stock prices. Event studies focusing on compositional changes in the S&P 500 index, for

Transcription

1

2

3 I. Inroducion There is srong empirical evidence ha demand and supply shocks can affec individual sock prices. Even sudies focusing on composiional changes in he S&P 500 index, for example, find ha announcemens of addiion o he index increases share prices and delising decreases share prices. 1 Shleifer (1986) and Harris and Gurel (1986) inerpre his as evidence ha demand curves for socks slope down. Lynch and Mendenhall (1997) and Beniesh and Whaley (1996) examining daa since announcemen and delising days were separaed, suppor downward-sloped demand curves, bu also find ha some of gains or losses due o index fund purchases or sales on he day of an addiion or deleion is reversed he following day. The lieraure on block purchases and sales of sock generally finds evidence of emporary price pressure on individual securiies condiional upon unusual demand or supply. 2 Lakonishok, Shleifer and Vishny (1992) find ha socks purchased by insiuions in a quarer show a significan rise. Chan and Lakonishok (1993, 1995) documen price reacions o insiuional rades. The implicaion of hese sudies is ha share prices may no represen fundamenal values bu depend significanly upon how much money is invesed and when. Recen research by Warher (1995) and Zheng (1997) suggess ha hese supply and demand effecs may aggregae o he level of he sock marke iself. Zheng (1998) uses invesmen secor flows, including households, from 1952 o 1996 and finds ha quarerly insiuional demand shocks, bu no household demand shocks, are conemporaneously correlaed o sock marke reurns. Using monhly reurns and flows for he muual fund indusry 1 See Shleifer (1986), Garry and Goezmann (1986) Harris and Gruel (1986), Dhillon and Johnson (1991) and Lynch and Mendenhall (1997). 2

4 from 1984 o 1993, Warher (1995) finds ha reurns and flows are srongly conemporaneously correlaed. He argues ha his is no due o DeLong e al (1990) feedback rading, i.e. o ha invesors chasing he marke by increasing inflows when he marke is rising and increasing ouflows when he marke is falling. His argumen is based on finding no lagged relaionship beween marke reurns and subsequen monh ne flows. Wihou daily daa, however, he canno rejec he hypohesis ha he feedback rading is simply occurring a a higher frequency ha he has daa. Also Fridson and Jonsson (1995) show, wih quarerly daa, ha ne inflows in highyield bond muual funds end o improve liquidiy and reduce bond liquidiy risk premia. A he very long horizon, papers by Bakshi and Chen (1994) and Consaninides, Donaldson and Mehra (1998) provide heory and evidence ha he age disribuion of he populaion is a significan deerminan of he equiy premium and hence sock prices relaive o he riskless asse. Presumably he mechanism for such fundamenal shifs in he equiy premium is he gradual shif in individual demand for equiies wih aging. In his paper, we use daily daa on hree S&P index funds o invesigae he relaionship beween demand shocks and he movemen of he S&P 500 index. We focus on index funds because heir goal is o simply rack he S&P. As a resul, index fund managers have lile incenive o delay or spread-ou orders o avoid adverse rading condiions. While economic gains may be had by delayed rading, racking error is increased as well. Thus, for S&P index funds, is possible o precisely idenify he day on which flows could affec he marke. In addiion, we know ha speculaive flows are no based on beliefs abou manager skill, bu abou beliefs abou he sock marke iself. A hird advanage of focusing solely on S&P index 2 See Scholes (1972), Holhausen, Lefwich and Mayers (1984) and Michaelson and Parch (1985), for example. 3

5 funds is ha we know exacly wha porfolio of socks are demanded on ha day and we have an index ha precisely maches ha porfolio. Shiller (1997) poins ou a problem endemic o all sudies correlaing fund flows wih sock price dynamics. Unless i is possible o prove ha he proceeds from sales of fund shares are no re-invesed in equiies, hen i is difficul o infer ha flows ou of muual funds in fac represens a shif in senimen abou he sock marke. Wihou individual accoun informaion, we canno address his problem direcly, since we do no know wheher he inflows derived from sales of equiies, or from a cash accoun. In he work ha follows, we implicily assume ha purchases and sales reflec re-balancing owards or away from equiies. Despie he limiaions of using aggregae daa a he fund level, our analysis of daily index fund flows indicaes a srong conemporaneous correlaion beween fund inflows and S&P marke reurns. We also documen a srong negaive correlaion beween fund ou flows and S&P marke reurns wih he excepion of ouflows from a back-end load fund. These effecs may be inerpreed in wo ways. Eiher invesor supply and demand affecs S&P marke prices, or invesors condiion heir demand and supply on inra-day marke flucuaions. To sor ou hese effecs, we examine railing invesor reacion o marke moves. Our resuls sugges he marke reacs o daily flows. A lack of any nex-day reacion o posiive marke moves suggess ha invesors are no chasing shor-erm rends. We also invesigae wheher index invesor demand shocks are permanen or emporary. Demand shocks migh simply exer emporary price pressure on S&P 500 socks a he end of he day ha reverse hemselves in he nex rading session. This would be consisen wih he findings by Harris and Gurel (1986) and Lynch and Mendenhall (1997) and ohers ha sock price moves on a lising or delising day parially reversed hemselves in he nex session. 4

