NBER WORKING PAPER SERIES ASSET FIRE SALES (AND PURCHASES) IN EQUITY MARKETS. Joshua D. Coval Erik Stafford

Transcription

1 NBER WORKING PAPER SERIES ASSET FIRE SALES (AND PURCHASES) IN EQUITY MARKETS Joshua D. Coval Erik Safford Working Paper hp:// NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachuses Avenue Cambridge, MA May 2005 We hank Malcolm Baker, Ken Froo, Bob Meron, André Perold, David Scharfsein, Andrei Shleifer, and Tuomo Vuoleenaho for valuable commens and discussions, and seminar paricipans a Harvard Business School and he Universiy of Virginia. We also hank Joshua Polle and Ken French for providing some of he daa used in his sudy. The views expressed herein are hose of he auhor(s) and do no necessarily reflec he views of he Naional Bureau of Economic Research by Joshua D. Coval and Erik Safford. All righs reserved. Shor secions of ex, no o exceed wo paragraphs, may be quoed wihou explici permission provided ha full credi, including noice, is given o he source.

2 Asse Fire Sales (and Purchases) in Equiy Markes Joshua D. Coval and Erik Safford NBER Working Paper No May 2005 JEL No. G14, G32, G20 ABSTRACT This paper examines asse fire sales, and insiuional price pressure more generally, in equiy markes, using marke prices of muual fund ransacions caused by capial flows from 1980 o Funds experiencing large ouflows (inflows) end o decrease (increase) exising posiions, which creaes price pressure in he securiies held in common by hese funds. Forced ransacions represen a significan cos of financial disress for muual funds. We find ha invesors who rade agains consrained muual funds earn highly significan reurns for providing liquidiy when few ohers are willing or able. In addiion, fuure flow-driven ransacions are predicable, creaing an incenive o fron-run he anicipaed forced rades by funds experiencing exreme capial flows. Joshua D. Coval Harvard Business School Boson, MA and NBER jcoval@hbs.edu Erik Safford Harvard Business School Boson, MA esafford@hbs.edu

3 This paper assesses he coss of asse fire sales in equiy markes. Financial disress is cosly whenever a firm s pas financing decisions inerfere wih curren operaions. One siuaion where his arises is when capial providers force a firm o quickly sell specialized asses. Because he sale is immediae, he liquidiy premium can be large, resuling in ransacion prices ha are subsanially below heir fundamenal values. Equiy markes are relaively liquid, bu no perfecly so. Largely because of he high liquidiy in equiy markes, many firms ha specialize in equiy invesing are willing o allow capial providers o wihdraw heir capial on demand. Evidence presened in his paper suggess ha even in he mos liquid of markes, asses someimes sell a fire sale prices. Shleifer and Vishny (1992) analyze he equilibrium aspec of asse sales and describe a siuaion where liquidiy can disappear, making i very cosly for someone who is forced o sell. They essenially argue ha asse fire sales are possible when financial disress clusers hrough ime a he indusry level and firms wihin an indusry have specialized asses. When a firm mus sell asses because of financial disress, he poenial buyers wih he highes valuaion for he specialized asse are oher firms in he same indusry, who are likely o be in a similarly dire financial siuaion, and herefore will be unable o supply liquidiy. Insead, liquidiy comes from indusry ousiders, who have lower valuaions for he asse, and hus bid lower prices. This sory can be recas easily in a capial marke seing. Here, he firms are professional invesors, who follow somewha specialized invesmen sraegies. In his conex, specializaion refers o concenraed posiions in securiies ha have limied breadh of ownership, and imporanly, have significan overlap wih ohers following a similar sraegy. 1 Specializaion is common in invesmen managemen, wih many professional invesors focusing on a single or limied number of invesmen sraegies. Meron (1987) and Shleifer and Vishny (1997) presen models of invesmen managemen ha rely on specializaion o derive limied arbirage. 1 For example, merger arbirage is a specialized invesmen sraegy followed by many professional invesors, requiring relaively large posiions in socks ha evenually are held mainly by merger arbirageurs. 1

4 Financial disress occurs when a firm sruggles o make paymens required by is liabiliies, which for a financial firm arises when invesors redeem heir capial. When capial is immediaely demandable, a poorly performing muual fund wihou significan cash reserves has no choice bu o sell holdings quickly, which will be cosly if oo many ohers are selling he same posiions a he same ime. If immediacy is no required, he seller can wai for beer pricing, raise addiional funds, or poenially renegoiae exising claims. However, hese are no legiimae opions for a poorly performing fund. Renegoiaion wih large numbers of claim holders is infeasible, and raising addiional funds from new invesors, while exising invesors wan ou, is highly unlikely. Regulaions preven muual funds from raising funds by shor selling oher securiies, and binding margin consrains are likely o resric shor selling by severely underperforming hedge funds. 2 Accurae assessmen of asse fire sale coss requires considerable ransparency in he decisions of he firm and is invesors, whereas mos seings in which asse fire sales are cosly are likely o be highly opaque. The primary challenge in measuring he coss of asse fire sales is ha disinguishing financial from economic disress requires idenifying asse sales ha are a direc consequence of he financing decisions of he firm. In many corporae seings, financial difficulies and economic difficulies coincide over muli-year periods, making causaliy difficul o assign. Addiionally, efficien esimaion of coss requires precise measuremen of fair asse value, which can be a challenge in environmens characerized by illiquidiy and declining prices. The focus of his paper is on he asses held by open-ended muual funds. The openended muual fund srucure produces a highly ransparen firm wih invesmen decisions ha 2 The asse fire sale sory is similar o he price pressure hypohesis of Scholes (1972), where sock prices can diverge from heir informaion-efficien values because of uninformed shocks o excess demand o compensae hose who provide liquidiy. The asse fire sale sory idenifies forced selling by disressed muual funds as one paricular ype of uninformed shock, and explains why hose who provide liquidiy during such a crisis are likely o demand addiional compensaion. 2

