inEquityOwnershipandChangesintheMarketValueoftheFirm by JohnJ.McConnellPurdueUniversity HenriServaesLondonBusinessSchoolandCEPR andKarlV.LinsUniversityofUtah November14,2003 ______________________________________ WethankseminarparticipantsatBI(NorwegianSchoolofManagement),LondonBusinessSchool,andStockholmSchoolofEconomicsmentsandsuggestions,andSergeySanzharandPedroSaffiforexcellentresearchassistance.PartofthisworkpletedwhilethesecondauthorwasBertilDanielssonVisitingScholarattheStockholmSchoolofEconomics. ChangesinEquityOwnershipandChangesintheMarketValueoftheFirm Abstract Westudythestockpriceresponsetoannouncementsofsharepurchasesbycorporateinsidersovertheperiod1994through1999.Thecross-sectionalvariabilityintheresponseisconsistentwithacurvilinearrelationbetweenfirmvalueandinsiderownership,wherethevalueofthefirmfirstincreases,thendecreasesasinsiderownershipincreases.Theseresultsareconsistentwithacausalinterpretationoftherelationshipbetweeninsiderownershipandfirmvalue.Theresultsoffurthertestsareinconsistentwithaninterpretationthatthefirmsinoursamplearemovingtowardanewequilibriumownershiplevelorthatinsidersarepurchasingsharestosignalthatthefirmisundervalued. ChangesinEquityOwnershipandChangesintheMarketValueoftheFirm
1.IntroductionMorck,Shleifer,andVishny(MSV)(1988),McConnellandServaes(1990,1995), HermalinandWeisbach(1991),Holderness,KrosznerandSheehan(1999)andothersdocumentastatisticallysignificantcross-sectionalcorrelationbetweenshareownershipbycorporateinsiders(usuallydefinedasmanagersandmembersoftheboard)andcorporateperformance,whereperformanceismeasuredeitherasTobin’sQorreturnonassets.Thisobservedempiricalrelationshiphasoftenbeeninterpretedtomeanthatownership“matters”andthatachangeinshareownershipbyinsiderscanbeusedtochangecorporatevalue.Suchinterpretationshavebeencriticizedforignoringthepotentialendogeneitythatmayarisewhenexternalpressurespushfirmstowardoptimalownershipstructuresthatjointlyoptimizeoverownershipandvalue. SuchcriticismshavetheiroriginsinDemsetz(1983)whoarguesthattheobservedlevelofshareownershipbyinsidersandfirmperformanceistheeofmarketforcessuchthateachfirm’sownershipstructurewillbeoptimalforthatfirm.Ifso,changesinownershipcannotbeusedtoenhancecorporatevalue.Hefurtherarguesthatanyobservedcross-sectionalempiricalrelationshipbetweenthelevelofinsidershareownershipandfirmperformancemustbespurious.StudiesbyDemsetzandLehn(1985),AgrawalandKnoeber(1996),LodererandMartin(1997),Cho(1998),DemsetzandVillalonga(2001),Himmelberg,Hubbard,andPalia(1999),andColes,Lemmon,andMeschke(2003)supportDemsetz’criticismempirically. Inthisstudy,weexaminedirectlytherelationshipbetweenchangesinshareownershipbyinsidersandchangesinthevalueofthefirm.Inparticular,weexaminechangesinfirmvaluearoundannouncementsthatinsidershavepurchasedshares.Weusethecurvilinearrelationship
2 reportedbyMcConnellandServaes(1990)asthestartingpointforouranalysis.McConnelland Servaesestimateaquadraticrelationshipas:Tobin’sQ=a+b1(Insiderownership)+b2(Insiderownership)
2 +c1(Otherownership)+c2(Controlvariables).
(1) whereTobin’sQisthemarketvalueofthefirmdividedbythereplacementvalueofassets, insiderownershipisthefractionofsharescontrolledbyofficersanddirectors,andother ownershipisownershipbylargeblockholdersandinstitutionalinvestors.Theyreportthatb1is positiveandsignificantandthatb2isnegativeandsignificantincross-sectionalregressions. Ifweepttherelationshipin(1)asbeingliterallytrue,achangeinownershipby insidersgives: Tobin’sQ+∆Q=b1(Insiderownership+∆Insiderownership)+b2(Insiderownership+∆Insiderownership)2+c1(Otherownership) +c2(Controlvariables).
(2) Subtracting(1)from(2)gives:∆Q=b1(∆Insiderownership)+b2(∆Insiderownership)
2 +2b2(∆InsiderownershipxInsiderownership)
(3) whereinsiderownershipismeasuredbeforethechange.Thepredictedsignofb1ispositive;the predictedsignofb2isnegative;andthepredictedsignofthecoefficientofthethirdtermis negativeandtwicethemagnitudeofthecoefficientofthesecondterm.Henceforth,wereferto thecoefficientofthethirdtermasb3.Thefirsttwotermsofequation(3)saythatanincreasein shareownershipbyinsidersgivesrisetoanincreaseinfirmvalueuptoapointafterwhichvalue declineswithfurtherincreasesininsidershareownership.Thethirdtermsaysthattheincrease infirmvalueassociatedwithanincreaseininsiderownershipdependsupontheinitiallevelof 3 insiderownership.Notethattheincreaseinvalueduetoanincreaseinshareownershipbyinsidersesprogressivelysmallerthehighertheinitiallevelofinsiderownership.Thatis,themarketvalueeffectassociatedwithagivensharepurchasebyinsidersissmallerthehighertheinitiallevelofinsiderownership,andcanbenegativeforsufficientlyhighinitiallevelsofinsiderownership. Wetestthepredictionsof(3)usingstockpricechangesaroundannouncementsofsharepurchasesbymanagersandmembersoftheboardasaproxyforchangesinfirmvalue(i.e.,asaproxyforchangesinQ).WeperformthesetestswithinsidersharepurchasestakenfromThomsonFinancialfortheperiod1994through1999.Whenannouncementperiodexcessstockreturnsareregressedagainstthechangeinthefractionofsharesownedbyinsiders,thechangeinthefractionofsharesownedbyinsiderssquared,andthecrossproductofthechangeinandtheinitiallevelofinsiderownership,theresultsarelargely,butpletely,consistentwiththeliteralinterpretationof
