DO DEMAND CURVES FOR SMALL STOCKS SLOPE DOWN? Arnold R. Cowan Iowa State University. JEL Classification: G12, G14

Transcription

1 DO DEMAND CURVES FOR SMALL STOCKS SLOPE DOWN? Ernes N. Bikimirov Brock Universiy Arnold R. Cowan Iowa Sae Universiy Bradford D. Jordan * Universiy of Kenucky JEL Classificaion: G12, G14 Published in The Journal of Financial Research Vol. XXVII, No. 2 Pages Summer 2004 * We hank Adiya Kaul, Joseph Kushner, Rick Mendenhall, and Jeff Wurgler for helpful commens and suggesions.

2

3 Absrac Socks added o he S&P 500 generally experience posiive abnormal reurns following he announcemen. Several compeing explanaions exis for his reacion, bu small sample sizes and oher issues make i difficul o disinguish among hem. We examine his subjec using he small-cap Russell 2000 index, which has a number of advanages over he S&P 500 in his conex. Our primary finding is ha socks added o or deleed from he Russell 2000 experience significan changes in sock price and rading volume, bu he effec is ransiory. The resuls suppor he price pressure hypohesis.

4

5 I. Inroducion We examine securiy reurns, rading volume, and insiuional ownership for firms ha are added o and deleed from he Russell 2000 index from 1991 o We have wo main goals. Firs, a large body of lieraure examines sock price reacions for large-cap socks added o he S&P 500 index. In conras, few sudies examine small-cap indexes such as he Russell 2000; herefore, his sudy appears o be one of he firs o examine exensively he effec of index lising on smaller socks. Second, numerous heories have been advanced o explain he posiive abnormal reurns observed for socks added o he S&P 500, bu limied sample sizes and oher issues make i difficul o disinguish among hem. In conras, he Russell 2000 is reformulaed on a regular basis wih hundreds of changes every year, hereby allowing for much larger samples. The Russell 2000 index is a subse of he Russell 3000, a value-weighed index of he 3,000 larges U.S. socks. The Russell 3000 is divided ino he Russell 1000, which conains he larges 1,000 socks, and he Russell 2000, which conains he remaining 2,000 smaller socks. The Russell 2000 small-cap index is widely followed, and i is he sandard benchmark for evaluaing he performance of small-cap muual funds. I is also he underlying index mimicked by funds such as he Vanguard Small-Cap Index Fund. The way he Russell 2000 is formed is imporan. A company is added o he Russell 2000 because is marke value falls and i is shifed ou of he Russell 1000, or because is marke value rises sufficienly. Similarly, a company is deleed from he Russell 2000 if is marke cap eiher rises or falls sufficienly. In our analyses, we focus primarily on socks ha are ruly added o or deleed from he Russell 2000 as opposed o being shifed ino or ou of he Russell Based on a sample of 4,321 socks ha are added, we find evidence of significan posiive

6 abnormal reurns in he period surrounding reconsiuion. From day -20 o day 0, where day 0 is he reconsiuion dae, he cumulaive abnormal reurn is 1.89% (significan a he.0001 level). However, negaive abnormal reurns are observed pos-even, and he overall effec is essenially zero. For 3,092 firms ha are deleed, we find a significan negaive cumulaive abnormal reurn of -3.47% (significan a he.0001 level) over he (-20, +0) window. These negaive abnormal reurns are offse in he days following he even, and he ne effec is boh economically and saisically negligible. Overall, we find lile or no evidence of permanen effecs on he values of firms ha are added o or deleed from he Russell 2000 index. In ligh of he relaively large samples we consider and he highly ransiory naure of he abnormal reurns, our resuls suppor he pricepressure hypohesis (Harris and Gurel 1986). Of he various heories inended o explain sock price reacions o index inclusion or deleion, his is he only one ha predics a purely emporary effec. In essence, index inclusion creaes a emporary excess demand for added socks. The reverse is rue for deleions. However, once all porfolio rebalancing is complee, prices reurn o equilibrium levels. Our conclusions regarding he ransiory naure of he Russell 2000 effec are reinforced by our analysis of abnormal rading volume. For boh addiions and deleions, here is a clear (and significan) spike in rading volume on he reconsiuion dae. For he addiions group, abnormally high volume exiss for several days before and afer he reconsiuion dae, reaching a maximum of almos 40% above prediced on day 0. In conras, for deleions, day 0 is he only day wih significan posiive abnormal volume. The possibly permanen effecs we find from he reconsiuion of he Russell 2000 concern insiuional ownership. Firms ha are added o he index gain, on average, eigh 2

