Dynamics Between Equity Holdings and Returns

Transcription

1 Dynamcs Beween Equy oldngs and eurns Peo S. alev Emly Lo Monash Unversy P. Joam Weserholm Unversy of Sydney 5 January 2007 Absrac Ths paper nvesgaes he neracons beween changes n share ownershp srucure and equy reurns. We only model prce movemens and rades of hree dfferen nvesor classes: foregn nsuons domesc nsuons and real nvesors on he elsn Soc Exchange over he perod of 2000 ll Conssen wh Grnbla and eloharu (2000) we fnd ha foregn nsuons end o engage n momenum based radng sraeges whle domesc nsuons and real nvesors are conraran. Foregn nsuons and real nvesors respond o he same day soc prce movemens and hose n he mmedaely precedng nervals. Negave feedbac radng by domesc nsuons s evden up o lags of several days. Across all hree nvesor classes nraday ne rade s shown o have some mpac on fuure prce movemens. Ths appears o be due prmarly o a prce pressure effec whch reverses n laer nervals. These fndngs add o he undersandng of he causal relaon beween nsuonal rade and reurns. JEL Classfcaon: TBA eywords: ownershp srucure nsuonal nvesors momenum negave feedbac radng Correspondng auhor: Peo alev Deparmen of Accounng and Fnance Faculy of Busness and Economcs Monash Unversy; P.O.Box 97 Caulfeld Eas VIC 345 Ausrala. Tel: Fax: ; Emal: Peo.alev@BusEco.Monash.edu.au We han he Nordc Cenral Secures Deposory and euers (SICA) for provdng he daa.

2 Dynamcs Beween Equy oldngs and eurns 5 January 2007 Absrac Ths paper nvesgaes he neracons beween changes n share ownershp srucure and equy reurns. We only model prce movemens and rades of hree dfferen nvesor classes: foregn nsuons domesc nsuons and real nvesors on he elsn Soc Exchange over he perod of 2000 ll Conssen wh Grnbla and eloharu (2000) we fnd ha foregn nsuons end o engage n momenum based radng sraeges whle domesc nsuons and real nvesors are conraran. Foregn nsuons and real nvesors respond o he same day soc prce movemens and hose n he mmedaely precedng nervals. Negave feedbac radng by domesc nsuons s evden up o lags of several days. Across all hree nvesor classes nraday ne rade s shown o have some mpac on fuure prce movemens. Ths appears o be due prmarly o a prce pressure effec whch reverses n laer nervals. These fndngs add o he undersandng of he causal relaon beween nsuonal rade and reurns. JEL Classfcaon: G0 eywords: ownershp srucure nsuonal nvesors momenum negave feedbac radng

3 INTODUCTION In hs paper we am o conrbue o he undersandng of he causal relaon beween nsuonal equy holdngs and reurns. Snce movemens n soc prces are caused by nvesors radng decsons he characerscs of he nvesors mang hese radng decsons would be expeced o have an mpac on soc prces. There s a growng body of leraure ha sees o documen and explan he lnages beween changes n he holdngs of dfferen nvesor classes and soc prce movemens. owever such sudes have generally been hampered by a lac of precson n he avalable daa. Ths paper examnes he relaon beween equy holdngs and reurns usng a unque daase from he elsn Soc Exchange (EX). Smlar o Grffen arrs and Topaloglu (2003) we sudy he neracon beween equy reurns and changes n ownershp srucure. Specfcally he prmary obecve of hs sudy s o deermne he drecon of causaly beween hese wo varables of neres down o nra-day accuracy and hus help resolve conradcons n pror research. We separaely analyze each of he hree followng nvesor classes: foregn nsuons domesc nsuons and real raders. A secondary research ssue addressed n hs sudy s he herd behavor whn each of hese classes ha s he propensy of nvesors o rade n he same drecon. In recen years here has been a sgnfcan growh n he presence of nsuonal raders n equy mares. Consequenly he bul of exan research s focused on nsuonal radng. In parcular he posve correlaon beween nsuonal radng acvy and soc reurns s well documened n he leraure. For nsance n an early sudy lemosy (977) examnes he mpac of ne nsuonal ownershp changes on reurns n he surroundng monhs and fnds a conemporaneous relaon. More recenly Nofsnger and Sas (999) repor ha for her sample of NYSE-lsed secures here s a posve correlaon beween he wo varables of neres bu he drecon of causaon remans ambguous. In oher words we are faced wh he queson of wheher soc prce movemens drve nsuonal radng decsons or vce versa. The former s commonly arbued o herdng behavor ha s ceran rader classes respondng n he same manner o sgnals. erdng n response o pror reurns s nown as feedbac radng whch can be posve or negave dependng on he drecon of rade (see among ohers DeLong Shlefer Summers & Waldman 990; ong & Sen 999). For nsance Laonsho Schlefer and Vshny (992) examne he quarerly holdngs for a large sample of penson funds and repor ha such nsuonal nvesors do no rade n response o prce changes. We do no dsngush beween foregn and domesc real raders as he former accoun for a neglgble percenage of shareholdngs and rade. 2

