NBER WORKING PAPER SERIESINVESTMENT SENTIMENT MEASURESLily QiuIvo WelchWorking Paper 10794 BUREAU OF ECONOMIC RESEARCH1050 Massachusetts AvenueCambridge, MA 02138September 2004We especially thank Charles Lee, Stephen Ross, Andre Shleifer, and Richard Thaler for making theirdata generously available to us. Umit Gurun provided newer data, and seminar participants atMchigan and Rochester had to suffer through a presentation involving some of the tables in thispaper. The views expressed herein are those of the author(s) and not necessarily those of theNational Bureau of Economic Research. ©2004 by Lily Qiu and Ivo Welch. All rights reserved. Short sections of text, not to exceed twoparagraphs, may be quoted without explicit permission provided that full credit, including © notice,is given to the source.
Investor Sentiment MeasuresLily Qiu and Ivo WelchNBER Working Paper No. 10794September 2004JEL No. G12, G14ABSTRACTThis paper compares investor sentiment measures based on consumer confidence surveys withmeasures extracted from the closed-end fund discount (CEFD). Our evidence suggests that these twokinds of sentiment measures do not correlate well with one another. For a short 2 4 year period inwhich we have direct investor sentiment survey data from UBS/Gallup, only the consumerconfidence correlates well with investor sentiment. Further, only the consumer confidence basedmeasure can robustly explain the small-firm return spread and the return spread between stocks helddisproportionately by retail investors and those held by institutional investors. Surprisingly, there iseven a hint that the consumer confidence measure can explain closed-end fund IPO activity, whilethe CEFD cannot. In sum, our evidence supports the view that sentiment plays a role in financialmarkets, but that the CEFD may be the wrong measure of QiuDepartment of EconomicsBox BBrown UniversityProvidence, RI 02912 Ivo WelchYale UniversitySchool of Management46 Hillhouse Haven, CT 06520-8200and NBERivo_welch@
AIntroductionThebehavioralfinancetheoryofDeLong,Shleifer,Summers,andWaldmann(1990),changesinnoisetradersentimentmustbedifficulttopredict,,thetheorypredictsthatsentimentcaninfluencesecuritypricingundertwonecessaryconditions:[1]theassetsareheldpredominantlybysentiment(noise)traders,and[2],Shleifer,andThaler(1991),henceforthLST,-causeindividualinvestorswerealreadyknowntodisproportionallyholdclosed-endfunds(henceforth,CEF),LSTinterprettheclosed-endfunddiscount(henceforth,CEFD)asa(neg-ative),Shleifer,andThaler(1991)thenwringfurtherimplicationsandempiricalsupportfromthisinsight:(.,moreoptimism)’stendtoappearinLST’,Shleifer,andThaler(1991)alsodiscountotherpossiblefactorsdeterminingtheCEFD,firstandforemostagency(transaction),(2002),-known(andamusing)exchangeintheJournalofFinancein1993,Chen,Kan,andMiller(1993)pointoutthatthecorrelationbetweentheCEFDandthesizespreaddeclinedinthelatterhalfofLST’,“Che’KM”pointoutthattheexplanatorypoweroftheCEFDforsmall-firmreturns1Lee,Shleifer,andThaler(1991)canalsohintthatthetheoryisnotinconsistentwithanegativeCEFDuponfundinception,followedbyasharpdropfromapremiumintoadiscount,andthenapositivedriftinCEFDreduction[toaccountfortheneedtoofferapositiveexpectedrateofreturn].However,thedynamicsareweak:theprocessbywhichthenegativediscountbecomespositiveisnotclear,
,Chopra,Lee,Shleifer,andThaler(1993)respondthatallsmallfirmsaregenerallynoise-tradersensitivewithlowinstitutionalownership,thatexpectingtofindaneffectaftersplittingsubsamplesagainisaskingfortoomuch,,,Lee,Shleifer,andThaler(1991)isaseminalpaper,bothinthenoveltyofitsideasanditssubsequentimpact:asofMarch2004,“investorsentiment”finds53matches,comparedto78forthephrase“APT.”Ifanything,-seriessentimentevidence,anddosoprimarilybyborrowingideasandconceptsofhowtotestforinvestorsentimentfromLST.