5 are easy o idenify and monior. The opened muual fund is also exremely relian on ouside capial o fund is invesmen opporuniies only he occasional back-end load sands beween ouside capial providers and heir capial. Monhly reporing of oal ne asses allows real ime measuremen of he pressures ha ouside capial providers place on he firm. Moreover, because of high rading frequency in public markes, deviaions in ransacion prices from fair values can be accuraely assessed via he racking of pos-sale reurns. On he oher hand, he sock marke environmen is a relaively hospiable one for asse sales. Wih high ransacion volumes and low execuion coss, a disressed seller of a lised equiy migh expec o find many willing buyers. In addiion, muual funds ha selec he open-ended organizaional form do so precisely because hey view he poenial coss of his srucure o be low. 3 Thus, our focus is on a seing where asse fire sales are unlikely, bu where high ransparency permis hem o be properly deeced should hey exis. 4 To examine empirically asse fire sales in equiy markes and he effecs of insiuional price pressure more generally, we consruc a sample of siuaions where we suspec widespread muual fund rading of individual socks caused by capial flows. Fundamenal value is no immediaely observable, bu by sudying sysemaic paerns in abnormal reurns over ime, we can idenify deviaions beween ransacion prices and fundamenal value ex pos if we find evidence of significan price reversals following forced ransacions. We aemp o disenangle price pressure from informaion effecs by focusing on siuaions where he fire sale sory predics ha muual fund sales are moivaed by necessiy, as opposed o opporunisic informaion-based rading. In paricular, we focus on muual funds sock ransacions ha are forced by financial disress, and herefore unlikely o reveal much new informaion abou he individual securiies being sold, and where here is considerable overlap in he holdings among 3 Sein (2004) presens a model where compeiion pushes muual funds oward he open-end form even hough his severely consrains heir abiliy o conduc arbirage rades. 4 See Andrade and Kaplan (1998), Asquih, Gerner, and Scharfsein (1994), Gilson (1997) for sudies of financially disressed firms. See Pulvino (1998) for a sudy focusing on asse fire sales in he used aircraf marke. 3

6 poorly performing funds. The empirical resuls provide considerable suppor for he view ha concenraed muual fund sales forced by capial flows exer significan price pressure in equiy markes, ofen resuling in ransacion prices far from fundamenal value. We find ha poor performance leads o capial ouflows for muual funds, he mos serious of which, we consider financial disress. This corroboraes previous research, which finds a srong relaion beween muual fund flows and pas performance. 5 Ineresingly, funds who find hemselves in he boom decile of capial flows end o be less diversified han oher funds, holding nearly 25% fewer securiies han a ypical fund. The analysis also indicaes ha flows ino and ou of muual funds do indeed force rading. Muual funds in he boom decile of capial flows are roughly wice as likely o sell, or eliminae holdings, han funds experiencing normal flows. This forced rading is especially cosly for muual funds when here is significan overlap wih he securiies held by oher funds experiencing ouflows, where ransacions appear o occur far from fundamenal value. We esimae ha invesors providing liquidiy o he disressed funds earn significan abnormal reurns over he subsequen monhs. Moreover, we show ha forced ransacions are predicable, which creaes an opporuniy for fron-running. An invesmen sraegy, which shor sells socks mos likely o be involved in fire sales, and buys ahead of anicipaed forced purchases, earns average annual abnormal reurns well over 15%. Ineresingly, we find ha exreme inflows can be cosly for muual funds oo. Funds experiencing large inflows end o increase heir exising posiions, creaing significan price pressure in he socks held in common by hese funds. Like he asse fire sales, hese inflowdriven purchases produce rading opporuniies for sraegy ousiders. Over he sample period, 1980 o 2003, exreme inflows o muual funds are considerably more common han ouflows, making inflow-driven purchases a more frequen and severe siuaion. 5 See for example, Ippolio (1992), Chevalier and Ellison (1997), and Sirri and Tufano (1998) 4

7 Finally, we documen ha concenraed muual fund buying and selling caused by capial flows is highly relaed o he momenum effec found in equiy reurns, bu ha a considerable abnormal reurn remains afer conrolling for momenum. Because muual fund flows are sensiive o pas performance, he socks ha muual funds are forced o sell end o overlap wih he socks idenified as good shors by a momenum sraegy. This paper is organized as follows. Secion I describes he daa. Secion II presens evidence on he exisence and magniudes of price pressure and asse fire sales in equiy markes. Secion III examines he incenives for providing liquidiy during crisis periods and for fronrunning. Secion IV discusses he persisence of insiuional price pressure, and Secion V concludes. I. Daa Descripion A. Muual Fund Holdings, Reurns, & Flows Mos of our analysis relies on a merger of he wo major muual fund daabases ha have been used exensively in he lieraure: he Specrum muual fund holdings daabase and he CRSP muual fund monhly ne reurns daabase. Comprehensive descripions of boh can be found in Wermers (1999), who conducs a similar merge. Our merge procedure is as follows. Firs, funds are mached by name. To make sure we have idenified he iming of changes in holdings accuraely, we only include fund-quarer observaions repored across adjacen quarers when holdings changes are required. Fundquarer observaions where he value of he holdings differs subsanially from ha repored in he CRSP daabase ne of he cash posiion are removed. Because our focus is on US equiy funds, we exclude funds wih an invesmen objecive code indicaing any of he following: inernaional; municipal bonds; bonds and preferred; or meals. Finally, because he number of mached funds is significanly lower in he 1980s han in he 1990s, we ofen emphasize he subperiod from 1990 hrough 2003, in our analysis. 5

