university of illinois library AT URBANA-CHAMPAIGN BOOKSTACKS

Transcription

1

2 university of illinois library AT URBANA-CHAMPAIGN BOOKSTACKS

3 CENTRAL CIRCULATION BOOKSTACKS The person harging this material is responsible for its renewal or its return to the library from whih it was borrowed on or before the Latest Date stamped below. You may be harged a minimum fee of $75.00 for eah lost book. Theft, mutilation, and underlining of books are reasons for disiplinary ation and may result in dismissal from the University. TO RENEW CALL TELEPHONE CENTER, UNIVERSITY OF ILLINOIS LIBRARY AT URBANA-CHAMPAIGN SEP APR When renewing by phone, write new due date below previous due date. L162

4 Digitized by the Internet Arhive in 2011 with funding from University of Illinois Urbana-Champaign

5 330 STX B385 Faulty Working Paper :1 76 opy THE THE Institutional Trades and Intra-Day Stok Prie Behavior Louis K. C Chan Department of Finane University of Illinois Josef Lakonishok Department of Finane University of Illinois Bureau of Eonomi and Business Researh College of Commere and Business Administration University of Illinois at Urbana-Champaign

6

7 BEBR FACULTY WORKING PAPER NO College of Commere and Business Administration University of Illinois at Urbana-Champaign November 1992 Institutional Trades and Intra-Day Stok Prie Behavior Louis K. C. Chan Josef Lakonishok Department of Finane

8

9 Forthoming, Journal of Finanial Eonomis Institutional Trades and Intra-Day Stok Prie Behavior Louis K. C. Chan Josef Lakonishok August 1992 Both authors are from the College of Commere, University of Illinois at Urbana-Champaign, Champaign, IL We thank Gil Beebower and Vasant Kamath from SEI for providing us with the data and for sharing their insights on various aspets of trading. We also thank Bill Bryan, Eugene Fama, Ken Frenh, Charles Lee, Rihard Leftwih, Harold Mulherin, Mithell Petersen, Mark Ready, Jay Ritter, Andrei Shleifer and an anonymous referee for helpful suggestions, and Rohit Gupta and Peng Tu for researh assistane. This paper has been presented at the Amsterdam Institute of Finane, the Conferene on Seurity Markets Transation Costs at Vanderbilt University, the CRSP seminar at the University of Chiago, INSEAD, the Institute of Quantitative Investment Researh (Europe), the NBER Summer Conferene on Behavioral Finane, Ohio State University, Tel Aviv University, the University of Illinois and the University of Wisonsin-Madison. Computing support was provided by the National Center for Superomputing Appliations, University of Illinois at Urbana-Champaign

10

11 Abstrat This paper examines the prie impat of institutional stok trading, using a unique data set whih reports the transations (large and small) made by 37 large institutional money management firms. The diretion of eah trade and the identity of the management firm behind eah trade are known. While institutional trades are assoiated with some prie pressure, we find that the average prie impat is small. There is also a marked asymmetry between the prie impat of buys versus sells. We relate our findings to various hypotheses on the elastiity of the demand for stoks, the ost of exeuting transations and the determinants of market impat. While market apitalization and relative trade size influene the market impat of a trade, the dominant influene is the identity of the money manager behind the trade.

12

13 This paper uses a unique data set to examine the effets of institutional trading on stok pries. This data set reports the transations made by a sample of 37 large institutional money management firms over the ourse of two and a half years. Eah transation is expliitly identified as a purhase or sale by the money management firm in question; in addition, we are able to distinguish between the trades of different management firms. There are more than one million transations, both small and large, involving issues listed on the New York and Amerian Stok Exhanges. These transations aount for 5 perent of the dollar value traded on these two exhanges over the sample period. We ompare the prie at whih eah transation is exeuted with a variety of benhmark prie measures on the date of the trade. Traking the intra-day behavior of pries around institutional trades allows us to evaluate the prie impat of institutional trades and helps us to disentangle alternative explanations for the soures of suh market impat. Trading on equity markets has beome inreasingly dominated by institutional investors. Shwartz and Shapiro (1990) estimate that in 1989 about seventy perent of trading volume on the New York Stok Exhange is aounted for by member firms and institutional investors. In light of their importane, the impat of institutional trading on stok pries has been the subjet of inreased attention. Some have suggested, for example, that the inreased onentration of trading inreases intra-day prie swings (Report of the New York Stok Exhange's Panel on Market Volatility and Investor Confidene, 1990). Trades in a partiular stok may be generated under a variety of investment styles (ative or passive, value or growth) as well as a variety of order plaement strategies (market or limit orders), on the part of both buyers and sellers. Most of these fators are not diretly observable at the time of the trade. Nonetheless, the general pereption is that institutional investors on average trade frequently in large amounts with aordingly large prie impat. Early studies on the effets of institutional trading are unable to distinguish between institutional and non-institutional transations. Hene,