7 insiuional owners, represening an increase of abou 20%. Percenage ownership by insiuions rises by 4%. Boh he change in he number of insiuional owners and he percenage held by insiuions are significan. The opposie paern is observed for firms ha are deleed. Thus, here is evidence of a porfolio rebalancing beween insiuional and noninsiuional invesors ha occurs because of he reconsiuion of he Russell 2000 index. However, our conclusions regarding changes in insiuional ownership are empered by he possibiliy ha such ownership is relaed o size, a leas in erms of he number of owners, and firms are added o, or deleed from, he Russell 2000 based on changes in size. II. Hypoheses As numerous sudies documen, posiive, significan abnormal reurns are associaed wih he announcemen of new addiions o he S&P 500 index (e.g., Shleifer 1986; Beneish and Whaley 2002). A number of compeing hypoheses are offered o explain sock marke reacions o changes in he S&P 500 index. These hypoheses differ from each oher no only in he reason for abnormal reurns bu also in heir prediced duraion. They can be grouped ino four basic caegories as follows: 1. The price pressure hypohesis, advanced by Harris and Gurel (1986), is he only heory ha predics a emporary increase in he price of a newly added sock. According o his hypohesis, for socks added o he S&P 500, he inflow of buy orders by large index funds pushes he price of a sock above is equilibrium level. Passive sellers are araced by he price increases caused by he increased demand. These passive suppliers of liquidiy are compensaed for heir liquidiy service when prices rerea o heir fullinformaion levels. 3

8 2. Under he imperfec subsiues (or downward-sloping demand curve) hypohesis, formalized by Shleifer (1986), as soon as one of wo similar socks is added o he index, hese socks are no longer close subsiues for each oher. Index funds place excess demand on he sock added o he index, and, assuming downward-sloping demand curves, equilibrium prices rise when demand curves shif o compensae. 3. Under he cos reducion hypoheses, socks added o he S&P 500 become less expensive for invesors o rade, boh in erms of bid-ask spreads as liquidiy increases (Amihud and Mendelson 1986) and he coss of acquiring qualiy informaion as analys following grows (Goezmann and Garry 1986). These cos reducions lead o increases in value for added socks. 4. The informaion signaling hypohesis (e.g., Jain 1987) is based on he fac ha he S&P Index Commiee expressly exercises judgmen regarding financial soundness in choosing companies o avoid excessive urnover in he index. Assuming ha S&P has superior informaion and experise in evaluaing companies gained from raing bonds, is selecions may reveal new and valuable informaion o he marke. In each case, he auhors of hese sudies presen evidence ha is generally consisen wih he respecive hypoheses. However, i is difficul o disenangle he compeing explanaions. Kaul, Mehrora, and Morck (2000) provide he cleares evidence o dae by sudying he effec of a preannounced increase in he weighs for 31 socks in he Torono Sock Exchange (TSE) 300 index in They find evidence of a significan, nonransiory 2.3% average excess reurn during he even week. Because he changes o he index were known in advance, he even sudied by Kaul, Mehrora, and Morck was free of informaion effecs. Furher, because he socks were already in 4

9 he index, he poenial cos reducions are no likely o be significan, and, in fac, Kaul, Mehrora, and Morck find no change in spreads. Finally, conrary o he price pressure hypohesis, he gains were no reversed. Thus, he evidence in Kaul, Mehrora, and Morck suppors he imperfec subsiues hypohesis. While he even sudied by Kaul, Mehrora, and Morck has a number of advanages, here are noneheless some poenial concerns. The sample size is small, he even was a one-ime occurrence, and, consequenly, he even dae is he same for all 31 socks. All of he socks had heir weighs increased, so he effec of a decrease could no be sudied. Finally, previous sudies have examined he effec of being added o or deleed from an index, as opposed o a shif in weighs. Sricly speaking, an addiion or deleion is a change in weighs (eiher from or o zero), bu i is noneheless conceivable ha marke reacion could be differen for he wo ypes of changes. As a means o discriminae furher among he compeing hypoheses, we consider he Russell The Russell 2000 small-cap index is widely followed, and i is he sandard benchmark for evaluaing he performance of small-cap muual funds. I is also he underlying index mimicked by funds such as he Vanguard Small-Cap Index Fund. The Russell 2000 has a number of advanages over he S&P 500 for research on he effec of index membership. Firs, addiions o he S&P occur irregularly and infrequenly. In conras, he Russell 2000 is reconsiued on an annual basis (since 1990), and abou 500 companies are replaced in he index each year. Much larger sample sizes are hus available. Furhermore, mos S&P 500 sudies focus on new addiions only because he majoriy of S&P 500 deleions involve 5