4 Grnbla Tman and Wermers (995) and Ca aul and Zheng (2000) also analyze quarerly holdngs daa bu fnd ha nsuonal nvesors employ momenum-based radng sraeges. In oher words hey posve feedbac rade. Furhermore hs causal relaonshp s un-dreconal wh radng acvy havng no mpac on fuure prce movemens. An alernave explanaon for co-movemens beween equy holdngs and reurns s ha ceran mare parcpans are relavely more nformed and hus nfluence fuure prces va her rades (see Glosen & Mlgrom 985; e & Peron 2004 among ohers). Danel Grnbla Tman and Wermers (997) for example sudy he performance of several muual fund porfolos over a 30-year meframe. The fndngs ndcae ha sample fund managers are able o ancpae secury prce movemens a leas some of he me (See Chen Jegadeesh & Wermers 2000; Grnbla & Tman 993; Pnnuc 2003; Wermers 2000 among ohers). owever Pnnuc (2003) cauons ha such fndngs may be drven by prce pressure from nsuonal radng raher han any nformaonal effec. Ths may occur for example when here s large ne nsuonal acvy on one sde of he mare. Due o he demand for lqudy prces n he conemporaneous or mmedaely subsequen perods may ncrease even n he absence of any prvae nformaon. Indeed n her nvesgaon of quarerly nsuonal holdngs and ne radng acvy Gompers and Merc (200) documen fndngs ha srongly reec boh he feedbac rade and nformed radng hypoheses n favor of he explanaon ha nsuonal radng acvy exacs pressure on equy prces. Also n conras o many of he sudes dscussed above oher researchers have repored emprcal evdence suggesng ha causaly s b-dreconal ha s boh feedbac rade and nformaon/prce pressure play a role n explanng he relaon beween he wo varables of neres (for example see Denns & Srcland 2002; Grnbla & eloharu 2000; Sas Sars & Tman 200). Thus alhough much research neres has been focused on hs area resuls from exan papers are no unform. I s plausble ha hs mgh be due o he daa consrans ha have largely lmed analyses o quarerly ownershp daa from he U.S. mares. Accordngly several recen papers have addressed hs ssue usng eher mproved mehodologes or daases. From quarerly daa Sas Sars and Tman (200) employ a mehodology whch allows hem o nfer he relaon beween nsuonal holdngs and reurns over more precse me perods. esuls ndcae b-dreconal causaly wh prce pressure beng of greaer explanaory mporance han nformed radng acvy. Furher almos all he quarerly covarance s arbuable o nraday prce changes bu he lac of daa precludes any furher analyses. 3

5 Emprcal evdence derved usng hgher frequency daa s sparse. An excepon s Grnbla and eloharu (2000) who employ a unque dealed daase of nvesor holdngs on he elsn Soc Exchange (EX). The fndngs show ha foregn nvesors a maory of whch are large overseas nsuons engage n posve feedbac radng. Furher hey exhb superor performance ha canno be compleely explaned by momenum profs. In oher words hese nvesors may also possess an nformaonal advanage. Grffen e al. (2003) analyses NASDAQ ransacon daa over a 0-monh perod 2 wh broer-ype denfers for each rade sde. In conras o Grnbla and eloharu (2000) he analyss reveals ha whle nsuons engage n momenum-based sraeges such rades do no nfluence fuure prce movemens. In summary whle pror evdence provdes a srong ndcaon ha here s a posve relaon beween nsuonal radng volume and reurns he cause of hs comovemen s an ssue ha sll needs o be resolved. I s somemes argued ha local nvesors are more nformed due o her proxmy o he mare (see for nsance Choe ho & Sulz 2005; Ln 2006; chards 2005). As such we dfferenae beween foregn and domesc nsuons. Furhermore alhough nsuons form an ncreasngly large proporon of he equy mares n erms of boh ownershp and radng acvy we also consder he behavor of anoher mporan class of mare parcpans ndvduals or real nvesors. ere resuls are more conssen across pror sudes. Frs such nvesors end o negave feedbac rade ha s hey employ conraran radng sraeges (Grnbla & eloharu 2000; Jacson 2003; anel Saar & Tman 2005; chards 2005). Second real nvesors are relavely less nformed and end o mae less profable radng decsons (see for example Barber Lee Lu & Odean 2004; Barber & Odean 2000). Ths does no necessarly mean ha he causal relaonshp beween real rades and soc prce movemens wll be un-dreconal. anel e al. (2005) for nsance fnd srong emprcal evdence showng ha ne ndvdual radng volume drves secury reurns n subsequen perods. Snce s no very plausble ha ndvduals have an nformaonal advanage over he much larger nsuonal raders on he NYSE he auhors conclude ha hs co-relaon s lely due o a emporary prce pressure effec. erdng behavor ha s he exen o whch nvesors are radng n he same drecon also relaes o co-movemens beween equy holdngs and reurns. Noe ha feedbac radng he endency of nvesors o reac n he same way o pror reurns s self a ype of herdng behavor. owever herdng 2 May 2000 o February 200 4

6 can be due o facors oher han a smlar response o prce movemens (Froo Scharfsen & Sen 992; Sas 2004). Boh he nformed and unnformed have been shown o herd alhough for dfferen reasons. As an example fund managers may rade n he same drecon based on he same se of prvae nformaon (rshleger Subrahmanyam & Tman 994) or o preserve her repuaons (Scharfsen & Sen 990). Conversely ndvduals may be nfluenced by senmen or oher psychologcal effecs (Shller 984; Shlefer & Summers 990). Our paper conrbues o he leraure n he followng four areas. Frs whle mos exan sudes only nvesgae eher nsuonal or real radng behavor we examne he lnages beween equy prce movemens and he rades of hree dfferen nvesor classes: foregn nsuons (FI) domesc nsuons (DI) and real raders (T). Second we buld on he mehodology developed by Grffen e al. (2003) and apply hs o a unque nra-day daase. Our analyss dffers from Grffen e al. (2003) n ha whle hey only dfferenae beween nsuonal and ndvdual raders we analyze foregn and domesc nsuons separaely. Thus resuls from hs sudy conrbue no only o academc wor on nsuonal rade bu also o he home bas leraure. Thrd we help resolve he nconssences n pror fndngs by usng a daase ha allows us o dfferenae beween wo conflcng hypoheses namely wheher soc prce movemens drve nsuonal radng decsons or vce versa. Fourh he models employed also allow us o documen paerns n radng decsons whn each nvesor class of neres. Specfcally we es for herdng acvy a boh daly and nraday level. There are hree man fndngs n hs sudy: Frs on a daly bass here s evdence of feedbac rade for all hree nvesor caegores of neres. Conssen wh Grnbla & eloharu (2000) we fnd ha foregn nsuons end o engage n momenum based radng sraeges whle domesc nsuons and real nvesors are conraran. Foregn nsuons and real raders appear o respond o same day soc prce movemens and hose n he nervals mmedaely precedng. On he oher hand domesc nsuons negave feedbac radng s evden up o a lag of several days suggesng ha here s a lag n he response of such nvesors o sgnals conaned n pas reurns. Moreover wh he excepon of he larger foregn nsuonal raders here s no evdence of nraday feedbac radng. Second he emprcal evdence wealy suppors he hypohess ha ne daly radng acvy drves reurns ha s an nformaon advanage and/or a prce pressure effec. owever hs mpac s almos neglgble for he domesc nsuons wheher on he conemporaneous or lagged radng days. Across all hree nvesor classes nraday ne rade s shown o have some mpac on fuure prce movemens. Ths 5

