Institutional trading and Abel Noser data

25 Pages·2017·0.61 MB·English

by  Gang Hu

Journal of Corporate Finance 52 (2018) 143–167
ContentslistsavailableatScienceDirect
Journal of Corporate Finance
journal homepage: www.elsevier.com/locate/jcorpfin
Institutional trading and Abel Noser data
T
⁎
Gang Hu , Koren M. Jo, Yi Alex Wang, Jing Xie
HongKongPolytechnicUniversity,SchoolofAccountingandFinance,HungHom,Kowloon,HongKong
ARTICLE INFO ABSTRACT
Keywords: WesurveythegrowingacademicliteratureusingAbelNoserdata,including55publicationsthus
InstitutionalInvestor far. We analyze publication patterns to explore how the availability of a specialized micro-
Trading structuredatasetpropagatesacrossdifferentareaswithinfinanceandintootherdisciplinessuch
AbelNoser as accounting. Of note, we identify corporate finance and accounting as the most under-re-
ANcerno searchedareasthatofferpromisingopportunitiesforfutureacademicresearchusingthedata.To
Accounting
provideguidanceforresearchersinterestedinusingAbelNoserdata,weanalyzeinstitutional
CorporateFinance
trading using transaction-level data spanning more than 12years and covering 233 million
Investments
MarketMicrostructure transactions with $37 trillion traded. We provide background information on the origin and
historyofthedata,offersuggestionsforcleaningandusingthedata,anddiscuss(dis)advantages
JELclassification:
ofAbelNosercomparedtootherdatasourcesforinstitutionaltrading.Wealsodocumenttwo
G12
G14 simple facts: 1) institutional trade sizes decline dramatically over time, rendering trade size-
G23 basedinferencesofinstitutionaltradesproblematic;2)weestimatethatAbelNoserdatacover
M40 12%ofCRSPvolumeoveroursampleperiodand15%for1999–2005,significantlyhigherthan
the estimate in Puckett and Yan (2011) for 1999–2005: 8%, a widely quoted number in the
literature.Thisbackgroundshouldproveusefulforresearchersseekingtoaddressanumberofas
yetunexploredissues,especiallyincorporatefinanceandaccountingresearch.
1. Introduction
Institutionalinvestorsplayanincreasinglyimportantroleinallaspectsoffinancialmarkets.Inthispaper,wefocusonapro-
prietarydataset ofinstitutional trading transactions: AbelNoserdata (also knownasAbel/NoserorANcernodata). Academic re-
searchusingAbelNoserdatadatesbacktoBlume(1993).1Wesurveythegrowingacademicliterature,including47publishedpapers
andeightforthcomingpaperssofar,usingAbelNoserdatatoaddressvariousresearchquestionsinmarketmicrostructure,corporate
finance,investments,andaccounting.
Weanalyzepublicationpatternsandshowhowtheavailabilityofaspecializedmicrostructuredatasetpropagatesacrossdifferent
areas within finance and into other disciplines such as accounting. Based on our analysis of publication patterns, we discuss po-
tentiallyfruitfuldirectionsforfutureresearch,andidentifycorporatefinanceandaccountingastwounder-researchedareasthatoffer
themostpromisingopportunitiesforfutureresearchusingAbelNoserdata.WebelievethatthetrendoffinancestudiesusingAbel
Noserdatawillremainstrongintheforeseeablefuture,asevidencedbytheten2018orforthcomingpublications.Withinfinance,
corporatefinanceappearstobethemostunder-researchedareagiventhatitisalargefieldwithrichliteratureandvariousinteresting
⁎Correspondingauthor.
E-mailaddresses:[email protected](G.Hu),[email protected](K.M.Jo),[email protected](Y.A.Wang),
[email protected](J.Xie).
1Table5containsthelistofpublicationsusingAbelNoserdata.Anup-to-datelistofpublicationscanbefoundattheAbelNoser(ANcerno)Data
Page:http://ganghu.org/an
https://doi.org/10.1016/j.jcorpfin.2018.08.005
Received18December2017;Receivedinrevisedform8August2018;Accepted10August2018
Available online 16 August 2018
0929-1199/ © 2018 Elsevier B.V. All rights reserved.
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
corporate events and contexts. However, only 11 out of the 55 publications so far are in corporate finance, even less than 15 in
microstructure.Thismaymeanthatcorporatefinance,andsimilarlyaccounting,couldofferthebestpotentialforfutureresearch
usingAbelNoserdata.
Moreover,AbelNoserdataiswell-suitedforstudyingcorporatefinancetopics.OneofthemainadvantagesofAbelNoserdatais
itshighfrequency.ThisfeaturemakesAbelNoseranidealsourceforin-depthexaminationsofinstitutionalinvestors' tradingbe-
havior around various corporate events. An example of this line of academic inquiries is Bethel et al. (2009), one of the earliest
publicationsusingAbelNoserdata,anditispublishedintheJournalofCorporateFinance.Theauthorsexamineinstitutionaltrading
aroundmergersandshareholdervotingoutcomes,andfindevidenceforanactivemarketforvotingrightsaroundmergerevents.Ina
recentworkingpaper,LiandSchwartz-Ziv(2018)examinehowshareholdervotesandtradesarerelatedinabroadercontextalso
usingAbelNoserdata.Bothstudieshighlighttheadvantageofusingdetailedinstitutionaltradingrecordstogainfurtherinsightsinto
importantcorporatefinanceissuesraisedinpriorstudies,suchasGillanandStarks(2000),andBethelandGillan(2002).
Several other studies also highlight the fact that Abel Noser data are uniquely well-suited to tackle various corporate finance
researchtopics,comparedtoalternativedatasourcessuchasSEC13Ffilings,whichareonlyquarterly.Chemmanuretal.(2009)
explicitlyidentifySEO(seasonedequityoffering)allocationsusingAbelNoserinstitutionaltradingrecordsforthefirsttimeinthe
literature. Ahern and Sosyura (2015) study sensationalism in media coverage amid merger rumors. They find that institutional
investorsarenetsellersinthetargetfirmwithmergerumors,suggestingthatinstitutionsprovideliquiditytoindividualswhobuy
targets upon merger rumors. Henry and Koski (2017) examine whether skilled institutions indeed exploit positive abnormal ex-
dividendreturns,andfindthatinstitutionsconcentratetradingaroundcertainex-datesandarecapableofidentifyingex-dayevents
withhighertradingprofit.
ToprovideguidanceforresearchersinterestedinusingAbelNoserdata,wefirstprovidebackgroundinformationontheorigin
andhistoryofAbelNoserdata.Wedescribethedataandanswerimportantquestionssuchaswhyinstitutionalinvestorsprovidetheir
tradingdatatoAbelNoser,andwhyAbelNoserpr`ovidesinstitutionaltradingdatatoacademia.Wediscussvariousdataissuesand
offersuggestionsforcleaningandusingthedata.WethendiscussadvantagesanddisadvantagesofAbelNoserdatacomparedto
otherpotentialdatasourcesforinstitutionaltrading:Plexus,SEI,SEC13Ffilings,CRSP/Thomsonmutualfundholdings,TAQ,CAUD,
TORQ,NASDAQclearingrecordsdata,andNASDAQPostData.ThemaindisadvantageofAbelNoserdatabaseisthatitcoverstrades
fora subset of institutional investors. On the other hand, Abel Noser data offer clear advantages over the above alternative data
sourcesforinstitutionaltrading.Themainadvantagesincludehighfrequency,longandcontinuoustimeseries,detailedandaccurate
informationonspecifictransactionssuchasside(buyversussell),andtypeandidentitiesofsampleinstitutionsallowingforcross-
sectionalandinstitution-specificanalyses.