14 2 they fous on blok trades, trades over 10,000 shares (e.g., Kraus and Stoll (1972)). These and later studies (Holthausen, Leftwih and Mayers (1987, 1990), Ball and Finn (1989)) also suggest several reasons why trades might affet pries. An impat of trades on pries is onsistent with the presene of new information onveyed by transations (Kyle (1985), Easley and O'Hara (1987)), or with the existene of various kinds of market fritions. Suh fritions might reflet the existene of different forms of liquidity osts, inluding the osts of proessing orders (Demsetz (1968)) or ompensation for inventory imbalanes (Amihud and Mendelson (1980), Ho and Stoll (1987)). Alternatively, the market prie of a stok may be affeted by shifts in exess demand beause investors do not reognize the existene of lose substitutes for an individual stok. Measuring the impat of trades on pries thus allows an evaluation of the importane of these different effets on the flow demand/supply shedules for stoks. Numerous studies doument the inability of portfolio managers to outperform various passive benhmarks, despite the onsiderable effort to analyze and selet stoks (Brinson, Singer and Beebower (1991), Fama (1991), Lakonishok, Shleifer and Vishny (1992)). This "implementation shortfall" may be due to the ostliness of atually exeuting trades (Perold (1988)). Indeed, the heavy expenditure of resoures by institutions on trading failities and personnel suggests that exeution osts might be non-negligible and, moreover, that they are potentially ontrollable (Bodurtha and Quinn (1990)). There is, aordingly, great interest in omparing the exeution performane of different money management firms. In evaluating the profitability of various trading rules, researhers also find it neessary to adjust for the osts of trading. In omparison with the literature on blok trades, our results provide evidene from a distintive sample of trades. Previous studies use the tik test to lassify trades as buyer- or seller-initiated. Blok trades at a prie below (above) the prie prior to the blok are onsidered to be seller- (buyer-) initiated. Zero tik trades are in general not lassified. Holthausen, Leftwih and Mayers (1987) find that the tik rule properly lassifies only 52.8 perent of a sample of trades known to be buyer-

15 initiated. Lee and Ready (1991) also explore the auray of the tik test. The earlier studies thus measure the impat of large trades from the perspetive of the relatively more impatient party (i.e., the one willing to trade on an uptik or downtik). There are, of ourse, varying shades to a trader's impatiene to trade. Our results apture the traes of institutional trading ativity on stok pries, aross the spetrum of degrees of impatiene to trade, inorporating many different trading strategies and many different investment styles. What we are after in this paper, therefore, is the average market impat inurred by institutions when altering the omposition of their portfolios. Moreover, our study of differenes aross managers an shed some light on the impat of the degree of impatiene on exeution osts. Our findings suggest that both institutional purhases and sales are aompanied by some prie pressure relative to the opening prie on the trade date. However, there is a marked asymmetry between the behavior of pries after purhases and after sales. After buys, the stok prie ontinues to appreiate; in ontrast, the prie almost fully reovers to its prior level. As a result, the average prie hange weighted by the dollar size of the trade (the prinipal-weighted average) from the open to the lose on the trade day for buys is 0.34 perent, while the orresponding average prie hange for sells is perent. This asymmetry is intriguing, and we provide several onjetures as to its soure. The prie impat of transations is related to the market apitalization of the stok, and to the relative diffiulty of the trade. But an even more dominant influene on the trade's prie impat is the identity of the money management firm behind the trade, suggesting onsiderable differenes aross management firms. While our findings indiate that institutional trades are assoiated with some impat on stok pries, the magnitude of the effet pales in omparison to the figures reported by previous authors. Kraus and Stoll (1972), for example, find that large buy bloks are assoiated with a prie hange (from the prior losing prie to the lose on the trade date) of 1.41 perent, while our findings (based on both small and large transations) suggest that the open-to-lose average prie hange for buys is only 0.34 perent. The low magnitude of the prie impat also has strong