7 appears o be due prmarly o a prce pressure effec whch reverses n laer nervals. Fnally n erms of he secondary hypohess we fnd evdence of herdng behavor across all hree nvesor classes a boh daly and nraday frequences. The remander of he paper s organzed as follows: Secon 2 conans a descrpon of he EX and daases used are nroduced n Secon 3. Emprcal mehodology ulzed s se ou n Secon 4 whle resuls of he analyses are dscussed n Secon 5. Fnally Secon 6 provdes concludng remars. 2 TE FINNIS MAET The elsn Soc Exchange (EX) s he prmary exchange n Fnland. As wh mos maor exchanges EX-raded secures nclude shares bonds and equy dervaves such as warrans. In addon radng acvy n Fnnsh dervaves on he EUEX s seadly ncreasng. Inroduced n 999 he Euro replaced he Fnnsh Mara as he EX radng currency. As of January 2006 here were 37 frms lsed on he EX. 3 Togeher wh equy rghs and seven across-lsed Swedsh frms hese mae up 29 raded share ssues wh a oal mare value of approxmaely 203 bllon Euros. Ths fgure represens a saggerng ncrease of over 2000% from a mare capalzaon of 0 bllon Euros n 992. Thus alhough sll relavely small s apparen ha he EX s a seadly growng mare. In lne wh hs radng mes on he EX have been exended from 7 o 8 hours whn our sample perod. The EX s essenally an order-drven mare ha s rades are execued by machng ncomng orders o ones on he opposng sde of he order boo wh openng and closng prces beng deermned by means of a sngle prce call aucon. A sgnfcan feaure of he EX s he domnance of one company Noa. A he exreme secures n Noa accouned for almos 80% of he oal mare capalzaon. Whle hs percenage has sgnfcanly decreased (o approxmaely 35% n 2005) Noa sll represens a maor par of he EX n erms of boh mare value and radng acvy. The EX was chosen prmarly because he rchness and deph of he daa avalable allows a horough nvesgaon of he research ssues. Despe hs he use of Fnnsh daa may sll be open o some crcsm. For example due o he sze of he mare and he domnance of Noa resuls may no be seen as beng as wdely represenave as sudes ulzng US.-based daases. ere s mporan o 3 Ths has snce ncreased o approxmaely 50 frms as a May

8 reerae ha daa avalably s very much an ssue. To our nowledge s no currenly possble o oban such daa for any oher maor exchange n he U.S. or elsewhere. Moreover whle he EX n self s sll a relavely small mare here has been subsanal growh n recen years. I also currenly forms par of he OMX a group of sx exchanges ha also ncludes he Copenhagen Soc Exchange and he Socholm Soc Exchange. As of January 2006 mare capalzaon of he OMX was 743 bllon Euros. ence can be argued ha he EX s becomng a sgnfcan mare n he regon f no nernaonally. The Fnnsh mare s also conducve o he focus of he curren research because here s a consderable amoun of foregn acvy. In parcular durng he sample perod of foregn nsuons accoun for abou 20% of radng volume. A any gven pon whn hs sample perod such nvesors also made up a sgnfcan percenage of he ownershp srucure. As such on he whole he benefs of he daase ouwegh s dsadvanages and wll ad n provdng new nsgh no he research ssues. 3 DATA The obecve of hs sudy s o nvesgae he relaon beween changes n share ownershp and prce movemens of shares ha rade on he EX beween May 2000 and December Two raw daases are ulzed n he nal analyss one drecly from he exchange and he oher from he Nordc Cenral Secures Deposary (NCSD). The former comprses nraday ransacon daa mesamps prces and volumes for each rade as well as he broer denfers assocaed wh each rade sde. The laer records every change n share holdngs all deposs whdrawals and adusmens ha occur durng he sample perod. The deph of nformaon capured faclaes a classfcaon of radng acvy accordng o nvesor-specfc characerscs. Dealed descrpons of every feld n he avalable daa and explanaons for he assocaed codes are oulned n Appendx. Mare ndces used n he analyss are obaned from he Secures Indusry esearch Cenre of Asa-Pacfc (SICA). A machng process ha combnes relevan nformaon from boh ses of raw daa (for a dealed dscusson see Lnnanmaa 200) s hen appled o oban a fnal ransacon-by-ransacon daase ha ncludes nvesor-ype nformaon for each rade-sde. In addon furher prelmnary wor s done o classfy he nvesors no foregn nsuons domesc nsuons and real raders. The daly daa are hen analyzed o denfy secures ha are suable for he sudy. Frs snce he focus of he curren 7

9 emprcal chaper s on raded shares all oher nsrumens are elmnaed. 4 Ths leads o a sample sze of 76 observaons. Followng hs he daa are examned o denfy hnly raded secures. In order o oban resuls ha are comparable across he hree nvesor classes such secures are defned as hose ha have a leas an average of one rade per nvesor ype per day durng he perod beng suded. Based on hs creron he sample se s furher reduced o 05 secures. Noe ha he sample sll ncludes socs ha are no lsed across he enre sample perod. Where daa are avalable he model wll be appled o hese socs whn sample perod. For he nraday analyss we sar wh a sample of he 25 mos acve secures n erms of radng volume. A dffculy arses when mos of he radng volume on any gven day s arbuable o us one nvesor class as he ne change n holdngs wll hen be zero for each class. If hs occurs on oo many days across he sample perod here wll no be suffcen observaons o properly mplemen he nraday model. Ths leads o he furher excluson of 5 secures. We hus have a fnal sample sze of 05 for he daly analyss and 20 for he nraday model. Transacon daa for he fnal sample are furher organzed by excludng rades recorded for non-radng days 5 or ousde of normal mare hours. For ndvdual socs a fler s also appled o remove days around 6 soc spls and secondary ssues as such evens may resul n he emprcal models capurng a conemporaneous relaon beween he wo varables ha mgh no oherwse be evden. The fnal daase s comprsed of approxmaely 230 daly and 0965 fve-mnue observaons for each sample soc ha s lsed across he enre perod. 4 METODOLOGY Emprcal esng of he hypoheses s carred ou usng several vecor-auoregressve (VA) models. In he curren wor he man obecve s o deermne he lead-lag beween he ne radng acvy of each nvesor class and prce movemens. To accomplsh hs a VA model s bul n whch he varables of neres are he change n percenage ownershp by each nvesor class and soc reurns whn each nerval respecvely. To fully capure he dynamcs of he relaonshp beween our varables of neres he analyss s frs performed usng daly daa wh and whou he ncluson of a conemporaneous 4 For example exclusons nclude rghs ssues. 5 Non-radng days refer o days on whch he mare s closed such as Easer and Chrsmas. 6 Specfcally rades on he day pror o and on he day afer hese evens are excluded. 8