(Wehavelittletosayaboutthecross-sectionalcovariationinCEFDexploredinLee,Shleifer,andThaler(1991).),ourpaperhasanunfairadvantageoverLST—,wecanexpandoursamplefrom1985to2002(16years),almostdoublingthesample,,webelieveitisdifficulttofurthervalidatetheCEFDsentimentinterpretationusingotherfinancialmeasures:itisalwaysrelativelymorelikelythatsomefinancialphenomenonstealssignificancefromtheCEFDbecauseit,too,,“direct,”-sumptions,buttheassumptionsaredifferent:•FinancialMeasures:(theproxyextractionfromtheCEFD)withotherimplications(.,therateofreturnonassetsdisproportionallyheldbynoisetraders).TotheextentthatothercostscanmattertotheCEFD(.,agencycosts)orthatsmartertradersholdeitheraparticularCEFortheunderlyingassets,—theCEFDisessentiallyabook-to-marketratio—itisalsorelativelymorelikelythatanothertheorycouldeventuallyofferanexplanationforbothfindings,•SurveyMeasures:Thesurvey-basedsentimentindexesrequireanidentificationofconsumersasbeingtheindividualretailinvestorsthatDeLong,Shleifer,Summers,2Forexample,itcouldbethatthereisatime-varyingpremiumtoliquidityandagencycosts,(1997)andBerkandStanton(2004)haverecentlyproposedarationalexplanationforsomeofthetime-patternintheCEFD.(ItdoesnotexplaintheoriginalCEFpremiumorthecorrelationbetweenthesizeorretailstockpremiumandsentiment.)3
andWaldmann(1990)andLee,Shleifer,andThaler(1991),—butprobablyunlikely—,-tunately,wehavesomedatatotestthis:someregularsurveysofinvestorshavere-centlyappeared,,wefindthatchangesintheUBS/Gallupsurveyofinvestors’sentimentindexcorrelatesstronglypositivelywithchangesintheconsumerconfidenceindexes,,theUBS/.(Unfortunately,theUBS/Gallupinvestorsurveyhastooshortahistorytopermitustorelyonitforourfullinvestigation.)Ourdataspansfromthe1960stotheearly2000s,dependingondataseries,::Ingeneral,wefindthatthetwokindsofmeasures—survey-basedandCEFD-based—-tentand/,theUBS/Gallupsurveymeasureofinvestorsentiment—boththatofrelativelywealthyinvestorsandrelativelypoorinvestors—-FirmReturnSpread:Lee,Shleifer,andThaler(1991)suggestthatasentimentmeasureshouldcorrelate(contemporaneously)“small-firmreturnspread.”•DecreasesintheCEFDcorrelatestatisticallysignificantlypositivelywiththesmall-firmspread,,therelationshiphasweakenedsignificantlyafter1985.•IncreasesintheMichiganconsumerconfidenceindexcorrelatestatisticallysig-nificantlypositivelywiththesmall-firmspread,
•IfJanuaryobservationsareexcluded,theMichiganconsumerconfidenceindexperformsequallywell,buttheCEFDlosesallsignificance.•Onanannualfrequency,despiteasmallnumberofobservations,thecorrelationsremainpositiveandstatisticallysignificantforbothsurveyindexes.•BoththeMichiganconsumerconfidenceandtheCEFDindexofferuniquestatis-ticallysignificantexplanatorypowerforthesmall-firmspread—-basedmeasuresperformbetterinthesecondhalfofthesamplethantheCEFD-basedmeasure,therelationshipbetweenthesmallfirmspreadandthesurvey-basedsentimentmeasure“feels”-StockReturnSpread:-tutionalholdingsfrom13(f)filings,3andentertainedthreedifferentmeasuresofthe“retail-stockreturnspread.”Sortedintermsofrelevancepredictedbythesentimenttheory,weexplore:(a)Thereturnonstockswithzeroinstitutionalholdingsminustheaveragereturnonallotherstocks.(b)Thereturnonstockswithzeroinstitutionalholdingsminusthereturnonstocksinthetopdecileofinstitutionalholdingsoftheremainingstocks.(c)Thereturnonstockswithzeroinstitutionalholdingsandwithinthissetofstocksrankedinthelowerhalfofdollartradingvolume,minusthereturnonstocksinthetoptwodecilesofinstitutionalholdings,•’spointofview.•TheMichiganconsumerconfidenceindexcorrelatessignificantlypositivelywiththethreeretail-stockreturnspreads,andinstrengthaspredictedbythetheory.•Itisimportanttocontrolformarket-widemovementsintheseregressions,be-causeretailstocksperformedworsewhenthestockmarketperformedbetter,(Rule13(f)),institutionalinvestmentmanagerswhoexerciseinvestmentdiscretionoveraccountswithpubliclytradedsecurities(section13(f)securities)andwhoholdequityportfoliosexceeding$100millionarerequiredtofileForm13(f),000sharesand/orwithamarketvalueover$200,