8 Muual fund flows are esimaed using he CRSP series of monhly oal ne asses (TNAs) and reurns. The ne flow of funds o muual fund during monh is defined as: FLOW = TNA TNA 1+ R ) (1) 1 ( FLOW flow, = (2) j TNA 1 where TNA is he CRSP TNA value for fund j a he end of monh, and R is he monhly reurn for fund j over monh. In order o mach wih he quarerly holdings daa, we sum monhly flows over he quarer o calculae quarerly flows. Mos of he analysis involving muual fund flows uses he dollar value of FLOW as a percenage of beginning of period TNA as in equaion (2). B. Measuring he Relaion beween Fund Performance and Flows I is well documened ha capial flows o and from muual funds are srongly relaed o pas performance (e.g. Sirri and Tufano (1998)). We use a simple Fama-MacBeh (1973) syle regression model o forecas fund flows based on pas reurns and lagged flows. 4 k = 1 k 8 flow = a + b flow + c R (3) k h= 1 h h In paricular, each quarer or monh,, we esimae a cross-secional regression as in (3). We hen calculae he ime series average of he coefficiens and repor -saisics using he ime series sandard error of he mean. Expeced flows are calculaed as he fied values using he ime series average of he coefficiens. We esimae he regression coefficiens using a sub-sample of he quarerly muual fund observaions ha we view as having he mos reliable daa. In paricular, we impose he following daa requiremens: Fund mus have a some poin had, TNA j, > $ 1M 6

9 A some poin, fund mus have had a leas 20 holdings TNA Changes in TNA canno be oo exreme, in paricular: 0.50 < < 2. 0 TNA CRSP 1 TNA, Daa from CRSP and Specrum canno be oo differen: < j < 1. 3 Specrum 1.3 TNA j, 1 The coefficiens from he sub-sample regressions are used o calculae expeced flows for all funds, including hose excluded from he esimaion. Resuls are no meaningfully alered, if he funds excluded from he regressions are compleely dropped from he analysis alogeher. Table 1 repors regression resuls. As expeced from previous research, here is a srong relaion beween muual fund flows and boh lagged flows and lagged reurns. Quarerly muual fund flows are highly significan in explaining fuure flows for up o a full year, while quarerly fund reurns are imporan deerminans of fuure flows for wo years. The resuls are largely consisen wih pooled regression resuls. The main disincion is ha he explanaory variables in he Fama-MacBeh regression focus on explaining cross-secional differences in flows whereas he pooled regression coefficiens mus also accoun for ime-series variaion in overall flows. As a resul, he Fama-MacBeh coefficiens are esimaed more precisely, relaive o he number of observaions. C. Fund Behavior in Response o Financial Pressure Our noion of a sock fire sale requires ha several differen owners, who are each experiencing financial disress, conemporaneously sell he securiy. Muual funds experiencing significan ouflows have no choice bu o sell some of heir holdings o cover redempions unless hey have excess cash or can borrow. Typically, borrowing is difficul and, because mos funds are evaluaed agains all-equiy benchmarks, few mainain significan cash balances. Moreover, shor selling oher securiies is usually no feasible. 6 Therefore, he immediae selling 6 The Invesmen Company Ac of 1940 prevens muual funds from shor selling and buying securiies on margin. 7

10 of some exising holdings is he only opion. In addiion, i is imporan ha here are many sellers relaive o poenial buyers. A single fund selling when ohers are willing and able o provide liquidiy is unlikely o produce a fire sale price. Only when many funds are forced o sell he same securiy should we expec o see significan price pressure. Table 2 provides an overview of fund behavior in response o financial pressure. In Panels A and B, we sor funds ino deciles according o acual and expeced quarerly flows. We hen calculae he fracion of a given fund s posiions ha are mainained, expanded, reduced, or eliminaed during he given quarer and average hese values across funds wihin each decile. We also repor for each decile he average 12-monh fund reurn and he average number of holdings. In Panels C and D, we average a he holding level, reporing he fracion of holdings wihin each decile ha are mainained, expanded, reduced, or eliminaed. As Table 2 makes clear, fund flows have a srong impac on fund rading behavior. In paricular, funds experiencing (or expeced o experience) large ouflows are far less likely o expand or mainain exising posiions and far more likely o reduce or eliminae posiions. For insance, a holding ha ranks in he op decile of fund ouflows has a 54.8% chance of being reduced or eliminaed by is fund, whereas a holding in he boom decile has a 20% chance of being reduced or eliminaed. A similar paern exiss for holdings of funds ha are expeced o experience ouflows during he curren quarer. Holdings of funds in he op decile of expeced ouflows are 48.4% likely o be reduced or sold versus 27.4% for holdings by funds in he boom decile. Finally, funds experiencing ouflows end o be less well diversified han hose experiencing inflows. Funds in he op decile of ouflows average 82.8 holdings, whereas funds in he boom decile average Alhough he number of holdings is an imprecise measure of diversificaion, i is consisen wih he idea ha specialized funds may be more sensiive o he coss of financial disress. Figure 1 displays he average endency of ne adjusmens o exising posiions as a funcion of capial flows. This is merely a graphical represenaion of some of he daa from Table 2. Consisen wih he fire sale sory, he funds wih he mos significan ouflows are very 8