10 companies ha reorganize, merge, or go bankrup. 1 However, many companies ha are removed from he Russell 2000 each year say in business. A second advanage o he Russell 2000 is ha i is formulaed by simply ranking all U.S. companies based on heir marke value on May 31 each year. However, reconsiuion of he index does no occur unil June 30, a monh afer he members are deermined. Thus, here is no poenial announcemen effec on he reconsiuion day since changes are generally known. 2 Furher, if here are liquidiy enhancemens or decreased informaion coss for socks added o he index, prices should adjus well in advance of he official reconsiuion dae. The Russell 2000 has a hird advanage in ha a sock is added eiher because is relaive marke cap grew sufficienly or else because is relaive marke cap shrank, in which case i is shifed ou of he Russell 1000 index. Similarly, companies are deleed eiher because hey become oo big, and are hus shifed o he Russell 1000, or because hey become oo small. As a resul, i is possible o separaely sudy new lisings and delisings, as well as shifs beween indexes. Also, relaive o he case of TSE weigh adjusmens considered by Kaul, Mehrora, and Morck (2000), he Russell 2000 offers some significan poenial benefis. As discussed in he nex secion, our sample size is considerably larger, and we have muliple even daes. We are able o sudy separaely index shifs as well as rue addiions and deleions o he index. Finally, 1 Several sudies examine deleions from he S&P 500 (e.g., Goezmann and Garry 1986; Beneish and Whaley 2002). The larges deleions sample among hese sudies is 49, compared o over 3,000 considered in his sudy. 2 Since 1989, S&P has had a policy of announcing changes a leas five days in advance, when possible, so he Russell 2000 is no unique in his regard. Unlike he S&P 500, however, addiions and deleions occur on regularly scheduled daes, and he companies added or eliminaed can be predicaed wih relaively high precision even before he announcemen dae. 6

11 he reconsiuion of he Russell 2000 is a regularly scheduled even, as opposed o a one-ime even, and i is very well known and closely moniored by marke paricipans. 3 III. Sample Selecion This sudy exends from 1991 hrough Socks ha are added o or deleed from he Russell 2000 are placed ino four groups as follows: 1. Pure addiions are socks added o he Russell 2000 index ha were no previously in he Russell Pure deleions are socks deleed from he Russell 2000 index ha are no shifed o he Russell Upward shifs are socks ha were shifed from he Russell 2000 o he Russell Downward shifs are socks ha were shifed from he Russell 1000 o he Russell In each year of he period, we examine every addiion o and deleion from he Russell 2000 index and classify each ino one of he four groups. For a company o be included in he sample, i mus be possible for us o deermine unambiguously wheher or no he sock was a member of he Russell 1000 before i was added o or afer i was deleed from he Russell Cases in which i is unclear because of evens such as mergers, name changes, and delisings are discarded. 3 To our knowledge, no published sudies exis using he Russell 2000, bu we have recenly become aware of hree working papers (Reed, 2000; Madhavan, 2001; and Chen, 2002). The objecives of hese sudies are comparable o ours, bu we provide a more comprehensive analysis. Firs, we use a more sophisicaed even-sudy mehod (Madhavan uses raw reurns; Chen and Reed use simple marke-adjused reurns). Second, wo of hese papers (Madhavan and Reed) do no disinguish beween shifs and pure addiions or deleions. Finally, we examine daily rading volume (only Reed does his) and changes in insiuional ownership (only Chen examines his issue). 7

12 Because many companies are added o and deleed from he Russell 2000 each year, our sample sizes are large. For pure addiions and pure deleions, we have 4,321 and 3,092 observaions, respecively. The corresponding numbers for upward shifs and downward shifs are 861 and 875, respecively. The reason we have many more pure addiions han pure deleions is ha deleions are frequenly due o evens such as mergers and bankrupcies, in which cases he necessary daa are no available. In our empirical analyses, we presen resuls for all four groups in all cases. However, in erms of he compeing hypoheses we discussed previously, i is no clear wha is prediced for he upward shifs and downward shifs groups since socks are simply moved from one index o anoher. For his reason, we will focus primarily on he pure addiions and deleions groups in our discussions. IV. Abnormal Reurn Analysis We sudy reurns behavior in he period surrounding reconsiuion of he Russell 2000 index. Because of he way he index is formed, a sandard even sudy approach using a pre-even esimaion period o esimae marke model parameers is problemaic. Specifically, for he pure addiions and he upward shif groups, socks in he samples are more likely o have experienced a recen run-up in value han socks in he overall populaion. The reverse is rue for he pure deleions and downward shifs groups. Beyond his, a pre-even esimaion period decreases he sample size for pure addiions because many of hem are recen iniial public offerings. A pos-even esimaion period inroduces a survivorship bias for deleions from he index and decreases he sample size for deleions. Finally, he use of simple marke-adjused reurns may no produce powerful ess. 8