:,onanannualbasis,——,ourevidencesuggeststhatsentimentplaysaroleinfinancialmarkets,:Ourpaperhasnothingtosayaboutcausality—whethersentiment(intheformoftheCEFDorintheformofconsumerconfidence)“drives”,bothmaybedrivenbyanunderlyingunmeasuredvariable,sentiment,:whenthestockmarketdrops,itmaycauseadropininvestorandconsumerconfidence,whichinturncancauseafurtherdropinthestockmarket,—theclosed-endfunddiscount(“financialmeasure”)andtheconsumerconfidence(“surveymeasure”)—forwhichwehavereasonablycompletemonthlydataseries.[InsertTable1(DescriptiveStatistics)abouthere]Table1liststheunivariatecharacteristicsofourseries,
ATheClosed-EndFundDiscountIndexesOurfirstmeasureisthetraditionalinvestorsentimentmeasureinthefinanceliterature,whichisbasedontheclosed-endfunddiscount(CEFD).BothLee,Shleifer,andThaler(1991)andRoss(2002)generouslysharedthisdatawithus;,-ante,,(2002)closed-endfunddataandtheLee,Shleifer,andThaler(1991)data,thecorrelationbetweentheirvalue-weightedmea-suresis95%,Thevalue-weightedregressioncoefficientsareRossCEFDMeasuret=−+·LSTCEFDMeasuret+Noiset.(1)Thecorrelationbetweenthetwosources’equal-weightedCEFDmeasuresis“only”83%,,,,aroundthebreakpointswherethetwoseriesoverlap,%%%%%...............%%%%%Canweaveragethetwoseries?ThedifferencebetweentheRossandLSTserieshasamedianof–%,ameanof–%,%,andaninterquartilerangeof–%to+%.Further,%;%.Therefore,aroundthe1980break,%%%,—sowecantalkaboutsentimentim-provementsandsentimentincreases—byusingthenegativeoftheCEFD,“d.”“vw.”(“ew.”)denotesvalue-weighted(equal-weighted).Therefore,
-weightedandequal-weighteddiscounts,,werelyonequal-weightedCEFDchanges,,bothequal-weightedandvalue-weighted,inoursamplewasaround10%,rangingfromabout+25%(inmid1979)to–14%(attheturnof1968/69).Theaveragesentimentchangewasjustaboutzero,withatypicalmonth-to-monthstandarddeviationof2%.DropsintheCEFDinexcessof8%occurredinMarch1968,November1976,%occurredinNovember1967,:TheMichiganConsumerConfidenceIndex,,:1."(andyourfamilylivingthere)arebetterofforworseofffinanciallythanyouwereayearago?"2."Nowlookingahead—doyouthinkthatayearfromnowyou(andyourfamilylivingthere)willbebetterofffinancially,orworseoff,orjustaboutthesameasnow?"3."Nowturningtobusinessconditionsinthecountryasawhole—doyouthinkthatduringthenexttwelvemonthswe’llhavegoodtimesfinancially,orbadtimes,orwhat?"4."Lookingahead,whichwouldyousayismorelikely—thatinthecountryasawholewe’llhavecontinuousgoodtimesduringthenextfiveyearsorso,orthatwewillhaveperiodsofwidespreadunemploymentordepression,orwhat?"5."Aboutthebigthingspeoplebuyfortheirhomes—suchasfurniture,arefrigerator,stove,television,,doyouthinknowisagoodorbadtimeforpeopletobuymajorhouseholditems?"Answersarecodedonascalefrom1(good)to5(bad),andaveraged(equal-weighted).(2000).Anexcerpt:Themonthlysurveyofconsumersisanongoingnationallyrepresentativesurveybasedonap-proximately500telephoneinterviewswithadultmenandwomenlivinginhouseholdsinthecoterminousUnitedStates(48StatesplustheDistrictofColumbia).,,andthetotalsampleforanyonesurveyisnor-mallymadeupof60%newrespondents,and40%
,thesampledesignsupportsthepoolingofuptosixoftheindependentmonthlysamplestoachievelargersamples,,thesurveydocumen-tationgivesgreatemphasistodemographicsampling,samplingerror,samplecoverageandnon-responseerrors,sampleweightingquestionnairedesign,telephoneinterviewing(andinterviewertraining),codingmethods,,(survey)isrunbyNFOResearch,Inc.,ofGreenwich,,000households,ofwhichroughly3,,adifferentpanelof5,:’’’’’,therearethreeresponseoptions:positive,negative,,thepositivefigureisdividedbythesumofthepositiveandnegativetoyieldaproportion,whichthesurveycallsthe"relative",:thepresentsituationindex;theaverageofindexesforquestions1and3;theexpectationsindex:theaverageofindexesforquestions2,4,,oftenduringthemonthofsampling,andafinal(thusrevised)—andborneoutinourlaterdataanalysis—(andespeciallytheindividual’sowncondition),,forourpurposes,
Table1showsthattheaverageMichiganconsumerconfidenceindexisaround90,,,thefirstdifferenceintheMichiganconsumerconfidenceindex,;andthefirstdifferenceintheConsumerBoardcon-sumerconfidenceindex,-to-monthstandarddeviationofaround3,,andthemostpronouncedimprovementsinJan2004,Mar1991,andNov1992.[InsertFigure1(TimeSeriesofSentimentMeasures)abouthere],changesintheseindexes(althoughwerelyonmonthlytests,notannualtests),(forValidation)Wealsohavesomeothersurveydata,whichsufferfrominsufficientlylong,insufficientlydense,and/,theyarenotsuitableasprimarydataseries,