11 likely o reduce heir exising posiions. However, he figure clearly shows ha here is a similar effec caused by inflows. Funds in he op decile of capial flows end o increase heir exising posiions. This is ineresing, because unlike he firms who mus sell in he face of ouflows, hese funds have more opions. These funds can accumulae cash or purchase securiies ha hey do no currenly own, neiher of which is feasible for firms facing ouflows. D. Idenifying Fire Sales A naural measure of asse fire sales migh be o deermine he number of shares sold due o capial ouflows and scale by he number of shares ousanding or rading volume. Iniial ess using such a measure delivers resuls ha are economically large, bu of mixed saisical reliabiliy. There are wo reasons why share-based measures may fail o properly idenify asse fire sales. Firs, as argued above, fund-level responses o ouflows are unlikely o resul in any significan price pressure unless a sock finds iself wih many forced sellers and few willing buyers. A large invesor can orderly liquidae a large posiion, bu a large number of small invesors canno orderly liquidae a similar size aggregae posiion. This makes i imporan for a measure o emphasize commonaliy in he capial flows o invesors of a paricular securiy. A second weakness of share-based measures of asse fire sales is ha hey are highly sensiive o reporing errors in fund holdings. Because funds are only required o repor heir holdings on a semi-annual basis, Specrum relies on volunary disclosure o fill in off-quarer holdings. This resuls in a non-rivial frequency of errors in repored holdings and places a significan burden on share-based measures o idenify hem as such. In view of he above consideraions, our procedure for classifying muual fund sock sales as asse fire sales (and purchases) is as follows. Firs, we idenify ransacions by funds ha are likely forced by capial flows. In paricular, we define quarerly forced sales (buys) as decreases (increases) in holdings by funds experiencing concurren ouflows (inflows) greaer han 5%. 9

12 The difference beween forced buys and forced sales is hen scaled by he oal number of muual fund owners, o form a variable we call PRESSURE. 7 PRESSURE i, = ( Buy > ) ( < ) j i, flow 5% Sell j i, flow 5% Own i, 1 Buy i, equals one if fund j increased is holding in sock i during quarer, and zero oherwise. Sell i, is defined similarly based on decreases. Own i,-1 equals one if fund j owns sock i a he beginning of quarer. Socks wih PRESSURE -15% are deermined o be fire sale socks, and hose wih PRESSURE 25% are deermined o be inflow-driven sock purchases. 8 Table 3 displays summary saisics for he fire sale and inflow-driven purchase samples. One of he mos sriking paerns from he able is ha he number of muual funds ha can be linked o CRSP sock daa increases 10-fold over he sample period. Anoher persisen paern is ha average flows ino muual funds are consisenly posiive over his period. Average fund flows are posiive in all bu 2 of he 24 years and increase subsanially over he sample period. Average annual flows are abou 3.5 imes larger in he 1990s han hey are in he 1980s, which does no accoun for he dramaic increase in he number of funds. Panel A of Table 3 reveals ha he number of fire sale socks varies quie a bi hrough ime, averaging abou 2% of he firms lised in CRSP, and ranging from 0.1% in 1983 o nearly 9% in Average quarerly flows for muual funds involved in fire sales are consisenly around -10% of oal ne asses. These funds end o have prior reurns ha have underperformed he overall marke, especially in he 1990s. The socks hemselves do no ell a simple sory. In he 1980s, he fire sale socks end o have been relaively srong performers prior o he fire sale even, whereas in he 1990s, he fire sale socks have done abou as well as he average holding of he fund. j (4) 7 We require a leas 10 muual funds owners before we calculae he PRESSURE variable. 8 The cuoffs of -15% and 25% approximaely correspond o he 5 h and 95 h perceniles of he PRESSURE variable, respecively. 10

13 Panel B of Table 3 repors a sample summary of he inflow-driven purchases. The average quarerly inflows o he muual funds ha are involved in hese ransacions consisenly average around 20% of TNA. These funds end o have ouperformed he overall marke prior o he even, and end o increase heir holdings ha have ouperformed by an even larger margin. The individual socks ha comprise he inflow-driven purchases have average annual reurns prior o he even of nearly 50%. Again, he number of socks involved in inflow-driven purchases varies considerably hrough ime, averaging around 6%, and increasing over he sample period. II. Price Effecs of Muual Fund Sales & Fire Sales In general, deecing price pressure effecs around muual fund sock ransacions is problemaic because of he simulaneous effecs of price pressure and informaion revelaion. In an aemp o disenangle price pressure and informaion effecs, we examine sock price changes around widespread forced and unforced muual fund sales, and look for evidence of sock price drops followed by a significan price reversal. 9 If muual funds bring informaion ino prices hrough heir rading, hen we should see a price drop in he period where hey are selling heavily, and hen no drif in abnormal reurns following he rades. However, if muual fund rading is driven by necessiy raher han informaion, and if his forced rading resuls in fire sale prices, hen we should see a significan price drop over he period where hey are being forced o sell, followed by a period of posiive abnormal reurns compensaing hose who provided liquidiy in he crisis period. Table 4 displays monhly abnormal reurns around various ypes of muual fund sock ransacions. Monhly abnormal reurns are calculaed using simple ne-of-marke reurns, where he CRSP value-weighed index proxies for he marke porfolio. In he spiri of Fama and 9 This empirical approach is similar o he one used by Michell, Pulvino, and Safford (2004) who sudy price pressure around mergers. 11