10 9 varable and hen repeaed a an nraday frequency of fve mnues. All VA models are esmaed usng ordnary leas squares regresson (OLS). 4. Daly VA Model Ths nal model comprses a sysem of wo separae equaons. The frs has reurns as he dependen varable whle changes n nvesor-class holdngs are used as he dependen varable for he second. For boh equaons he ndependen varables are lagged reurns and lagged values of he holdngs change seres. These are defned as follows: TSO SELL BUY Equaon () where :Change n holdngs over perod; BUY : Toal nvesor class buy-sde volume n soc whn perod ; SELL : Toal nvesor class sell-sde volume n soc whn perod ; TSO : Toal shares of secury on ssue as of he begnnng of he day. For comparson purposes he model s esmaed separaely for each nvesor caegory. Lags are deermned usng a ch-squared dsrbued es sasc 7 and he sysem s defned as follows for each nvesor class : urns ET ET ET e Equaon (2) where l: Number of lags se o a mnmum of one and a maxmum of fve; : Error erms whch are assumed o be normally dsrbued wh zero mean and consan varance; ET: Close-o-close reurns for soc. 7 Lag lenghs are also checed usng he Schwarz Informaon Creron.

11 0 4.2 Granger-Causaly Tess In esmang he model dscussed above as wh mos nvesgaons no me seres lead-lag behavor (Sas & Sars 997) s dffcul o fnd any heorecal suppor allowng a clear predcon of he sze and drecon of he coeffcens a each lag. Therefore n order o more clearly dfferenae beween our wo conflcng hypoheses Granger-Causaly (GC) ess (969) are employed. GC ess do no nfer causaly per se bu raher gve an ndcaon of he exen o whch lagged values of each varable affecs movemens n he oher. So for example f changes n he level of foregn nsuonal holdngs n a parcular secury conan nformaon ha s useful n predcng fuure prce movemens foregn nsuonal radng acvy s sad o granger-cause reurns. I should also be noed ha Granger- Causaly can be b-dreconal. Thus n he example above prce movemens can also concurrenly Granger-cause foregn nsuonal rade. Besdes allowng a clearer ndcaon of he lead-lag relaon GC esng also faclaes easer comparson across dfferen secures n he sample se as well as beween nvesor caegores. 4.3 Modfed Daly VA Model When a secury s very acvely raded any such lnage may be more evden over fner me nervals. Concepually f nvesors do n fac adus her radng decsons n response o secury prces movemens hey should do hs on a connuous nraday bass raher han us a he end of each radng day. In oher words any relaon beween he wo varables of neres may be conemporaneous ha s occur whn he same radng day. To ae hs no accoun he nal VA model s modfed as follows: urns ET ET ET ET δ δ e Equaon (3) where l: Number of lags se o a mnmum of and a maxmum of 5; : Error erms whch are assumed o be normally dsrbued wh zero mean and consan varance; δ: eflecs any same day relaon beween he wo varables.

12 4.4 Inraday VA Model I s expeced ha for more lqud secures any lnage beween nvesor radng acvy and share prce movemens occurs conemporaneously. A prelmnary analyss of he nraday daa ndcaes ha many of he sample secures are no suffcenly lqud on a mnue-o-mnue bass. A fve-mnue frequency s chosen as s fne enough o reflec any nraday co-movemens beween he wo varables of neres. Thus boh ne rade flows and share prce reurns are aggregaed over he chosen nerval lengh. Furher as prevously dscussed he analyss s performed on a subse of 20 acvely raded socs. 5 ESULTS FINDINGS AND DISCUSSION Table provdes summary sascs for our sample se. The percenage of oal shares on ssue held by each nvesor ype and mare capalzaons are calculaed as of January 2000 whle sascs on radng acvy are aen across he enre sample perod from January 2000 o 3 s December Noa s by far he leadng secury on he EX. A he sar of he sample perod hs soc alone accouns for 68% of he sample se n erms of mare capalzaon. owever he domnance of Noa declned durng he sample perod o 42% of oal sample soc value as of December In addon a small percenage of sample secures form a maor par of he EX wh he bul of oal mare capalzaon and radng acvy concenraed whn he op 25 sample socs. A brea-down of ownershp nformaon by nvesor classes ndcaes ha foregn nsuonal (FI) raders appear o be he maor players n he mare accounng for 70% of oal shareholdngs whle 0% s held by domesc nsuons and 20% by real raders. On he oher hand radng volumes are relavely evenly dsrbued across he hree nvesor classes. <Inser Table abou here> Due o he fac ha he model s esmaed for each nvesor class n every sample secury full resuls are oo cumbersome o faclae meanngful nerpreaon. Summary ables are hus provded for each relevan se of resuls. Cross-seconal averages of he esmaed coeffcens and adused 2 s as well as he percenage of coeffcens ha are sgnfcan a each lag are presened and dscussed below. Table 2 provdes fndngs for foregn nsuonal radng acvy Table 3 for rades of domesc nsuons