ValueLabel1verypessimistic2somewhatpessimistic3neither4somewhatoptimistic5veryoptimistic6don’tknow7refused8noanswerAnotherquestion(S5)providesaclassificationintoinvestorswithmorethan$100,000instockandbondinvestments(henceforthtermed“wealthy”),andinvestorswithless(henceforthtermed“poor”).Thetotalnumbersare22,687“wealthy”investor-monthsand29,987“poor”“veryoptimistic”as+2,a“somewhatoptimistic”as+1,asomewhatpes-simisticas–1,anda“verypessimistics”asa–
AnevenmorerecentsurveyistheInvestor1-YearConfidence(Semi-Annualto2002,MonthlyThereafter),:.(Theindividualserieshaswaytoofewobservations.)ThefirstsurveyoccurredinOctober1989,thenwassemi-annualuntilJanuary2002,,thetypicalShillerindexwasabout75,,,-fore,,administeredbytheInter-UniversityConsortiumforPoliticalandSocialResearch(ICPSR)attheUniversityofMichi-gan,theNationalOpinionResearchCenter(NORC)attheUniversityofChicago,(mnemonic“HAPPY,”identifiedasGeneralHappiness)is“Takenalltogether,howwouldyousaythingsarethesedayswouldyousaythatyouareveryhappy,prettyhappy,ornottoohappy?”Thereare3answers(“veryhappy”,“prettyhappy”,“nottoohappy”)inadditionto“don’tknow”and“noanswer.”Therewereabout40,000responsesinthesurvey,,(30%to35%,).However,theycorrelateperverselywiththenegativeoftheclosed-endfunddiscountindexes(–43%,)—whentheCEFDturnsmorebullish,-Sahalia,Parker,andYogo(2004)exploresthecombinedUSsalesgrowthforTiffany(since1960),Saks(since1991),Bulgari(since1992),Gucci(since1991),Hermes(since1992),LVMH(since1993),andWa-terfordWedgwood(since1994).5TheShillerindexdoesdisplaysomestrangefeaturesinitssemiannualperiod:itwasverybearishwhenmarketsweregenerallyheldtobeveryexuberant,.,,itcorrelatesnegativelywithanyotherconfidenceandsentimentindex,://
Wefindthatluxurygrowthcorrelatespositivelywithbothfinancialandsurveysen-timentincreases,(2004)Itisessentiallythefirstprincipalcomponentofsixsentimentmeasures:SENTIMENT=−·ClosedEndFundDiscount(t)+·NYSETurnover(t-1),logged,detrended+·NumberofIPOs(t)(2)+·AverageFirstDayIPO(t-1)Return+·ShareofEquityinTotalAggregateIssuing−·DividendPremiumPD−ND(t-1).(TheirdividendpremiumisthelogdifferenceofM/BratioofpayersminusM/Bratioofnonpayers.)Notsurprisingly,theBakerandWurglerindexcovariespositivelywiththeclosed-endfunddiscount(about25%correlation).ItalsocovariesequallywellwiththeCon-ferenceBoardconsumerconfidenceindex(28%)—butitdoesnotcovarywiththeMichiganconsumerconfidenceindex(–2%).EDependentVariables:StockReturnDataOurthreedependentvariablesarethe(contemporaneous)-capitalizationstocksandthelargestcapitalizationstocks,-to-monthstandarddeviationwasaround7%.Retail-StockReturnSpreadTheinstitutionalholdingportfolioswereformedfromThom-son13(f)datareports,eachquarterend(March,June,September,andDecember),;forexample,January,February,,;,forwhichthetheorypredictsprogressivelyhigherexplanatorypower:-institutionalhold-ingstocksminustherateofreturnonstockswithinstitutionalholdings.(Port-foliosarealwaysequal-weighted.)13
-institutionalholdingsdecileportfolio,-tradingvolumezero-institutionalholdingstocks,“retail-stockreturnspreads.”Thecorrelationamongthesethreemeasuresisbetween73%and90%.Table1showsthatretailstocksper-formedaboutthesameasinstitutionalstocks,-seriesstandarddeviationisrelativelysmall,onlyabout3%to5%-firmreturndatacomesfromaconsiderablylongertimespan(1965–2003)thantheretail-stockreturndata(1980–2003).Thereforeresultsbasedonretail-stockreturnspreadsmaybelessreliablethanresultsbasedonsmallstockspreads.[InsertFigure2(TimeSeriesofReturnSpreadsandS&P500PercentChanges)abouthere]Figure2plotsthetime-seriesofthe(logof)theS&P500,thesmall-stockreturnspread,andthethirdretail-stockreturnspreadaftertheS&P500percentchangehasbeenhedgedout(.,in-sampleregressionresiduals).Thetwoseriesdocovary,,-hedgeretailandlow-dollar-trading-volumestocksdidworstin2000,&P500ascontrolformarketconditions.(Itmakesnodifferencewhetherdividendsareincludedornot,orwhetheranothermarket-index,suchastheCRSPvalue-weightedindex,isused.),Shleifer,andThaler(1991),thesmall-firmreturnspreadisnotcorre-latedwiththeoverallstockmarketrateofreturn—inoursample,itscorrelationwiththeS&P500rateofreturnis+2%.Therefore,incorrelationswiththesmallfirmspread,,ourretail-stockreturnspreadshaveaveryhighnegativecorrelation(around–33%)withtherateofreturnontheS&,controlfortheoverallmarketisimportantinexplainingtheretail-stockreturnspread,,%