(3).Wefindthatb1ispositiveandsignificantlydifferentfromzeroandthatb2andb3arenegativeandsignificantlydifferentfromzero.Inmostcases,wecannotrejectthehypothesisthatthecoefficientoftheinteractionterm(i.e.,b3)istwicethemagnitudeofb2atthe10%levelofsignificance.However,inmostregressions,thepointestimateofb3isactuallysmallerthanb2.Forexample,inourbasecaseregression,b2=–1.58andthecoefficientoftheinteractiontermis–1.10. Anaggressiveinterpretationofourresultsisthatownershipmattersexactlyasspecifiedbyequations(1)and(3)andthatanincreaseinshareownershipcaneitherincreaseordecreasevaluedependinguponthefirm’sinitialownershipstructure.Alessaggressiveinterpretationisthatequations(1)and(3)providearoughapproximationoftherelationshipbetweeninsiderownershipandfirmvalue,but,nevertheless,anincreaseininsiderownershipcanleadtoan 4 increaseinfirmvalueuptoapointafterwhichfurtherincreasesininsiderownershipreducevalue.Aweakinterpretationisthattherelationshipbetweenchangesinownershipandchangesinvalueistheendogenouseofmarketforcespeleachfirmtowardsanoptimalownershipstructureandthechangesthatweobservearesimplyaconsequenceofthatprocess.Weareinclinedtowardthelessaggressiveinterpretation,butwementonotherinterpretationsandpresentteststhatattempttodistinguishamongthem. Theremainderofthispaperanizedasfollows.Thenextsectiongivesanoverviewofrelatedliteratureandmotivatesourempiricalanalysis.Section3describesourdata.Section4presentsourbasicempiricalresults.Section5presentsvarioustestsofrobustnessofourbasicresults.Section6considersthequestionofwhethertheinsidersharepurchasesinoursampleshouldbeviewedasmovingfirmstowardtheiroptimalequilibriumlevelsofinsidershareownership.Section7addressesthequestionofwhethertheresultsshouldbeinterpretedtomeanthatmarketparticipantsmerelyviewsharepurchasesbyinsidersasasignalthatthefirmisundervalued.Section8summarizesourfindingsandsetsforthourconclusions.
2.BackgroundTheoreticianshavemadepellingcaseforthepropositionthatownershipofshares bycorporatedecisionmakers(i.e.,managersandmembersoftheboardofdirectors)canhaveanimportantinfluenceonthewayinwhichthefirmismanagedand,therefore,onthefirm’sobservedmarketvalue(Baumol(1959),JensenandMeckling(1976),Demsetz(1983),Stulz(1988)).Whilesupportingthistheory,Demsetz(1983)hasarguedthatnorelationshipbetweenownershipstructureandvaluewillbeobservedempiricallyor,ifoneis,itisspurious.Thatis,ifeachfirmhasoptimizeditsownershipstructure,inacross-sectionoffirms,anycorrelation 5 betweenmarketvalueandownershipstructuremustbeidental.And,thus,theempiricalrelationshipshouldnotbeinterpretedasanindicationofacausalrelationshipbetweenownershipstructureandfirmvalue.Forexample,inacross-sectionoffirms,iffirmswithhighownershipbyinsidershappentohavehighmarketvaluesandviceversa,thatshouldnotbeinterpretedtomeanthatlowvaluefirmscanincreasetheirmarketvaluesbyrequiringthatinsidersownmoreshares.Indeed,ifeachfirmisatitsoptimum,anychangeinownershipshouldresultinadeclineincorporatevalue. Empirically,DemsetzandLehn(1985)useasampleof511paniesforwhichtheyhaveownershipdatafor1980toexaminetherelationshipbetweencorporateprofitabilityandthefractionofoutstandingsharesheldbyfive(20)shareholders.Theyfindnocorrelationbetweenprofitabilityandownershipconcentration.Theyconcludethattheirresultsareconsistentwiththeargumentthatownershipstructureisendogenous.LodererandMartin(1997)considertheroleofinsideownershiponthevaluecreatedincorporateacquisitions.Thoughtheyfindastatisticallysignificantcorrelationbetweenvalueandownership,theexplanatorypoweroftheirregressionissmallandthestatisticalsignificancedisappearsaltogetherinasimultaneousequationsspecification.SimilartoLodererandMartin(1997),AgrawalandKnoeber(1996),Cho(1997),andDemsetzandVillalonga(2001)eachuseasystemofequationstoexaminethedeterminantsofinsiderownershipandfirmvaluesimultaneouslyforvarioussamplesofU.S.firms.Thelattertwostudiesconcludethatcorporatevalueand/orprofitabilityaffectsthelevelofinsideownership,butnotviceversa.Thefirststudyreportsnoconnectionbetweenownershipandvalue. Himmelberg,Hubbard,andPalia(HHP)(1999)usefirmfixedeffectsinaregressionmodelinwhichfirmvalueisregressedagainstinsiderownership.They,too,reportno 6 significantrelationbetweeninsiderownershipandfirmvalue.(However,Zhou(2001)takesHHPtotask.Hepointsoutthatinsidershareownershipadjustsslowlythroughtimesuchthatevenifthereisarelationshipbetweenfirmvalueandinsidershareownership,afixedeffectsmodelwillobscuretherelationshipinacross-sectionalregression.)Coles,Lemmon,andMeschke(2003)constructastructuralmodeltostudythedeterminantsofownership.Theyarguethataspuriouscurvilinearrelationshipbetweenfirmvalueandinsiderownershipmayemergeincross-sectionalregressionseventhougheachfirmisatitsoptimalownershiplevelandtheypresentdatatosupporttheirargument.ThesestudiesallsupportDemsetz’contention. Ontheothersideofthecoin,MSV(1988)useasampleof371Fortune500firmstoestimateapiecewiselinearregressionwithTobin’sQasthedependentvariableandthefractionofsharesownedbymembersofpany’sboardofdirectorsastheindependentvariableofinterest.Theyfindasignificantcorrelationinwhichvalueinitiallyincreaseswithownershipofsharesbytheboardupto5%andthendeclinesuptoownershipof25%afterwhichtherelationshipagainespositive.Theyhypothesizethatthisnonlinearrelationshipisduetotheinteractionoftwooffsettingfactors--anincentiveeffectandanentrenchmenteffect.Theyallowforthepossibilitythattheobservedrelationshipiscausalwith“toomuch”ownershipcausingadeclineinvalue. Insimilarspirit,McConnellandServaes(1990,1995)estimatetherelationshipbetweenTobin’sQandthefractionofsharesheldbymanagersanddirectors(i.e.,corporateinsiders),large-blockshareholders,andinstitutionalinvestorsforlargesamplesofNYSEandAMEXfirmsfortheyears1976,1986,and1988.Amongotherthings,theyreportasignificantcurvilinearrelationshipinwhichthevalueofthefirmfirstincreasesandthendecreasesasinsiderownershipincreases.Theycautiouslyinterprettheirresultsasbeing“…consistentwith 