13 and Table 4 for rades of real nvesors. For each able coeffcens of neres and he hypohess esed are se ou n Fgure below. Fnally summary resuls for assocaed Granger-Causaly ess are oulned n Table 3. Unless oherwse saed all references o coeffcens and p-values ndcae crossseconal aggregae values. TADE ETUNS TADE (erdng) (Feedbac Trade) ETUNS (Informaon/PP) Fgure 5. Tradng Acvy of Foregn Insuons When foregn nsuonal radng acvy s modeled as he dependen varable all cross-seconal mean coeffcens are posve. owever he magnude of coeffcen means decreases from a pea of 0.6 a he frs lag o 0.02 for lag fve. The proporon of he sample wh sgnfcan 8 coeffcens also vares nversely wh lag lengh from almos 30% of he ndvdual soc s o only 8.4% for 5. Moreover he p-values show ha only s sgnfcan a he % level. Ths ndcaes ha here s a posve relaon beween lagged reurns and he radng decsons of foregn nsuons ha s such raders do a leas o some exen feedbac rade by buyng (sellng) n response o pror upward (downward) soc movemens. In oher words conssen wh he fndngs of Grnbla & eloharu (2000) FI nvesors end o engage n momenum-based radng sraeges. As would be expeced any such reacon o pas reurns occurs qucly and dsspaes as longer nervals are consdered. On he oher hand all he cross-seconal average s are also close o zero and p-values show ha none are sgnfcan a he % level. Ths suggess ha FI radng acvy does no have any meanngful effec on fuure soc prce movemens. owever he summary GC es sascs show ha reurns granger-cause changes n FI holdngs as well as vce versa bu he proporon of sample socs wh a sgnfcan es sasc s hgher for he former. Thus FI radng acvy may nfluence prce movemens parcularly a closer lags bu any such mpac wll be mnmal. 8 A he 0% level of sgnfcance 2

14 Tesng for conemporaneous effecs reveals posve cross-seconal average δs whch are sgnfcan a he % level of sgnfcance. For each equaon n he sysem hs coeffcen s larger and has a hgher percenage of posve and sgnfcan ndvdual soc coeffcens as compared o eher or across all fve lags. Thus mos co-movemens occur whn he same radng day. Also he mean δ s much larger han ha of δ. Taen ogeher such resuls sugges ha causaly s b-dreconal bu provde more suppor for he feedbac radng hypohess over a prce pressure/nformaon effec. Fnally cross-seconal mean s from he nraday analyss are posve bu close o zero and nsgnfcan across all lags. On he oher hand s are on average posve up o 4 bu hs coeffcen s negave a lag fve. Furher he p-values ndcae ha a he 0% level only s emprcally sgnfcan. GC ess show b-dreconal causaly bu greaer suppor exss for he hypohess ha reurns granger-cause changes n holdngs han for vce versa. Thus n conras o resuls from he daly analyss he prce pressure/nformaon hypohess plays a greaer role n explanng nraday comovemens beween he wo varables of neres. The reversal n he sgn of a longer lags suggess ha s more lely ha hs s due o a prce pressure effec. <Inser Table 2 abou here> 5.2 Tradng Acvy of Domesc Insuons The second nvesor class of neres s comprsed of local Fnnsh nsuons. ere s negave across all fve lags. None of he relevan p-values are sgnfcan a he % level bu he aggregae coeffcens a he frs and ffh lag are found o be emprcally sgnfcan a he 0% level. These resuls sugges a negave relaon beween ne domesc nsuonal (DI) radng acvy and pror soc reurn movemens. In oher words DI raders end o engage n negave feedbac rade (.e. conraran radng sraeges) by buyng (sellng) n response o pas prce decreases (ncreases). Furher n conras o resuls based on FI rades hs response does no appear o be as concenraed n he mmedae subsequen nerval and may perss no longer lags. When analyzng he mpac ha DI rades have on share prce reurns n followng perods we fnd ha none of he aggregae s are sgnfcan a even he 0% level. Moreover hese coeffcens are very close o zero across all lags. Thus he negave relaon beween DI radng acvy and equy reurns s wholly arbuable o negave feedbac rade. Ths fndng s suppored by he GC es sascs whch show ha reurns granger-cause ne DI rade bu no vce versa. 3

15 In addon nraday co-movemens are of relavely lesser mporance when analyzng ha of local Fnnsh nsuons as he cross-seconal means of boh δ coeffcens are negave and neher s sgnfcan even a he 0% level. Smlarly whle all of he cross-seconal or coeffcens from he nraday model are negave he p-values show ha none are sgnfcan even a he 0% level. GC ess however show un-dreconal causaly from ne DI rades o reurns. Thus whle he nraday relaon here s almos neglgble any such co-movemens are arbuable o a prce pressure/nformaonal effec. Snce here s no evdence from prevous analyses conduced ha such raders have any nformaonal advanage he former explanaon s more plausble. Such fndngs are n conras o hose of pror sudes whch sugges ha local raders may have an edge (Choe e al. 2005). Ths may be a feaure of he EX where foregn raders end o be larger nsuons. <Inser Table 3 abou here> 5.3 Tradng Acvy of eal Traders eal raders form he hrd and fnal class under nvesgaon. Cross-seconal s are negave for he frs four lag lenghs consdered and sgnfcan a he 0% level for and 3 ; coeffcens aggregaed across he enre sample se are negave and sgnfcan from lags one o hree. owever wh he excepon of hese coeffcens are on average very close o zero across all lags. GC es sascs show ha reurns granger-cause changes n real holdngs as well as vce versa. Thus can be seen ha here s a negave correlaon beween ne real radng acvy and equy prce movemens wh emprcal evdence supporng b-dreconal causaly. Ths conraran radng s mos dsnc n he subsequen perod and s sll evden up o a lag of several days whereas any nfluence ha such raders have on fuure reurns s only apparen for he frs lag. Furher boh δ and δ are negave and sgnfcan a he % level ndcang ha real rades affec reurns as well as vce versa. The former coeffcen s much larger n absolue value erms. Such fndngs have several mplcaons agan ndcang ha feedbac rade plays a relavely more mporan role n explanng hese co-movemens. For he nraday model cross-seconal mean s are negave close o zero and nsgnfcan across all lags. however s negave and sgnfcan a he 5% level 4