13 To address hese problems, we follow Hillion and Vermaelen (2001) by combining he Ibboson (1975) reurns across ime and securiies (RATS) approach wih he Fama-French (1993) ime-series model of he evoluion of securiy reurns. To do his, we run he following cross-secional regression on each day in even ime: R j - R f = α + β (R m - R f ) + s SMB + h HML + ε j. (1) For day, R j is he rae of reurn of he j h sock, R f is he one-monh Treasury bill rae, and R m is he Cener for Research in Securiy Prices (CRSP) value-weighed reurn on all NYSE, AMEX, and Nasdaq socks. The Fama-French (1993) marke-wide facors SMB (small minus big) and HML (high minus low) are consruced using six value-weighed sock porfolios formed on size and book-o-marke, respecively. SMB is he average reurn on he hree smallsock porfolios minus he average reurn on he hree big-sock porfolios. HML is he average reurn on he wo value-sock porfolios minus he average reurn on he wo growh-sock porfolios (Fama and French 1993 provide greaer deail on he consrucion of he facors). The esimae of he inercep α represens he average abnormal reurn for day. The RATS approach wih Fama-French facors has several advanages: 1. I does no require a separae esimaion period, hereby reducing he impac of any run-up or run-down bias. 2. I conrols for size and book-o-marke effecs. 3. I provides a large sample size compared o mos oher esimaion mehods. Conrolling for size and book-o-marke effecs may be paricularly imporan in sudies of he Russell 2000 index simply because he socks examined are small and may have book-o-marke characerisics ha differ from he populaion as well. For example, socks ha fall from he 9

14 Russell 1000 o he Russell 2000 have suffered declines in marke cap and may hus have relaively low marke-o-book raios. Table 1 conains he resuls of he even sudies for all four groups. Day 0 is he reconsiuion day. The average abnormal reurns ( AR ) are in Panel A, and he cumulaive abnormal reurns ( CAR 1, 2 ) are in Panel B. For he pure addiions group, here is a clear clusering of posiive abnormal reurns in he period surrounding reconsiuion. As shown in Panel A, hese firms gain, on average, an abnormal.92% ( = 4.36) on he reconsiuion day. From Panel B, he cumulaive abnormal reurn over he 21-day period running from day -20 o day 0 is 1.89% ( = 4.45). By way of comparison, earlier sudies of he S&P 500 generally find abnormal reurns of approximaely 3% for addiions; however, one recen sudy (Beneish and Table 1 abou here Whaley 2002) suggess ha he effec is more han wice his size over he period. For he pure addiions group, Panel B of Table 1 also shows cumulaive abnormal reurns are generally negaive following reconsiuion. In fac, for he 20-day period exending from day +1 o day +20, he cumulaive abnormal reurn is -1.63% ( = -3.96). Thus, he posiive abnormal reurns observed for he (-20, 0) window are almos precisely offse in he (+1, +20) window. The cumulaive abnormal reurn over he (-20, +20) window is an insignifican.26% ( =.44). 4 Turning o he pure deleions group, negaive abnormal reurns are generally observed on he days preceding reconsiuion. On he reconsiuion day alone, however, he average 4 We have performed robusness ess, including an analysis of possible even-induced shifs in variance, using a variey of procedures, and we find resuls similar o hose repored here. 10

15 abnormal reurn is an insignifican -.11% ( = -.35). Panel B shows ha he cumulaive abnormal reurns for his group are generally posiive following reconsiuion, and, as wih he pure addiions group, he pre-even abnormal reurns end o be offse in he pos-even period. In conras, firms deleed from he S&P 500 generally experience negaive and significan abnormal reurns, alhough he magniude varies from sudy o sudy because of small sample sizes. The abnormal reurns for he wo pure groups are ploed in Figure I for he (-20,+20) window. As shown, in boh cases, i is clear ha reconsiuion has an effec, bu, as indicaed by our analysis in Table 1, i is equally clear ha he effec is ransiory. In ligh of he large sample sizes, he almos complee absence of a permanen effec consiues evidence in favor of he price pressure hypohesis. Figure I abou here As previously discussed, we have no paricular predicions for firms ha are shifed beween he Russell 1000 and 2000 because hese firms merely exchange one index affiliaion for anoher, bu we repor he resuls for hese groups for he sake of compleeness. Table 1 shows ha he upward shifs group experiences relaively small negaive abnormal reurns preeven and similarly small posiive reurns pos-even. There is some evidence of a modes, permanen increase in value. For example, he CAR over he relaively long (-40, +40)-day window is 2.59% ( = 1.87). Similarly, he downward shifs group appears o experience a small, nonransiory decrease in value. The CAR over he (-40, +40)-day window is -2.82% ( = -2.16). The resuls for he wo shifs groups provide weak evidence for permanen effecs. Why his occurs is a maer of conjecure given he exising lieraure. More imporanly for us, he 11