6 Oherwise demand shocks can be permanenly incorporaed in asse prices. Indeed, evidence poining owards his direcion has been found by Shleifer (1986) who has shown ha he S&P inclusion price effec is permanen. Also Garry and Goezmann (1986) did no find any evidence of rebound in prices. Unlike all hese sudies we do no focus on he lising-delising phenomenon, concenraing, insead, on he analysis of he effecs of clearly idenified liquidiy shocks (flows ino invesmen funds) on asse pricing in general. This provides us wih an opporuniy o answer wo ineresing quesions. Firs wha is he driving force behind he increase in he relaive value of he S&P 500 index in he recen years. The S&P index has ouperformed broader sock indexes, recenly causing analyss o quesion wheher his is due o fundamenal differences in index composiion, or o invesor preference for he index iself. We give evidence o suppor he invesor preference hypohesis. Second, we can es a long-sanding ene of he financial microsrucure lieraure, ha demand shocks due o liquidiy rading are emporary and end o mean rever over ime. In paricular, in heir paper Campbell, Grossman and Wang (1992) show ha asses reac o informaion shocks over ime in a differen way, according o he naure of he shocks. Prices mean-rever if he informaion shock is ransiory and idiosyncraic. No mean reversion akes place if he public flow of informaion is abou permanen news. Usually idiosyncraic shocks are loosely idenified wih liquidiy rading, while permanen shocks are raced back o more sysemaic srucural facors (e.g. dividends or earning announcemens). However a formal es o idenify permanen vs. emporary shocks is difficul o perform. Our daase provides us wih he opporuniy o es his imporan issue. We examine shocks o prices originaed by demand flows ino index funds (ypically liquidiy rading ypes of shocks). If hey are emporary, hen we expec hem o rever overnigh or in he following days, resuling in a negaive correlaion beween flows and 5

7 fuure reurns. If he shocks are permanen hey should be incorporaed ino prices and no reversion or lagged relaion in reurns should exis. To furher examine his effec, we also consider he relaed behavior of he S&P fuures index. The quesion of wheher liquidiy shocks are ransiory is a fascinaing one. If he recen growh in he S&P has been enirely due o uninformed inflows, his lends suppor o he Bakshi and Chen (1994) and Consaninides, Donaldson and Mehra (1998) hesis abou he changing equiy premium, as well as revealing he mechanism of he change. Finally, wha influences demand? Research by Cohen (1998) suggess ha individuals have a endency owards posiive-feedback rading, bu oher facors beyond pas price dynamics may affec demand. Boh Cohen (1997) and Zheng (1997) find evidence ha bond yields affec equiy purchase and sale decisions. Recen empirical sudies by Sirri and Tufano (1993) and Chevalier and Ellison (1995), carried ou wih low frequency (quarerly) daa and using ne inflows 3 in he funds, show ha invesors seem o reac non-linearly o fund performance, asymmerically flocking in when performances have been high and saying in afer bad performance. These resuls seem o sugges performance-chasing behavior by invesors and low risk aversion. In his paper, we will show ha, a higher frequency (daily) and breaking down flows ino inflows and ouflows, he behavior for invesors in index funds urns ou being compleely differen. I appears ha only invesors ouflows are affeced by pas performance, while inflows are no. Also, he invesors behavior appears o be mainly moivaed by risk aversion, insead of reurn-chasing 4. 3 Given ha a separae breakdown ino inflows and ouflows is no available for muual funds, such auhors consruc flows as he proporional growh in oal asses under managemen for he fund in he considered period. 4 An alernaive hypohesis is ha if he invesor has consan relaive risk aversion and wishes o mainain consan proporion of risky/riskless asses, he will sell equiy when is value rises. In his case we would expec a negaive relaionship beween asses demand and asse reurns. 6

8 Furhermore, among he differen poenial moivaing facors we focus on exper recommendaions. Using he average marke-iming newsleer recommendaion over he period, we find ha invesors appear o reac o exper advice abou he marke. Bullish newsleer senimen is associaed wih greaer inflows and smaller ouflows, bu, again, more han by he average recommendaion, invesors seem mosly influenced by uncerainy. Indeed, dispersions of analyss opinions reduce inflows and increase ouflows. II. Daa Descripion II.1 Funds Our sudy focuses on hree Fideliy index funds: he Sparan U.S. Equiy Index ($12.9 Billion), he Sparan Marke Index ($ 5.8 Billion) and he VIP Marke Index ($3.0 billion) over he period 1993 o he presen. We have daily dollar-valued share purchases and share redempions for each of he hree funds, as well as heir daily reurns. Togeher, he hree funds represen approximaely 20% of he muual fund money indexed o he S&P while hey are no as large as he Vanguard Marke Index Trus ($64 Billion) heir asses are considerable. The Sparan U.S. Equiy Index is a no-load fund, he Sparan Marke Index has a 1/2 % charge for redempion wihin 90 days and he VIP Marke Index is sold principally hrough insurance channels. Also here is a srong difference beween he wo Sparan indexes in erms of minimum iniial invesmen ha is equal o 10,000 Dollars for he Sparan marke Index (1/2 % load fund) and equal o 100,000 Dollars for he Sparan U.S. equiy Index (no-load fund). The difference in iniial invesmen provides a way of esing invesors behavior according o heir wealh. We expec ha, in general, richer invesors will have a lower sensiiviy o shor erm marke swings 7

9 eiher because heir sensiiviy o risk is lower 5 or because heir invesmen horizon is longer, or because hey are beer informed on long run rends. This effec should overcome he one due o he absence of back-end fee radiionally considered in he lieraure (Chordia (1996)). And Indeed, we find ha ouflows are no sensiive o reurns in he fund wih he highes invesmen requiremen (Sparan U.S. Equiy Index) There are no precise numbers for how much muual fund money is indexed o he S&P 500, bu i is probably beween $80 o $100 billion as of mid The Sparan U.S. Equiy index is he second-larges S&P index fund, and The capializaion of he S&P in mid-1998 was abou $8,900 billion and so he Fideliy funds hold no more han abou 1/4 of a percen of he oal capializaion of he S&P porfolio. Never-he-less, hey may represen he marginal invesors in S&P 500 socks only a small porion of shares of socks in he S&P 500 (1/4 percen) urn over each day. 7 The correlaion in invesor flows across he Fideliy funds is high suggesing ha he iming of inflows and ouflows of hese funds is likely o be represenaive of he oher index muual funds. Even if he invesors in our sample were perfecly represenaive of he invesors in oher index muual funds, hey represen only a fracion of he equiy indexed o he S&P. A subsanial porion of insiuional money is indexed. To he exen ha insiuional managers 5 This can be explained if invesors have CARA uiliy funcion. 6 This esimae is based upon a lis of 19 index funds managing $78 billion as of May, 1998, lised by Index Funds Online (hp:// This is cerainly an underesimae, since only one of he hree funds in our sample is included in he lis. Of he 19 funds, Fideliy Sparan Marke Index Trus is he second larges, behind he Vanguard Marke Index Trus. CNNfn esimaed in June 1997 ha 59 S&P Index funds managed $83 billion in asses (hp://europe.cnnfn.com/yourmoney/9706/24/yomo_indexfunds/). 7 No daily urnover informaion for he S&P 500 socks is available, bu dollar-valued urnover informaion abou he NYSE is probably fairly represenaive since S&P 500 socks represen mos of he capializaion of he NYSE (more han 90%). NYSE capializaion as of 6/98 was $9,900 billion and rading aciviy for he firs six monhs of 1998 was $3,474 billion, suggesing 60% of share ransaced per year, or abou 1/4 percen. Thus he daily volume of he S&P 500 8