7 thehypothesisthatcorporatevalueisafunctionofthestructureofequityownership”(McConnellandServaes(1990),p.595).HermalinandWeisbach(1991),Holderness,KrosznerandSheehan(1999),andHabibandLjungqvist(2003)alsoreportevidenceofasignificantrelationshipbetweeninsiderownershipandthevalueofthefirmforpubliclytradedpanies.AndersonandReeb(2003)takeaslightlydifferenttack.Theyconsider141“familyfirms”amongtheFortune500.SimilartoMcConnellandServaes,theyreportasignificantcurvilinearrelationshipbetweenfirmvalueandfamilyownership. Ontheinternationalfront,LaPorta,Lopez-de-Silanes,ShleiferandVishny(LLSV)(2002),Claessens,Djankov,FanandLang(CDFL)(2002),andLins(2003)examinethecorrelationbetweenvalueandownershipstructureacrossavarietyofcountries.Theyfocusonthedistinctionbetweenownershipofvoting/controlrightsandownershipofcashflowrights.LLSVandCDFLreportthatfirmvalueincreasesasmanagementownershipofcashflowrightsincreases;Linsreportsthatvaluedeclinesasmanagementownershipofcontrolrightsincreases. OurstudyislargelymotivatedbytheresultsinMcConnellandServaes(1990,1995)inthatweaskwhetherchangesinthefractionofsharesownedbyinsidersexhibitarelationshipwithchangesinthevalueofthefirmthatisconsistentwiththecurvilinearrelationshipdocumentedbythemfortherelationshipbetweenthelevelofinsiderownershipandtheleveloffirmvalue.McConnellandServaes,inturn,motivatetheirstudyonthebasisofStulz(1988),althoughtheiranalysiscouldalsohavebeenmotivatedbytheargumentsputforthbyMSV(1988). Inparticular,Stulz(1988)constructsamodelinwhichthetakeoverpremiumthatabiddermustpaytogaincontrolincreasesasthefractionofsharesownedbymanagementincreases,buttheprobabilitythatthetakeoverwilleeddeclines.Initially,anincreasein 8 ownershipbyinsidershastheeffectofincreasingthevalueofthefirm(i.e.,increasingtheexpectedvalueofthetakeoverpremium)butbecauseincreasingownershipreducesthelikelihoodofaessfultakeoveroffer,valueeventuallyreachesapeak,thendeclines.Thatis,thefirm’svaluereflectstheintersectionoftwoopposingforces.Similarly,MSV(1988)presentanargumentinwhichvalueisafunctionoftwoopposingforces--theincentiveeffectsofownershipandtheentrenchmenteffects.Theydonotspecifyacurvilinearrelationship,buttheirargumentallowsforthatpossibility.Aswedescribedattheoutset,weusestockreturnsfromaneventstudyaroundsharepurchasesbyinsiderstoinvestigatetherelationshipbetweenchangesinshareownershipandchangesincorporatevalueasspecifiedinequation
(3).
3.DataDataoninsiderpurchasesaretakenfromThomsonFinancialovertheperiod1994 through1999.IndividualsdefinedbytheSecuritiesandExchangeCommission(SEC)asinsidersarerequiredtoreportanypersonaltradesinthesharesoftheirfirmstotheSECbythe10thofthemonthfollowingthetradedate.Thisincludesopenmarketpurchasesandsales,sharesacquiredandsoldthroughtheexerciseofoptions,andavarietyofothertypesoftransactions. Fortheanalysesinthispaper,wefocusonasubsetofthereportedtrades.First,weincludeonlypurchasesofatleast10,000shares.Second,weconsideronlyopenmarketpurchases.Acquisitionsthroughtheexerciseofoptionsaremorelikelytobeanticipatedbythemarketbecauseinformationonoptionholdingsispubliclyavailable.Moreover,OfekandYermack(2000)documentthatwhenexecutivesexerciseoptionstoacquirestock,nearlyallofthesharesaresoldshortlythereafter.Thus,theincreaseininsiderownershipesabout 9 throughoptionexerciseisunlikelytobepermanent.Forthisreason,wealsoexcludesales,manyofwhicharelikelytoberelatedtooptionexerciseand,therefore,arelikelytohavebeenanticipatedbythemarket.Third,weexcludetradesbyindividualswhoareconsideredbytheSECtobeinsiders,butwhoarenotofficersormembersoftheboard.Forexample,ownersofmorethan10%ofthesharesofpanyaredeemedinsidersbytheSECeveniftheyarenotpartofthemanagementteamortheboard.Fourth,itissometimesthecasethatoneinsiderreportsapurchaseandanotherreportsasaleonthesameday.Thesedaysareremovedfromouranalysis;onlydayswhennoinsidersalesarereportedareconsideredinthisstudy.Ifmorethanoneinsiderfromthepanyreportsapurchaseonthesameday,wesumthosetradesandusethetotalastheinsiderpurchasesonthatday. DataonthelevelofshareownershipbyofficersandmembersoftheboardofdirectorsaretakenfromCompactDisclosure.CompactDisclosuregatherssuchdatafromannualcorporateproxystatements.Inaddition,withineachyear,weupdatethelevelofinsiderownershipaftereachinsidertrade.However,atthebeginningofeachyear,weresetthelevelofinsiderownershipusingCompactDisclosure. Todeterminethereliabilityofthisdatabase,weselected200firmsatrandomacrossallyearsandsoughttohandcollectownershipdatafromcorporateproxiesforthesepanies.Wewereabletocollectdatafor172ofthem.1WeparedtheinsiderownershipfromtheproxieswiththeinsiderownershipreportedbyCompactDisclosure.Inmostinstances,insiderownershipispreciselythesamefromthetwosources.Further,thecorrelationcoefficientbetweeninsiderownershipfromthetwosourcesis0.92,andthemeansandmediansoftheownershiplevelsarenearlyidentical.Intheteststhatfollow,forthe172firmsforwhichwe 1Wedonothaveesstoproxiesfor28ofpanies. 