16 fac ha he pure groups and he shif groups have somewha differen behavior means ha i is imporan o separae hem, paricularly for he pure addiions and he downward shif groups. V. Trading Volume Analysis We analyze rading volume around he reconsiuion dae using procedures similar o hose in Campbell and Wasley (1996). Our purpose is o deermine if abnormal volume is associaed wih he abnormal reurns found in our previous secion. A brief descripion of our mehods for calculaing abnormal volume follows. Firs, we calculae he log-ransformed percenage of shares ousanding: 5 100ni, V i, = ln( ), (2) S i, where n i, is he number of shares raded for securiy i on day, and S i, is he number of shares ousanding on day. Nex, we esimae marke model abnormal rading volume as: 6 AV i, i, ( i i m, = V α + β V ), (3) where α i and β i are compued via ordinary leas squares. The marke model volume for day is calculaed by using all Nasdaq socks excluding American Deposiary Receips (ADRs): V N =, (4) 1 m, Vi, N i= 1 5 Ajinkya and Jain (1989) and Cready and Ramanan (1991) show he imporance of log ransformaion of he volume daa o approximae a normal disribuion. As in Cready and Ramanan, we add o he daily percenage of shares ousanding o accommodae zero volume. 6 The esimaion period runs for 200 days from -479 o -280 days before he reconsiuion day, and abnormal volumes are calculaed for he 81 days cenered on he reconsiuion day. 12

17 where N is he number of Nasdaq socks excluding ADRs on day. For robusness checks, we use differen es saisics, esimaion periods, and proxies for he marke. For example, we use boh a parameric -es and a nonparameric rank es. We repor only he rank es saisic because Campbell and Wasley (1996) find ha he nonparameric es saisic is always more powerful in deecing abnormal rading volume han he parameric es saisic. In addiion o he pre-even esimaion period, we also invesigaed a pos-even esimaion period saring 280 days afer he reconsiuion day. As for he proxy for marke rading volume, we also use he combined volume of Nasdaq, NYSE, and AMEX socks. Neiher he esimaion period nor he choice of overall marke volume affecs he general resuls. To save space, he main resuls, which are repored in Table 2, use he pre-even esimaion period and Nasdaq socks in he calculaion of he marke rading volume. 7 Table 2 abou here As shown in Table 2, all four groups exhibi significan increases in rading volume on he reconsiuion day. The change in rading volume ranges from he smalles significan increase of 12.66% (r = 3.10) for he pure deleions group o he larges significan increase of 38.43% (r = 5.52) for he pure addiions group. The pure addiions group experiences significan posiive abnormal volume (a he 5% level or beer) every day from -3 o +4. In conras, he pure deleions group has posiive abnormal volume only on he even day. On every oher day, abnormal volume is negaive and insignifican. 7 The reason we consider muliple definiions of marke volume is he well-known issue concerning he abulaion of NYSE versus Nasdaq volume due o he double couning of rades ha involve a marke maker/specialis. 13

18 Table 2 also shows ha socks ha are shifed o he Russell 1000 experience significan posiive abnormal volume for several days surrounding he even day, and abnormal volume is posiive on every day we examine. Socks ha shif downward experience a very brief period of abnormal posiive volume; oherwise, hey generally display insignificanly negaive values. VI. Insiuional Ownership In his secion, we examine changes in insiuional ownership for firms ha are added o or deleed from he Russell 2000 index. For he mos par, our mehods are an exension of hose used by Prui and Wei (1989) in heir analysis of he S&P 500. We collec daa on insiuional holdings from S&P s Securiy Owner s Sock Guide. Because hese daa mus be hand-colleced, an analysis of our full sample of over 9,000 company-evens would be very ime consuming o prepare. To creae a more manageable experimen, we randomly seleced 30 companies for each group each year during he period, giving us a relaively large sample of 1,200 companies, wih 300 in each of he four groups. We calculae he average of boh insiuional share holdings and he number of insiuional invesors in he April and May preceding reconsiuion. We calculae he same averages in July and Augus following reconsiuion. We hen es for significan differences. Specifically, he percenage of shares held by insiuional invesors before reconsiuion, PNHOL i, pre, and subsequen o reconsiuion, PNHOL i, pos, are calculaed for each firm i as follows: PNHOL, = i pre INSHOL SHROUT i, pre i, pre and PNHOL, = i pos INSHOL SHROUT i, pos i, pos, (5) 14