16 wh p-values also ndcang ha hs coeffcen s no emprcally sgnfcan a he oher lags. Moreover he sgn on he coeffcen reverses a longer lags wh 4 and 5 beng on average posve. These fndngs ndcae ha smlar o he resuls found usng FI rades nraday correlaon beween ndvdual radng acvy and equy reurns s due prmarly o a prce pressure mpac. <Inser Table 4 abou here> 5.4 Furher Dscusson Frs evdence of feedbac radng behavor s found across all hree classes of neres. Feedbac radng occurs n a posve drecon for foregn nsuonal raders and n a negave drecon for DI and ndvdual nvesors. In oher words he former engage n momenum-based radng sraeges whle he laer are conraran. The cross-seconal mean and δ coeffcens from he basc VA model are larges for foregn nsuons and smalles for ndvduals. Ths ndcaes ha he radng decsons of FI nvesors are he mos sensve o movemens n soc prces followed by domesc nsuons. On he conemporaneous day and whn he subse of hghly lqud secures however appears ha real nvesors negave feedbac rade o a larger exen han DI raders. Second whle here s a leas wea evdence of holdngs changes drvng fuure reurns across all classes hs effec s almos neglgble when consderng DI rades. Surprsngly beween he remanng wo nvesor classes real radng acvy generally seems o conan more nformaon abou fuure and same day prce movemens. Conssen wh he argumens pu forward by anel e al. (2005) hs may be prmarly due o a prce pressure effec when he conraran behavor of such raders leads o hem nadverenly acng as lqudy provders raher han any nformaonal advanage. In erms of our secondary hypohess boh he VA and GC ess provde srong emprcal suppor for nraday herdng acvy across all hree nvesor classes of neres. Specfcally all mean crossseconal coeffcens are posve and sgnfcan a he % level. Furher GC es sascs show ha ne radng acvy conans nformaon relang o fuure radng behavor whn each nvesor class. I should be noed ha wh he excepon of FI raders we fnd lle evdence of nraday feedbac rade ha s herdng canno be explaned by nvesors smply respondng n he same way o sgnals conaned n prce changes. The cause of such nraday herdng behavor parcularly for DI and ndvdual raders hus remans a queson for fuure research. 5

17 <Inser Table 5 abou here> 6 CONCLUSION Ths paper examnes he dynamcs of he relaon beween radng acvy and soc prce movemens for hree dfferen nvesor classes on he EX. These nvesor classes are foregn nsuons domesc Fnnsh nsuons and ndvdual real raders. The man fndngs are as follows. Frs on a daly bass here s evdence of feedbac rade for all hree nvesor caegores of neres. owever whle foregn nsuons end o engage n momenum based radng sraeges domesc nsuons and real nvesors are conraran. Furher analyss of he conemporaneous co-relaon ndcaes ha foregn nsuons and ndvdual raders appear o respond o same day soc prce movemens and hose n he nervals mmedaely precedng. On he oher hand domesc nsuons negave feedbac radng s evden up o a lag of several days suggesng ha here s a lag n he response of such nvesors o sgnals conaned n pas reurns. Moreover analyses a a fner frequency show ha wh he excepon of foregn nsuonal raders here s no evdence of nraday feedbac radng. Second we fnd wea emprcal evdence n suppor of he hypohess ha ne daly radng acvy drves reurns ha s a prce pressure and/or nformaon advanage effec. owever hs mpac s almos neglgble for he domesc nsuons wheher on he conemporaneous or lagged radng days. Across all hree nvesor classes nraday ne rade s shown o have some mpac on fuure prce movemens. The fndngs from hs nraday analyss sugges ha hs s prmarly due o a prce pressure effec whch reverses n laer nervals. There are hus sysemac paerns n he behavor of he nvesgaed nvesor classes; As hey nerac hese effecs reverse ou over me. Fnally n erms of he secondary hypohess we fnd evdence of herdng behavor across all hree nvesor classes a boh daly and nraday frequences. The observed dfferences n radng sraegy beween nvesor classes can easly be raonalzed. Domesc nsuons have he mos mmedae access o nformaon on domesc companes and have resources o ae he role of value raders. Domesc nsuons become conraran when hey rade o resore fundamenal values of he secures hey rade. Ths s conssen wh our observaon ha domesc nsuons follow negave feedbac sraeges wh up o several days lag. Foregn nsuons have less drec access o nformaon abou companes n a foregn mare and may also perceve ha here s a moral hazard rs n ha hey have lle conrol over how a foregn company s managed. In 6

18 hs suaon he mos raonal sraegy for foregn nvesors s o engage n momenum radng nvesng n companes ha have a proven posve developmen n prce. Ths s conssen wh our observaon ha foregn nsuons follow a momenum sraegy where hey ac on shor-erm sgnals. Domesc real raders may conss of wo groups: one less nformed and one beer nformed abou asse values and order flow. Less nformed real raders end o change her lm orders oo slowly endng up on he conra sde o shor nermedae erm momenum raders whch s conssen wh he prce pressure hypohess. Beer-nformed real raders ae he conra sde o momenum raders when hey feel ha asse values have moved away from fundamenal values. Ths lne of reasonng s conssen wh emprcal evdence n he leraure whch ndcae ha real raders can be profable followng a conraran sraegy. anel e al. (2005) for nsance fnd srong emprcal evdence showng ha ne ndvdual radng volume drves secury reurns n subsequen perods. A possble lmaon of he curren research s ha he resuls obaned usng daa from a relavely small exchange may no mprove our general undersandng of he relaon beween equy holdngs and reurns. The small sample sze of lqud secures may also be seen as a lmaon. owever as prevously menoned whle he EX s small compared o oher global exchanges has become a mare of mporance n he European regon and some socs are mporan componens n a global porfolo. Furher he unqueness and deph of he daa ouwegh hese dsadvanages. Alhough he analyses conduced have shed some lgh on he research ssues here are sll quesons ha reman unanswered. Frs we fnd ha co-relaon beween ne nvesor class specfc rades and reurns n he followng nervals are manly due o a prce pressure effec parcularly on an nraday bass. Does hs mean ha none of hese hree nvesor ypes has an nformaonal advanage? Fuure research could am o deermne he relave proporon of new nformaon assmlaed no prces by he rades of each nvesor class. 9 ecall also ha whle we fnd lle evdence n suppor of nraday feedbac rade here s srong emprcal suppor for herdng acvy across all hree nvesor ypes of neres. Anoher poenal ssue for fuure research s hus deermnng he causes of such herdng behavor. In oher words whn a radng day wha s ha drves nvesors o rade n he same drecon f no n response o a common sgnal conaned n prces. 9 In oher words relave conrbuon o he prce dscovery process. 7