10 collecteddatafromproxies,weusetheproxydata.Forallothers,weusetheCompactDisclosuredata. Theannouncementperiodexcessreturn(APER)foreachpurchaseputedbysubtractingthereturnofthevalue-weightedCRSPIndexfromthereturnofpany’sstockforthedayonwhichtheinsiderreportedthetradetotheSECandthefivefollowingdays.ThesesixdailyexcessreturnsaresummedtogivetheAPERfortherelevantinsidersharepurchase.Weusethissix-dayintervalbecausethetradeinformationusuallydoesnotenterthepublicdomainforseveraldaysafteritisfiledwiththeSEC(LakonishokandLee(2001)).Toavoidproblemswithoutliers,weremoveanyAPERthatisnotwithinthreestandarddeviationsofthemeanAPER.Wealsodeletefirmswhosesharepricewasbelow$2atthetimeoftheannouncementandcaseswherethepriceatwhichtheinsiderboughtdifferedbymorethan20%fromtheclosingpriceonthedayofthetrade.Finally,wefindthatinabout11%ofthecases,insidersreporttradesaftertherequiredreportingdeadline.Suchtradesareretainedinthesampleiftheyarereportedwithin90daysofthereportingdeadline.Otherwise,theyareremovedfromthesample. Table1providessummarystatisticsofthedata.Oursampleincludes4,141differentpurchasesbyinsidersrepresenting1,700panies,orabout2.5tradespany.Themeanandmedianpurchaseare61,158and20,000shares,respectively,whichrepresent0.42%and0.15%ofpany’soutstandingstock(recallthatwedroppedalltradesoflessthan10,000shares).Thus,thetypicalpurchaseinoursampleisnotsmall.Indollarvalueterms,themeanandmedianpurchasesare$898,783and$213,750,respectively.Beforethepurchase,meanandmedianinsiderownershipwerenottrivialat19.88%and13.83%oftotaloutstandingshares.Thepurchasesarespreadreasonablyevenlythroughtimealthough1998hasamodest 11 bulgerelativetootheryears.ThefinalrowofpanelofTable1givesthemeanandmediansix-dayAPERsof0.94%and0.35%,respectively.Botharehighlystatisticallysignificantwithp-values<0.001.
4.Adirecttestoftherelationshipbetweeninsiderpurchasesandchangesinfirmvalue Weestimatethefollowingregressionmodel:APER=b1(∆Insiderownership)+b2(∆Insiderownership)
2 +b3(∆InsiderownershipxInsiderownership)+e
(4) whereAPERisaproxyforthechangeinQ.Wetestthehypothesisthatb1ispositive;b2is negative;andb3,thecoefficientoftheinteractionofthechangeininsiderownershipwiththe initiallevelofinsiderownership,isnegativeandtwicethesizeofb2.Thedependentvariableis thesix-dayAPERaroundannouncementsofinsiderpurchases.Theindependentvariablesare thenumberofsharespurchaseddividedbythenumberofsharesoutstanding(i.e.,theincreasein thefractionofsharesheldbyinsiders),theincreaseinthefractionofsharesownedbyinsiders squared,andthecrossproductoftheinitiallevelofinsiderownershipandtheincreaseinthe fractionofinsidershareownership.(Hereafter,werefertothefractionofsharesownedby managementandtheboardasinsiderownership.)Aswenoted,insidertradesmustbereportedbythe10thdayofthemonthfollowingthetrade.Notsurprisingly,manytradesformanypaniesarereportedonthe10thofeachmonth.Additionally,tradesbymorethanpanyaresometimesreportedonthesamedayevenwhenitisnotthe10th.Becausethe excessreturnsarelikelytobecorrelatedfortradesthatarereportedonthesameday,weinclude anindicatorvariableforeachreportingday.Thatis,weestimateamodelwithreporting-day fixedeffects.Thereare934reportingdaysinthesample. 12 Theresultsoftheregressionarepresentedincolumn1ofTable2.Asshowninthetable,b1ispositiveandsignificantlydifferentfromzero(p-value=0.04);b2isnegativeandsignificantlydifferentfromzero(p-value=0.07);andb3isnegativeandsignificantlydifferentfromzero(p-value=0.07).Additionally,anF-testindicatesthatb3isnotsignificantlydifferentfrom2b2atanytraditionallyeptablelevelofstatisticalsignificanceforrejectionofanullhypothesis(p-value=0.26).Theseresultsareconsistentwithaliteralinterpretationofequations(1)and
(3).Inparticular,anincreaseinshareownershipbyinsidersmaygiverisetoeitheranincreaseordecreaseinthevalueofthefirmdependingupontheinitiallevelofinsiderownership--atalowinitiallevelofinsiderownership,thevalueofthefirmincreases;atahighinitiallevelofinsideownership,thevalueofthefirmfalls.Inmanyrespects,theseresultsarepowerfulevidenceinsupportofthehypothesisthatchangesinshareownershipbyinsiderscananddoincreasecorporatevalueuptoapointafterwhich“toomuch”insiderownershipcananddoesreducevalue.Theevidenceispowerfulbecauseannouncementperiodreturnsarenotoriouslynoisysuchthatdiscoveryofanystatisticallysignificantrelationshipbetweenannouncementperiodexcessreturnsandapre-specifiedindependentvariablecanbecauseforcelebration.Holdingasidethatobservation,however,theresultsarenotperfect.Theflyintheointmentisparativemagnitudesofthepointestimatesofb2andb3.Asshowninthetable,thepointestimateofb3isactuallylessthanthepointestimateofb2.Obviously,eitheralowerestimatedvalueforb2orahigherestimatedvalueforb3orbinationofthetwowouldmakeforevenpellingevidence. AvisceralindicationoftherelationshipimpliedbytheestimatedcoefficientscanbeseeninFigure1whichshowstheeffectonfirmvalueof1%and5%increasesininsiderownershipgivendifferentlevelsofinitialownership.Theresultsareofthereporting-day 13 fixedeffects(i.e.,ofthereporting-dayindicatorvariables).Thegraphillustratesthatthestockpricereactiontoapurchaseofadditionalsharesbyinsidersispositiveatlowlevelsofinsiderownership,butdeclinesasinitialinsideownershipincreasesandisnegativeathighinitiallevelsofinsiderownership.Theinflectionpointoftheimpliedrelationshipbetweenfirmvalueandinsiderownershipisaround50%insiderownership.