14 MacBeh (1973), we calculae average abnormal reurns each monh and hen use he ime series of mean abnormal reurns for saisical inference o conrol for poenial cross-secional dependence in he monhly abnormal reurns. 10 The fire sale sample is seleced according o he procedure described in he previous secion, where socks wih widespread selling by disressed funds are considered fire sale socks. Table 4 also repors he average abnormal value of our pressure variable for easy comparison o he associaed monhly abnormal reurns. In Panel A, he paern in average abnormal reurns around he widespread selling of socks held by disressed muual funds is sriking (see Figure 2 for a graphical represenaion). We find significanly negaive abnormal reurns in he monhs of forced selling and hose immediaely surrounding hem. Over he quarer where, on ne, a leas 15% of he owners are disressed sellers of he same sock 11, he average abnormal sock reurn is -10.1% wih a -saisic of There is some spillover in sock price performance ino he nex monh, wih average abnormal reurns of -1.4%, as ne forced selling remains high. Over he quarer where fire sales are occurring, roughly 20% of he owners are on average ne forced sellers of each sock, and his amoun of forced selling pressure coninues ino he subsequen monh. Imporanly, he downward paern in abnormal reurns evenually reverses once he ne forced sales dissipae. From monhs 4 o 12 following he fire sale quarer, average abnormal ne forced selling pressure rereas o under 2%, and sock prices for he fire sale socks rebound 7.74%% over his period, wih a -saisic of The rebound does no fully cover he losses associaed wih he iniial price drop, bu represens a significan cos of a fire sale. This evidence suggess ha widespread forced selling by disressed muual funds exers significan downward price pressure on he individual socks sold, well beyond any conemporaneous informaion effecs. 10 This procedure gives equal weigh o each monhly observaion, raher han each individual observaion. When individual observaions are given equal weigh and assumed independen, he paerns are more pronounced wih highly significan es saisics. 11 Monhs -2, -1, and 0 define he even quarer. 12

15 In Panels B and C, we explore he role of each ingredien of he fire sale widespread ne selling pressure, by consrained funds. When here is widespread selling by unconsrained funds, he sriking fire sale paern in abnormal reurns is no presen. Panel B of Table 4 (see also Figure 3) repors CAARs around similarly widespread muual fund sales, bu by funds ha are no necessarily forced o sell. In paricular, his sample is again idenified using a pressure variable, bu one alered by removing he condiion on flow from he calculaion. Again, here is a significan price drop over he quarer of widespread selling, wih some coninuaion ino he nex quarer. However, he reversal is much more modes. In fac, he fire sale socks ha are par of his uncondiional sample are driving mos of he reversal. This is consisen wih volunary muual fund rading bringing informaion ino prices, and on average, prices adjusing o an appropriae level. In Panel C of Table 4 (see also Figure 4), we repor resuls for a sample of muual fund sales by disressed funds, where he selling is isolaed. We idenify he sample based on he following condiion: -15% < PRESSURE < 0. In he case of isolaed disressed selling, here is a modes, bu saisically significan, average abnormal sock price drop of -2.52% (-saisic = -6.66) over he even quarer, wih mos of his drop being reversed over he subsequen quarers. Alhough he paern in CAARs is saisically significan, he economic magniudes are fairly small. The key o he reversal appears o be ha he selling is widespread among muual funds ha mus immediaely sell due o capial ouflows. Moreover, he effec seems o be increasing in he number of ne sellers and in he level of disress. Figure 5 illusraes he comparaive saics wih nine graphs represening various proporions of ne sellers wih varying degrees of disress. In paricular, he op row displays unconsrained funds, he middle row shows funds wih flows > 5%, and he boom row repors funds wih flows > 10%. The proporion of ne sellers varies across he columns, wih a leas 5% of owners selling in he firs column, a leas 15% of owners selling in he second column, and a leas 25% of owners selling in he hird column. The cener graph corresponds o he fire sale sample (resuls in Panel A of Table 4). 13

16 The fire sale paern becomes more exreme moving down and/or o he righ, suggesing ha he inensiy of fire sales is increasing in boh of hese facors. For example, when we hold he capial flows hreshold consan a flows >5% and increase he proporion of ne sellers hreshold from a leas 15% o a leas 25% he average abnormal reurn from monh -2 o +3 is % (-saisic = -7.58) wih a reversal of 15.01% (-saisic = 3.84) over he nex 3 quarers. Likewise, holding he proporion of ne sellers hreshold a a leas 15% and increasing he flows hreshold o flows >10% resuls in an average abnormal reurn from monh -2 o +3 of % (-saisic = -6.36) wih a full reversal of 14.57% (-saisic = 4.93) over he nex 3 quarers. The resuls also reveal somehing abou which socks are sold in response o significan ouflows. The abnormal reurns for he fire sale socks prior o he acual sale are close o zero (slighly posiive, bu saisically insignifican). This suggess ha hese socks have performed relaively well when compared o he marke, and exremely well relaive o he disressed funds average holdings. For he mos par, he disressed funds have been significanly underperforming he marke. In paricular, he cumulaive average abnormal reurn over he nine monhs prior o he fire sale quarer for he disressed funds is -4.5% (-saisic = -7.40). Thus, i appears ha in crisis periods, sruggling funds sell relaive winners raher han downside momenum socks. We are presumably observing oucomes from opimized behavior, suggesing ha he sale of loser socks during hese crisis periods would resul in more severe price discouns. Finally, he persisence in ne forced selling is ineresing. No all funds experience ouflows a he exac same ime. The iniial widespread forced sales can creae an exernaliy for he marginal funds ha would no face capial wihdrawals in he absence of his price pressure. However, heir performance is sufficienly affeced by he fire sales hemselves, ha hey face ouflows he subsequen quarer, which forces hem o sell wih a lag. This is relaed o an effec described in Michell, Pulvino, and Safford (2002) creaed by margin requiremens. A firm up agains a financing consrain (i.e. facing a margin call) may have o sell, and in so doing, 14