10 behave differenly from individual invesors, we migh expec he flows o muual funds o be only parly correlaed o flows by insiuions. Thus, we expec o find modes price effecs condiional upon fund flows. Share purchases, due o auomaic dividend re-invesmen, are reaed separaely. There is a srong increasing rend in he flows for each of he funds hrough ime. This rend is undoubedly due o boh he growing populariy of indexing as well as he effors by Fideliy o arac cusomers. I may also simply reflec increasing demand for muual funds in general and for equiies in paricular. Alhough his is an ineresing issue, in his paper we are concerned wih very shor-erm shifs in demand. Thus, i is imporan o conrol for hese long-erm effecs. In he analysis below, we normalize he flows o accoun for boh marke growh and fund growh. Marke growh is imporan since he oal capializaion of he S&P 500 grows over he inerval of sudy. A $100 million demand shock in he early par of he sample period canno be equaed wih a $100 million demand shock in he laer par of he period, given ha he S&P has averaged more han 20% reurn in he period. Simply dividing by he capializaion, however, does no accoun for he fac ha we are using he funds as proxies for daily demand shocks by oher index fund cusomers. While Fideliy s index funds are no as large as Vanguard s Marke Index Trus, paricularly early in he sample, we expec heir flows o have been correlaed, and hus o provide us informaion abou wha index cusomers behavior in general. Thus we also normalize by dividing sandardized flows by heir 180-day moving average. In effec, we examine he behavior of local deviaions in inflows and ouflows expressed as percenage marke share. mus currenly be over $20 billion. While volume canno be equaed exacly wih he percenage of he company ha 9

11 II.2 Sock marke variables: S&P 500 and S&P 400 spo and fuures We consider boh he spo and fuures oal reurn on he wo indexes, he S&P500 and he S&P400. The socks ha compose he S&P400 baske are approximaely he nex 400 bigges afer hose represened in he S&P500. Alhough no formally esed in his paper, we expec hem o be subjec o he same ype of sysemaic shocks and, on average, o reac o hem similarly due o similar beas 8. We rea fuures and spo prices as conemporaneous end-of-day observaions, however as Miller, Muhuswamy and Whaley (1994) poin ou, his may no be a well-founded assumpion, due o non-rading or delayed rading in he spo. They sugges ha he reversion in he fuures basis may be due no o arbirage, bu raher o sable spo prices. To address his, we separae he basis ino he spo and fuures prices and find evidence consisen wih heir asserion ha daily basis reversion is no arbirage-relaed. We also consider he oal volume of ransacions in he S&P500 index, he open ineres on he fuures, he marke capializaion and he implied volailiy. In paricular, implied volailiy is calculaed as he volailiy on he S&P500 index derived by he opions on he S&P fuures. Call opions a he money for he shores mauriy period are used. Boh open ineres and implied volailiy can be defined eiher as measures of invesors uncerainy or as measures of dispersion of beliefs. Open ineres, in paricular, has been idenified wih dispersion of beliefs. We will show how our analysis can lead o a reformulaion of such a definiion. In paricular we will show ha invesors reac more srongly o risk han o expeced reurn. This is changes hands in a given period, i does indicae he order of magniude. 8 The fac ha we consider indexes helps us as i eliminaes mos of he idiosyncraic effecs specific o any sock. (We hank R. Shiller for helpful discussion on his poin). 10

12 consisen wih recen resuls (Busse (1997)) ha, using a frequency of daa analogous o ours (daily), show how muual fund managers exploi his feaure explicily iming volailiy 9. II.3 Marke- Timing Newsleer Daa Forbes columnis Mark Hulber provided us wih a daase compiled by he Hulber Financial Diges covering 101 invesmen newsleer during he period June 1980-December This daabase has been sudied carefully in Harvey and Graham (1996). They find no evidence of marke-iming abiliy by managers bu Graham (1998) finds some evidence ha newsleers herd in heir recommendaions. This daabase allows us o address he issue of how analyss' repors affec financial markes. The "recommendaions" conained in he daase are defined as a proposed porfolio composiion, expressed as a percenage of socks vs. -bills o be held in he porfolio. The average recommendaion changes daily, alhough mos newsleers are published weekly or a longer inervals. Thus, even hough all newsleer publishers may have changed heir opinions a once on a given day, i akes abou a week for hese opinion changes o find heir way o prin. A, he same ime, he daase has been consruced in such a way ha each new observaion is added o he raw file he dae Hulber receives i in he mail or over he phone for leers wih free holines, raher han he dae published on he newsleer. This should properly accoun for he ime i akes o reach he invesors. To allow for addiional delay in he ime i akes o reach invesors (on op of he ime already allowed for o reach Hulber), we considered he average value of he recommendaions a and In paricular, using daily daa, Busse shows ha volailiy iming is an imporan facor in he reurn generaing process of he fund indusry as a whole, leading o varying degrees for individual funds and significanly conribues o radiional fund performance esimaes. 11