Wefurtheranalyzeinstitutionalinvestors'tradingbehaviorusingAbelNoserdata.Oursamplecontains232.6millioninstitutional
trading transactions, 1.26 trillion shares and $37.5 trillion traded from January 1999 to September 2011. Abel Noser provides
institutionaltradingdatafrom1997to2015.However,dataforthefirsttwoyearsareverysmallandafewvariableswereremoved
after September 2011, including a key institution identifier. Section 3.1 provides further discussions about sample start and end.
UsingAbelNoserdata,wedocumenttwosimplefacts.First,wedocumentthatinstitutionaltradesizesdeclinedramaticallyover
time,consistentwiththefindingsinChordiaetal.(2011).Thissignificantdeclinerenderstradesize-basedinferencesofinstitutional
tradesproblematic,supportingthefindingsinCreadyetal.(2014).
Second,weputanupperboundandfourprogressivelytighterlowerboundsonAbelNoserdata'scoverageofCRSPvolume.The
lowestlowerboundfollowsthesamemethodologyinPuckettandYan(2011),andweareabletoarriveatasimilarestimateasthat
inPuckettandYan(2011).Ourotherthreelowerboundsmodifythemethodologyandofferhigher(tighter)estimates.Usingournew
method,weestimatethatAbelNoserdatacover12.3%~12.6%ofCRSPvolumeovertheentiresampleperiodfromJanuary1999to
September2011,and14.9%~15.3%for1999–2005.Theseestimatesaresignificantlyhigherthanthe8%estimateinPuckettand
Yan(2011)for1999–2005,awidelyquotednumberintheacademicliterature(see,e.g.,Brownetal.(2014),Creadyetal.(2014),
LjungqvistandQian(2016),andHenryandKoski(2017)).
Overall,ourpapershouldproveusefulforresearchersseekingtoaddressanumberofunexploredissuesusingAbelNoserdata,
especiallyincorporatefinanceandaccountingresearch.Therestofthepaperisorganizedasfollows.Section2describesAbelNoser
dataandcomparesitwithvariousotherdatasourcesforinstitutionaltrading.Section3discussesdataissuesandpresentsempirical
results.Section3.1.11presentsananalysisofpublicationpatternsusingAbelNoserdata.Section4containsabriefsurveyofthe
growingacademicliteratureusingAbelNoserdata,including55publicationssofarandselectedworkingpapers.Section5concludes
withadiscussionofdirectionsforfutureresearch.
2. Datasourcesforinstitutionaltrading
Inthissection,wefirstdescribeAbelNoserinstitutionaltradingdata.Wethendiscusstheadvantagesanddisadvantagesofusing
AbelNoserdatatostudyinstitutionaltradingcomparedtootherpotentialdatasourcessuchasPlexus,SEI,SEC13Ffilings,CRSP/
Thomsonmutualfundholdings,TAQ,CAUD,TORQ,NASDAQclearingrecordsdata,andNASDAQPostData.
2.1. AbelNoserdata
Thedescriptionsanddiscussionsinthispaperarebasedonknowledgeaccumulatedthroughthefirstauthorspendingmorethan
twoyearsworkinginsideFidelity'sGlobalEquityTradinggrouponsimilartradingdataanddirectlyinteractingwithAbelNoserasan
outsideconsultingservice.ItisalsobasedonnumerouscompanyvisitsandconversationswithAbelNoseroveraperiodofmorethan
144
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
tenyearsasanacademicsubscriberofthedatalater.
2.1.1. WhoisAbelNoser?(Abel/Noser,ANcerno,andAbelNoser)
LocatedinNewYorkCity,AbelNoserisabrokeragefirmthatalsoprovidestransactioncostanalysistoinstitutionalclients.The
firm was co-founded by Stanley Abel and Gene Noser in 1975 with a goal of providing low cost trading to institutional clients.
ConfusionoftenexistsregardingthetermsofAbel/Noser,ANcerno,andAbelNoser.ThefirmwasoriginallynamedAbel/NoserCorp.
anditprovidesbothbrokerageservicesasasell-sidefirmandtradingcostmeasurementservicestobuy-sideinstitutionalinvestors.
Someoftheirtransactioncostanalysisclientsraisedconcernsaboutconfidentialityoftheirtradingdataandpotentialmisuseoftheir
databythebrokeragesideofAbelNoser.Inresponsetothisconcernandthegrowthoftransactioncostanalysisservices,thefirmset
upAbelNoserHoldingsLLCastheparentcompany,andcreatedawholly-ownedsubsidiarynamedANcernoLtd.,separatefromits
brokeragebusinesstospecializeintransactioncostanalysis(ANforAbelNoser,andcerno,Latinforexamineanddiscern).However,
thisnamechangecausedconfusioninthemarketplaceintermsoftherelationshipbetweenANcernoandAbelNoser.SoANcerno
changeditsnameagaintoAbelNoserSolutions,renderingANcernodefunct.AbelNoserSolutions,justlikeitspredecessorANcerno,
isstillwholly-ownedbyAbelNoserHoldingsandseparatefromthebrokeragebusiness.TheaboveiswhywerefertothedataasAbel
Noserdata.
2.1.2. WhydoinstitutionalinvestorsprovidetheirtradingdatatoAbelNoser?
Theanswertothisquestionistwofold,differentforthetwomaintypesofinstitutionswhoprovidetheirtradingdatatoAbel
Noser:investmentmanagersandplansponsors.InvestmentmanagersaremutualfundfamiliessuchasFidelityInvestments,Putnam
Investments,andLazardAssetManagement.ExamplesofplansponsorsincludetheCaliforniaPublicEmployees'RetirementSystem
(CalPERS),theCommonwealthofVirginia,andUnitedAirlines.
Plan sponsors, who fall under the jurisdiction of the Department of Labor (DoL) rather than the Securities and Exchange
Commission(SEC),aresubjecttotheEmploymentRetirementIncomeSecurityAct(ERISA),whichmandatesthatplansponsorsneed
todemonstrate “bestexecution”fortheir trading transactions conducted onbehalf ofplanparticipants. Inorder to fulfillthis re-
quirement,manyplansponsorssubscribetoanexternaltransactioncostanalysisservice,suchasAbelNoserorPlexus.Therefore,one
mightcharacterizeplansponsors'presenceinAbelNoserdataas“semi-voluntary.”
On the other hand, investment managers subscribe to Abel Noser's services on a voluntary basis. For each individual trade,
transaction cost savings may appear to be small. On the other hand, total dollar trading costs, especially for large investment
managers,canbesignificant.Inaddition,sinceitiswell-documentedthatinvestmentmanagers'abnormalreturnperformanceor
alphaonaverageisminimalornon-existent,potentialtradingcostsavingsmaybecomemoreimportantrelativetoinvestmentalpha.
So investment managers also subscribe to external transaction cost analysis service providers such as Abel Noser. Some large in-
vestmentmanagers,suchasFidelityandStateStreet,evencreatedtheirowninternalgroupsdedicatedtoanalyzingandlowering
trading costs. Forexample, Fidelity created aTrading Techniques &Measurement (TT&M)group around 2000,part ofits Global
EquityTrading,whichhandlesallFidelitymutualfunds'equitytradingtransactions.Atitspeak,TT&Mhadmorethantenfull-time
employees,includingseveralPh.D.’sandanex-MITstatisticsprofessor.