19 eferences Barber B. M. Lee Y. Lu Y. & Odean T. (2004). Who gans from rade? Evdence from Tawan. Worng Paper Unversy of Calforna Davs. Barber B. M. & Odean T. (2000). Tradng s hazardous o your wealh: The common soc nvesmen performance of ndvdual nvesors. Journal of Fnance 55(2) Ca F. aul G. & Zheng L. (2000). Insuonal radng and soc reurns. Worng Paper Unversy of Mchgan. Chen. L. Jegadeesh N. & Wermers. (2000). The value of acve muual fund managemen: An examnaon of he socholdngs and rades of fund managers. Journal of Fnancal and Quanave Analyss Choe. ho B.-C. & Sulz. M. (2005). Do domesc nvesors have an edge? The radng experence of foregn nvesors n orea. evew of Fnancal Sudes 8(3) Danel. Grnbla M. Tman S. & Wermers. (997). Measurng muual fund performance wh characersc-based benchmars. Journal of Fnance DeLong B. J. Shlefer A. Summers L.. & Waldman. J. (990). Posve feedbac nvesmen sraeges and desablzng raonal speculaon. Journal of Fnance 45(2) Denns P. J. & Srcland D. (2002). Who blns n volale mares ndvduals of nsuons? Journal of Fnance 57(5) Froo. A. Scharfsen D. S. & Sen J. (992). erd on he sree: Informaonal neffcences n a mare wh shor-erm speculaon. Journal of Fnance

20 Glosen L.. & Mlgrom P.. (985). Bd as and ransacon prces n a specals mare wh heerogeneously nformed raders. Journal of Fnancal Economcs 4() Gompers P. A. & Merc A. (200). Insuonal nvesors and equy prces. Quarerly Journal of Economcs Granger C. W. J. (969). Invesgang causal relaons by economerc models and cross-specral mehods. Economerca Grffen J. M. arrs J.. & Topaloglu S. (2003). The dynamcs of nsuonal and ndvdual radng. Journal of Fnance 53. Grnbla M. & eloharu M. (2000). The nvesmen behavor and performance of varous nvesor ypes: A sudy of Fnland's unque daa se. Journal of Fnancal Economcs Grnbla M. Tman S. & Wermers. (995). Momenum nvesmen sraeges porfolo performance and herdng: A sudy of muual fund behavour. Amercan Economc evew 85(5) Grnbla M. & Tman S. D. (993). Performance measuremen whou benchmars: An examnaon of muual fund reurns. Journal of Busness 66() rshleger D. Subrahmanyam A. & Tman S. D. (994). Secury analyss and radng paerns when some nvesors receve nformaon before ohers. Journal of Fnance 49(5) ong. & Sen J. (999). A unfed heory of underreacon momenum radng and overreacon n asse mares. Journal of Fnance Jacson A. (2003). The aggregae behavour of ndvdual nvesors. Worng Paper London Busness School. 9

21 anel. Saar G. & Tman S. D. (2005). Indvdual nvesor radng and soc reurns. Worng Paper Unversy of Texas. e B. & Peron. (2004). ow nformed are acvely radng nsuonal nvesors? Evdence fom her radng behavour before a brea n a srng of consecuve earnngs ncreases. Journal of Accounng esearch 42(5) lemosy. C. (977). The mpac and effcency of nsuonal ne radng mbalances. Journal of Fnance Laonsho J. Shlefer A. & Vshny. W. (992). The mpac of nsuonal radng on soc prces. Journal of Fnancal Economcs Ln W. (2006). Performance of nsuonal Japanese equy fund managers by nvesmen syle. Journal of Porfolo Managemen 7(). Lnnanmaa J. (200). Tracng ndvduals o ransacon daa. Worng Paper elsn School of Economcs and Busness Admnsraon. Nofsnger J.. & Sas. W. (999). erdng and feedbac radng by nsuonal and ndvdual nvesors. Journal of Fnance 54(6) Pnnuc M. (2003). An examnaon of he performance of rades and soc holdngs of fund managers: Furher evdence. Journal of Fnancal and Quanave Analyss 38(4) chards A. (2005). Bg fsh n lle ponds: The radng behavour of foregn nvesors n Asan emergng equy mares. Journal of Fnancal and Quanave Analyss Scharfsen D. S. & Sen J. C. (990). erd behavour and nvesmen. Amercan Economc evew 80(3)

22 Shller. (984). Soc prces and socal dynamcs. Broongs Papers on Economc Acvy Shlefer A. & Summers L.. (990). The nose rader approach o fnance. Journal of Economc Perspecves Sas. W. (2004). Insuonal herdng. The evew of Fnancal Sudes 7() Sas. W. & Sars L. (997). eurn auocorrelaon and nsuonal nvesors. Journal of Fnancal Economcs Sas. W. Sars L. & Tman S. (200). The prce mpac of nsuonal radng. Worng Paper Unversy of Texas. Wermers. (2000). Muual fund performance: An emprcal decomposon no soc-pcng alen syle ransacon coss and expenses. Journal of Fnance

23 TABLE : Summary Sascs Ths able provdes cross-seconal aggregae summary sascs for he fnal se of sample frms. Mare Cap s he mare capalzaon as of January 2000 he begnnng of he sample perod. Volume Traded s he oal volume raded over he 5-year sample perod. FI DI and ET refer o he percenage of oal shares on ssue held by foregn nsuons domesc nsuons and real nvesors respecvely. FI Vol DI Vol and ET Vol refer o he percenage of oal volume raded arbuable o foregn nsuons domesc nsuons and real nvesors respecvely. FIT DIT and et are he percenage of rades made by each of hese nvesor classes. oldngs nformaon s presened as of January 2000 whle all radng sascs are aen across he perod from January 2000 and December Panel A gves sascs for all sample socs; Panel B gves cross-seconal sascs excludng Noa. Panel A Sas Mare Cap. FI DI ET Volume Traded FI Vol DI Vol ET Vol FIT DIT et Mean Medan SDev Max Mn Sew ur Panel B Sas Mare Cap. FI DI ET Volume Traded FI Vol DI Vol ET Vol FIT DIT et Mean Medan SDev Max Mn Sew ur