13 We use he sub-sample covering he period January 1993-December 1997 and consruc he average recommendaions by he newsleers as well as he cross secional sandard deviaion. Given ha he disribuions are skewed and he mean may no be he correc locaional measure o use, we also use he median recommendaion. III. Tess and Resuls The firs goal is o es if fund inflows move he index or if changes in he index deermine invesors flows or boh. To his purpose, we regress he S&P500 reurns on funds flows. In addiion, we reverse he regressions and examine he response of flows o lagged S&P 500 reurns. In addiion we run an explici es of causaliy o es he direcion of causaliy. Nex, we explore he poenial moivaion for rading by regressing inflows and ouflows on he newsleer recommendaion series. Finally, we consider he exen o which he S&P price changes due o demand shocks are permanen or emporary by using he changes in he S&P 500 fuures prices in place of spo and by considering wheher price reversals in he basis are due principally o he spo reversing or he fuure reversing. The logic of using he fuures price is ha i represens a fuure claim on he S&P. If changes are expeced o rever before he exercise dae, hen he fuures price will no reac o dynamics due o demand. All he specificaions have been esimaed using a Cochrane-Orcu correcion for auocorrelaion. Also, o check he robusness of our esimaions, all he funcional specificaions have been re-esimaed using a Whie-correcion for he errors o conrol for heeroskedasiciy and auocorrelaion. We checked if flows and reurns are unsaionary, using Dickey-Fuller ess for uni roos. The ess rejec a 99% confidence he exisence of uni roos for all he considered series. 12

14 III.1 Spo Reurns on Flows We are ineresed in seeing if here is any relaionship beween each fund s flows and marke reurns. To pursue his goal we look a boh he relaionship beween funds' flows and he S&P500 reurn, and he one beween flows and he difference beween S&P500 reurns and S&P400 reurns. The implied assumpion is ha if here is any srong underlying relaionship beween funds' flows and asse reurns, i should manifes iself mosly on he socks composing he index and no on he oher socks. We herefore conduc a se of linear regressions of boh S&P 500 reurns and of he difference beween S&P 500 and S&P 400 reurns on normalized conemporaneous inflows and ouflows from each of he hree funds. The specificaion of his es is sraighforward. Since he fund flows are only proxies for index fund flows in general, we do no allow for ime-variaion in coefficiens due o changes in fund size. We esimae he equaions: R = α + β * + ε I and R = α + β * + ε O where R is he daily reurn on he S&P, while I and O represen normalized daily fund inflows and ouflows respecively. We sar simply, focusing on he conemporaneous relaionships, even if i is conceivable here is some complex, lagged srucural relaionship among reurns and flows ha would demand a more sophisicaed specificaion. Tables 3-4 show he resuls of he firs se of regressions of S&P 500 reurns on fund inflows and ouflows. For consisency wih Warher (1995) we repor resuls for he flows normalized by he railing 160-day moving average of he S&P 500 capializaion. To es for he robusness of he resuls o he ype of sandardizaion echnique used, we re-esimaed all he 13

15 funcional specificaions using a differen sandardizaion procedure based on he daily innovaion componen of he flows. In paricular, we used he following sandardizaion procedure: SFlows RFlows MAFlows RMAMKT RMAMKT =. where SFlows are sandardized flows, RFlows are he raw flows, before sandardizaion, MAMKT is he rolling moving average of marke capializaion in he previous 160 days and MAFlows is he moving average of he flows in he previous 160 days. Given ha he resuls agree, we repor only one specificaion. Each fund is repored separaely. While he explanaory power of he regressions is modes 10, he resuls are consisen across funds -- he coefficien on conemporaneous flows is posiive and significan a he 5% level. The resuls show a posiive correlaion beween inflows and reurns and a negaive one beween ouflows and reurns. The resuls for he specificaion where he difference beween S&P 500 and S&P 400 reurns and funds' flows agree wih he previous ones, showing a posiive relaionship beween inflows and reurns differenials beween S&P500 and S&P400 index reurns and a negaive one beween hem and ouflows (Tables 22 and 23). We will come back more in deph in he end of he paper. 10 The low explanaory power of he regressions, boh when regressing marke reurns on flows and when regressing flows on pas reurns is due o he limied sample available. I is hard o expec ha he behavior of a single fund could affec he marke index, explaining a large porion of is variaion. This is confirmed by he fac ha, running he same regressions on he aggregae flows of he hree funds added ogeher, he explanaory power rises. On he oher hand, in erms of explaining funds inflows wih marke reurns, here are idiosyncraic effecs ha ge washed ou once more funds are pooled ogeher. We repor he disaggregaed resuls because hey provide an ineresing view on how invesors heerogeneiy (capured by differences in minimum invesmen requiremens among funds) affec heir invesmen behavior. 14

16 To es he direcion of causaliy more formally, we run a Geweke-Meese-Den (1983) es of causaliy, using marke reurns and flows ino he funds. The resuls are represened in Table 5. This es is robus o auocorrelaion and o misspecificaion in he deerminaion of he correc number of lags 11. Boh ypes of specificaions rejec he null of no-causaliy from flows o marke reurns, while fail o rejec he null of no-causaliy in he oher direcion (from marke reurns o funds inflows). Of ineres in is he lack of significance for he ouflows on he second fund. This is he fund wih a minimum iniial invesmen requiremen equal o 100,000 dollars. We expec ha he invesors in such a fund are more sophisicaed and wealhier. This should reduce heir sensiiviy o marke shor-erm swings and, herefore, he sensiiviy of ouflows o marke negaive reurns. These resuls are consisen wih Warher's evidence using aggregae flows a he monhly level for all muual funds. Wih he benefi of daily daa, however, we can rejec he hypohesis ha invesors are simply chasing he marke inra-day. One approach o measuring he effecs of demand shocks is based on he assumpion ha posiive feedback rading, if i exiss, canno be solely confined o one day -- feedback rading mus also happen on he day following a big marke move, if i occurred conemporaneously. To see if we can rejec posiive feedback rading a daily inervals, we nex regress flows on lagged reurns. III.2 Fuures Reurns on Flows 11 We consider boh he S&P500 reurns and he difference beween S&P500 and S&P400, as well as he aggregae flows for he hree funds 15