2.1.3. WhydoesAbelNoserprovideinstitutionaltradingdatatoacademia?
WhyisAbelNoserfriendlytoacademiaandpositivelyinclinedtoprovidetheirproprietarydatatoacademicresearchers?Perhaps
ithassomethingtodowithitsacademically-connectedroots.GeneNoser,oneoftheco-foundersandChairmanEmeritusofthefirm,
isoneoftheauthorsofaJournalofFinancepublicationontradingcosts:Berkowitzetal.(1988).
Being a consulting service provider, it is possible that Abel Noser values the credibility and publicity brought about by the
association with academia. For example, Plexus, one of Abel Noser's main competitors of in the transaction cost analysis service
providerspace,alsousedtoprovidedataforacademiaresearch.Inearlieryears,however,theprovisionofdataforacademicresearch
wasquiteadhocandthesampleperiodsprovidedwereshortandsometimesscattered:afewweeksormonths'datasometimeswith
breakswithinthesampleperiods,forexample,thethree-monthAbelNoserdatausedinBlume(1993).
WhenandwhydidAbelNoserstartsupplyingcomprehensiveinstitutionaltradingdatatoacademicresearcherssystematically,
andforcontinuousandlongsampleperiods?Thisislargelyduetotheeffortsbythefirstauthorofthispaper.During2000to2003,
thefirstauthorwasanemployeeatFidelity'sTT&Mgroup,whowasthelargestclientofAbelNoser.Infact,Fidelity'sTT&Mgroup
helpedAbelNosersettingupitsinstitutionaltradingdatastructureandtradingcostsmeasurementframework(“specs”),whichwere
significantlydifferentfromtheircompetitor:Plexus'.WiththestrongsupportofTT&Mgroup'shead,whoholdsaPh.D.fromMIT
himself,thefirstauthorsuccessfullyconvincedAbelNosertoprovide“all”theirdataforhisdoctoraldissertation,afterreplacing
identities of sample institutions with numeric codes. This happened in 2002, after Abel Noser and the first author signed a con-
fidentialityagreementthatyear.Allthreechaptersofthefirstauthor'sdissertationuseAbelNoserdata.Thedissertation,Hu(2005),
becamepublicafterasuccessfuldefenseatBostonCollegein2005andwasthebasisforthreeeventualpublications:Huetal.(2008),
Hu(2009),andChemmanuretal.(2010).
It is worth noting that there were non-trivial initial costs involved for Abel Noser in providing its trading data for academic
research, e.g., extracting, anonymizing, uploading, and updating the data, and providing additional information such as a “data
dictionary” and cross-reference tables so that researchers can understand and make good use of the data. Having incurred these
significant initial costs mainly at the urging by the first author, Abel Noser opened up and started providing the data to other
academicresearchers.
Unlikemostotherdatavendors,itisclearthatAbelNoser'smotivationinthisendeavorisnotfinancial–inearlyyears,thedata
145
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
Fig.1.Typicalinstitutionaltradingprocess.Thisfiguredepictsatypicalinstitutionaltradingprocess.ItisamodifiedreproductionofFig.3inHu
(2009).Theportfoliomanagerandthetraderareinsidethesamebuy-sideinstitutionalinvestor.P ,VWAP,andP aremarketprices,whereasP
pre post E
istheinstitutionalinvestor'sownexecutionprice.
wereprovidedfreeofchargewhenaskedbyanyacademicresearcher,onlyconditionalonsigningaconfidentialityagreement.Abel
Nosersoonrealizedthereisahigherdemandthantheyexpected.Inordertocoveroverheadandothercosts,theystartedcharginga
nominal$500annualsubscriptionfeewhichincludesallhistoricaldata.Laterthesubscriptionfeegraduallyincreasedovertheyears,
butstillonlytocovercostsinvolved.
2.1.4. DescriptionofAbelNoserdata
Foreachclienttradingtransaction,theAbelNoserdatabasecontains107differentvariables.Numbersofvariablesavailablevary
slightlyovertimesinceAbelNoseraddedandremovedvariablesincertainyears.Inaddition,thereareabouttencross-referencefiles
thatcontainadditionalinformationabouttheidentitiesofclient,manager,broker,andthestock.Forbrevity,wedonotlistall107
variablesandthesereferencefileshere.Wefocusourdiscussionsoncommonlyusedvariables.
BeforewediscussvariablesprovidedbyAbelNoser,itisusefultoelaborateonthetradingprocesstobetterunderstandhowAbel
Nosercollectsthesevariables.Fig.1depictsatypicalinstitutional tradingprocess.Thisprocessstartswiththeportfoliomanager
(PM)makinginvestmentdecisions,whichstocktobuyorsell,andthequantitytobetraded.ThenthePMsendstheordertoatrader
insidethesameinstitution.Withinputsfromthetrader,thePMalsodeterminesatradinghorizon,whichisoftenatradingday.The
trading horizon can also be shorter than a trading day, or it can span multiple trading days. Fig. 1 and our description of the
investmentandtradingprocessisamodifiedversionofFig.3andtheaccompanyingdescriptioninHu(2009),ascitedinHenryand
Koski(2017).Tradingdataforinvestmentmanagerclientsarereceiveddirectlyfromtheseclients'OrderDeliverySystem.AbelNoser
usesalognumbertoidentifyeachbatchoftradesreceivedfromclients.Tofacilitatetradingcostanalysisoftradesbyclients,Abel
Noserprovidesinformationonvariousbenchmarkpricesandvolumescorrespondingtoeachtrade.ThisiswhyAbelNoserdatahave
somanycolumnsorvariables.
EachclienttradingexecutioncorrespondstooneobservationinAbelNoserdata,whichincludethreemaincategoriesofvariables:
first,identitiesofmarketparticipantsincluding:clientcode,clienttypecode,clientmgrcode,clienttdrcode,andclientbkrcode.Theclientcode
is a unique numeric ID for each of Abel Noser's institutional clients. The clienttypecode identifies the type of institutional clients:
investmentmanagers(clienttypecode=2),plansponsors(clienttypecode=1),andbrokers(clienttypecode=3).AbelNoserdatadonot
contain many broker clients and the data coverage tends to be sporadic. Thus, empirical analysis of institutional trading should
normallyexcludeobservationswithclienttypecode=3andmissingclienttypecode.OthercodesarediscussedlaterinSection3.1.The
secondcategoryofvariablescontainsspecificinformationforeachtransaction,includingthesymbol,CUSIP,side,price,volume,and
commissionusd.SymbolandCUSIPidentifythestocktraded.Side,price,volume,andcommissionusdspecifywhetherthetradeisabuyor
sell,theexecutionprice,thenumberofsharestraded,anddollarcommissionspaidonthetransaction.Thethirdcategoryofvariables
containsreferenceandmarketbenchmarkinformation(marketpriceandmarketvolume)forvarioustimehorizonsandaggregation
levelsforeachtransaction.
146
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
2.2. Otherdatasources
Inthissub-section,wedescribevariousotherpotentialdatasourcesforinstitutionaltrading:Plexus,SEI,SEC13Ffilings,CRSP/
Thomsonmutual fundholdings, TAQ, CAUD, TORQ, NASDAQ clearingrecords data, andNASDAQ PostData. We alsodiscuss the
advantagesanddisadvantagesofusingAbelNoserdatatostudyinstitutionaltradingcomparedtothesealternativedatasources.