17 adversely impac prices for ohers following a similar sraegy. Here, he financing consrain is no only binding, bu also acually ighening, as capial providers wihdraw funds. Ineresingly, his sequence of evens creaes persisen mispricing, which can ge worse before being eliminaed, and does no presuppose irraional invesors or managers. This exernaliy can be an imporan impedimen o arbirage in imperfec capial markes (Shleifer and Vishny (1997)). III. Incenives for Providing Liquidiy & Fron-Running The even-ime analysis presened in he previous secion suggess ha here may be a srong incenive o provide liquidiy a imes of widespread selling by financially disressed muual funds. In oher words, he buyers in asse fire sales are receiving aracive prices for providing liquidiy when few ohers are able or willing. In addiion, because capial flows are predicable, here may also be an incenive o remove liquidiy in anicipaion of forced sales by fron-running he disressed muual funds. We invesigae boh of hese incenives by sudying he porfolio reurns o invesors following hese invesmen sraegies. A. Invesmen Reurns following Asse Fire Sales The resuls displayed in Table 4 sugges ha he buyer in an asse fire sale will, on average, be compensaed for providing liquidiy. The compensaion is realized as prices reurn o heir informaion-efficien values in he subsequen monhs, and can be deeced in he form of posiive abnormal reurns. To measure invesmen performance o buyers in asse fire sales, we calculae he calendar-ime porfolio reurns o an invesmen sraegy ha buys all socks idenified as fire sale socks wihin he pas year, bu no wihin he mos recen quarer. The consrain ha he fire sale has no occurred wihin he mos recen quarer ensures ha his is a feasible invesmen sraegy in erms of all required informaion being publicly available. A secondary benefi of his consrain is ha i avoids he spillover of fire sales ino he subsequen monh. Measuring abnormal reurns requires a model of expeced reurns. We repor resuls 15

18 using hree differen models: CAPM, Fama-French 3-facor model, and a 4-facor model ha includes momenum (see Fama and French (1993) for a descripion of he consrucion of he facors). 12 Rp Rf = a + b [ Rm Rf ] + s SMB + h HML + u UMD + e (5) Table 5 repors calendar-ime porfolio regressions of he invesmen sraegy described above for boh equal-weigh and value-weigh porfolios. The inercep from hese regressions represens he average monhly abnormal reurn, given he model. The inerceps repored in Panel A, range from a mere 8 basis poins (bps) per monh (-saisic = 0.14), o jus under 1.0% per monh (-saisic = 2.12), and are remarkably similar beween equal- and value-weighed porfolios. However, only he equal-weigh porfolio has a saisically significan inercep, and his only afer conrolling for momenum. This is an exremely volaile sand-alone invesmen sraegy. The negaive loading on momenum is ineresing, suggesing ha hese socks covary significanly wih downside momenum socks, bu are no performing poorly hemselves. Anoher ineresing feaure of hese porfolios is he large esimaed coefficien on he marke excess reurn. The second panel in Table 5 displays performance resuls for an invesmen sraegy ha piles on o he disressed selling ha occurred in he prior monh. This sraegy, raher han providing liquidiy, is compeing for liquidiy, and herefore exacerbaing he price pressure. This is arguably a less feasible sraegy requiring immediae access o he muual fund holdings daa a he end of each quarer, bu sill condiional only on previous ransacions. The economic magniude of he abnormal reurns is very large, and usually associaed wih srong saisical reliabiliy. Five of he six invesmen sraegies resul in saisically significan abnormal reurns a radiional significan levels, ranging from -0.79% per monh (-saisic = -1.77), o -2.72% per monh (-saisic = -3.95). Conrolling for momenum reduces he inerceps by roughly half. 12 The facors are from Ken French: hp://mba.uck.darmouh.edu/pages/faculy/ken.french/daa_library.hml. 16

19 In many regards, his invesmen sraegy is very similar o a momenum sraegy. In boh cases, he porfolio conains socks ha have performed poorly, and coninue o underperform. However, he mechanism ha creaes he low reurns appears o be he coninued forced selling by disressed muual funds in one case, while unspecified in he case of momenum. Again, he bea esimaes are very large. In he CAPM regressions, he esimaed bea is 1.7 for boh he equal- and value-weighed porfolios. The final panel in Table 5 shows he resuls from combining he wo sraegies described above ino a long-shor invesmen sraegy. The same basic paern emerges. The reurns are somewha more exreme wih he value-weighed porfolios; exremely large in economic erms, wih annualized abnormal reurns ranging from 27.9% (-saisic = 3.11) o 45.4% (-saisic = 5.23), and fairly srong saisical reliabiliy. Noe ha because of combining wo sraegies ino a long-shor sraegy, and requiring a leas 10 firms in each of he long and shor porfolios, ha he number of monhs ha his sraegy is feasible drops o only 95 monhs ou of a possible 168 monhs. Therefore, he abnormal reurn esimaes oversae he rue invesmen reurns by ignoring he capial coss of sanding idle. 13 B. Invesmen Reurns following Inflow-Driven Purchases The evidence in Table 2 and Figure 1 sugges ha funds wih significan capial inflows end o increase heir exising holdings; much like he funds experiencing significan ouflows end o reduce heir exising posiions. In oher words, funds facing significan inflows behave as if hey oo are consrained. As in he fire sale sory, if many funds are simulaneously forced o buy he same securiies when few ohers are able o sell, ransacion prices may occur a a price significanly above fundamenal value, wha we will call inflow-driven purchases. 13 Myron Scholes describes liquidiy-relaed sraegies like his as fire-saion sraegies, where much of he ime is spen sanding around waiing for an even, while coss are incurred coninuously. 17