17 When new money flows in, he fund manager can eiher buy socks or ake a long posiion in he fuures marke. In he former case, his could drive up fuures prices. Given he low ransacion coss and high liquidiy of he fuures marke and he possibiliy of higher leverage ha such a marke offers, i is likely ha his can be a preferred direcion of invesmen for fund managers. Therefore, analogous o wha we did for he spo marke, we regress fuures reurns on conemporaneous inflows and ouflows, according o: R = α + β * + ε I in he case of he inflows, and R = α + β * + ε O in he case of he ouflows. We consider he reurns on he fuures on he S&P500, he difference beween such fuures and he spo index S&P500 and he difference beween he fuures on he S&P500 and he fuures on he S&P As wih he spo price, all he regressions are significan, showing a posiive correlaion beween inflows and fuures reurns and a negaive one beween ouflows and fuures reurns. This ells us ha he wo markes are linked eiher by fund manager aciviy or by arbirage across hem. We repor he key resuls in Tables 6-7. In paricular hree resuls are worh menioning. Firs, he daa ell us ha fund managers may inves a leas par of heir new money in he fuures markes aking long posiions. Bu he behavior in he fuures markes changes according o he naure of he flow. New inflows appear o affec firs he fuures marke. Therefore he reurn differenial beween fuures and spo marke is posiively correlaed o inflows. Ouflows, on he conrary, appear o induce he managers o firs liquidae heir spo posiions. This induces a posiive correlaion beween ouflows and he fuures-spo reurn differenials. In erms of 16

18 differenial wih respec o he S&P400 fuures, he resuls are more uncerain. They are significan only if we consider he oal inflows aggregaed and no, as we have done up o now, fund by fund. In his case hey show a posiive relaionship beween inflows and reurns differenial beween S&P500 and S&P400. III.3 Flows on Reurns To invesigae he effecs of pas reurns on he decision o inves in funds, we consider wo specificaions. The firs is a regression of inflows (ouflows) on conemporaneous reurns and pas (one-period lagged) inflows (ouflows). The second is a regression of inflows (ouflows) on muliple lagged reurns and inflows (ouflows). The firs specificaion is given by: I = + β * R + γ * I 1 α + ε for he inflows and: O = + β * R + γ * O 1 α + ε for he ouflows. Lagged flows are used o conrol for rend. Two specificaions are considered: one wih reurns on he S&P 500 index and one wih he difference beween hese reurns and hose on he S&P 400 index. Given ha he resuls agree we repor and describe he former, referring o he las secion of he paper for a proper descripion of he laer. In he second specificaion, we es he lagged influence of reurns on flows, esimaing he following equaions: I = α + N k = 0 β * R k N k + k = 1 δ k * I k ε 12 See noe 6. 17

19 for he inflows and O = α + N k = 0 β * R k N k + k = 1 δ k * O k ε for he ouflows. Here N represens he number of days we included in he regression. To keep he specificaion parsimonious and, a he same ime, o include he salien lags, we consider six lags (N=5). This covers he full rading week and picks up he reurns of he las day of he previous week. The resuls of he regression of flows on S&P500 reurns are repored in Tables 8 and They show a srong posiive correlaion of inflows on conemporaneous reurns and a negaive one beween ouflows and conemporaneous reurns. This specificaion confirms he resuls of he previous secions. I also allows us o beer conrol for rends in he flows by explicily modeling he emporal dependence by adding of he lagged dependen variable. Analogous resuls are repored for he relaionship beween funds flows and fuures marke reurns (no repored in he ables). Resuls for lagged reurns are repored in Tables 10 and 11 for he spo marke case and in Tables 12 and 13 for he fuures marke case. Noice ha lack of explanaory power of he regression and he absence of significance for he explanaory variables rule ou any causal relaionship from pas reurns o funds inflows. The behavior of he inflows for all he hree funds sugges ha when invesors decide o inves, hey do no chase he rend. In he case of he ouflows, on he conrary, a relaionship seems o exis for he firs and hird fund, while no relaionship is deecable for he second fund. This is jusifiable in erm of is peculiar fee 18

20 srucure. The exisence of a saisically significan relaionship beween ouflows and pas reurns and he lack of any relaionship beween hese and inflows, seems o sugges an asymmeric behavior by invesors who reac more quickly o bad han o good news. I is ineresing o noe ha lagged ne inflows (no repored in he ables) are never significan, eiher considering he hree funds separaely, or aggregaing he daa 14. This should sound as a warning owards ha par of he lieraure ha, lacking a separae availabiliy of daa on inflows and ouflows, has relied on ne flows and poin oward he need o separaely analyze inflows and ouflows. In general he resuls seem o sugges he lack of sophisicaed rend-chasing sraegies by he invesors, and he exisence of a cerain degree of over-reacion by risk-averse invesors o bad news. I is worh noicing ha hese resuls conras wih he resuls found in sudies wih lower frequencies (quarerly) where i seems ha invesors reac more o posiive reurns ha o negaive ones. Also, using daily daa, we hus confirm Warher s hypohesis ha feedback rading is no driving he relaionship beween monhly flows and reurns (a leas for he case of inflows). The higher frequency daa allow us o clearly pin down he direcion of causaliy, from inflows o marke reurns and no vice-versa. The resuls on he funds flows-fuures reurns in a following secion srongly confirms his hypohesis, and suggess ha he effec of fund share purchases and sales is upon he spo as well as he fuures marke. III.4 Explaining Fund Flows 13 We esed for he robusness of he esimaions, using differen sandardizaions of he flows. As he resuls agree we do no repor hem, bu are available upon requess from he auhors. 14 This fis wih he previous resuls ha, for he ess of Granger causaliy find a more significan relaionship for he ouflows han for he inflows. 19