2.2.1. PlexusandSEI
SeveralearlymicrostructurestudiesuseinstitutionaltradingdataprovidedbyPlexus,forexample,KeimandMadhavan(1995,
1997)andConradetal.(2001,2003).Irvineetal.(2007)usePlexusdatainnovativelytostudytipping.Plexusdataareperhapsthe
closestinnaturetoAbelNoserdata.Infact,AbelNoserandPlexuswerecompetitorsinthemarketplaceofprovidingtransactioncost
analysisservices.Asmentionedbefore,themainadvantageofAbelNoserdataisthelongandcontinuoustimeserieswhichallowsfor
analysesofmanyinterestingcorporatefinanceandinvestmentresearchtopics.Ontheotherhand,Plexusdataweretypicallypro-
vided on a much more ad hoc basis and for much shorter and scattered sample periods. Importantly, Plexus has long stopped
providingitsdatatoacademiasinceitsacquisitionbyInvestmentTechnologyGroup(ITG)in2005.ChanandLakonishok(1993,
1995)usedataprovidedbySEICorporation,whicharesimilartoPlexusandAbelNoserdataaswell,butitalsostoppedproviding
datatoacademiaalongtimeago.
2.2.2. SEC13Ffilings
Underthe1934SecuritiesAct,allinstitutionalinvestorswithmorethan$100millionin13(f)securities(mostlypubliclytraded
equity,butalsoconvertiblebondsandoptions)mustfileSECForm13F.Thedisclosureofportfolioholdingsisatthemanagement
companylevelonaquarterlybasis,e.g.,mutualfundmanagementcompaniesandhedgefunds.Afundmanagementcompany(e.g.,
FidelityandVanguard)managingmultipleindividualfundswillreportaggregatedholdingsbyalloftheseindividualfunds.
The date when the Form 13F is filed with the SEC is referred to as the “filing date,” and the quarter-end date on which the
portfolioisdisclosedasthe“quarter-endportfoliodate.”Normally,theSECrequiresthatthemaximumlagbetweenthetwodates
shouldbe45calendardays.However,subjecttoapprovalbySEC,institutionscandelayorpreventdisclosureofcertainholdings,
usuallyuptooneyear,fromthedaterequiredfortheoriginal13Ffilings.Such“hiding”ofholdingsisdisclosedinanamendmentto
theoriginal13Ffilingafterarequestisdenied,oraftertheconfidentialityperiodexpires.Forfurtherdetaileddiscussionsofhiding
holdingsby13Finvestors,refertoAgarwaletal.(2013)andAragonetal.(2013).
TheadvantagesofAbelNoserdatacomparedto13FfilingsishighlightedinChemmanuretal.(2009)forinstitutionaltrading
aroundcorporateeventssuchasseasonedequityofferings(SEOs):withdailytradingrecordsfromAbelNoser,theauthorsareableto
identityallocations andtrading ofSEOsbyeach institution.The advantage ofAbelNoserisalsohighlighted byPuckett andYan
(2011) for investments topics: they find that institutional investors earn significant abnormal returns on their trades within the
tradingquarter(namelyinterimtrading),whichareundetectedby13Fdata.ThemaindisadvantageofAbelNoserdatabaseisthatit
doesnotcoverallinstitutionalinvestorsinthemarket,whilethe13Fdatabasehasmorecompletecoverageofinstitutionalinvestors
sincetheseinvestorsarerequiredtoreporttheirholdingsonaquarterlybasis.
2.2.3. CRSP/Thomsonmutualfundholdings
TheSecuritiesExchangeActof1934andtheInvestmentCompanyActof1940mandatedmutualfundstoreporttheirholdingsto
theSECofdates(“reportdates”).Thesedisclosuresmustincludeallofthefund'sportfoliopositions,including,butnotlimitedto
commonequities,preferredequities,options,bonds,andshortpositions.PriortoMay2004,theSEConlyrequiredmutualfundsto
filetheirportfolioholdingstwiceayearusingthesemi-annualN-30Dform.SinceMay2004,theInvestmentCompanyActof1940
mandatesthatindividualmutualfundsdisclosetheirportfolioholdingsquarterlyinFormsN-CSRandN-Qwithadelayofnolonger
than60days.ThedatethisportfolioinformationisfiledwithEDGARisknownasthe“filedate,”andthedifferencebetweenthefile
dateandreportdateisoftenknownasthe“reportingdelay.”Agarwaletal.(2015)andSchwarzandPotter(2016)providedetailed
descriptionsofreportingrequirementsformutualfunds.
The most widely used mutual fund holding data is the Thomson Mutual Fund Holdings (Thomson) database, which contains
portfolio holdings since January 1979. This data has been used by majority of existing mutual fund research papers (e.g., Zheng
(1999)andKacperczyketal.(2008)).PriortoitsownershipbyThomson,thisdatabasewasknownasCDAInvestmentTechnologies
data(Danieletal.(1997)),aswellastheCDASpectrumholdingsdatabase(Wermers(1999)).Itcontainsmanymutualfundport-
foliosreportedasofdatesnotmandatedbythe1940InvestmentCompanyAct.Startinginthesecondhalfof2003,CRSPMutual
FundDatabase's(CRSP)portfoliosbecameavailable.TheCRSPMutualFundDatabase'scoverageisfromJanuary2001topresent.
ThemainadvantageofAbelNoserdatacomparedtoCRSPorThomsonMutualFundHoldingsisitsdailyfrequencyofreporting.
Withsemi-annualorquarterlyportfoliodisclosuresfromCRSPorThomsonMutualFundHoldings,researcherscannotobservethe
interimtradingbetweentwoconsecutivereportdates.Kacperczyketal.(2008)conductathoroughanalysisofunobservedactionsof
mutualfundsleadingtoreturngap—thedifferencebetweenthereportedmutualfundreturnandthereturnonaportfoliothatinvests
inthepreviouslydisclosedfundholdings.ThemaindisadvantageofAbelNoserdatabaseisthatitdoesnotcoverallfundsinthe
market,whileCRSPandThomsonMutualFundHoldingshavebettercoverageintermsofmutualfundssincethesefundsarerequired
toreporttheirholdings.Inaddition,AbelNoser'sfund-levelidentifiersandinformationcanbeproblematicattimes,thoughthey
havebeenusedinprioracademicstudies.
147
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
2.2.4. TAQ
TheTradeandQuote(TAQ)databasecontainsintradaytransactionsdata(tradesandquotes)forallsecuritieslistedontheNew
YorkStockExchange(NYSE),AmericanStockExchange(AMEX),aswellasNASDAQNationalMarketSystem(NMS)andSmallCap
issues.
ThemaindisadvantageofTAQcomparedwithAbelNoseristhatTAQdatabasedoesnotidentifythetraderanddoesnotclassify
directionoftrades(buysversussells).PioneeredbyLeeandReady(1991),researchstudiesanalyzingthesedataapplyanalgorithm
to assign the side ofthe trade depending on whether thetrade price is closer to bid orask. However,the Lee and Ready (1991)
algorithmhasbecomeunreliableinmorerecentyearsduetosub-pennytradingandshallowdepthsatthequotedbidandask.Onthe
otherhand,AbelNoserdatacontainaccurateinformationonwhethereachtransactionisabuyorasellbysampleinstitutions.Thus,
noinferencealgorithmisneeded.
