Conflict_of_Interest

96www.cfapubs.org©2010CFAInstituteFinancialAnalystsJournalVolume66Number4©2010CFAInstituteConflictsofInterestandAnalystBehavior:EvidencefromRecentChangesinRegulationArmenHovakimianandEkkachaiSaenyasiriRegulationFDmadeanalystslessdependentoninsiderinformationanddiminishedanalysts’motivestoinflatetheirforecasts.TheGlobalResearchAnalystSettlementhadanevenbiggerimpactonanalystbehavior:Themeanforecastbiasdeclinedsignificantly,whereasthemedianforecastbiasessentiallydisappeared.Theseresultsaresimilarforallanalysts.urinvestigationoftheimpactofrecentchangesinregulationonanalysts’forecastingbehaviorfollowsanumberofstudiesthatarguedthatanalystsweremotivatedtoproduceresearchreportsthatdidnotreflecttheirtrueopinions.Analyststendedtomakeexcessive“buy”recommendationsandinflatedearningsforecastsforseveralreasons,twoofwhichgainedconsiderableattentionfromregulatorsintheUnitedStates.First,analystsmayhavefeltcompelledtofavormanagersincoveredcompaniesinordertogainprivilegedaccesstoinformationflow(Lim2001).Second,althoughanalystsaresupposedtoprovideinvestorswithaccurateandtruthfulresearchreports,conflictsofinterestcouldoccurbecauseanalysts’compensationwastiedtoprofitsgeneratedfrominvestmentbankingbusinessandbrokeragecommissions(LinandMcNichols1998;Carleton,Chen,andSteiner1998).Intheearlypartofthefirstdecadeofthiscentury,inanefforttorestorepublicconfidenceinU.S.capitalmarkets,U.S.regulatorsenactedseveralrulesandregulations,prosecutedanalystswhoseresearchreportsweretaintedbyconflictsofinterest,andfinedbanksthatfailedtopreventresearchanalysts’conflictsofinterest.Twoofthemainregulatorydevelopmentsduringthisperiodwere(1)RegulationFairDisclosure(RegFD),whichbecameeffectiveon23October2000,and(2)theGlobalResearchAnalystSettlement(GlobalSettlement),whichwasannouncedon20December2002.1Althoughtheprimarygoalsofthesetworegulatoryactionsaredifferent,theybothhavethepotentialtoimprovethequalityofanalystforecasts.OneofthestatedgoalsofRegFDistoprohibitprivatecommunicationbetweencompaniesandanalysts,therebyhelpingtoleveltheplayingfieldsothatmarketparticipantscanhaveequalaccesstoinformationandmakinganalystslessdependentonsuchcommunication.Inprohibitingcompaniesfromselectivelydisclosingprivateinformationtoanalysts,RegFDmayreduceanalystforecastbiasbyeliminatingtheincentiveforanalyststoinflatetheirearningsforecastsinordertogainaccesstoinsiderinformation.TheGlobalSettlementisanimportantenforcementagreementbetweenU.S.regulatorsand12largeinvestmentbanks(theBig12banks)designedtoeliminateresearchanalysts’conflictsofinterest.Ifsuccessful,theGlobalSettlementshouldreduceoptimisticbiasinanalystforecasts.OurstudyconsideredwhetherthesetwoactionsbyU.S.regulatorsreducedthebiasinanalysts’earningsforecastsdocumentedinpreviousstudies.Wefocusedonannualearningsforecastbiasforseveralreasons.First,investorsmayuseanalystforecaststoformexpectationsofearningsandcashflows,bothofwhichareimportantinputsforstockvaluationmodels.Inflatedearningsforecastscandrivestockpricesabovetheirfairvaluesifinvestorsfailtoadjustforthebias.2Second,giventheflurryofnewregulations,regulatorsclearlyconsideranalystbehavioranimportantfactorinmaintaininginvestorconfidenceinfinancialmarkets.Regulationiscostlybecauseofthesignificantexpensesassociatedwithanalyzingproblematicsituationsanddevelopingremedies.Moreover,restrictionsandreportingrequirementsimposedonvariousmarketparticipantsresultinongoingcompliancecosts.Thesecostscanbejustifiedonlyifthenewregulationshelpreduceanalysts’conflictsofinterestandtherebygenerateanimportantbenefitforfinancialmarkets.ArmenHovakimianisprofessoroffinanceatBaruchCollege,NewYorkCity.EkkachaiSaenyasiriisassistantprofessoroffinanceatProvidenceCollege,Providence,RhodeIsland.OJulyAugust2010www.cfapubs.org97ConflictsofInterestandAnalystBehaviorThird,moststudiesthathaveexaminedtheimpactofRegFDandtheGlobalSettlementonanalystbehaviorfocusedonforecastaccuracyandforecastdispersion(Bailey,Li,Mao,andZhong2003;Agrawal,Chadha,andChen2006).3Theseaspectsofanalystbehavior,however,arelittleaffectedbyconflictsofinterest,thefocusofourstudy.Otherstudieshaveexaminedforecastbias.Clarke,Khorana,Patel,andRau(2006)foundthattheGlobalSettlementhadnoimpactonrelativebiasinanalystforecasts.FocusingontheimpactofRegFDonbiasinquarterlyearningsforecastsbetweenOctober1999andDecember2001,MohanramandSunder(2006)foundthattheseforecastsbecamemoreoptimisticafterRegFDbutattributedtheincreasetounexpectedlylowrealizedearningsduringthe2001recession.Ourlongerstudyperiod(1996–2006)allowedustocontrolformacroeconomicconditionsinourregressionanalysis.Furthermore,weexaminedlongerterm(upto24months)earningsforecastsinwhichtheforecastbiasismoreapparent(Richardson,Teoh,andWysocki2004).AlthoughHerrmann,Hope,andThomas(2008)foundsomeevidenceofdeclineinforecastbiasfollowingRegFD,theyfocusedoninternationallydiversifiedcompaniesonly;weexaminedallU.S.companies,andourprimaryfocuswasonchangesinforecastbiasaftertheGlobalSettlement.Lastly,theabilityofanalyststoforecastearningsaccuratelycanbeeasilyandstraightforwardlyverifiedbecauseactualearningsareobservedattheendoftheforecastperiod.Barber,Lehavy,McNichols,andTrueman(2006)studiedthechangeindistributionofstockrecommendationsmadefrom1996to2003.Theyfoundthatthepercentageofbuysdecreasedstartinginmid2000.4Howunbiasedthenewdistributionofstockrecommendationsis,however,remainsuncertain.Butweknowthatthebiasshouldbezeroattheaggregatelevelwhenanalystsmaketheirforecastsonthebasisoftheirtrueopinions.InstitutionalBackgroundHistorically—andespeciallybeforerecentregulations—analystshavetendedtomakeundulyoptimisticearningsforecasts.Inthissection,wediscussthepossiblereasonsforthisoptimisticbiasandthepotentialimpactsoftherecentregulationsonsuchbias.WhyDoAnalystsMakeOveroptimisticEarningsForecasts?Anumberofstudieshavedocumentedthatanalystsregularlymakeoveroptimisticearningsforecasts(Brown1997;Chopra1998;Beckers,Steliaros,andThomson2004).Optimisticbiastendstobelargerforlongertermforecastsandsmallerforforecastsmadeclosertotheearningsannouncementdate.Thisphenomenonisusuallyreferredtoasthewalkdowntrend(Richardson,Teoh,andWysocki2004).Severalexplanationshavebeenofferedforanalystoptimism.First,analystsmaybeinfluencedbyconflictsofinterestiftheircompensationistiedtoinvestmentbankingfeesandbrokeragecommissions.LinandMcNichols(1998)foundthatanalystsaffiliatedwithunderwritersmakemorefavorablestockrecommendationsandlongtermearningsgrowthforecaststhananalystsnotsoaffiliated.AgrawalandChen(2005)discoveredthatoptimisminlongtermearningsgrowthforecastsishighwhenanalystsworkforfinancialinstitutionswhoserevenuescomemainlyfrombrokeragebusiness.Carleton,Chen,andSteiner(1998)foundthatstockrecommendationsmadebybrokeragefirmsaremoreoptimisticthanthoseofnonbrokeragefirms.UsingAustraliandata,Jackson(2005)notedthatoptimisticanalystsgeneratemoretradesfortheirbrokeragefirmsthandolessoptimisticanalysts.Chan,Karceski,andLakonishok(2007)showedthatanalysts’earningsforecastsareinfluencedbytheirdesiretowininvestmentbankingclients.Doukas,Kim,andPantzalis(2005)reportedthatstockswithexcessanalystcoverageyieldlowerfuturereturns,consistentwiththeconflictofinteresthypothesis.HongandKubik(2003)foundthatbrokeragehousesrewardoptimisticanalysts;optimisticanalystsatlowstatusbrokeragehousesaremorelikelytomoveuptohigherstatusbrokeragehousesthanarelessoptimisticanalysts.Second,analystsmayfeelcompelledtomaintaingoodrelationswithcompanymanagementinordertogainaccesstoinsiderinformationthatcanhelpimprovetheaccuracyoftheirforecasts(Lim2001).Third,analystsmaytendtocoverstocksforwhichtheyhavepositiveviewsanddroporavoidstocksforwhichtheyhavenegativeviews,whichcaninduceaselfselectionbias(McNicholsandO’Brien1997).Fourth,analystsmayhaveacognitivebiasthatleadsthemtooverreacttogoodearningsinformationandunderreacttobadearningsinformation(EasterwoodandNutt1999;Nutt,Easterwood,andEasterwood1999).Finally,thewalkdowntrendmaybedrivenbythe“earningsguidancegame,”inwhichanalystsissueoptimisticforecastsatthestartofthefiscalyearandthenrevisetheirestimatesuntilthecompanycanbeattheforecastattheearningsannouncementdate(Richardson,Teoh,andWysocki2004).98www.cfapubs.org©2010CFAInstituteFinancialAnalystsJournalRecentRegulations.BeforeRegFD,analystsandinstitutionalinvestorsoftenhadaninformationaladvantageoversmallinvestorsthroughprivatecommunicationswithmanagementandconferencecallsinwhichcompanymanagersdiscussedpastperformanceandprovidedguidanceonfutureprospects.Suchtimelyinformationgavetheseinvestmentprofessionalsanunfairadvantagethatallowedthemtotradestocksprofitablyattheexpenseofuninformedinvestors.Togainaccesstothisinformationflow,analystsmayhavehadtomaintaingoodrelationswithinsidersbymakingoptimisticforecastsandbuyrecommendationsintheirresearchreports.Analysts’excessivelyoptimisticviewsofthestocksweremisleadingandcontributedtothedeteriorationofinvestorconfidenceincapitalmarketintegrity.ThroughRegFD,whichwasintroducedinOctober2000,theU.S.SECintendedtoimprovefairnessandrestorepublicconfidenceinthemarketsbyrequiringU.S.publiccompaniestodisclosematerialinformationsimultaneouslytoallmarketparticipants.Othersourcesofconflictsofinterest,however,remainedunaddressedbyRegFD.Forinstance,analystscouldbepressuredtomakeoptimisticforecastsandbuyrecommendationsinordertofavorinvestmentbankingclientsandgeneratetradingvolume.TheSECandsuchselfregulatoryorganizations(SROs)astheNationalAssociationofSecuritiesDealers(NASD;nowtheFinancialIndustryRegulatoryAuthority[FINRA])andtheNYSEpaidsignificantattentiontothisissueandintroducedanumberofnewrulesandregulationstocurbthenegativeconsequencesoftheseconflictsofinterest.TheSarbanesOxleyActof2002(SOA),alsoknownasthePublicCompanyAccountingReformandInvestorProtectionActof2002,becamelawon30July2002.TheSOAisabroadpieceoflegislationthatcoversvariousbusinesspractices,includingauditorindependence,corporateresponsibility,enhancedfinancialdisclosure,analysts’conflictsofinterest,andcorporateandcriminalfraudaccountability.TheSOAamendedtheSecuritiesExchangeActof1934bycreatingSection15D,whichrequiresFINRAandtheNYSEtoadoptrulesreasonablydesignedtoaddressresearchanalysts’conflictsofinterest.TocomplywiththeSOA,theNASDreleasedRule2711(ResearchAnalystsandResearchReports)andtheNYSEamendeditsRule351(ReportingRequirements)andRule472(CommunicationswiththePublic).Mostprovisionsoftheseruleswentintoeffecton9July2002.Theserulesmitigateanalysts’conflictsofinterestbyseparatingresearchanalystsfromtheinfluenceoftheinvestmentbankingandbrokeragebusinesses.Researchanalysts’compensationcannolongerbetiedtotheperformanceofthesebusinesses.Inaddition,analystsarerestrictedfrompersonaltradinginthestockstheycover.On6February2003,theSECadoptedRegulationAnalystCertification(RegAC).5RegACprovidesguidelinesforproperdisclosureofpotentialconflictsofinterestofsellsideanalysts,includingtheirassociationwithinvestmentbankingclientsandthestructureoftheircompensation.Regulatoryobjectiveshavealsoreceivedsupportfromrigorousenforcementactions.FollowingajointinvestigationbytheSEC,NASD,NYSE,andNewYorkStateAttorneyGeneral,10largeU.S.andmultinationalinvestmentbanksagreedtopayafineof$1.435billionintheGlobalResearchAnalystSettlementfortheirfailuretoadequatelyaddressresearchanalysts’conflictsofinterest.Announcedon20December2002,thetermsoftheGlobalSettlementinitiallycovered10banks.6Thefinalagreementwasannouncedon28April2003.Twomorebanksreachedsettlementson26August2004.7TheGlobalSettlementandtheSROrulessharethesamespiritinthattheirmutualobjectiveistoeliminateanalysts’conflictsofinterest.Theintroductionoftheserulesandregulationsallowsustodifferentiateamongthealternativeexplanationsforanalystforecastbiasproposedintheliterature.First,areductioninforecastbiasafterRegFDwouldsupporttheargumentthatanalystswereoveroptimisticowingtotheirneedforinsiderinformation,especiallyifsuchareductionwerestrongerforinformationallymoreopaquecompanies.Second,areductioninbiasaftertheGlobalSettlementandRule2711wouldbeconsistentwiththehypothesisthatanalystbehaviorwasundulyinfluencedbyconflictsofinterest.8Incontrast,selfselectionandcognitivebiasesmayexisteveninaworldwithoutconflictsofinterest.Therefore,ifthesebiasesarethemainreasonsforanalysts’overoptimisticforecasts,thentheseregulatorychangesshouldhavenoeffectonforecastbias.9SampleandVariablesWedownloadedsellsideanalysts’earningsforecastsforfiscalyearenddatesbetween1996and2006fromtheDetailfileoftheIBESdatabase.Weusedforecastsforcurrentandsubsequentyearearningspershare(EPS),whicharemadefortheupcomingandfollowingyears’earningsannouncementdates.10Figure1illustratesthetimelineofanalystforecasts.TheearliestanalystforecastsforaspecificfiscalyearendEPSaremade24monthsbeforetheforecastfiscalyearend(inforecastmonth–23).ForeachEPS,analystscanJulyAugust2010www.cfapubs.org99ConflictsofInterestandAnalystBehaviormakemultipleforecastsoverthecourseofthenext24months.Someanalystsmaycontinuetomakeforecastsaftertheforecastfiscalyearendsbecausecompaniesannouncetheirannualearningsafteradelayofseveralmonths.BecausethelengthoftheEPSannouncementdelaycouldbeaffectedbyhowhighorlowtherealizedEPSisrelativetotheconsensus,weretainedonlythoseforecastsmadenomorethanonemonthaftertheforecastfiscalyearend(inforecastmonth+1),whichleftuswithatotalof2,297,792forecasts.Foreachforecast,IBESprovidesactualearnings,forecastdate,forecastperiod(fiscalyear)end,earningsannouncementdate,analystcodeidentity,brokercodeidentity,andnumberofanalystsusedforconsensuscalculation.11WeusedtheIBESBrokerTranslationfiletoconvertbrokercodesintobrokers’names,whichweusedtoidentifyanalystswhoworkedfortheBig12banks.StockpricesarefromtheIBESSummaryfile.12WedownloadedrealGDPgrowthratesfromthewebsiteoftheU.S.BureauofEconomicAnalysis.WedownloadedSICcodesfromtheCRSPmonthlyfile.Wedefinedanalystforecastbias,thefocusofouranalysis,astheaverageanalystforecasterrorandcalculateditasfollows:(1)(2)and(3)whereAj,t=theactualearningspershareforcompanyjinfiscalyeartFj,t,m,i=theaverageofannualearningsforecastsforfiscalyearendtofcompanyj,madeinmonthmbyanalystiKj,t,m,i=thenumberofforecastsmadeinmonthmbythesameanalystiforthesamecompanyjandfiscalyeartIj,t,m=thenumberofanalystsmakingforecastsinmonthmforcompanyjandfiscalyeartPj,t1=thestockpriceofcompanyjoneyearbeforethefiscalyearendt13NotethatallEPSforecastsmadeforthesamecompanyandthesamefiscalyeararenormalizedbythesamestockprice.Usingthesamestockpriceasthedenominatorguaranteesthatanychangesinforecastbiasacrossforecastmonths(m)aretheresultofchangesinanalystforecasts,notofchangesinthestockprice.InourcalculationsaccordingtoEquations1–3,weusedonlynewforecastsmadeinmonthm.Staleforecastsfromearliermonths(m1,etc.)werenotcarriedoverintomonthm.Inotherwords,eachforecastparticipatedinthecalculationoftheforecastbiasonlyonce,inthemonthinwhichtheforecastwasmade.Inoursample,anaverageanalystmade4.5forecastsforeachannualEPS.BecauseforeachannualEPSwetracked25monthforecasts(frommonth–23tomonth+1),theimplicationisthatanaverageanalystinoursamplemadeaforecastforeachcoveredcompanyaboutonceeverysixmonths.Tominimizetheinfluenceofoutliersandmisreporteddatainouranalysis,wereplacedwithmissingvaluesanyextremeobservationsofforecastbias,companysize,markettobookratio,thenumberofstocks,andthenumberofindustryanalystsfollowing.14WedroppedfromthesampleallforecastsmadeinOctober2000andDecember2002(1.5percentofoursample)andobservationswithmissingvaluesofanyrelevantvariable.WewereFigure1.TimelineofAnalystForecastsYearinWhichActualEPSIsCalculatedEarningsForecastsCanBeMadeatAnyDaybeforeEADStockPrice(Pt−1)Month−23FiscalYearEnd(t−2)FiscalYearEnd(t−1)Month0Month+1FiscalYearEnd(t)orForecastPeriodEndDateEarningsAnnouncementDate(EAD)Month−11BiasjtmjtmjtjtFAP,,,,,,,=−()−1001FIFjtmjtmjtmiiI,,,,,,,,==11FKFjtmijtmijtmikkK,,,,,,,,,,,==11100www.cfapubs.org©2010CFAInstituteFinancialAnalystsJournalleftwith1,586,000individualanalystforecasts,whichweusedtocalculate434,268averageforecasterrors.Foreachfiscalyearandforeachofour7,315samplecompanies,oursamplecontainedupto25monthlyobservationsofforecastbias(Biasj,t,m).Table1reportsthesummarystatisticsfortheoverallsampleof434,268observationsandforeachofthethreesubperiods.TheperiodbeforeRegFDrepresents53percentofoursampleobservations,withtheperiodbetweenRegFDandtheGlobalSettlementandtheperiodaftertheGlobalSettlementrepresenting18percentand29percentofthesampleobservations,respectively.Themeanforecastbiasacrossallsampleobservationsis1.39percentofstockprice.Thisresultisconsistentwithpriorevidencethatanalysts’forecastsareoptimisticallybiased(Brown1997;Chopra1998).NosignificantdifferenceexistsbetweenthemeanforecastbiasbeforeRegFD(1.72)andthemeanforecastbiasbetweenRegFDandtheGlobalSettlement(1.97).ThemeanforecastbiasismorethanfourtimessmalleraftertheGlobalSettlement(0.41),withthedifferencestatisticallysignificantatthe1percentlevel.Theaveragemarketcapitalizationofcompaniesinoursamplewas$4.5billion,andtheaveragemarkettobookratiowas3.57.Onaverage,8.41analystscoveredacompanyinanyparticularmonth.Theanalystsinoursampleworkedforbrokersthat,onaverage,eachemployed65.7analysts.Atypicalanalystfollowed16.30stocksfrom4.78industriesand,atthetimeoftheforecast,hadbeenintheIBESdatabasefor6.24yearsandmakingforecastsforthecoveredstockfor2.5years.Around17percentofforecastsweremadeforcompanieswithnegativeearnings,and36percentofforecastsweremadeforcompanieswhoseearningsweredecliningrelativetoearningsinthepriorfiscalyear.TestResultsInthissection,wepresenttheresultsoftheunivariatetestsandoftheregressionanalysisoftheeffectsofRegFDandtheGlobalSettlementonbiasinanalystforecasts.UnivariateResultsbyForecastMonth.Table2presentsthemedianforecastsbythemonthinwhichtheforecastsweremadeandbythefiscalyearforwhichtheyweremade.Thenumbersintheleftmostcolumnrepresentthemonth(relativetothefiscalyearend)oftheforecast.ThenumbersinthetoprowrepresentthefiscalyearsforwhichtheTable1.SummaryStatisticsDescriptionVariableNumberofObservationsMeanNumberofObservationsMeanBeforeRegFDBetweenRegFDandGSAfterGSBeforeRegFDBetweenRegFDandGSAfterGSForecastbiasBias434,2681.39231,09677,305125,8671.721.970.41RegFDindicatorRegFD434,2680.18231,09677,305125,8670.001.000.00GlobalSettlementindicatorGlob434,2680.29231,09677,305125,8670.000.001.00CompanycharacteristicsAnalystcoverageNumA434,2688.41231,09677,305125,8678.218.238.88Marketcap($millions)CompanySize434,2684,470.00231,09677,305125,8673,480.005,250.005,800.00MarkettobookratioMB434,2683.57231,09677,305125,8673.783.473.23NegativeEPSEPSLoss434,2680.17231,09677,305125,8670.160.260.14DecliningEPSEPSDecline434,2680.36231,09677,305125,8670.370.450.27LitigationLitigation434,2680.27231,09677,305125,8670.250.300.27LaborintensiveLabor434,2680.61231,09677,305125,8670.600.630.63AnalystcharacteristicsCompanyspecificexperienceYearStk434,2682.50231,09677,305125,8672.552.432.46GeneralexperienceYearIBES434,2686.24231,09677,305125,8676.456.195.87No.ofstockscoveredNumStk434,26816.30231,09677,305125,86718.1814.3114.06No.ofindustriescoveredNumInd434,2684.78231,09677,305125,8675.464.153.93BrokersizeBrokerSize434,26865.70231,09677,305125,86754.9889.0371.06Note:Thistablepresentsthesummarystatisticsfortheoverallsampleandforthethreesubperiods.JulyAugust2010www.cfapubs.org101ConflictsofInterestandAnalystBehaviorforecastsweremade.Forexample,forecastsmadeinSeptember2000forthefiscalyearendedDecember2000(i.e.,threemonthsbeforethefiscalyearend)areinrow–3andcolumn00.ThetwosolidlinesseparatetheforecastsmadebeforeandafterRegFDandtheforecastsmadebeforeandaftertheGlobalSettlement.Thesixbottomrowspresentforecastbiasforeachfiscalyearaveragedacrossallforecastmonths,alongwiththerealizedearningspershare,averageforecasts,annualstockreturns,andrealGDPgrowthrates.15Toalignfiscalyearenddateswithannualvariables,suchasrealGDPgrowthrates,weusedonlyforecastsforcompanieswithDecemberfiscalyearends.ForeachyearbeforetheGlobalSettlement,themedianforecasterrorsaresignificantlypositive.Furthermore,foreachyearbeforetheGlobalSettlement,weobservethewalkdowntrendwithforecastbiassteadilyd