21 Alhough we find no suppor for posiive feedback rading, he general quesion of wha moivaes he flows in and ou of index funds is an imporan one. We would expec ha mos of he inflows and ouflows are he resul of liquidiy rading by uninformed invesors. We examined he daa for evidence of seasonaliy a he monhly and daily levels o see if inflows occurred a regular inervals. We expeced, for example, ha end of year and beginning of monh inflows would be sronger due o deposi of paychecks and bonuses. We do no repor he resuls of his because we found virually no seasonal variaion in flows. The only evidence of seasonaliy we found was in he srengh of he coefficiens in he regressions hemselves. Flows responded more srongly in January, as well as on Mondays, for example. We do no know why. III.5 Risk and Uncerainy The analysis so far has mainly focused on he firs momen of reurns. We have shown, ha, unlike previous lieraure assumed, people reac more o negaive han o posiive performance, suggesing ha invesors behavior is srongly driven by risk aversion. We now go more in deph ino his aspec of invesors behavior, looking direcly a how invesors' behavior is affeced by risk. In paricular, we will focus on invesors reacion o risk. We will also look a how exper opinion (i.e. analyss repors) affec invesors behavior: he resuls indicae ha, even in his case, he second momen (i.e.: volailiy and dispersion of beliefs) affec invesors more han average reurns. Volailiy is a naural variable o use o explain flows, since we expec risk o influence invesor porfolio choice. No only do we have curren volailiy o use as an explanaory variable for flows, bu we also have measures of fuure volailiy. In addiion o risk measures per se we also have measures correlaed o "disagreemen" among paricipans in he financial 20

22 markes. In paricular, he radiional financial microsrucure lieraure usually idenifies dispersion of beliefs wih open ineres and risk wih implied volailiy. Our analysis offers us he opporuniy o reconsider such idenificaion. We herefore look a he relaionship beween flows and implied volailiy and open ineres on he fuures on he index. To conrol for spurious correlaion, open ineres is considered boh before and afer sandardizaion by he mean value of he marke capializaion. In paricular, we es: I = + β * R + γ * I 1 α + δoi + ε and O α + β R + γ * O + δ * OI + ε = * 1 where OI represens he open ineres on fuures on he S&P 500. Beyond checking he robusness of earlier resuls he objecive of his es o see how he decision o inves in index funds is affeced by he global level of uncerainy prevailing in he marke. For his second purpose we also use he implied volailiy, esimaing direcly: I = α + β + γ * IV + ε * I 1 and O = + β * O 1 α + γiv + ε In his case he implied volailiy (IV) direcly proxies for he level of marke uncerainy and riskiness. Tables repor he resuls of he regressions of flows on conemporaneous open ineres (Tables 14 and 15) and implied volailiy (Tables 16 and 17) of he S&P 500 index. The coefficiens on implied volailiy are always highly significan and posiive, for boh inflows and ouflows. A Granger causaliy es beween flows and volume and volailiy, (no repored in 21

23 he able) indicae a wo-way-causaliy beween hem. This would fi wih he inuiion ha fuure marke risk, as well as dispersion in curren beliefs, accenuae he index fund flows. On he conrary, open ineres affecs only ouflows. Also, alhough no repored, he added absolue values of daily inflows and ouflows are srongly correlaed o volume, volailiy and open ineres. These resuls seem o sugges ha open ineres could be a good proxy of marke riskiness, while implied volailiy would proxy for dispersion of beliefs. Indeed, higher riskiness should affec ouflows, wih a negaive or zero effec on inflows, while dispersion of beliefs should increase boh inflows and ouflows, as i increases boh he people who flock in he marke under he expecaion of higher reurns and he people who leave i, fearing lower reurns. And his is indeed wha he resuls sugges. We will see laer how dispersion in he beliefs of he financial analyss, again play a sronger role han heir average recommendaions. All his seems o porray a clear picure of invesors risk aversion, very differen from he one radiionally observed a lower frequency. III.6 Marke iming newsleers Nex, we look a he relaionship beween marke iming newsleers and invesmen flows ino he funds. Some fund flow is likely due o speculaive rading as opposed o liquidiy rading. Wha informaion do hese speculaors use? We consider a number of poenial informaion sources for speculaive rading: macro-economic daa and exper recommendaions. Alhough Graham and Harvey (1996) found lile evidence ha hese newsleers as a whole had any iming abiliy, hey never he less appear o generae self-susaining profis 22

24 hrough subscripions. In order o deermine he influence of exper opinion on he inflows and ouflows, firs we regress flows on he mean value of recommendaions: F = α + MR + Info + ε and on heir cross-secional dispersion: F = α + SR + Info + ε Here MR and SR are, respecively, he mean value of he recommendaions and heir dispersion (calculaed as sandard deviaion) 15. Boh he specificaions are esed alone and including addiional "informaion variables" (Info). These variables have been widely used in he lieraure. In paricular, a number of papers have poined ou ha macroeconomic variables may help explain variaions in he equiy premium in he shor run. Ferson and Schad (1995) poin ou ha performance evaluaion of fund managers should ake ino accoun he value of condiioning on macroeconomic condiioning. Graham and Harvey (1996) explicily es he iming abiliy of newsleers by condiioning upon macroeconomic variables. We use hese variables o conrol for oher addiional sources of informaion invesors can base heir behavior on. The naural candidaes we consider are he 3-monh ineres rae, he qualiy spread, he erm spread beween long and shor erm bonds and he dividend yield. I is a very srong es of he impac of he newsleers, as addiional explanaory power would imply an informaion se no direcly conained in ha of he oher exising variables and, herefore, superior capaciy as leading indicaor. The average of he conemporaneous and lagged mean value is used o capure effecs due o he lagged arrival of repors As menioned before, o properly accoun for he ime i akes o reach he invesors, boh MR and SR are calculaed as he average beween he values a ime and Alernaively, we also esimaed he same regressions he median value. Given ha he resuls for he mean and median recommendaions coincide we will refer only o hose using he mean. 23