When using TAQ, to proxy for investor type, researchers commonly use trades cutoffs by dollar size (Lee and Radhakrishna
(2000)), block trades (Kraus and Stoll (1972) and Bozcuk and Lasfer (2005)), or a more sophisticated daily trades and quarterly
holdingsmappingprocedureproposedbyCampbelletal.(2009).However,assuggestedbyCreadyetal.(2014),inferringinvestor
typebased on tradesizeis notreliable. However,Abel Noserdata contain institutional trades with type andidentities ofsample
institutions.ThemainadvantageofTAQcomparedwithAbelNoseristhatTAQdatacontainallmarkettrades,whileAbleNoserdata
onlycovertradingbyasubsetofinstitutionalinvestors.
2.2.5. CAUDandTORQ
TheNYSE'sConsolidatedEquityAuditTrailData(CAUD)containdetailedinformationonallordersexecutedontheexchange,
bothelectronicandmanual(thosehandledbyfloorbrokers).Thisdatabaseincludesprogramtradingandindexarbitragetrading.
One of the fields associated with the buyer and seller of each order, Account Type, specifies whether the order comes from an
institutionalinvestor.TheAccountTypedesignationofindividualinvestorordershasitsoriginsintheaftermathofOctober1987
stockmarketcrash.Kanieletal.(2008),Kanieletal.(2012),andHendershottetal.(2015)containdetaileddiscussionsandanalyses
ofCAUDdata.
Thetrades,orders,reports,andquotes(TORQ)databasebytheNewYorkStockExchange(NYSE)containsdetailedinformation
ontrades,orders,quotes,andauditreportsassociatedwithtradesforasize-stratifiedsampleof144NYSE-tradedequitysecurities
fromNovember1,1990,toJanuary31,1991.Thereareapproximatelyfifteenfirmsrandomlyselectedfromwithineachcapitali-
zationdecile.Thedatabaseidentifiesthepartyinitiatingthetrade(individualversusinstitution,forexample)andthedirectionofthe
trade(buyversussell).Thisallowsthepreciseidentificationofthesevariableswithoutrelyingonalgorithmsintendedtoinfertrade
directionandtraderidentity.SeeSiasandStarks(1997),Kavajecz(1999),LeeandRadhakrishna(2000),andChakravarty(2001)for
furtherdescriptionsoftheTORQdatabase.
ThemainadvantagesofAbelNoserdatacomparedtoCAUDandTORQisthatAbelNosercoversstocksfromallstockexchanges
inU.S.andcontainsidentities andtypesofsample institutional investors,allowing forinstitution-specificempirical analysis. The
disadvantageofAbelNoserdatabaseisthatitdoesnotcoveralltradesortradersinthemarket,whileCAUDandTORQhavebetter
coverageintermsoftraderssincethesedataaredirectlyfromtheexchange.
2.2.6. NASDAQclearingrecordsdataandNASDAQPostData
NASDAQclearingrecordsbasedtradingdataconsistoftradingbybrokeragehousesinallNASDAQ-listedfirmsfromJanuary2,
1997,toDecember31,2002.Thedataincludethedate,time,tickersymbol,tradesize,andpriceofeachtransactionforeachstock.
These clearing records also include market maker IDs from the settlement process allowing trading volume to be assigned to in-
vestmentbanks.Thedatacanidentifythedirectionofeachtrade(buyorsell).Thedataalsocontainseparateprincipal/agentflagsto
identifywhetherthepartiesaretradingfortheirownaccountorforaclient.Thedataincludebothtradesreported“tothetape”(tape
report)andunreportedNASDAQclearingrecords(nontapereport).ThemainadvantageofAbelNoserdatacomparedtoNASDAQ
clearingrecordsbasedtradingdataisthatresearcherscanseparatelytracktradingrecordsofeachsampleinstitutioninAbelNoser,
butnotinthelatter.ThemaindisadvantageofAbelNoserisitscoverageofonlyasubsetofinstitutionalinvestorsandhencemarket
dollartradingvolume.AccordingtoGriffinetal.(2012)whouseNASDAQclearingrecordsdata,thedatacover77.8%ofNASDAQ
tradingvolume.PleaseseeSection3.1.8belowforourresultsonAbelNoserdata'scoverageofCRSPtradingvolume.
NASDAQPostDatacontaintradedate,tickersymbol,marketmakerIDandtype(includingacodeforECNs),numberofshares
traded,dollarvolumeoftrade,numberoftrades,andaveragetradesizeforallNASDAQmarketmakers.Eachtradeisdividedinto
buy,sell,andcrossedtrades.PostDataareattributedtoparticularmarketmakers.JuergensandLindsey(2009)use136daysofthe
datafor3874NASDAQ-tradedstocksforallquotingmarketmakers.SinceNASDAQPostDataisabouttradingbymarketmakers,itis
quitedifferentfromthesetofinvestorscoveredbyAbelNoserdata,whicharemainlyeitherinvestmentmanagersorplansponsors.
Tosummarize,AbelNoserdata,thoughonlycontaintradesforasubsetofinstitutionalinvestors,offerclearadvantagesoverthe
abovelonglistofalternativedatasources.Themainadvantagesincludehighfrequency(atleastdaily),longandcontinuoustime
series,detailedandaccurateinformationonspecifictransactionssuchasside(buyversussell),andtypeandidentitiesofsample
institutionsallowingforcross-sectionalandinstitution-specificanalyses.
3. Dataandresults
Inthissection,wediscussvariousAbelNoserdataissues(Section3.1)andpresentresultsofourempiricalanalysisoninstitutional
tradesizesandAbelNoserdata'scoverageofCRSP(Section3.1.8).
148
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
3.1. Dataissues
3.1.1. Samplestartandend
Asmentionedearlier,AbelNoserprovidesinstitutionaltradingdatafrom1997to2015.However,dataforthefirsttwoyearsare
verysmallcomparedtolateryearsandtheremightbedataqualityissuesasmentionedbyAbelNoserduringthosetwoyears.Sowe
startoursampleperiodinJanuary1999.AbelNoserprovidesdataonaquarterlybasiswithatimelagaroundoneyear.Sometimein
2017,AbelNosercompletelystoppedprovidingdataforacademicresearch.Sothedataendedaround2015.Moreimportantly,Abel
Noserremovedclientcode inthedata afterSeptember2011 along withseveral othervariables. Initially,AbelNoseralsoremoved
clienttypecode. After pleading directly to upper management by the first author of this paper, Abel Noser agreed to add back cli-
enttypecode,butnotclientcode,fordataafterSeptember2011.
Clientcode is a key variable that separately identifies trades from different institutions. As discussed earlier, one of the main
advantages of Abel Noser data is the ability to separately identify and track trades for different institutions allowing for cross-
sectional and institution-specific analyses. Therefore, Abel Noser data after September 2011 are much less useful for academic
researchers,especiallyformostcorporatefinance,investments,andaccountingtopics,whereseparatelytrackingtradesfordifferent
institutions enables much more thorough and interesting investigations of relevant research questions. Abel Noser data after
September2011maystillbeusefulforcertainmarketmicrostructureandinvestmentsstudiesthatonlyaimtogenerallyexamine
institutionaltrades,forexample,astudyofinstitutionaltradingcoststhatdoesnotrelyonknowingsomethingaboutthetrader's
characteristics. Thoughstilluseful,AbelNoserdata afterSeptember2011 arelessunique comparedtoalternative sources forin-
stitutionaltradingsuchasNYSECAUDdiscussedearlier.Table5listssampleperiodsforall55publicationstodateusingAbelNoser
data.Notsurprisingly,sampleperiodsforalmostallpublications,includingforthcomingpapers,fallwithinthesampleperiodweuse:
January1999toSeptember2011(12yearsand9months).