19 where INSHOL i, pre and i pre SHROUT,, and INSHOL i, pos and SHROUT i, pos, are he insiuional holdings and oal shares ousanding for he pre- and pos-reconsiuion periods, respecively. To ge a sense of he exen of insiuional holdings, Table 3 presens some descripive saisics for he percenages of insiuional share holdings and numbers of insiuional invesors before and afer reconsiuion for he four groups for he wo pre-reconsiuion monhs, April and May, and wo pos-reconsiuion monhs, July and Augus. The percenages of insiuional share holdings are in Panel A, and he numbers of insiuional invesors are in Panel B. Table 3 abou here As shown in Table 3, he larges changes occur for he pure addiions group. Average insiuional ownership in hese companies rises from 27% o 32% (Panel A), and he average number of insiuions grows from 36 o 48 (Panel B). The pure deleions group experiences a decline in boh quaniies. Percenage ownership falls from 32 o 30, and he number of insiuions drops from 44 o 40. Table 3 also highlighs a noiceable difference beween he pure and shif groups. Firms in he shif groups have roughly hree o four imes as many insiuional owners, presumably because of heir greaer size. Percenage ownership by insiuions is roughly 50% greaer as well. Firms ha are shifed o he Russell 1000 gain in insiuional ownership, while firms ha shif o he Russell 2000 experience a decrease. To evaluae he significance of he changes illusraed in Table 3, we calculae he difference beween he percenage of insiuional holdings before reconsiuion (using he April/May period) and he percenage held afer reconsiuion (using he July/Augus period). This difference is calculaed for each firm in he sample and hen averaged across all firms 15

20 wihin a group. Sandard -ess and sign ess are used o evaluae differences in he mean beween he pre- and pos-reconsiuion periods. The same analyses are performed using he number of insiuional owners. The resuls are repored in Table 4. Table 4 abou here For he pure addiions group, Table 4 repors ha he mean increase in he percenage of insiuional holdings, 3.89%, is significan ( = 8.38), and a similarly significan 71.33% of he firms in his group regiser an increase (z = 7.39). The mean number of insiuional invesors rises by 8.37, and 78.33% of he sample regisers an increase, boh of which are significan a he.0001 level. In broader erms, Table 4 shows ha every change examined is saisically significan. Taken ogeher, he resuls in Tables 3 and 4 show ha insiuional holdings end o change around index reconsiuion. Pure addiions and upward shifs gain insiuional invesors, while pure deleions and downward shifs lose hem. A parial explanaion for hese findings may have o do wih changes in firm size. One consisen resul in Table 4 is ha firms ha increase in value and hereby eiher ener he Russell 2000 or move o he Russell 1000 also experience increased insiuional ownership. The reverse is rue for firms ha decrease in value. I is plausible ha he number of insiuional owners would vary direcly wih firm size; however, i is less clear ha he percenage of he sock owned would also behave his way. Furhermore, he changes in insiuional ownership in Table 4 occur over a relaively shor ime (April/May o July/Augus), and mos of he firms in our sample will no have experienced a paricularly dramaic shif in size over ha inerval. 16

21 VII. Summary and Conclusions A significan body of research documens ha socks added o he S&P 500 index generally experience posiive abnormal reurns following he announcemen. Various hypoheses have been advanced o explain his reacion, bu, o dae, small sample sizes and oher issues make i difficul o disinguish among hem. In an effor o disenangle he compeing explanaions, we invesigae securiy reurns, rading volume, and insiuional ownership for firms added o or deleed from he small-cap Russell 2000 index. This index has a number of advanages over he S&P 500 in his conex. Firs, addiions o he S&P occur irregularly and infrequenly, bu he Russell 2000 is reconsiued on an annual basis (since 1990), and abou 500 companies are replaced in he index each year. Furhermore, mos S&P 500 sudies focus on new addiions only because deleions usually involve companies ha reorganize, merge, or go bankrup. However, a large number of companies ha say in business are removed from he Russell A second advanage o Russell 2000 is ha i is formulaed by simply ranking all U.S. companies based on heir marke value on May 31 each year. However, reconsiuion of he index does no acually occur unil June 30, a monh afer he members are deermined. Thus, here is no poenial announcemen effec on he reconsiuion day since changes are generally known. Furher, if here is enhanced liquidiy or decreased informaion coss for socks added o he index, prices should adjus well in advance of he official reconsiuion dae. Anoher imporan aspec of he Russell 2000 is ha a sock is added eiher because is relaive marke cap grew sufficienly or else because is relaive marke cap shrank, in which case i is shifed ou of he large-cap Russell 1000 index. Similarly, companies are deleed eiher 17