25 The resuls are repored in Tables 18 and 19 for he average invesors recommendaions and in Tables 20 and 21 for he dispersion of analyss beliefs. They show ha recommendaions o inves ino socks increase inflows ino index funds, reducing ouflows. The resuls are srongly more significan for he ouflows (excep for he second fund for which he specific ype of exi fee affecs he resuls). The inflows are significan only for he firs fund. Ineresingly, he flows seem o be more srongly affeced by he dispersion of opinion among analyss, han by heir average recommendaions. In paricular a higher dispersion of newsleer opinions lowers he inflows and increases he ouflows. This confirms he earlier resuls ha risk averse invesors end o reac more o uncerainy and risk and less according o a reurn-chasing behavior. In general, analysis of he invesmen iming newsleer recommendaion seems o sugges ha he markes follow analyss' suggesions, increasing he invesmen ino he funds and reducing heir divesmens from hem, whenever analyss' repors "agree" on suggesing o inves ino socks. III.7 Permanen or Temporary Price Effecs? An imporan issue is wheher hese uninformed invesor demand shocks are emporary or permanen and if emporary, can hey be arbiraged by program raders. S&P index funds in general promise invesors end-of-day pricing. Tha is, hey promise invesors ha share redempions and purchases will be execued a same-day closing prices. We expec ha he funds hus minimize racking error by enering marke on close orders for he ne dollar value of heir flows. By is very naure, however, a marke-on-close price is no arbirageable agains he 24

26 fuures marke. The quesion is how his change of he marke-on-close price will affec fuure prices, ha is, if i will rever he following day. Prices are assumed o mean-rever if he informaion shock is a ransiory idiosyncraic shock, while no mean reversion akes place if he public flow of informaion is abou permanen news. Using flows and reurns we can es if he idiosyncraic demand shocks due o funds flows are permanen or emporary or permanen. If hey are emporary, we expec hem o rever over nigh or in he following days, originaing a negaive correlaion beween flows and fuure reurns. If he shocks are, insead, permanen hey should be incorporaed ino prices and no correlaion should be found in he following days. To es his hypohesis we looked a he relaions beween funds flows and index reurns. In paricular, we run he following regressions: R = α + N k = 0 β k * R N k + k = 0 δ k * I k ε and R = α + N k = 0 β * R k N k + k = 0 δ k * O k ε where R are he percenage daily reurns on he S&P 500, boh spo and fuures. For our experimens we consider hree inervals: he inerval marke opening-o-marke closing, he inerval closing of he previous day-closing of he following day, he inerval beween marke closing of he previous day and marke opening of he following day. Then we consider he relaionship beween flows and marke reurns for he successive N days, where N is assumed o be equal o 5, o consider all he weekly effecs. The raionale behind hese ess is he idea ha if here is mean reversion in reurns, we would expec o find correlaion beween pas flows and reurns. If, on he conrary, he demand shocks are permanen, no correlaion is 25

27 expeced beween pas flows and reurns. Therefore, esing he null hypohesis of no-mean reversion is equivalen o esing he hypohesis ha all he δs are equal o zero or no significan. The resuls (no repored as all of hem no saisically significan) hold for all he differen specificaions, wih boh spo reurns, fuures reurns and fail o rejec he null hypohesis of no-mean reversion. In paricular, hey show a srong conemporaneous relaionship beween flows and reurns, and no relaionship a all beween flows and fuure reurns. The lack of correlaion beween he overnigh price change and he flows in he previous day and beween hen flows and reurns in he following and he oher consecuive days (no repored in he ables) seem o confirm he hypohesis ha he shocks due o funds inflows are no emporary, bu are permanen. This es of marke reacion o liquidiy rading seems o rejecs he long-sanding ene ha liquidiy rading shocks mean rever and are emporary. One possible implicaion of he permanen naure of demand shocks is ha he growh of he sock marke during he las period of sudy may, in par, be a produc of demand, no economic fundamenals like dividend yields. To furher analyze his issue, we look a he relaionship beween he S&P500 and he S&P400 index. If index funds played any role, we would expec ha he difference in sock marke growh beween S&P500 and S&P400 be affeced by he behavior of he S&p500 index funds. As a firs preliminary es, we looked a he size and significance of he coefficiens represening he impac of flows in index funds on he S&P500 index, calculaed wih rolling regressions, over 600 day moving windows. Figure 1 repors he poin values of he coefficiens, as well as heir esimaed sandard deviaions. The increasing significance, as well as size of he impac of fund flows is sriking and can be due o he growing absolue size of he index fund indusry as well as o i increasing more imporan relaive role. 26

28 In Figure 2 and 3 hree indexes have been considered: S&P400 and S&P500 and a new index ha has been consruced deducing from he S&P500 index he par of i growh due o he impac of he ne (inflows ne of ouflows) index funds. The new index has been consruced calculaing he impac of he ne flows in index funds in he index, hrough a regression over he enire sample period (Figure 2) and hrough a series of rolling regressions of he index reurns on funds flows and using he esimaed coefficiens o calculae he par of he reurns due o he impac of funds flows (Figure 3). The resuling index has been consruced aking ou he cumulaed componen due o he funds. The figure is sriking: a large porion of he spread beween he wo indexes may be due o ne demand shocks. To es for he significance of he resuls, we run a series of boosrapping experimens. For he coefficien measuring he impac of fund flows on he S&P500 calculaed on he enire period (Figure 2) and for each coefficien calculaed for each single rolling regressions (Figure 3), we calculaed confidence bands boosrapped 5,000 imes and consrucing 95% inervals around he esimaed coefficiens. Using hese as new coefficiens, we hen calculaed he bands for he new consruced index (he so called S&P500 ne of fund effecs ). The bands are repored as Upper Band and Lower Band. The relaively narrow dimension of he bands wih respecs o he difference beween he new consruced index and he S&P ones, allows us o rejec he hypohesis ha he ne effec on demand on marke price level is zero. In shor, he imporan role played by he index funds is shown no only by he huge difference (-36%) beween he wo indexes ha can be explained in erms of funds flows (Figures 2 and 3), bu also by he increasing significance of he explanaory power of he coefficiens of funds flows on he index over ime (Figure 1). This sriking growh of 27