5.OtherSpecificationsAsisthecasewithmostempiricalstudies,implementationofourtestsrequiredthatwe makedecisionsregardingthedataandsampleasweprogressed.Inthissection,weexaminethesensitivityofourresultstoalternativechoices.5.1.Trimmingthedistributionofannouncementperiodexcessreturns Inourinitialregression,wetrimmedthedistributionofsix-dayAPERsatthreestandarddeviationsfromthemean.Wedidsobecausethefulldistributionincludesseveralextremeobservations--thehighestAPERis+95.93%andthelowestis–71.24%.InthesecondcolumnofTable2,weestimatetheregressionusingthefulldistributionofAPERs.Incertainrespects,theseresultsarestrongerthanthoseinthefirstcolumn.Inparticular,eachoftheestimatedcoefficientshasthepredictedsignsothatthebasicstructureofthemodelisthesameasinthebasecase,however,thep-valuesoftheestimatedcoefficientsaresmaller.Thatis,b1ispositiveandsignificantlydifferentfromzero(p-value<0.001);b2isnegativeandsignificantlydifferentfromzero(p-value=0.01);andb3isnegativeandsignificantlydifferentfromzero(p-value=0.01).But,withb2=–2.75andb3=–1.91,itisnowpossibletorejectthehypothesisthatb3=2b2atthe10%levelofsignificance(p-value=0.10).Thus,althoughthesignsofthecoefficients 14 areconsistentwiththebasic“story”behindequation
(1),b2is‘toobig’inabsolutetermsand/orb3is‘toosmall’tosupportaliteralinterpretationof
(1). Inourthirdregression,wegointheoppositedirection.Herewetrimthedistributionfurther.Specifically,weremoveAPERsiftheylieoutsideoftwostandarddeviationsofthemeanAPER.Inthiscase,theresultsareverysimilartothoseinthebasecaseregression.AsshowninthethirdregressionofTable2,b1ispositiveandsignificantlydifferentfromzero(pvalue=0.02);b2isnegativeandsignificantlydifferentfromzero(p-value=0.10);andb3isnegativeandsignificantlydifferentfromzero(p-value=0.03).Additionally,b2,withavalueof–1.21,isnotsignificantlydifferentfrom2b3,withavalueof–1.14(p-valueforthedifference=0.43).Still,b3isstubbornlylessthanb2.Thus,althoughtheresultscontinuetobeconsistentwiththetheorybehind
(1),theydonotperfectlysupportallofitsnuances.5.2Otherannouncementperiodintervals Inourbasecaseregression,weusethesix-dayintervalincludingtheannouncementdayandthefollowingfivedaystocalculateAPERs.WeexperimentedwithAPERscalculatedoverotherintervalsbetween4and10daysaroundannouncementsofinsidersharepurchases.Becausetheresultsareessentiallythesameasthoseincolumn1ofTable2,wedonotreporttheminatable.5.3.Otherreportinglags Inourbasecaseregression,weonlyincludedtradesiftheywerereportedwithin90daysofthereportingdeadline.Or,toputitslightlydifferently,weallowedpurchasestobeincludedeveniftheywerereportedwithalagofupto90days.Weexperimentedwithallowingtradestobeincludedwithlongerandshorterlagsbetweentherequiredreportingdateandtheactual 15 reportingdate.Regardlessofthecut-offfortheallowablereportinglag,theunreportedresultsoftheseregressionsareessentiallythesameasthoseincolumn1ofTable2.5.4.Trimmingthedistributionofshareownership Inourvariousregressions,thethirdexplanatoryvariableisthecrossproductoftheinitiallevelofinsidershareownershipandthechangeininsidershareownershipfromthatinitiallevel.Weattachsubstantialimportancetothemagnitudeofthecoefficientofthisvariablewhichdepends,perhapscritically,onthepre-purchaselevelofinsidershareownership.Toassurethatourresultsdoeaboutbecauseofpurchasesbyjustafewlarge-blockshareholders,weremoveobservationsinwhichtheinitiallevelofinsiderownershipismorethanthreestandarddeviationsabovethemean.Thistrimsthedistributionofpre-purchaseinsiderownershipat75.38%anddrops54observationsfromthesample.AsshowninthefourthcolumnofTable2,withthisspecification,thecoefficientestimatesarenearlyidenticaltothoseinthebasecaseregressionofcolumn1. Althoughwehavelessconcernaboutthebottomtailofthedistributioncausingapeculiare,weremovefirmsprisethebottom1%ofthedistributionofinsiderownershipalongwithfirmswithownershipmorethanthreestandarddeviationsabovethemean.2Wethenre-estimatetheregression.TheseresultsaregiveninthefifthcolumnofTable2.Again,theresultsarenearlyidenticaltothoseinourbasecaseregression.5.5.Forcingb3toequal2b2 Althoughourresultsareconsistentwithequation
(1),b2is‘toobig’and/orb3is‘toosmall’forthefittobeconsideredperfectineveryway.Onewaytodeterminewhetherthedataactuallyfitacurvilinearrelationshipistoestimateaconstrainedregressioninwhichb3isforced 2Wecannotcutoffownershipatthreestandarddeviationsbelowthemeanbecausethreestandarddeviationsislessthanzero. 16 tobeequalto2b2.Ifthisforcedrelationshipdoesnotfitthedata,theregressioncoefficientswill,then,notbestatisticallysignificant.AsshowninthesixthcolumnofTable2,suchisnotthecase.Eachofcoefficientsissignificantandeachhasthepredictedsign:b1is0.38(p-value=0.08);b2is–0.49(p-value=0.04);and,ofcourse,at–0.98(p-value=0.04),b3ispreciselytwiceb2. Thisregressionalsopaysabonus.Withthisspecification,whenwecalculatetheinflectionpointimpliedforequation
(1),itursat40%insiderownership--alevelveryclosetotheinflectionpointsestimatedbyMcConnellandServaes(1990,1995).Onepossibleinterpretationisthatwhenthechangeinownershipdataareforcedintoconformitywiththequadraticrelationshipin
(1),theresultsareverymuchinlinewithfindingsbasedonlevelsofinsidershareownership.5.6.Piecewiselinearspecification OurfindingssupportthegeneralpredictionbyMSV(1988)ofanonlinearrelationbetweenfirmvalueandinsiderownership.However,MSValsoreportaspecificpiecewiselinearrelationbetweenownershipandvalue.pleteness,weattempttoreplicatetheirfindingsusingourchangesinshareownershipdata.Wedefineexplanatoryvariablesthatallowustoestimateapiecewiselinearrelationbyidentifyingwhetherthechangeinownershipursforafirmwithpre-purchaseinsiderownershipintherangeof0%to5%(930observations),above5%andbelow25%(1973observations),andabove25%(1238observations).WethenestimatearegressioninwhichtheAPERisthedependentvariableandthethreechangeininsiderownershipvariablesofequation(3)plusthereportingdayindicatorvariablesaretheindependentvariables.Thecoefficientofthechangeinownershipforthe0%to5%ownershiprangeispositive,0.50,withap-value=0.59;thecoefficientofthechangeinownershipforthe 17 5%to25%ownershiprangeisalsopositive,0.50,withap-value=of0.05;andthecoefficientofthechangeinownershipfortheabove25%ownershiprangeisnegative,–0.17,withap-valueof0.17.Thepositivecoefficientsonthechangesinownershipupto25%andthenegativecoefficientonchangeinownershipabove25%ownershipareconsistentwiththecurvilinearrelationshipofMcConnellandServaes(1990).