FDependentVariable:,Shleifer,andThaler(1991),ourseriesofclosed-endfundIPOscomesnotjustfromtheCEFreturnsdatabase,butfromthemorecompleteThomsonFinancial’,thenumberof(domesticallyoriented),,[InsertTable2(SentimentMeasureValidations)abouthere]Table2showsthecorrelationamongmeasuresofsentimentchanges,-basedfinancialsentimentchangeshaveacorrelationofabout80%withoneanother,muchhigherthanthe52%correlationbetweentheMichiganandConferenceBoardsurvey-basedsentimentchanges.[InsertFigure3(-WeightedCEFD)abouthere](bothindexesareclosetorandomwalks),thetwoaredifferent.(Correlationsinlevelsareanecessary,butnotsufficientvalidationofsentimentmeasureidentification.)Figure3showsthatinthesamplebefore1985theinvestorsentimentwashighwhentheconsumerconfidencewashigh(upwardslopinglines,foroverallsampleandpre-1985),buttherelationshipreversespost1985(downwardslopingline).AninvestorcanthereforenotreliablyconcludefromthecurrentaverageCEFDwhereconsumerconfidencestands,
concernedwithconsumers’,,oursubsequentanalysisfocusesontheMichiganconsumerconfidenceindex—anoisybutreasonableproxyforUBS/,thefinancialCEFD-basedsentimentmeasuresdonotcorrelatestatisticallysignificantlywiththeUBS/,inordertoconsidertheCEFDaninvestorsentimentmeasure,anauxiliaryassumptionmustbethatthe(relevant)investorsdonotarticulatetheirsentimentinGallup’-FirmReturnSpread[InsertTable3(Small-FirmReturnSpread)abouthere]Table3explainsthesmall-firmreturnspread,.(Table3indicatedthatthetwomeasuresarealmostuncorrelated,whichmeansthatthecoefficientsandsignif-icancelevelsononeremainpracticallythesameifweexcludetheother.)Overthefullsampleperiod,%higherstandarddeviationincreaseinsmall-firmreturnspread,whileaonestandarddeviationhigherdecreaseintheMichigansurveyassociates“only”withan18%,,,,,-endfunddiscount,,comparedtotheMichiganindex,theConferenceBoardindexcorrelatesgenerallymorewithmacroeconomicvariablessuchasGDPchangesandunemploymentchanges,andgenerallylesswithfinancialvariablessuchasinterestratechanges,S&P500returns,excessreturns,
CExplainingTheRetail-StockReturnSpread[InsertTable4(Retail-StockReturnSpread)abouthere]Table4explainsthedifferencesbetweentherateofreturnonfirmsthathavenoinstitu-tionalholdings(13(f)filings)andtherateofreturnonfirmsthathaveinstitutionalholdings.(Bothportfoliosarethemselvesequal-weighted.)Thetableshowsthatwhenconsumersturnmorebullish,“retailstocks”outperform“institutionalstocks.”-stockreturnspreadpriorto1985,,,controlfortheoverallmarketrateofreturnisimportant.(OurresultsarevirtuallyidenticalifweusetheCRSPvalue-weightedstockmarketrateofreturnratherthantheS&P500percentchange.)RetailstockshadaninversecorrelationwiththeS&P500returninoursampleperiod,,[InsertTable5(Retail-StockReturnSpread,TopDecile)abouthere],,,,theCEFDremainsinsignificant.[InsertTable6(RetailLow-TradingStockSpread)abouthere]-heldandrarelytraded,(whichwouldmakemarketefficienciesmoredifficulttoarbitrage),whilethelatterarenowthetwotopinstitutionallyhelddeciles,butonlythosefirmsthatrankintheirdecile’,consistentwiththetheory,&P500tothelatterifwefirstrunaregressiontohedgeretailstockreturns,,theMichiganconsumerconfidencemeasureremainsstrongandsignificant,-stockspreadwithsuchaprocedure,therebyaskinghowmuchwecanexplainthatneithertheS&,onlyourfinalmeasureofretail-stockreturnspread(whichincorporatestradingactivity)