22 because hey become oo big, and are hus shifed o he Russell 1000, or because hey become oo small. Thus, we are able o sudy wo differen ypes of addiions and deleions. Our primary finding is ha significan changes in prices, rading volume, and insiuional ownership occur for socks added o or deleed from he Russell The effec depends on wheher a sock is a pure addiion or deleion as opposed o being shifed ino or ou of he Russell In large samples of pure addiions (n = 4,321) and pure deleions (n = 3,092), we find clear evidence of a significan price and volume reacion. Pure addiions gain value around he reconsiuion dae on abnormally high volume and also experience a significan increase in insiuional ownership. Similarly, pure deleions lose value, also on abnormally high volume. However, boh he price effecs and volume effecs are shor-lived and ransiory. This resul srongly suppors he price pressure hypohesis, indicaing ha he effec of index inclusion or deleion is simply a emporary even wih no permanen valuaion effec. 18

23 References Ajinkya, B. B. and P. C. Jain, 1989, The behavior of daily sock marke rading volume, Journal of Accouning and Economics 11, Amihud, Y. and H. Mendelson, 1986, Asse pricing and he bid-ask spread, Journal of Financial Economics 17, Beneish, M. D. and R. E. Whaley, 2002, S&P 500 index replacemens, Journal of Porfolio Managemen 29, Campbell, C. J. and C. E. Wasley, 1996, Measuring abnormal daily rading volume for samples of NYSE/ASE and NASDAQ securiies using parameric and nonparameric es saisics, Review of Quaniaive Finance and Accouning 6, Chen, H., 2002, On Russell index reconsiuion, Working paper, Universiy of Illinois a Chicago, Chicago, IL. Cready, W. M. and R. Ramanan, 1991, The power of ess employing log-ransformed volume in deecing abnormal rading, Journal of Accouning and Economics 14, Fama, E. F. and K. R. French, 1993, Common risk facors in he reurns on socks and bonds, Journal of Financial Economics 33, Goezmann, W. N. and M. Garry, 1986, Does delising from he S&P 500 affec sock price?, Financial Analyss Journal 42, Harris, L. and E. Gurel, 1986, Price and volume effecs associaed wih changes in he S&P 500 lis: New evidence for he exisence of price pressures, Journal of Finance 41, Hillion, P. and T. Vermaelen, 2001, Deah spiral converibles, Working paper, INSEAD. Ibboson, R. G., 1975, Price performance of common sock new issues, Journal of Financial Economics 2,

24 Jain, P. C., 1987, The effec on sock price of inclusion in or exclusion from he S&P 500, Financial Analyss Journal 43, Kaul, A., V. Mehrora, and R. Morck, 2000, Demand curves for socks do slope down: New evidence from an index weighs adjusmen, Journal of Finance 55, Madhavan, A., 2001, The Russell reconsiuion effec, Working paper, ITG, Inc., New York, NY. Prui, S. W. and K. C. J. Wei, 1989, Insiuional ownership and changes in he S&P 500, Journal of Finance 44, Reed, A. V., 2000, The life cycle of an arbirage opporuniy: Reconsiuions of he Russell 2000, Working paper, Universiy of Pennsylvania, Philadelphia, PA. Shleifer, A., 1986, Do demand curves for socks slope down?, Journal of Finance 41,

25 TABLE 1. Abnormal Reurns for Four Groups Around he Russell 2000 Index Reconsiuion Dae from 1991 o Panel A. Average ( AR ) Percenage Abnormal Reurns Trading Day Pure Addiions o he Russell 2000 Index (N = 4321) AR AR ) ( Pure Deleions from he Russell 2000 Index (N = 3092) AR AR ) ( Upward Shifs from he Russell 2000 o he Russell 1000 Index (N = 861) AR AR ) ( Downward Shifs from he Russell 1000 o he Russell 2000 Index (N = 875) AR AR ) ( ** *** *** *** *** *** ** *** ** ** *** *** *** *** *** *** *** *** *** *** *** *** *** ** *** ** *** ** *** ** ** *** *** *** ** *** **