20 We idenify inflow-driven purchases using he PRESSURE variable. In paricular, when PRESSURE > 25% we consider he sock o be involved in an inflow-driven purchase. To measure invesmen performance o sellers in inflow-driven purchases, we calculae he calendarime porfolio reurns o an invesmen sraegy ha buys all socks idenified as inflow-driven purchase socks wihin he pas year, bu no wihin he mos recen quarer. Again, we also examine he sraegy of piling on o he forced purchase, and a long-shor sraegy ha combines he wo sraegies. If asse fire purchases resul in significan price pressure, hen we should see posiive abnormal reurns o he piling on sraegy, followed by a reversal. Table 6 repors calendar-ime porfolio regressions of he invesmen sraegy described above for boh equal-weigh and value-weigh porfolios. The inerceps repored in Panel A represen he reversal. The average abnormal reurns are economically large and all are saisically significan, ranging from -0.47% per monh (-saisic = -2.15) o -0.84% per monh (-saisic = -3.41). Ineresingly, he monhly abnormal reurns remain saisically significan afer conrolling for momenum. The second panel in Table 6 displays performance resuls for an invesmen sraegy ha piles on o he forced buying ha occurred over he prior monh. Like he earlier piling on sraegy, his one compees for liquidiy, and herefore inensifies he price pressure. Again, he economic magniude of he abnormal reurns is very large, wih mixed saisical reliabiliy. Four of he six invesmen sraegies resul in saisically significan abnormal reurns, ranging from 0.46% per monh (-saisic = 1.55), o 0.97% per monh (-saisic = 2.59). The value-weigh porfolios end o produce less exreme abnormal performance, and lower levels of saisical significance. Panel C of Table 6 repors he resuls from combining he previous wo sraegies ino a single long-shor sraegy. This sraegy produces very large average abnormal reurns, all of which are saisically significan a radiional significance levels. The average abnormal reurns range from 1.11% per monh (-saisic = 3.00) o 1.76% per monh (-saisic = 4.82). The sraegy is highly relaed o momenum, as boh he equal- and value-weighed porfolios have 18

21 large posiive coefficiens on he momenum facor, bu inerceps from hese regressions are well over 1% per monh, suggesing ha his effec is somewha disinc from he momenum effec. C. Invesmen Reurns o hose Anicipaing Asse Fire Sales and Inflow-Driven Purchases The evidence presened so far, sugges ha here is a powerful incenive o ry o anicipae widespread forced buying and selling by consrained muual funds. There is a real possibiliy ha his is feasible because capial flows o muual funds are reasonably well explained by lagged flows and reurns. Using he regression model presened in Table 1, we forecas expeced flows o muual funds and idenify anicipaed fire sale and inflow-driven purchase socks using he procedure described in Secion II, subsiuing expeced flows for acual flows. 14 E [ PRESSURE ] i, + 1 ( Own + > ) ( + ) i, E[ flow 1] 95 percenile Own j i, E[ flow 1] 5 percenile Own i, < j = j (6) Table 7 repors calendar-ime porfolio regressions for an invesmen sraegy ha invess in socks where a fire sale or inflow-driven purchase is anicipaed. Specifically, anicipaed fire sale socks are idenified as hose wih an expeced pressure below -5% (repored in Panel A), while anicipaed inflow-driven purchase socks are hose wih an expeced pressure greaer han 10% (repored in Panel B). 15 The socks mos likely o be involved in fire sales have low subsequen reurns, resuling in negaive abnormal reurns, alhough saisical significance is mixed. The socks ha are mos likely o be involved in inflow-driven purchases have very high 14 Noe ha expeced flow cuoffs are se o he 5 h and 95 h perceniles of expeced flow levels, as opposed o fixed a 5% ouflows and inflows. 15 These cuoffs reflec roughly he 10 h and 90 h perceniles of he PRESSURE variable. We use he more liberal cuoffs o ensure ha a sufficien number of socks are in he anicipaed fire sale and inflow-driven purchase porfolios. 19

22 subsequen reurns, resuling in large abnormal reurns, which are generally saisically significan. Finally, he long-shor sraegy ha combines hese wo sraegies produces very large abnormal reurns, ranging from 0.96% per monh (-saisic = 1.61) o 2.36% per monh (-saisic = 3.60), alhough saisical significance is mixed. Conrolling for momenum reduces he abnormal reurns by roughly half. Socks expeced o be involved in inflow-driven purchases are he primary conribuors o he anicipaive sraegy. As a resul, he porfolio does no rely heavily on shor-selling, increasing he feasibiliy of implemening he sraegy. This is also consisen wih previous research documening ha he flow-performance relaion for muual funds is sronger for inflows han for ouflows, which allows for more accurae forecass of expeced inflows han he forecass of expeced ouflows. IV. Persisence of Insiuional Price Pressure A. The Role of Informaion & Agency Coss Care mus be aken in inerpreing he calendar-ime porfolio regression analysis. The esimaes of average abnormal reurns ignore wo imporan coss (Meron (1987)). Firs, he esimaes of abnormal performance exclude he coss of generally becoming informed abou he invesmen sraegy. Our esimaes are based on he period from 1980 o 2003, coinciding wih he availabiliy of he necessary daa. Imporanly, prices were se over his period by invesors who did no have access o hese daa, and herefore, heir assessmens of he risks and reurns associaed wih he sraegy were surely less precise han hose presened in he ables. Even now, he resuls, while economically enicing, are only marginally significan when aken as a whole. Second, once he general sraegy is recognized as poenially profiable, here are he informaion acquisiion coss associaed wih he individual securiies, which have no ye been borne by he evenual liquidiy providers. These coss oo, are no included in he measured abnormal reurns. 20