3.1.2. Potentialsamplebiases
SinceAbelNoserprovidestradingcostanalysisforclients,presumably,theseinstitutionscaremoreabouttheirtradingexecution
costs in general. Therefore, sample selection biases, if any, should mostly concern microstructure studies. On the other hand, for
researchquestionsincorporatefinance,investments,andaccounting,wecannotthinkofobvioussampleselectionbiases.Appendices
ofPuckettandYan(2011),Anandetal.(2012),andJame(2018)containdiscussionsofissuesrelatedtopotentialsurvivorshipand
selectionbiasesofAbelNoserdata.ThesepapersshowthatAbelNoserinstitutionsonaveragedonotdifferfrom13Finstitutionsin
stockholdings,returncharacteristics,andstocktrades,althoughtheyarelargerinsize.
3.1.3. MatchingwithCRSP
Abel Noser data contain three variables that identify each stock: symbol, CUSIP, and stockkey. Both symbol and CUSIP are as
providedbyAbelNoserinstitutionalclients.Complicatingthings,differentinstitutionsmayhavedifferentvariationsforthesame
stock's symbol and CUSIP, e.g., the number of characters in CUSIP may vary and symbol may be different for the same stock for
differentinstitutions.StockkeyisAbelNoser'sownstockidentifier.However,wehaveobservedthatstockkeymaycontainmatching
errorsandmayinfactbereusedovertimefordifferentstocks,thoughwehaveobservedthatthequalityofstockkeyhasimproved
overtheyears.InordertomatchwithCRSPPERMNO,wethereforerecommendfirstcleanupsymbolandCUSIPprovidedbyAbel
Noser,andthenusethemto matchwith CRSP.Thismatchingprocess isnon-trivial andmayinvolve partialmanualmatchingto
ensurematchingquality.
3.1.4. Datarepetitionandlognumber
WhenAbelNoserreceivesabatchoftradingdatafromaclient,typicallymonthly,itassignsauniquelognumberforeachbatchof
datareceived.Sometimesclientsmaysend“corrections”forearlierbatchesoftradingdata.Inotherwords,AbelNoserdatamay
containrepetitionsoftradingdataforthesameclientinstitutionoverthesametimeperiod.Tocounterthisproblem,researchers
usingAbelNoserdatawillneedtoremovethesedatarepetitionsbymakinguseofthelognumber,asitisuniqueforeachbatchof
clienttradingdatareceivedbyAbelNoser.
3.1.5. Intradaytimestamps
AbelNoserdatacontainseveralintradaytimestampsmarkingthetimesatdifferentpointsinthetradingprocessasdepictedin
Fig.1:orderentry(whentheportfoliomanagerenterstheorder),orderplacement(whentheinstitutionaltraderplacestheorder
withanexternalbroker),andorderexecution(whenthetradeorderisexecuted).Wheninstitutionalclientsdonotprovidethesetime
stamps,AbelNosersetsentryandplacementtimestothemarketopen(9:30am),andexecutiontimestothemarketclose(4:00pm).
MostacademicstudiesusingAbelNoserdatasofardonotmakeuseofintradaytimestamps.Onereasonisthatthesetimestamps
default to market open or close quite frequently, especially during earlier years. For example, Choi et al. (2017) state that “a
significantfractionoftradesarerecordedasoccurringattheopeningoftrading”(footnote6).Anotherpossiblecomplicationthat
mayaddfurthernoiseinintradaytimestampsisthesplittingandpackagingoftradeorders.However,itdoesnotmeanthatintraday
timestampscouldnotbeusefuljustbecausetheymightbenoisy.
In a recent paper, Huang et al. (2018a) make an important contribution by carefully analyzing the patterns of intraday time
stampsinAbelNoserdata.Theauthorspresentconvincingevidencethattheseintradaytimestamps,thoughnotwithoutlimitations
andcouldbenoisy,arelikelytobeauthenticandvalid.Huangetal.(2018a)firstshowthatthereareindeedalargepercentageof
placementtimesatthemarketopenandexecutiontimesatthemarketclose,thoughthesepercentagesdeclineoverthesampleperiod
149
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
significantly.Theyreportthatforplacementtimes,thepercentagestartsatover80%atthebeginningofthesample,butdeclinesto
about40%attheend,andthefullsamplepercentageoftradesplacedatthemarketopenis52%.ThisisconsistentwithAbelNoser's
institutionalclientsprovidingintradaytimestampstoagreaterdegreeovertime.Importantly,Huangetal.(2018a)findthatwhen
placementtimesarenotatthemarketopen,thepatternofplacementtimesduringthetradingdayappearstobeveryrealisticand
reasonable:hightradingvolumeafterthemarketopen,smoothtradingvolumeduringthetradingdayexceptforsmallhourlyspikes,
andincreasedtradingvolumeasthemarketcloseapproaches.TheirfindingssuggestthatthoughintradaytimestampsinAbelNoser
dataarefrequently atthemarketopen orclose,whenthey arenotatthemarketopen orclose,theyarelikely tocontain useful
information.2
3.1.6. Identitiesofsampleinstitutions
Thisisanimportantdataissue,sinceifarmedwithidentitiesofsampleinstitutions,researcherscanthenmatchAbelNoserdata
withotherpubliclyavailabledatasourcessuchasSEC13Ffilingtogainmanymorevariablesandadditionalcharacteristicsofsample
institutions.Theoriginaldatawereanonymous,thatis,thoughthereisaclientcodethatseparatelyidentifieseachsampleinstitution,
noinstitutionidentitiesinformationwereprovided.However,bycumulatingtradingbyAbelNosersampleinstitutionsforeachstock
and comparing trading patterns with quarterly portfolio holdings changes from public 13F filings, one may be able to “reverse
engineer”anduncoveridentitiesofAbelNoserinstitutions.Huetal.(2009),anearlyworkingpaperversionofHuetal.(2018),was
thefirsttoimplementsuchanon-trivialalgorithm,whichwaslaterusedinChemmanuretal.(2010).TheInternetAppendixofHu
etal.(2018)describesthismatchingalgorithmindetail.Notethatthesepapersonlyusetheinferredidentifyinformationtoclassify
institutions(e.g.,transientinstitutions)withoutrevealinganyinformationonanyspecificsampleinstitution.
Ontheotherhand,sometimeduring2010–2011,AbelNoserdataincludedaMasterManagerXreffilecontainingcross-reference
identityinformationforsampleinstitutions. Thisfilewasonlyavailableduringthisperiodandnotfordatasubscribersbeforeor
after.Theabovetimeperiodreferstotradingdataperiod,andsinceAbelNoserprovidestradingdatatoacademicresearchwitha
ninetotwelvemonths'timelag,oneneedstobeasubscriberofAbelNoserdatasometimeduring2011–2012tohavegainedaccessto
thiscrucialpieceofinformation.Withthiscross-referencefile,thereverseengineeringalgorithmimplementedinHuetal.(2009,
2018)isnolongernecessary,andresearcherscanmatchAbelNoserdailytradingdatawith13Fquarterlyholdingsdata,see,e.g.,
Choietal.(2017).
However,therevelation ofAbelNoserinstitution identities mayindeedbe oneofthemainreasons whyAbelNoserremoved
clientcodeinthedataafterSeptember2011.Asdiscussedpreviously,sincethemotivationforAbelNosertoprovidedataforacademic
researchisnotfinancialgainstobeginwith,thecostforthefirmtostopprovidingdatatoacademiaissmallandintangible.In2017,
AbelNosercompletelystoppedprovidingdataforacademicresearch.Asaresult,fornewresearcherstogainaccesstoAbelNoser
data,onesolutionmightbetoworkwithresearcherswhohaveusedthedatabefore.Table5containsalistofauthorswhohave
publishedacademicstudiesusingAbelNoserdatasofar.
