenedby additionto gender,Infobaseprovidesdataon maritalstatus,the presenceof children,age, and presentdescriptivestatisticsin TableI, PanelA. Thesedatarevealthatthe womenin our sampleare less likelyto bemarriedand to meanand medianagesof the men and womenin our sampleare womenreportslightlylowerhouseholdincome, althoughthe differenceisnot Throughoutthe paper,portfoliovaluesare reportedin BOYSTABLEIDESCRIPTIVESTATISTICSFORDEMOGR APHICSOFFEMALEANDMALEHOUSEHOLDSV ariableAllhouseholdsMarriedhouseholdsSin glehouseholdsWomenMenDifference(women men)WomenMenDifference(women men)WomenMenDifference(women men)PanelA:InfobasedataNumberofhousehold s8,00529,659NA4,89419,741NA2,3066, ($000) >$125, :Self-reporteddataNumberofhouseholds2,63 711,2261,7077,7006522,184 Networth($000) (%) (%) rmforwhichweareabletoidentifythegenderof thepersonwhoopenedthehousehold s ,children,age, ,wherethetoprangeisgreaterthan$125, $125, (%)istheproportionofthemarketvalueofcomm onstockinvestmentatthisdiscountbrokerage rmasofJanuary1991tototalself-reportednet worthwhenthehouseholdopenedits ECONOMICSIn additionto the datafromInfobase,we also havea limitedamountof self-reporteddatacollectedat the timeeachhousehold rst openedan accountat the brokerage(andnot subsequentlyupdated),whichwe summarizein TableI, PanelB.