CEFDbasedsentimentindexnowalmostcomesinsignificantinthesecondsubsample(andisstatisticallysignificantatconventionallevelsonaone-sidedtest),butitremainsinsignificantintheoverallsample.[InsertTable7(RetailLow-TradingStockSpread,Value-Weighted)abouthere]Table7showsthattheeffectisrobustifweusevalue-weightedportfolios,,thefindingsarealsorobustifweexcludeallJanuaryobservations.[InsertTable8(RetailLow-TradingStockSpread,SizeControlled)abouthere],,:especiallypriorto1985,-turnsmeansthatouroverallsample’sconsumerconfidencevariableisstillsignificantlypositive,,aboutone-thirdoftheeconomicinfluenceoftheconsumerconfidenceindexonretailstocksisduetoitsabilitytoexplainsmallstockreturns,,-plorethesmallstockreturnspread,,theretail-stockreturnspreadisjustbarelyexplainablebyconsumerconfidenceinaone-sidedtest().Insum,weinterprettheoverallevidencetosuggestthatconsumerconfidenceindexesplayaroleinexplainingtheperformanceofsmallstocks,retail-heldstocks,,
DPersistence(Prediction)Wewouldarguethatareasonablesentimentindexcouldbeinfluencedbyrecentpositivestockreturns—andespeciallyrecenthighoverallstockmarket(portfolio)returns,andhave(mild)-statisticsonthecorrelationsbetweenchangesinoursentimentmeasures,andourmonthlyratesofreturnofinterest:LagofSmallStockReturnSpread–5–4–3–2–10+1+2+3+4+–––++++–++–+––++–––++-AdjustedLow-TradeRetailStockReturnSpread–5–4–3–2–10+1+2+3+4+––++++–+++––+–+–+–––++&P500PercentChange–5–4–3–2–10+1+2+3+4++––+++++––++–––+++––++(sentimentanticipatesreturn)(returnanticipatessentiment)InaGrangercausalitysense,significantnumbersontheleftimplythatthesentimentindexpredicts(influences)thereturn,numbersontherightimplythatthesentimentindexispredicted(influenced)-adjustmentinthemiddlepanelisdonebyhedgingoutthein-sampleS&500returnviaregression.(Thenumbersaresimilarforotherretailspreads.),Durell(2001)hasworkedonsimilarquestions,(heexploresthere-lationbetweenthemarketandtheconsumerconfidenceindexinmoredetailthanouronesubsectionhere),butalsoinsomefindings—,mostimportantly,—andsomeothermilddifferencesinresults—maybepartlyduetodifferencesinspecifications(hetendstouselonger-termreturns),partlyduetodata(heusesonlytheConferenceBoardindexandonlyonecomponentthereof),andpartlyduetosample19
period(hehasoverlappingdatafortheCEFDandtheConferenceBoardforonly7years,1978–1985).EClosed-EndFundStartups[InsertTable9(MonthlyClosed-EndDomesticFundStartups(IPOs))abouthere],liketheCEFDsentimentmeasureitself,,thisshouldnotbetoosurprising:ittakestimetostartupafund,andasinglemonth’sbullishnessisnotlikelytotranslateintoimmediatefundstartups(althoughfundscouldbe“waitinginthewings”untilinvestorsturnmorebullish).[InsertTable10(AnnualClosed-EndDomesticFundStartups(IPOs))abouthere],wedonothavemuchannualdata,,morebullishconsumersmaybeableto!Thecorrelationdropsjustbelowtwo-sidedstatisticalsignificanceifweincludebothmeasures,,,thecorrelationsofbothbullishvariablesdrop,onlongerhorizons(upto18months),thecorrelationsincrease.(Ifwelogthedependentvariable,weloseanother20basispointsontheT-statistics,therebydroppingbelowordinarystatisticalsignificance.)Nevertheless,thisremainsapuzzlingfinding—andtoogoodtobetrue:underalmostanyhypothesis,wewouldhaveexpectedclosed-endfundsnottostartupiftheclosed-endfunddiscountishigh,,wewerenotexpectingtheconsumerconfidencetoplaymuchofarole,,wewanttoreemphasizethatitislikelythatfundstartupscorrelatewiththeconsumerconfidenceonlybecausethe(consumer)
,theS&(of18%),followedbychangesintheCEFD-basedsentiment(about10-12%,althoughtherelationshipisanunstable0%before1985,and20%after1985),andfinallytheConferenceBoardindex(9%,notsignificantandunstable).Ofcourse,thisiscontemporaneous,—although,asforsmallfirmsandretailfirms,,,;forexample,1974–1975,,,changesinboththeMichiganandtheConferenceBoardconsumerconfidenceindexandchangesinGDPcorrelatesignificantlypositively(around10to15%).Whenweworkwithchangesinunemployment,wefindthat,althoughnotstrong,thereisasuggestion(sometimesstatisticallysignificant,sometimesnot):thoughwedonothavenumbersforemploymentofwealthyinvestors,itisneverthelesshardtoimagineanoise(retail),theConferenceBoardconsumerconfidenceindexincreaseswhenunemploymentdecreases.(TheMichiganindexdisplaynocorrelationwithunemploymentchanges.)-Frenchbook-marketfactor—availablefromKenFrench’swebsite—,theCEFDisitselfinessenceabook-marketderivedvariable,,