3.1.7. Clientmgrcode,Clienttdrcode,andClientbkrcode
Forinvestmentmanagers,clientmgrcodeidentifiesfunds,fundmanagers,orseparatelymanagedaccounts.However,itmaychange
over time for the same fund and is not very reliable, though it has been used in prior studies. For plan sponsors, clientmgrcode
identifiestheinvestmentmanagerorfundmanagingplansponsorclients'assets.Jame(2018)providesagoodexampleofusageof
this information to identify hedge fund trades in Abel Noser data. The clienttdrcode identifies the trader inside the institutional
investorhandlingthetradingtransaction(seeFig.1),anditmaybeusefulforcertainmicrostructurestudies.Theclientbkrcodeallows
researcherstoidentifythebrokerwhoexecutedthetradingtransaction.Anandetal.(2012)andChemmanuretal.(2015)aregood
examplesthatmakeuseofbrokerinformationinAbelNoserdata.
3.2. Results
Table1presentssummarystatisticsoftheAbelNosersample.FollowingPuckettandYan(2011),afterusingCUSIP,symbol,and
stockkeyprovidedbyAbelNosertomatchwithCRSPPERMNO,weonlykeepstockswithCRSPSHRCDequaltoeither10or11,i.e.,
U.S.ordinarycommonshares.Wealsoonlykeeptransactionsbyeitherinvestmentmanagers(clienttypecode=2)orplansponsors
(clienttypecode=1),andremovetransactionsbybrokers(clienttypecode=3)andtransactionswithmissingclienttypecode.Notsur-
prisingly,therearemoreplansponsors(740)thaninvestmentmanagers(399).Asdiscussedearlier,plansponsorsarethecoreAbel
Noserclientsintermsofnumbers,andhencetherearemoreplansponsorsthaninvestmentmanagersinthesample.Thenumberof
plansponsorsdropsfairlysteadilyoverthesampleperiod,withapeakof344in2002andalowof138in2011.Thenumberof
sampleinvestmentmanagersfirstincreasesprettysteadilyandsignificantlyfrom37in1999toapeakof157in2006and2007,and
thendropssteadilybutrelativelylesssignificantlyto121in2011.Wedonotknowtheexactreasonsforsuchfluctuationsinthe
numberofAbelNoserinstitutionalclients.Ourconjectureisthatdecimalization,andespeciallytheincreasingprevalenceofalgo-
rithmtradingwhereinstitutionssplittradesintotinypiecesanduseautomatedprogramsfortradingexecutionmayhaveresultedin
lesservalue-addedbyanexternalconsultantsuchasAbelNoser.Ifinstitutionsbreakupallormostoftheirtradesintoverysmall
piecesandspreadthemthroughouttheday,itwouldbeveryhardforanoutsideconsultanttoofferanyadditionalhelpinimproving
tradingexecutionstrategyandlowertradingcosts.Italsobecomeslessimportantforinstitutionstorelyonanoutsideconsultantto
2WethanktherefereefordeepeningourunderstandingofthereliabilityandusefulnessofintradaytimestampsinAbelNoserdata.
150
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
Table1
Summarystatistics.
Year Investment Plan #oftrades Sharestraded $Traded Commissions Sharesper $pertrade
managers sponsors (millions) (billions) ($billion) ($million) trade
1999 37 342 5.1 46.2 $2060 $1589 8985 $400,401
2000 43 328 6.5 62.6 $2762 $1941 9671 $426,574
2001 64 334 8.1 89.0 $2699 $2555 10,964 $332,446
2002 81 344 10.5 112.2 $2708 $4488 10,700 $258,287
2003 84 316 10.4 94.3 $2350 $3707 9052 $225,607
2004 117 285 12.7 91.5 $2615 $3444 7199 $205,662
2005 132 244 16.8 98.8 $3059 $3132 5869 $181,774
2006 157 242 28.4 120.0 $3860 $2990 4231 $136,098
2007 157 220 36.0 117.8 $4159 $2647 3269 $115,455
2008 151 182 30.3 138.4 $3960 $2898 4576 $130,915
2009 143 173 24.5 131.8 $2839 $2695 5389 $116,054
2010 139 168 24.9 98.8 $2619 $2140 3971 $105,314
1–9/2011 121 138 18.4 57.7 $1776 $1273 3136 $96,495
Overall 399 740 232.6 1259.2 $37,467 $35,499 6693 $210,083
ThistablereportssummarystatisticsofAbelNoserdatafromJanuary1999toSeptember2011.Overallnumbersforinvestmentmanagersandplan
sponsorsarethetotaluniquenumbersofinstitutionsoverthewholesample.FollowingPuckettandYan(2011),wematchAbelNoserdatawith
CRSPdailystockfileandonlykeepthosestockswithSHRCDequaltoeither10or11(i.e.,U.S.ordinarycommonshares).Thedollarvalueofeach
tradeequalstheproductofthetransactionpriceandthenumberofsharesboughtorsoldbytheinstitution.Commissionsandsharestradedare
directlyfromAbelNoserdataandsummedacrosstradesexecutedwithineachcalendaryear.Thelasttwocolumnsreporttheaverageofshares
(dollars)pertrade.
measuretheirtradingcosts.Wefurtherdiscusstheprevalenceofalgorithmtradingandtheresultingdramaticdeclineininstitutional
tradesizelaterinthissection.
Afterapplyingfiltersmentionedabove,AbelNoserdatacontain232.6milliontransactions, with1.3trillionsharesand$37.5
trilliontradedovertheperiodfromJanuary1999toSeptember2011.Sampleinstitutionscollectivelypaid$35.5billioninbrokerage
commissions.Thistranslatesinto2.8centspershareand9.5basispointsperdollartraded.Theseestimatesareinlinewithprior
studiesoninstitutionaltrading.Forexample,bothusingAbelNoserdata,Hu(2009)estimatesinstitutionalbrokeragecommissionsto
be11to12basispointsperdollartraded(Table2onpage430)foranearlierandshortersampleperiod,andAnandetal.(2012)
estimatecommissionspaidbyinstitutionstobealsoexactly2.8centspersharefortheirsampleperiodof1999to2008(Table1on
page565).Thesepershareorperdollarcommissionestimatescouldbebiaseddownwardsbecausebrokersmayactasmarketmakers
thereby profiting from bid-ask spreads and charge zero commission for certain NASDAQ trades. On the other hand, Anand et al.
(2012)notethatthereductioninspreadsthataccompanieddecimalizationin2001madetheNASDAQzerocommissionbusiness
modeluntenable,andinstitutionsbeganpayingcommissionsonNASDAQtrades(footnote8onpage566).
Table2reportsstatisticsofAbelNoserdataseparatelyforinvestmentmanagers(PanelA)andplansponsors(PanelB).Though
therearemoreplansponsorsthaninvestmentmanagers,thequantityandthesizeofinvestmentmanagers'tradestendtobelarger.
Specifically,investmentmanagersaccountfor191.1milliontransactions,with1.1trillionshares($31.5trillion)traded,and$30.2
billion in brokerage commissions paid; whereas plan sponsors only account for 41.4 million transactions, with 0.2 trillion shares
($6.0trillion)traded,and$5.3billionpaidforbrokeragecommissions.