29 explanaory power coincides wih boh he fas growh of he sock index and he developmen of he index fund indusry and he growh in size of he hree funds we considered. To es his hypohesis more rigorously, we esimaed he following specificaions: N SR = α + β * NF ε k = 0 N k = 0 k k + FR = α + β * NF ε k k + We considered alernaive specificaions, where SR is he difference beween he reurns on he S&P500 spo and he reurns on he S&P400 spo index and FR is he difference beween he reurn on he S&P500 fuures and he S&P400 fuures index. Given ha we are looking for a direc effec of invesmen in index funds on marke reurns, we consider he ne flows (NF ia given by Inflows minus Ouflows). In Tables 22 and 23 are repored he resuls of he effecs of flows on he S&P500-S&P400 spo and fuures spread. The resuls confirm he inuiion ha he spread beween S&P500 and S&P400, boh spo and fuures is srongly influenced by flows in index funds. The more money flows ino index funds, he more hese influence marke reurns increasing he value of he socks par of he S&P500 baske. V Conclusions Our analysis of high frequency flow of funds daa for a se of large S&P 500 index funds srongly suggess ha invesor demand influences sock reurns bu no vice-versa. Using higherfrequency daa and being able o consider boh inflows and ouflows separaely, we show ha invesors are more affeced by risk han by performance. We show ha hey end no o chase posiive rends in reurns, while hey over-reac o negaive reurns immediaely liquidaing heir 28

30 posiions in he funds. This asymmeric behavior is he inverse of he one radiionally found in he muual fund indusry a lower frequencies and raises ineresing quesions. In paricular, i is worh invesigaing if he difference in resuls are due o a srucural difference in behavior beween muual fund and index fund invesors, suggesing some sor of clienele hypohesis o es, or if hese resuls are driven by he higher precision provided by daily daa. In he laer case, he whole radiional analysis of invesors behavior has o be a bes reformulaed. We also find some oher ineresing deerminans of invesor demand, including risk variables and exper recommendaions. An analysis of he dynamics of he S&P 500 fuures basis suggess ha he variaions due o demand shocks are largely permanen. Also, he analysis of he effecs of funds flows on he difference in behavior beween he S&P500 and he S&P400 indexes, boh spo and fuures sheds some ligh on he driving force behind he srong growh experienced by he sock exchange during he las decade. 29

31 References (1) Bakshi, G. S. and C. Zhiwu, 1994, Baby Boom, Populaion Aging and Capial Markes, Journal of Business 67, (2) Beneish, M.D. and R. Whaley, 1996, "An Anaomy of he 'S&P Game': The Effecs of Changing he Rules", Journal of Finance December, (3) Busse, J. 1998, Volailiy Timing in muual funds: evidence from daily reurns, N.Y.U. Sern Business School, Mimeo. (4) Campbell, J., Grossman, S. and J. Wang., 1992 Trading volume and serial correlaions in sock reurns, NBER Working Paper #4193 (5) Chan, L. and J. Lakonishok, 1995, The Behavior of Sock Prices Around Insiuional Trades, Journal of Finance, (6) Chan, L. and J. Lakonishok, April 1993, Insiuional Trades and Inraday Sock Price Behavior, Journal of Financial Economics, 33(2), (7) Chevalier, J. and G. Ellison, Augus 1995, Risk aking by muual funds as a response o incenives, NBER Working Paper #5234. (8) Chordia T., 1996, "The srucure of muual fund charges" The Journal of Financial Economics, 41: (1) (9) Consaninides, G., J. Donaldson and R. Mahra, 1998, "Junior Can' Borrow: A New Perspecive on he Equiy Premium Puzzle," NBER Working Paper #6617 (10) De-Long, J. B., A. Shleifer, L. H. Summers and R. J. Waldmann, June 1990, Posiive Feedback Invesmen Sraegies and Desabilizing Raional Speculaion, Journal of Finance, 45(2), (11) Den, J., Geweke, R. and W. Meese (1983) Comparing Alernaive Tess of Causaliy in Temporal Sysems: Analyic Resuls and Experimenal Evidence Journal of Economerics, 21(2), pages (12) Ferson, W. and R. Schad, 1995, Measuring Fund Sraegy and Performance in Changing Economic Condiions, Journal of Finance (13) Ferson, W. and V. A. Warher, 1995, Evaluaing Fund Performance in a Dynamic Marke, Financial Analyss Journal. 30

32 (14) Fridson, M. S. and Jonsson, J. (1995), Spread versus Treasuries and he riskiness of highyield bonds, The Journal of Fixed Income, December, (15) Garry M. and W. Goezmann, 1986, Does de-lising from he S&P500 affec sock price?, Financial Analyss Journal, 42(2) (16) Graham, J.R and C., R.Harvey (1996) Marke Timing Abiliy and Volailiy Implied in Invesmen Newsleers' Asse Allocaion Recommendaions Journal of Financial Economics, 42(3), (17) Harris, L. and Gurel, 1986, "Price and Volume Effecs Associaed wih Changes in he Sandard and Poors 500 Lis -- New Evidence for he Exisence of Price Pressures" Journal of Finance 41: (4) (18) Lakonishok, J. and A. Shleifer, Augus 1992, The Impac of Insiuional Trading on Sock Prices, Journal of Financial Economics, 32(l), (19) Lakonishok, J., 1991, Window Dressing by Pension Fund Managers, American Economic Review, 81(2), (20) Lakonishok, J., A. Shleifer, and R. W. Vishny, 1991, Do Insiuional Invesors Desabilize Sock Prices? Evidence on Herding and Feedback Trading, NBER Working Paper: (21) Lynch, A. W. and R.R, Mendenhall, 1997, "New evidence on sock price effecs associaed wih changes in he S&P 500 index", Journal of Business 70: (3) (22) Shiller, R, (1998) Commen on Vincen Warher, "Has he Rise of Muual Funds Increased Marke Insabiliy?" Brooking-Wharon Papers (23) Shleifer, A., July 1986, Do Demand Curves for Socks Slope Down?, Journal of Finance, 41(3), (24) Sias, R.W. and Sarks, L.T. (1996) Reurn Auocorrelaion and Insiuional Invesors, Journal of Financial Economics, 46, (25) Sirri, E. R. and P. Tufano (1993), Buying and selling muula funds: flows, performance, fees and service, Working Paper, Harvard Business School. (26) Warher, V. A., 1995, Aggregae Muual Fund Flows and Securiy reurns, Journal of Financial Economics, 39, (27) Whaley, R. E. and Messod, 1996, D. An Anaomy of he "S&P Game": The Effecs of Changing he Rules, Journal of Finance; 51(5),