%.ThereisamildcorrelationbetweentheCEFDandmonthly1-yearinterestratechanges:whentheCEFDdecreases(morebullish),,,changesinbothconsumerconfidencemeasurescorre-latestronglypositivelywithchangesininterestrates:consumersturnmorebullishwheninterestratesrise,,wedetectnosolidsystematiccontemporaneousrelationshipbetweensentimentmeasuresandgen-eralIPOissuingactivity,,-endfunddiscountwithreturnsandourconsumerconfidencemeasure(),=−+(+)·+(−)·:−−+(−)·+(+)·s&:(3)OnlytheS&P500rateofreturnisstatisticallysignificant,,,()andthemarketreturnisirrelevant(t=−);after1985,thesmallstockreturnspreadbecomesinsignif-icant(),whiletheS&P500rateofreturnbecomesstronglysignificant(t=).Onanannualbasis,wefindoneoddcorrelation:inlevels,theCEFDcorrelateshighly(andstatisticallysignificantly),,
(2004),weareabletodosomepreliminaryexplorationofourrelationshipintheUnitedKingdom,becauseDimson,Nagel,andQuigley(2004)(1955-2001),+20%,amongthem35%inNovember1999,and24%−10%,whichwas−19%inDecember1999.(%.)Thesereturnswereabout4standarddeviationsofftheseriesmean,10whichleadsustobelievethatthe11/99to01/,,,-denceindicator(CC,series99),butalsoafinancialsituationindicator(FSI,series01),andageneraleconomicsituationindicator(ESI,series03).;thechangeinFSI(ESI)()().11Reportingallcoefficientsinpercent,explainingcontemporaneoussmallfirmexcessreturns,+·∆CC+(−)·RFTSE+ t-stat()()+() t-stat()()+()winsorizedat±2σ+·∆CC+(−)·RFTSE+ t-stat()()+(),,(CC)withchangesintheeconomicsituationindicator(ESI).,%%.11TheECalsopublishesforwardlookingstatistics,
+·∆FSI+(−)·RFTSE+ t-stat()()+() t-stat()()+()winsorizedat±2σ+·∆FSI+(−)·RFTSE+ t-stat()()+()Althoughthesigniscorrect,wehadexpectedmorestatisticalsignificance,,thatwillrequireananalysisbeyondwhatwecanaccomplishinourpaper—,
“behavioral”loosely,becauseitcouldalsobethatagencyissuesarethereasonwhyad-visorsplacetheirclients’trustfundsintotheseclosed-endfunds.)However,forfutureresearchstudies,ifaninvestorsentimentmeasureiscalledfor,-qualityones,
ReferencesAit-Sahalia,Yacine,JonathanParker,andMotohiroYogo,2004,LuxuryGoodsandtheEquityPre-mium,TheJournalofFinance59,,Malcolm,andJeffreyWurgler,2004,InvestorSentimentandtheCrossSectionofStockReturns,Workingpaper,,JonathanB.,andRichardStanton,2004,ARationalModeloftheClosed-EndFundDiscount,Workingpaper,,Nai-Fu,RaymondKan,,1993,ArethediscountsonClosed-endFundsaSentimentIndex,TheJournalofFinance48,795–,Navin,,AndreiShleifer,,1993,Yes,DiscountsonClosed-EndFundsareaSentimentIndex,TheJournalofFinance48,801–,RichardT.,2000,SurveysofConsumers,Workingpaper,,.,AndreiShleifer,,,1990,NoiseTraderRiskinFinancialMarkets,JournalofPoliticalEconomy98,703–,Elroy,StefanNagel,andGarrettQuigley,2004,-2001,FinancialAnalystsJournal?,,JohnA.,,2004,InvestorSentimentandtheClosed-endFundPuzzle:Out-of-sampleEvidence,EuropeanFinancialManagement10,235–,Alan,2001,StockMarketExpectationsandStockMarketReturns,Workingpaper,,.,AndreiShleifer,,1991,InvestorSentimentandtheClosed-EndFundPuzzle,TheJournalofFinance46,75–,StephenA.,2002,ANeoclassicalLookatBehavioralFinance;ClosedEndFunds,Workingpaper,,Matthew,1997,Closed-EndFundDiscountsinaRationalAgentEconomy,Workingpaper,
,CEFDBasedVariableMeanSdvMinQ1MedianQ3MaxρRange#−−−−−–−−−−−–−−−−−–−−−−–,SurveyBasedVariableMeanSdvMinQ1MedianQ3MaxρRange#–−−−−––−−−−–,SurveyBasedVariableMeanSdvMinQ1MedianQ3MaxρRange#−–−−−−−–−–−−−−−–−–−−−−−––−−−–ρRange#−−−–−−–−−−−–−−−−−–−−–-EndFundStartupsVariableMeanSdvMinQ1MedianQ3MaxρRange#–−−−–-endfunds,cefdistheclosed-endfunddiscount,michistheMichiganconsumerconfidenceindex,cbistheConferenceBoardconsumerconfidenceindex,gallupistheUBS/Galluppollofinvestors,
,(Value-WeightedCEFDDecreases)%**42378%**23179%**−4%−−13%−−0%−%−1%−%%%%(Equal-WeightedCEFDDecreases)−7%−−10%−−5%−%%%%−6%−%(MichiganConsumerConfidenceIncreases)%**32042%**9057%**%**%**%**%(ConferenceBoardConsumerConfidenceIncreases)%**%*%*%:-endfunddiscountbasedmeasure,ewdenotesthatitisequal-weighted,,