16 impliations for the muh-debated issue of the market impat ost of exeuting trades. A manager who gives up only an eighth of a point eah way would inur a round-trip ost of 0.68 perent on a typial stok. In ontrast, we are hard-pressed to find a round-trip market impat ost, in the aggregate over all trades, exeeding 0.36 perent. The modest estimates of market impat osts attest to the keenly ompetitive nature of the investment industry. The remainder of the paper is organized as follows. Setion 1 outlines alternative explanations for the prie impat of trades, and desribes the data. The empirial results are desribed in Setion 2. Setion 3 relates our results on the prie impat of trades to the question of measuring the osts of exeuting trades. A final setion provides the onlusions. 1. Preliminaries 1.1. Reasons for Prie Impat Sholes (1972), Mikkelson and Parth (1985), Harris and Gurel (1986), Shleifer (1986), Loderer, Cooney and Van Drunen (1991) examine the elastiity of the demand for stoks. Three potential explanations for prie hanges triggered by large trades are suggested in the literature: (i) short run liquidity osts, (ii) imperfet substitution, and (iii) information effets. Short run liquidity osts arise beause of the diffiulty in finding immediately willing buyers or sellers. Efforts to attrat buyers or sellers translate into prie onessions. In many large trades, blok traders provide some of the liquidity, and are ompensated at least in part by a prie onession. A timely return of pries following a trade to the prior equilibrium is onsistent with this explanation. Pries will also hange around large trades if there are no perfet substitutes for a partiular stok. Hene, a seller faes a downward sloping demand urve and a buyer an upward sloping supply urve. Thus, the seller in a large transation has to offer a disount to indue buyers to absorb the extra shares. Similarly, a premium has to be offered by the buyer in a large transation. It stands to reason that the slope of the demand and supply urves will depend on the length of the horizon, although this point has not been emphasized in the literature. The imperfet substitution hypothesis is

17 onsistent with permanent prie hanges or muh slower reversals following the trade, ompared to the preditions of the short run liquidity hypothesis. Permanent prie hanges aused by large trades are also expeted if the trades reveal private information that is subsequently impounded into the new equilibrium prie. Informed sellers believe that the stok is overpried, and informed buyers that the stok is underpried. The information effet depends on the identity of the buyer or seller and in many studies the size of a transation is used as a proxy for the information ontent of the trade Transation Data The data set used in this study onsists of the transations made by 37 large money management firms from July 1986 until the end of The data are provided by SEI Corp., a large onsulting firm in the area of finanial servies for institutional investors. For eah transation, the CUSIP number of the stok is reorded, along with the trade date, the number of shares, the dollar amount of the trade and dollar ommissions. Eah trade is identified as a purhase or a sale, and there is an indiator for the money management firm behind the trade. Eah money management firm is identified to us only by a numeri ode. We math up the SEI data on trades with the reord of transation pries provided by the Franis Emory Fith Co. Sine the SEI data do not ontain a time stamp, we annot identify the transation pries immediately before or after a speifi trade by a money manager. In order to understand how trades are exeuted, however, what we would atually like to know are the market onditions when the portfolio manager atually deided to trade (whih ould be muh earlier than the atual exeution of the trade). Suh information, of ourse, is not generally available. Table 1 provides some desription of our sample. We analyze 1,215,387 transations, amounting to a value traded of billion dollars. The sample is very large when ompared to those used in previous studies and aounts for about 5 perent of the dollar value traded on the NYSE and AMEX. As Lakonishok, Shleifer, Thaler and Vishny (1991) find, most of the trading ativity of institutions is in the largest stoks. The top deile by market

18 apitalization aounts for 48 perent of the trades and 61 perent of the dollar value traded. In ontrast, the bottom 40 perent of the stoks by market apitalization aount for 3.7 perent of the trades and only 0.6 perent of the dollar value traded. Previous studies on prie impat fous on trades exeeding 10,000 shares. Our results indiate that many of the institutional trades are atually quite small. From Panel A of Table 2, the average number of shares per trade is only 8400 for buys and 9100 for sells, and the medians are 2400 and 2700 for buys and sells, respetively. Moreover, 25 perent of the trades involve less than 1000 shares, and only about 20 perent of the trades involve more than 10,000 shares. The small size of a typial trade is surprising, given that our data ome from large money managers who are expeted to be involved in larger trades. The small trade sizes relative to typial holdings are onsistent with the view that managers trade strategially in order to redue the influene of short run liquidity osts, or information effets. Previous studies find that the number of bloks traded on a downtik substantially exeeds the number of bloks traded on an uptik. One explanation suggested for this phenomenon is that it is easier to sell large amounts than to buy large amounts. Therefore, we expet to find that sells are larger than the orresponding purhases. This is indeed the ase, although the differenes are quite small. For example, in the largest stoks, the mean number of shares traded is 8200 and 8700 for buys and sells, respetively. Panel B (Table 2) presents the distribution of the dollar value of trades. The median trade is less than $100,000 and only about 6 perent of the trades exeed $1,000,000. As the ompany size inreases, the trades also get large. In Panel C, we report the distribution of trade size relative to normal daily trading volume. The median is only 2 perent, indiating that a typial institutional trade is not a major event. However, as the size of the ompanies dereases, the typial institutional trade beomes a more signifiant event. For example, the median for buys is 0.24 in group 1, relative to 0.01 in group 4. The largest 1 perent of trades are many times