3.2.1. Institutionaltradesizes
Fig. 2 plots institutional trade size over time, measured as average shares per trade and average $ per trade, for all sample
institutions (Panel A), investment managers (Panel B), and plan sponsors (Panel C), respectively. The data plotted in Fig. 2 are
presented inthelast two columns inTables 1 and2.Across all threepanels in Fig. 2,there isa clear declining trend of average
institutionaltradesizeovertime.Forinstance,lookingat$pertradeinPanelA,averagetradesizedeclinedfrom$400thousandin
1999to$96thousandin2011,a75%reduction.Forinvestmentmanagers,averagetradesizedeclinedfrom$493thousandin1999
to$113thousandin2011.Similarlyforplansponsors,averagetradesizedroppedfrom$214thousandin1999to$57thousandin
2011.
Lookingataveragetradesizealone,thoughinformative,maynottellthewholestory,asthedistributionoftradesizecouldbe
highlyskewed,withasmallfractionoflargetradescontributingdisproportionallytothesamplemean.Thus,wefurtherexamine
percentilesofinstitutionaltradesizeinTable3.Someratherstrikingpatternsemerge.First,thedistributionoftradesizeappearsto
be highly skewed, with average trade size being much higher than median trade sizes. For example, the median $ traded for all
institutionsoverthewholesampleis$9222(Table3PanelA),whilethecorrespondingaverageis$210,083pertrade(Table1),more
than23timeslarger.Infact,theaverageismuchhigherthanthe75thpercentile–$52,602(Table3PanelA).
From 1999 to 2011, the median dollars (shares) traded declined from $58,900 (1600 shares) to $4708 (135 shares) for all
institutions, a 92.0% (91.6%) drop, and the median declined from $60,250 (1600 shares) to $9447 (295 shares) for investment
managers, a 84.3% (81.6%) drop. The decline is much more dramatic for plan sponsors: from $56,700 (1700 shares) to $194 (5
shares), a 99.7% (99.7%) drop. One possibility is that, since plan sponsors' assets are typically managed by external investment
managers, who trade on their behalf, it is possible that these trades were in fact fractions of larger trades executed by external
151
G.Huetal. Journal of Corporate Finance 52 (2018) 143–167
Table2
Investmentmanagersversusplansponsors.
Year #oftrades(millions) Sharestraded(billions) $Traded($billion) Commissions($million) Sharespertrade $pertrade
PanelA.Investmentmanagers
1999 3.4 36.9 $1694 $1276 10,741 $493,146
2000 4.5 51.6 $2321 $1611 11,462 $515,175
2001 5.7 67.2 $2055 $2084 11,883 $363,441
2002 7.5 89.7 $2158 $3600 11,894 $286,268
2003 8.1 79.5 $1991 $3173 9819 $245,925
2004 10.6 78.5 $2247 $3015 7373 $211,125
2005 15.1 86.5 $2686 $2815 5744 $178,373
2006 26.1 104.6 $3365 $2602 4013 $129,107
2007 32.5 96.7 $3389 $2260 2976 $104,359
2008 26.3 120.6 $3441 $2528 4586 $130,876
2009 20.3 115.5 $2472 $2363 5682 $121,653
2010 17.9 83.5 $2193 $1800 4650 $122,190
1–9/2011 13.1 47.9 $1472 $1069 3667 $112,563
Overall 191.1 1058.6 $31,484 $30,196 7268 $231,861
PanelB.Plansponsors
1999 1.7 9.3 $367 $313 5462 $214,291
2000 2.0 11.0 $441 $330 5577 $223,953
2001 2.5 21.8 $644 $471 8855 $261,282
2002 2.9 22.5 $550 $887 7644 $186,650
2003 2.3 14.8 $359 $534 6378 $154,755
2004 2.1 13.1 $368 $429 6304 $177,598
2005 1.8 12.3 $373 $317 6933 $210,686
2006 2.3 15.4 $496 $388 6693 $215,216
2007 3.5 21.1 $769 $387 5961 $217,276
2008 4.0 17.8 $519 $369 4505 $131,173
2009 4.1 16.3 $367 $332 3951 $88,584
2010 6.9 15.3 $426 $339 2209 $61,551
1–9/2011 5.3 9.8 $304 $204 1833 $57,059
Overall 41.4 200.5 $5982 $5302 5562 $169,237
ThistablereportsstatisticsoftransactionsinAbelNoserdataseparatelyforinvestmentmanagers(PanelA)andplansponsors(PanelB)from
January1999toSeptember2011.
investmentmanagersonbehalfofplansponsors.However,thiscannotfullyexplainthedramaticdecliningtrendovertimeforplan
sponsors and for the whole sample. Fig. 3 plots percentiles of $ principal traded over time for all sample institutions (Panel A),
investmentmanagers(PanelB),andplansponsors(PanelC),respectively.3
Oneplausibleexplanationforthisdramaticdeclineininstitutionaltradesizeisperhapstheincreasingprevalenceofalgorithm
tradingaspostulatedbyChordiaetal.(2011):“Algorithmictradingwasnonexistentintheearly1990sbutwasexpectedtorepresent
about half of the trading volume in 2010” (footnote 49 on page 261). In general, the declining trend of institutional trade size
documentedhereisconsistentwithandprovidesdirectsupportforthefindingsinChordiaetal.(2011),whofindthatmorefrequent
smallertradeshaveprogressivelyformedalargerfractionoftradingvolumeovertime,andthatturnoverhasincreasedthemostfor
stockswiththegreatestlevelofinstitutionalholdings.
Thissignificantdeclineininstitutionaltradesizerenderssize-basedinferencesofinstitutionaltradesproblematic.Pioneeredby
Cready(1988)andLee(1992),transactionsize-basedtechniquesarewidelyappliedintheacademicliterature:Creadyetal.(2014)
identifythat“over30publishedpapersemployingtransactionsize-basedtechniqueswith10ofthemappearinginyear2010orlater
(footnote1onpage878).”Less than$10,000 iscommonly usedasacutoffforsmall tradesdone bysmall (retail)investors.Our
evidencesshowthatthiscutoffisnolongerreliableduetothedramaticdeclineininstitutionaltradesizeovertheyears.Infact,the
median$tradedforoursampleinstitutionsis$9222overtheentiresampleperiod,whichmeansmorethanhalfoftheinstitutional
tradeswouldhavebeenmisclassifiedifoneweretoapplya$10,000cutoff.This$10,000cutoffmaybelessproblematicinearlier
years.Forexample,in1999,the25percentileforallAbelNosersampletradeswas$16,929.However,thenexttwelveyearssawa
dramaticdeclineandthe25percentilein2011wasonly$482.Overall,ourfindingsoninstitutionaltradesizesareconsistentand
supportiveofthefindingsinCreadyetal.(2014).
3.2.2. AbelNosercoverageofCRSP
AbelNoserdata'scoverageofCRSPvolumeisanimportantstatisticthatauthorsandrefereescareabout.PuckettandYan(2011)
statethat:“Onaverage,thistradingactivityaccountsforapproximately8%ofthedollarvalueofCRSPtradingvolumeduringthe
1999to2005sampleperiod.Assumingthatinstitutionalinvestors,inaggregate,areresponsiblefor80%ofCRSPtradingvolume,we
estimate that ANcerno institutions account for 10% of all institutional trading volume” (2nd paragraph on page 606, underlines
3Thepatternforsharestradeissimilar,andthefigureforsharestradedisomittedforbrevity.
152