24 23 TABLE 2: Foregn Insuonal Tradng Acvy Ths able provdes resuls from each of he hree models appled o foregn nsuonal radng acvy. Coeffcens are cross-seconal means aen across all socs n he sample. refers o he FI radng acvy and he soc reurns. are he coeffcens on lagged changes n holdngs and hose on lagged reurns. In addon adused 2 re gven for each regresson. Panel A shows resuls from he daly VA: urns ET ET ET ET δ δ e Panel B gves resuls for he modfed VA where δ s he coeffcen on he conemporaneous varable: urns ET ET ET ET δ δ e Panel C oulnes fndngs for he nraday model. P-values are gven n parenheses. Panel A: Daly VA Model Coeffcen p-value (0.84) (0.00) (0.00) (0.00) (0.9) (0.85) (0.00) (0.8) (0.26) (0.8) (0.66) %Sg Coeffcen p-value (0.0) (0.9) (0.4) (0.0) (0.50) (0.5) (0.00) (0.00) (0.00) (0.00) (0.20) %Sg Panel B. VA wh conemporaneous varable Coeffcen p-value (0.85) (0.00) (0.00) (0.00) (0.00) (0.5) (0.75) (0.00) (0.4) (0.8) (0.3) (0.64) %Sg Coeffcen p-value (0.0) (0.00) (0.34) (0.5) (0.27) (0.38) (0.25) (0.00) (0.00) (0.00) (0.00) (0.9) %Sg Panel C. Inraday Coeffcen p-value (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.4) (0.) (0.26) (0.78) (0.9) %Sg Coeffcen p-value (0.5) (0.07) (0.7) (0.) (0.92) (0.5) (0.00) (0.00) (0.0) (0.06) (0.32) %Sg

25 24 TABLE 3: Domesc Insuonal Tradng Acvy Ths able provdes resuls from each of he hree models appled o domesc nsuonal radng acvy. Coeffcens are cross-seconal means aen across all socs n he sample. refers o he FI radng acvy and he soc reurns. are he coeffcen on lagged changes n holdngs and hose on lagged reurns. In addon adused 2 are gven for each regresson. Panel A shows resuls from he daly VA: urns ET ET ET ET δ δ e Panel B gves resuls for he modfed VA where δ s he coeffcen on he conemporaneous varable: urns ET ET ET ET δ δ e Panel C oulnes fndngs for he nraday model. P-values are gven n parenheses. Panel A. Daly VA Model Coeff p-value (0.83) (0.00) (0.00) (0.00) (0.09) (0.28) (0.07) (0.38) (0.96) (0.8) (0.09) %Sg Coeff p-value (0.0) (0.48) (0.94) (0.69) (0.50) (0.3) (0.00) (0.00) (0.0) (0.0) (0.22) %Sg Panel B. VA wh conemporaneous varable δ Coeff p-value (0.84) (0.58) (0.00) (0.00) (0.00) (0.07) (0.24) (0.09) (0.42) (0.89) (0.8) (0.09) %Sg Coeff p-value (0.0) (0.52) (0.83) (0.62) (0.76) (0.50) (0.0) (0.00) (0.00) (0.0) (0.00) (0.2) %Sg Panel C. Inraday Model Coeff p-value (0.83) (0.00) (0.00) (0.00) (0.09) (0.28) (0.07) (0.38) (0.96) (0.8) (0.09) %Sg Coeff p-value (0.0) (0.48) (0.94) (0.69) (0.50) (0.3) (0.00) (0.00) (0.0) (0.0) (0.22) %Sg

26 25 TABLE 4: eal Tradng Acvy Ths able provdes resuls from each of he hree models appled o real radng acvy. Coeffcens are cross-seconal means aen across all socs n he sample. refers o he FI radng acvy and he soc reurns. are he coeffcen on lagged changes n holdngs and hose on lagged reurns. In addon adused 2 are gven for each regresson. Panel A shows resuls from he daly VA: urns ET ET ET ET δ δ e Panel B gves resuls for he modfed VA where δ s he coeffcen on he conemporaneous varable: urns ET ET ET ET δ δ e Panel C oulnes fndngs for he nraday model. P-values are gven n parenheses. Panel A. Daly VA Model Coeffcen p-value (0.23) (0.00) (0.00) (0.00) (0.0) (0.06) (0.03) (0.37) (0.06) (0.3) (0.70) %Sg Coeffcen p-value (0.0) (0.00) (0.04) (0.09) (0.78) (0.25) (0.00) (0.00) (0.02) (0.00) (0.26) %Sg Panel B. VA wh conemporaneous varable δ Coeff p-value (0.22) (0.00) (0.00) (0.00) (0.00) (0.00) (0.06) (0.0) (0.30) (0.03) (0.22) (0.72) %Sg Coeff p-value (0.0) (0.00) (0.03) (0.36) (0.58) (0.44) (0.29) (0.00) (0.00) (0.0) (0.00) (0.23) %Sg Panel C. Inraday Model Coeffcen p-value (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.5) (0.4) (0.6) (0.7) (0.25) %Sg Coeffcen p-value (0.5) (0.03) (0.82) (0.2) (0.59) (0.43) (0.00) (0.00) (0.0) (0.06) (0.32) %Sg

27 TABLE 5: Granger-Causaly Tess Ths able provdes he resuls of Granger-causaly (969) ess beween ne changes n holdngs for each nvesor class and soc reurns. refers o soc reurns FI o changes n foregn nsuonal holdngs DI hose n domesc nsuonal holdngs and ET hose for real holdngs. Sgnfcance s ndcaed as follows: * sgnfcan a he 0% level; ** sgnfcan a he 5% level; *** sgnfcan a he % level. Panel A gves resuls for he daly model and Panel B hose for he nraday model. Granger-cause eurns Granger-cause Granger-cause eurns eurns Granger-cause eurns Panel A. Daly FI Coeffcen *** *** %Sgnfcan Coeffcen 2.20 *** *** %Sgnfcan DI Coeffcen *** %Sgnfcan Coeffcen *** %Sgnfcan ET Coeffcen *** %Sgnfcan Coeffcen *** %Sgnfcan Panel B. Inraday FI Coeffcen *** 2.74 *** %Sgnfcan Coeffcen 4.20 *** *** %Sgnfcan DI Coeffcen ***.782 %Sgnfcan Coeffcen 3.03 *** *** %Sgnfcan ET Coeffcen ***.557 %Sgnfcan Coeffcen *** *** %Sgnfcan