6.Optimalinsiderownership6.1.Overview Aswenotedattheoutset,amajorconcernregardingthecross-sectionalregressionsofMSV(1988),McConnellandServaes(1990,1995)andothersderivesfromDemsetz’(1983)argumentthatownershipisdeterminedendogenouslyandallfirmsareattheiroptimalownershipstructuresallthetime.Ifso,norelationshipshouldbeobservedinacrosssectionalregressionofperformanceagainstownershipand,ifoneis,itmustbespurious.Thisobservationleavesunansweredthequestionofwhyinsiderstradeatalland,iftheydotrade,whysuchtradesareassociatedwithanincreaseinstockprices(atleastoverlowrangesofinitialinsiderownership).Thatis,ifallfirmsareattheiroptimalownershipstructures,arguablyanytradingshouldbeassociatedwithnegativestockpricereactions.Wedo,ofcourse,observesomenegativestockpricechangeswheninsidersbuymoreshares.Butthosenegativestockpricechangestendtobeconcentratedoveraspecificrangeofshareownership.Thatis,negativestockpricereactionsursystematicallywheninitialinsideownershipisalreadyhigh.Theseresultsaredifficulttoreconcilewiththepropositionthatallfirmsareattheiroptimalownershipstructuresallthetime. AvariationoftheDemsetzargumentisproposedbyCoreandLarcker(2002).Inparticular,theyproposethatownershipstructurestendtobeattheiroptimallevelsmostofthe 18 time,but,becausetherearecostsofadjustment,ownershipstructuressometimesdriftawayfromtheiroptima.Whenthisurs,andwhenthecost/benefitcalculusmakesitworthwhiletodoso,executiveseithervoluntarilypurchasemoresharesortheyareforcedtodosobypany’sboardofdirectors.Insupportofthisproposition,CoreandLarcker(2002)study195firmsthatadopted“targetownershipplans”overtheperiod1992through1997.Theyreportthatthesefirmshadlowmanagerialequityownershiprelativetopeersandpoorperformancerelativetopeerspriortoplanadoption.Aftertheplanadoptions,managersincreasedtheirshareownership,andthefirms’ountingandstockpriceperformanceimprovedaswell. AsthesizeoftheCoreandLarckersampleindicates,adoptionsofformaltargetownershipplansarerelativelyrare.Thatdoesnotmean,however,thatthesearetheonlyinstancesinwhichboardshaveexertedpressureonmanagementtopurchaseadditionalshares.Indeed,itispossiblethatmanyofthecasesofinsidersharepurchasesthatwestudyrepresentinstancesinwhichtheboardhasinformally-levelmanagersintobuyingmorestock.Itispossiblethattheboardrecognizedthatinsiderownershipwas‘toolow’andurgedofficerstopurchasemoreshares.Thus,thepurchasesbyinsidersthatweobservemaysimplybetradespushinginsiderownershipbacktowarditsoptimallevel. Inthissection,weaddressthequestionofwhethersharepurchasesbyinsidersshouldbeviewedasmovingfirmstowardtheiroptimalinsiderownershiplevels.Weundertakethisinvestigationfromseveralperspectives.First,weusethemodelproposedbyHHP(1999)todeterminewhetherthefirmsinoursamplecanbeviewedasexperiencinginsidersharepurchasesbecausetheirpre-purchaselevelsofinsidershareownershipwere“toolow.”WeusetheHHPmodelforthispurposebecauseofitsrelativelyrecentvintageandbecauseitiswidelyrecognized.Inparticular,weusetheHHPmodeltoestimateeachfirm’soptimalequilibrium 19 levelofinsiderownership.Wethenaskwhetherthefirmsinoursamplehaveaninsiderownership“deficit”relativetotheiroptimalequilibriumlevelssuchthattheinsidersharepurchasesthatweobservecanbeviewedasmovingthesefirmstowardstheiroptimalinsiderownershiplevels.Second,weaskwhetherannouncementperiodexcessstockreturnsarecorrelatedwiththeequilibriumownershipdeficitasdeterminedbytheHHPmodel.Thatis,doesthemarketrespondasifthepurchasesweremovingthefirmstowardstheiroptimalinsiderownershiplevels.Third,westepawayfromaspecificequilibriummodelofoptimalinsidershareownershipandaskwhetherinsiderpurchases,regardlessofwhethertheycanbeviewedaspurchasesthatmovethefirmtowardaspecifiedoptimalownershiplevel,arecorrelatedwithannouncementperiodstockpricechanges,ignoringthesquaredtermandthecross-productofthetwotermsinequation
(3).6.2.Anempiricalmodeloftheoptimallevelofinsiderownership Toconstructtheirmodel,HHPgathershareownershipdatafromproxystatementsforarandomsampleof600Compustatfirmsasof1982.Thesefirmsaretrackedthrough1992withownershipdatabeingcollectedeachyearfromproxystatements.Regressionmodelsof(atransformationof)managementshareownershipasafunctionofcertainfirmcharacteristicsareestimated.Thetransformationofmanagementshareownershipis: transformedinsideshareownership=log(insideownership/(1-insideownership)).