26 TABLE 1. Coninued. Panel B. Cumulaive Average ( CAR 1, 2 ) Percenage Abnormal Reurns Even Inerval Pure Addiions o he Russell 2000 Index (N = 4321) CAR 1, 2 ( CAR 1, 2 ) 1,2 Pure Deleions from he Russell 2000 Index (N = 3092) CAR ( CAR 1, 2 ) 2 Upward Shifs from he Russell 2000 o he Russell 1000 Index (N = 861) CAR 1, ( CAR 1, 2 ) 2 Downward Shifs from he Russell 1000 o he Russell 2000 Index (N = 875) CAR 1, ( CAR 1, 2 ) -10, *** *** *** ** -10, *** *** , *** *** , *** *** *** , ** ** , *** *** *** ** -20, *** *** , *** *** , *** *** *** -20, *** -40, *** *** , *** *** , *** *** , *** *** ** ** -40, ** Noe: This able repors abnormal and cumulaive abnormal reurns, expressed as percenages. We follow Hillion and Vermaelen (2001) by combining he Ibboson (1975) reurns across ime and securiies (RATS) approach wih he Fama-French (1993) ime-series model of he evoluion of securiy reurns. We run he following cross-secional regression on each day in even ime: R j - R f = α + β (R m - R f ) + s SMB + h HML + ε j. (1) For day, R j is he rae of reurn of he j h sock, R f is he one-monh Treasury bill rae, and R m is he CRSP value-weighed reurn on all NYSE, AMEX, and Nasdaq socks. The Fama-French marke-wide facors SMB (small minus big) and HML (high minus low) are consruced using six value-weighed sock porfolios formed on size and book-o-marke. SMB is he average reurn on he hree small-sock porfolios minus he average reurn on he hree big-sock porfolios. HML is he average reurn on he wo value-sock porfolios minus he average reurn on he wo growh-sock porfolios (Fama and French 1993 provide greaer deail on he consrucion of he facors). The esimae of he inercep α represens he average abnormal reurn for day. *** Significan a he 1% level. ** Significan a he 5% level.

27 TABLE 2. Average Percenage Abnormal Trading Volume ( AV ) and Rank Tes for Four Groups Around he Russell 2000 Index Reconsiuion Dae from 1991 o Trading Day Pure Addiions o he Russell 2000 Index (N = 2016) AV AV ) r ( Pure Deleions from he Russell 2000 Index (N = 3012) AV AV ) r ( Upward Shifs from he Russell 2000 o he Russell 1000 Index (N = 831) AV AV ) r ( Downward Shifs from he Russell 1000 o he Russell 2000 Index (N = 861) AV r AV ) ( ** ** ** *** *** ** *** *** *** *** *** ** ** ** ** ** Noe: Abnormal rading volume is esimaed using an equally weighed marke index wih all Nasdaq socks excluding ADRs as proxy for he marke rading volume and he 200 rading days esimaion period covering day 479 o day 280. *** Significan a he 1% level. ** Significan a he 5% level.

28 TABLE 3. Descripive Saisics for Percenage of Insiuional Share Holdings and Number of Insiuional Invesors Before and Afer he Reconsiuion of he Russell 2000 Index for Four Groups from 1991 o Panel A. Percenage of Insiuional Share Holdings Group Monh Parameer Average Median Maximum Minimum Sandard Deviaion Pure Addiions o he Russell 2000 Index (N = 300) April May July Augus Pure Deleions from he Russell 2000 Index (N = 300) April May July Augus Upward Shifs from he Russell 2000 o he Russell 1000 Index (N = 300) Downward Shifs from he Russell 1000 o he Russell 2000 Index (N = 300) April May July Augus April May July Augus

29 TABLE 3. Coninued. Panel B. Number of Insiuional Invesors Group Monh Parameer Average Median Maximum Minimum Sandard Deviaion Pure Addiions o he Russell 2000 Index (N = 300) April May July Augus Pure Deleions from he Russell 2000 Index (N = 300) April May July Augus Upward Shifs from he Russell 2000 o he Russell 1000 Index (N = 300) Downward Shifs from he Russell 1000 o he Russell 2000 Index (N = 300) April May July Augus April May July Augus

30 TABLE 4. Mean Differences of Insiuional Ownership Changes Following he Russell 2000 Index Reconsiuion Dae for Four Groups from 1991 o Group Parameer Mean Difference -Saisic () Percenage of Firms Regisering Increase Sign- Saisic (z) Pure Addiions o he Russell 2000 Index (N = 300) Pure Deleions from he Russell 2000 Index (N = 300) Percenage of Insiuional Holdings Number of Insiuional Invesors Percenage of Insiuional Holdings Number of Insiuional Invesors *** *** *** *** *** *** *** *** Upward Shifs from he Russell 2000 o he Russell 1000 Index Percenage of Insiuional Holdings (N = 300) Number of Insiuional Invesors *** *** *** *** Downward Shifs from he Russell 1000 o he Russell 2000 Index Percenage of Insiuional Holdings (N = 300) Number of Insiuional Invesors *** *** *** *** Noe: Insiuional ownership (percenage of insiuional holdings or number of insiuional invesors) difference is defined as he sock s average insiuional ownership in he wo-monh pre-reconsiuion period (April and May) subraced from he average insiuional ownership of he sock during he wo-monh pos-reconsiuion period (July and Augus). *** Significan a he 1% level.

31 FIGURE I. Cumulaive Average Abnormal Reurns for Pure Addiions and Deleions Around he Russell 2000 Index Reconsiuion Dae from 1991 o Cumulaive Average Abnormal Reurns Trading Days Pure Addiions o he Russell 2000 Pure Deleions from he Russell 2000