23 In some sense, we have documened ha equiies are in fac specialized asses. The fire sale sory begins wih he idea ha asses are specialized, which, in he shor run, fixes he pool of poenial buyers and sellers who fully value he asse. In he case of socks, where he informaion relevan for pricing is cosly o obain, specializaion will arise around who has and collecs regularly his informaion. In exreme siuaions, when a majoriy of he funds who are informed abou an individual sock are unable o volunarily rade, price seing may fall on funds who have no ye invesed in he relevan informaion. These coss are surely posiive, and poenially quie large. The evidence on funds experiencing significan inflows behaving as if hey are consrained is consisen wih hese coss being large. The fac ha hese addiional coss occasionally apply produces ime variaion in ransacion coss. Marke prices may no be perfecly efficien, in ha prices do no reflec all available informaion a every poin in ime, bu hey may well be wihin he bounds of ime-varying ransacion and informaion coss (a dynamic version of Grossman and Sigliz (1980)). An ineresing consequence of significan insiuional price pressure in conjuncion wih a srong fund performance-flow relaion is ha he price pressure can spillover ino subsequen periods (Shleifer and Vishny (1997), Shleifer (2000)). There are acually wo spillover effecs. One is he own-fund spillover, where a fund s buying or selling is exising posiions mechanically improves or degrades is own performance, which affecs capial flows in he subsequen period. Anoher spillover occurs across funds when he iniial insiuional price pressure affecs marginal funds ha would no face capial flows in he absence of his price pressure. However, heir performance is sufficienly affeced by he insiuional price pressure, ha heir capial flows are alered he subsequen quarer, forcing hem o ransac wih a lag. This sequence of evens resuls in persisen mispricing, which can ge worse before being eliminaed. 21

24 B. Implicaions An ineresing implicaion of he persisen insiuional price pressure relaes o performance evaluaion of fund managers. Over moderaely long periods of several quarers, he evidence suggess ha some porion of performance can, a imes, be aribued o price pressure, which will evenually reverse. In many applicaions of performance evaluaion, one would wan o conrol for his effec. Addiionally, o he exen ha fund managers undersand hese effecs, perverse incenives may exis o exploi he own-fund price pressure of rades o dress, or emporarily enhance, performance, and hereby induce subsequen flows. Since capial flows appear o be more sensiive o good performance han o poor performance, 16 he evenual reversal of he price pressure may no adversely affec capial flows enough o fully offse he iniial benefis. The likelihood of a single fund being able o creae sufficienly persisen price pressure as o induce meaningful capial inflows is low, bu a fund family may well be able o coordinae across is own funds o make such a sraegy worhwhile. The fac ha muual funds canno easily coordinae wih each oher heir conemporaneous selling of overlapping holdings, combined wih an ousider s abiliy o predic which funds will be forced o ransac gives rise o an incenive for predaory rading (Brunnermeier and Pederson (2005)). This can creae a siuaion where arbirageurs have an incenive o desabilize prices relaive o informaionally-efficien values by exploiing firms ha have chosen a capial srucure and organizaional form ha relies on immediaely demandable capial. C. Possibiliies for Fuure Research The cycle where capial flows can force widespread rading in individual securiies, resuling in insiuional price pressure, which in urn affecs fund performance and evenually feeds back ino capial flows, is inriguing. Two possible exensions may warran addiional 16 See for example, Chevalier and Ellison (1997), and Sirri and Tufano (1998). 22

25 research. Firs, he relaion o he momenum effec is enicing. This cycle may well describe he mechanics of why socks ha do well or poorly coninue o do so. The evidence presened here, suggess ha he simple liquidiy-moivaed sraegies we examine are highly correlaed wih momenum, bu offer abnormal reurns beyond he momenum facor. A second research avenue is o examine he role of his cycle in explaining he unusual pricing of echnology socks in he lae 1990s. A he ime, casual empiricism suggess ha focused secor funds holding concenraed posiions in echnology socks iniially ouperformed he broader indices, and consequenly received large inflows, which hey piled ino heir exising holdings. This, in urn, boosed heir performance and led o addiional inflows. V. Conclusion This paper sudies asse fire sales, and insiuional price pressure more generally, in equiy markes by examining a large sample of sock ransacions of muual funds. We find considerable suppor for he noion ha widespread selling by financially disressed muual funds leads o fire sale prices. Somewha surprisingly, we find ha funds wih large inflows behave as if hey oo are consrained o quickly ransac in heir exising posiions, on average buying more of wha hey already own. When inflow-driven purchases are widespread relaive o he poenial sellers of individual securiies, hese forced purchases also resul in persisen insiuional price pressure. These findings sugges ha even in he mos liquid markes here can be a significan premium for immediacy. The price effecs are relaively long-lived, lasing around wo quarers and aking several more quarers o reverse. This evidence adds o previous findings of price pressure effecs around index addiions and sock-financed mergers. Shor-run excess demand curves for socks appear o be less han perfecly elasic. Asse fire sales and inflow-driven purchases are probably he mos significan cos of financial disress for money managemen firms. Mos of hese firms have seleced an organizaional form ha allows capial providers o add or wihdraw capial on demand, 23

26 indicaing ha he expeced coss of demanding liquidiy are low. However, when many funds are forced o ransac he same socks a he same ime, he price impac can be subsanial. The exisence of insiuional price pressure in equiy markes is informaive abou he organizaion of money managemen firms and, in urn, he effec ha hese organizaions have on prices. Firs, i suggess ha he coss associaed wih being informed abou an individual securiy can be subsanial. Meron (1987) argues ha large fixed coss of becoming informed abou an invesmen opporuniy can iniially limi arbirage invesing, and once hey are borne, i can ake a while o learn how bes o exploi he opporuniy. Moreover, hese coss can lead firms o specialize. Specializaion limis heir abiliy o diversify, exposing hem o addiional risks, which Shleifer and Vishny (1997) describe as limis o arbirage. I cerainly appears ha many funds follow highly similar sraegies, such ha here are imes when many face redempions, and are conemporaneously forced o ransac he same securiies. In addiion, i seems ha i akes a while for forced ransacions o be undersood by sraegy ousiders, creaing ime variaion in ransacion coss, and allowing prices o remain apar from heir fundamenal value for several monhs. Imporanly, he asse fire sale sory provides a mechanism for raional mispricing. The marke is clearly somewha inefficien, in ha marke prices are no perfecly reflecive of all available informaion. However, he basis of his mispricing requires neiher irraional invesors nor managers. Prices evenually reflec available informaion, bu someimes wih a significan delay. 24