(4)Theexplanatoryvariablesintheirmodelsarelistedinourappendix.HHPestimatetheirmodelusingOLSregression.Theyestimatethemodelwithandwithout3-digitSICcodeindustryfixedeffectsandwithandwithoutfirmfixedeffects. UsingallfirmsavailableonbothCompustatandtheCompactDisclosuredatabasesovertheperiod1993through1999,wereplicatetheHHPprocedure.Weemploytheresulting 20 regressionmodelputetheoptimallevelofinsideownershipforthefirmsinoursamplethatexperienceinsiderpurchases.Theresultsofthisregressionallowustodeterminewhethertheinsiderpurchasesmovefirmstowardsorawayfromtheiroptimalinsideownershiplevelsordingtothismodelofoptimalownership.Injudgingwhetherinsiderpurchasesmovethefirmstowardorawayfromtheiroptimallevelsofinsiderownership,weemploytwoestimationsofthemodel--onewithoutandonewithindustryfixedeffects.WereportthecoefficientsoftheestimatedmodelsintheAppendix.3 Forbothmodels,parepre-purchaseinsiderownershipwiththe“optimal”levelofownershipintheyearbeforethepurchaseandtheyearofthepurchase.Thereareparisons:(1)theactualyear-before-purchaselevelofinsiderownershipv.thepredictedlevelestimatedwithoutindustryfixedeffects;(2)theactualyear-before-purchaselevelofinsiderownershipv.thepredictedlevelestimatedwithindustryfixedeffects;(3)theactualyear-ofpurchaselevelofinsiderownershipv.thepredictedlevelestimatedwithoutindustryfixedeffects;and(4)theactualyear-of-purchaselevelofinsiderownershipv.thepredictedlevelestimatedwithindustryfixedeffects.Thesearecalculatedforeachpurchaseandfortheaveragepurchaseperyearforeachfirm.Weconsiderthemeanandmedianvaluesofthedifferenceswhentheactualpre-purchasefractionofinsidershareownershipissubtractedfromtheoptimal(orpredicted)fractionofinsidershareownership.Werefertothisdifferenceastheinsidershareownershipdeficit.TheresultsarepresentedinPanelAofTable3. Regardlessofwhichregressionspecificationweconsider,themeandifferencebetweenthepredicted(i.e.,optimal)levelofinsiderownershipandtheactualpre-purchaselevelof 3Wedonotuseafirmfixedeffectsmodelforthisanalysisbecausewithafixedeffectsmodelestimatedoverthesampleperiod,thefirmfixedeffectcapturesthedegreetowhichthefirm’sownershipstructurediffersfromotherfirms’ownershipwiththesamecharacteristics,onaverage.Becausethepost-tradeownershiplevelisactuallyusedtoestimatethefixedeffectmodel,itwouldnotbesurprisingtofindthatfirmshave“toolittle”ownershipbeforeinsiderpurchases. 21 insiderownershipisnegativeandstatisticallysignificantlydifferentfromzero(allp-values<0.001).Thus,onaverage,insidershave“toomuch”ownershipinthesefirmspriortopurchases,withtheexcessbeingbetween2.02%and3.43%percentdependinguponwhichregressionspecificationisconsidered.Theseresultsareinconsistentwithanargumentthattheinsidersharepurchasesaremovingthesefirmstowardtheiroptimalinsiderownershiplevels. Whenweconsiderthemediansofthedifferencebetweentheactuallevelsofinsiderownershipandthepredictedlevels,aslightlydifferentpictureemerges.Themedianownershipdeficitispositive(i.e.,insidershave‘toolittle’ownership),albeitthedifferenceisfrequentlynotstatisticallysignificantlydifferentfromzero.Ifthemedianfirmisconsideredtobea“typical”firm,anargumentmightbemadethatthetypicalfirmisclosetoitsoptimalinsideownershiplevel.Ifso,thatstilldoesnotoffersupportforanargumentthatinsiderpurchasesaremovingthesefirmstowardtheiroptimalownershiplevels.Thatis,becausethetypicalfirmdidnothaveasignificantinsiderownershipdeficittobeginwith,therewaslittleneedforinsiderstopurchaseadditionalshares.Thus,themediansofthedifferencesbetweenthepre-purchaselevelofinsidershareownershipandtheoptimallevelofinsidershareownershipdonotindicatethatthesefirmswouldhavebeenexpectedtoexperienceadditionalsharepurchasesbyinsiders.Assuch,theydonotsupporttheargumentthatthesefirmshadtoolittleinsidershareownership. Onbalance,theresultsinPanelAofTable3indicatethatthepurchasesdonotappeartobeaimedatrestoringinsiderownershiptoitsoptimallevel,butitcouldbethatconsiderationofonlythemeansandmediansofthedistributionsismaskingacorrelationbetweenthesizeofthedeficitandthesizeofthepurchase.Thatis,itcouldbethatfirmswiththelargestownershipdeficitexperiencethelargestinsidersharepurchases.Toexaminethatpossibility,weestimateasimpleregressionwiththechangeinthefractionofsharesownedbycorporateinsidersasthe 22 dependentvariableandthepre-purchasefractionofsharesownedbyinsiderssubtractedfromthepredictedfractionofinsiderownershipastheexplanatoryvariable.Thatis,weregressthechangeinownershiponthedeficit.WeusethesamefourmeasuresoftheownershipdeficitasinPanelA.PanelBofTable3containstheresults. Thefindingsaretheoppositeofwhatwouldbeexpectedifthepurchasesweremeanttomovefirmstowardtheiroptimallevelsofinsiderownership.Therelationbetweentheamountofthepurchaseandtheownershipdeficitisnegative.Thisimpliesthatinsidersinfirmswith“toomuch”ownershipactuallybuymoresharesthaninsidersinfirmswith“toolittle”insiderownership.6.3.Thechangeinthevalueofthefirmandtheinsiderownershipdeficit