%−−*****%*******post19855%−−***%386−−−**%**%172−−−−−***Description:Thedependentvariable,,&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvariablesarenormalizedtoameanof0andastandarddeviationof1).(twostars)denotesignificanceatthe5%(1%)level,
-StockReturnSpread,Longsare13(f)%−−−−****pre198511%−−**%−−−−**Description:Thedependentvariable,,&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvariablesarenormalizedtoameanof0andastandarddeviationof1).(twostars)denotesignificanceatthe5%(1%)level,-StockReturnSpread,LongsareTopDecileof13(f)%−−−−****pre198514%54−−−**%−−***Description:Thedependentvariable,,isthemonthlyreturnonfirmswithno13Ffilings,&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvariablesarenormalizedtoameanof0andastandarddeviationof1).(twostars)denotesignificanceatthe5%(1%)level,
%−−******pre198516%−−****post198521%−−******Description:Thedependentvariable,,isthemonthlyreturnonfirmswithno13Ffilingsandwithinthiscategorythelowerhalfofdollartradingvolume,&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvariablesarenormalizedtoameanof0andastandarddeviationof1).(twostars)denotesignificanceatthe5%(1%)level,-TradingStockSpread,%246−−−−−****pre198510%−−−−**post198531%188−−−−−****Description:Thedependentvariableisthemonthlyreturnonfirmswithno13Ffilingsandwithinthiscat-egorythelowerhalfofdollartradingvolume,-weighted,&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvariablesarenormalizedtoameanof0andastandarddeviationof1).(twostars)denotesignificanceatthe5%(1%)level,
-TradingStockSpread,%−−*******pre198564%−−−−***post198531%−−******Description:Thetablediffersfromtheprevioustableinthatitincludesoneadditionaldependentvari-able,theexcessrateofreturnonsmallfirms().Thedependentvariable,,isthemonthlyreturnonfirmswithno13Ffilingsandwithinthiscategorythelowerhalfofdollartradingvolume,&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvari-ablesarenormalizedtoameanof0andastandarddeviationof1).(twostars)denotesignificanceatthe5%(1%)level,
-EndDomesticFundStartups(IPOs)-1%−−−−−−**-0%366−−−−−:Thedependentvariableisdescribedinthefirstcolumn,andiseitherthelevelofclosed-endfundstartupIPOs()orthethelevelofclosed-endfundstartupIPOs().&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvariablesarenormalizedtoameanof0andastandarddevi-ationof1).(twostars)denotesignificanceatthe5%(1%)level,-endDomesticFundStartups(IPOs)-6%−−**%−−***%−−**%26−−−−−:Thedependentvariableisdescribedinthefirstcolumn,andiseitherthelevelofclosed-endfundstartupIPOs()ortheannualdifferenceofclosed-endfundstartupIPOs().&,thesecondrowprintsthestandardizedcoefficient(bothdependentandindependentvariablesarenormalizedtoameanof0andastandarddevi-ationof1).(twostars)denotesignificanceatthe5%(1%)level,
-SeriesofSentimentMeasuresMichigan CC ()1970198019902000Negative of CEFD ()Ross DataLST Data1970198019902000Michigan CC Changes ()1970198019902000Negative of CEFD Changes ()Ross DataLST −−20−1001020−−
-SeriesofStockPrices,ReturnSpreads,andCEFStartupsLog Price, S&P5001970198019902000Small−Stock Return Spread ()1970198019902000Hedged Retail−Stock Return Spread1970198019902000CEF ()0246810−−−
-EndFundDiscountO*O*O*O*O*OO*O*O*OOO*OOO*OO******O**OOO*O*O**O2000*OO*OO***O**1998*OOOOO1*999****O**OO****Opo*sOt−*1*O98O*5*O***O*O***1984OO****O1997*O*O***O*OOO*1966*O*******OOO****OOO****OO***O*******O**OO****OO****O*OO*OO***1969****O**OO1968*OO*O*OO****O*198O6**O*O*********1*9881967OO*O9O*OOO****OOO*********O1996**OO**O1985O*******O****O*OOOOOOO*O*O*O******O***O*OOO**O*O***O**O*******O*OO1972*1994**1*990O***O*OO**O1987****O***O**OO1995***********OOO*O*O*****O*******O**197**7O*1991*O***O*O****1993*******aO*OO*1976*l datO***al*OOO*******O****O****OO*******O1983***1973******OO****O***1978OO1971******OO*****O*O**O**1970************1992*********O**pre−1985O**************O****O**O**O***O***1979*O********O***O*1981*********O1974********1975**1982*******1980**−− of CEFD (EW)Explanation:,:,,“in-sync.”36Michigan Consumer Confidence Index5060708090100110