WenowconsiderwhethertheAPERsaroundtheannouncementsofinsiderpurchasesareconsistentwiththeinterpretationthatinsidersharepurchasesaremovingfirmstowardtheiroptimalownershipstructures.Forthisanalysis,weregressAPERsagainstthreeindependentvariables:(1)thechangeininsideownership,(2)theinsiderownershipdeficit,and(3)theinteractionbetweenthechangeininsiderownershipandtheinsiderownershipdeficit.Weincludetheownershipdeficitintheregressiontoexaminewhetherthechangeinthevalueofthefirmdependsonwhetherthepurchasesurinfirmswith“toomuch”ratherthoseinthosewith“toolittle”insiderownership.Further,weinteractthesizeofthepurchasewiththesizeofthedeficit.Ourreasoningfordoingsogoesasfollows:wheninsidersbuysharestomakeupalargerdeficit,thestockpricereactionshouldbeespeciallylarge. Theoptimalownershiphypothesispredictspositivecoefficientsforeachofthevariables.TheresultsoftheregressionaredisplayedinPanelCofTable3.Allthecoefficientshavethepredictedsign,butonlythecoefficientofthedeficitismarginallysignificant,withap-valueof 23 0.09,andinonlytwoofthefourspecifications.Thisresultcouldbeconsideredweakevidenceinsupportoftheoptimalownershipexplanation.However,beforeembracingthatinterpretationtooenthusiasticallyanoteofcautionisappropriate. Whenafirmhastoolittleinsideownershipordingtotheoptimalownershipmodel(i.e.,thedeficitislarge),itisalsothecasethatsuchafirmislikelytohavealowlevelofprepurchaseactualownership.Indeed,thecorrelationbetweenthedeficitandpre-purchaseactualownershipis–0.83whenthedeficitputedwithoutindustryfixedeffectsand–0.87whenthedeficitputedwithindustryfixedeffects.Further,thecurvilinearrelationshipofequation(1)impliesthatsharepurchasesbyinsidershaveasmallerAPERwhenpre-purchaseinsiderownershipislarge.Thus,theapparentsignificanceofthepositivecoefficientsonthedeficitinPanelCofTable3mayeaboutbecausethedeficitsarecorrelatedwiththelevelofpre-purchaseinsiderownership.Toconsiderthatpossibility,weinsertthelevelofprepurchaseinsiderownershipintotheregressioninplaceofthedeficitandre-estimatetheregressions.Ineachofthefourregressions,thepre-purchaselevelofinsiderownershiphasanegativecoefficientandap-valuelessthan0.05.Wethenincludeboththepre-purchaselevelofinsiderownershipalongwiththedeficit.Inthreeofthefourregressions,thesignofthedeficitcoefficientisreversedandnoneofthecoefficientsofthedeficitaresignificant.Asregardsthelevelofpre-purchaseinsideownership,thesignsofthecoefficientscontinuetobenegativeandtwoofthefourhavep-valueslessthan0.05.Thus,thisanalysisfurtherweakensthealreadylimitedsupportforthehypothesisthatthestockpriceresponsethatpaniessharepurchasesbyinsidersrepresentsamovementtowardanoptimalownershipstructure. Whilesomeofthetestsinthissectionprovidesupportfortheargumentthatpurchasesbyinsiderscanbeviewedasmovingfirmstowardanoptimalownershipstructure,takenasawhole, 24 thesupportismodestandthecounterevidenceisplentiful:Theaverageinsiderpurchaseursinfirmsthatalreadyhave“toomuch”insiderownership,therelationshipbetweentheownershipdeficitandthemagnitudeofthesharepurchasesisnegative,andthestockpriceresponsearoundannouncementsofthepurchasesisnotcorrelatedwiththedeficit.6.4.Asimplemodelofinsidersharepurchasesandoptimalinsiderownership AnargumentmightbemadethatourfailuretofindasignificantrelationshipbetweentheAPERsandthechangeininsiderownershipursbecausewehaveusedaspecific,andincorrect,modelofequilibriumoptimalownership.Thus,ourfailureisnotduetothelackofacorrelation,butduetomodelmisspecification.Onewaytoaddresssuchaconcernistoabandonaspecificmodelandtomerelysearchforcorrelationbetweenthechangeinownershipandthestockpricechange. Assumingthatthepurchasesarenotanticipatedbymarketparticipants,allelseequal,largerpurchasesshouldhaveamoresubstantialstockpriceresponsethansmallerpurchasesbecause,presumably,theycorrectamoresubstantialdeviationfromtheoptimum.Toinvestigatewhetherlargerpurchasesdohavealargerimpact,weregressAPERsaroundannouncementsofinsidersharepurchasesagainsttheassociatedincreaseinthefractionofinsidershareownership.Wecontinuetoincludeindicatorvariablesforeverytradereportingday,butweexcludetheotherexplanatoryvariablesemployedpreviously. TheresultsofthissimpleregressionarereportedinPanelDofTable3.Thecoefficientontheincreaseininsiderownershipisnegative,albeitnotsignificantlydifferentfromzero(p- 25 value=0.71).Thisresultcontradictsthepropositionthatincreasesinshareownershipbyinsiderstypicallyrepresentmovementstowardanoptimallevelofshareownership.4
7.InsidersharepurchasesassignalsaboutthevalueofthefirmAfurtherpossibilityisthatpurchasesbytheinsidersaresimplyasignalthatthefirmis undervalued(Seyhun(1986))andthepositivestockpricereactiontotheannouncementismerelyareflectionofthissignal.Aswedemonstratedinsection6,wedonotfindapositiverelationbetweenthesizeofthepurchaseandthestockpriceresponse.Thus,weneedtodevelopamoreelaboratesignalingexplanationtoexplaintheobservedstockpricereaction. Onepossibilityisthatthemarginalinformationconveyedbythesignalisdecreasinginthesizeofthepurchasesuchthatthestockpricereactiondeclinesasthesizeofthepurchaseincreases.ThisexplanationisconsistentwiththeAPERbeingpositivelycorrelatedwiththesizeofthesharepurchaseasafractionoftotalshar