19 larger than the typial daily volume in small stoks, whereas in the largest stoks suh trades are typially less than 40 perent of the daily volume. Many studies fous on trades whih are larger than the typial trading volume. Our results indiate that suh trades are very unommon, at least in the more liquid stoks where most institutional holdings are onentrated. 2. Empirial Results 2.1. The Prie Impat of Institutional Purhases and Sales Table 3 summarizes the prie impat of institutional purhases (panel (A)) and institutional sales (panel (B)), together with the perentage ommission ost. For eah transation, the perentage return is alulated from the day's opening prie to the trade, and from the trade to the losing prie; the perentage return from the opening to the losing is also reported. 1 These orrespond, respetively, to the total, temporary and permanent effets on the stok prie on the trade date, as disussed in Holthausen et al. (1987). Further, to determine whether a typial institutional trade is fundamentally distinguishable from other trades, we ompare the transation prie in a stok to the volume-weighted average of all transation pries in the same stok on the trade date. In the subsequent disussion, we fous on the prinipal-weighted average of eah prie impat measure. This proedure follows the norm in the investment industry, and permits evaluation of the overall dollar amount of the prie impat. Pries for institutional purhases are 0.22 perent higher than the opening prie on the trade date on a prinipal-weighted average basis. Suh a differene amounts to eight ents per share, less than one tik, on a stok with a prie of $36.50 (the volume-weighted average prie over our sample period). The prie inrease from the open to the trade is onsistent with all three hypotheses outlined in setion 1.1. In part, the rise also reflets the average daily upward drift in pries, although this omponent is small the mean total perentage hange from the open to the lose on the Standard and Poor's Composite Index over this period is 0.06 perent. A final interpretation of the prie movement from the open to the trade is that

20 8 institutional money managers may be responding passively to hanges in the stok prie before initiating transations. Sharply at odds with the reversal predited by the short-run liquidity hypothesis, we find that there is a further prinipal-weighted average prie inrease of 0.12 perent from the trade to the losing prie. It is possible that the prie pressure after the trade is a result of follow-up trades in the same stok. These additional trades might be initiated by the same manager as part of a larger trading program, or by other managers, to the extent that they engage in "herding" behavior. For institutional purhases, the permanent prinipal-weighted prie hange from the open to the lose is 0.34 perent. The simple mean prie hange is lower (0.26 perent), and is onsiderably less than previous estimates of the prie impat of blok purhases. Kraus and Stoll (1972) and Holthausen et al. (1990) find that the average permanent prie hange is around one perent. There are several reasons why it is not surprising that earlier papers doument larger prie effets. These studies fous only on large blok transations. In addition, their reliane on the tik test to infer trade diretion results in the exlusion of bloks assoiated with zero prie tiks. Finally, it is also quite probable that the remaining transations (those assoiated with an up or down tik) represent trades initiated by relatively less patient investors (i.e., those willing to pay a larger prie onession in exhange for greater immediay). Aordingly, the average prie impat is likely to be larger in the ase of purhases or sales seleted on the basis of a non-zero tik, ompared to purhases or sales in general (whether initiated by the investor or not). Another possible reason why we find lower prie impat is that past studies of blok trading use data from earlier periods (no later than 1983). Dramati hanges have sine ourred in equity markets with respet to trading volume and tehnology, ommission rates, and the growth of hedging instruments. Table 3 also reports the median and other perentiles for eah measure of prie impat. Relative to the open or the lose, the median impat for buys is zero while the median permanent hange for buys is also zero. Evidently, the typial institutional purhase has little or no impat on

21 pries. However, the perentiles of the distribution of returns indiate that there is substantial dispersion aross trades with respet to their prie impat Another perspetive on the prie impat of institutional orders is obtained by omparing the trade prie to an average of transation pries from the same day. Berkowitz, Logue and Noser (1988) interpret the prie impat relative to the volume-weighted average prie as a measure of exeution ost. Using this benhmark, the dollar-weighted average impat is very small, at 0.02 perent. Similar values are obtained if the alulation of the volumeweighted prie exludes the trade under onsideration, or if the simple average prie is used as the benhmark. Indeed, the simple average impat is slightly negative, whih would imply, under the interpretation of Berkowitz, Logue and Noser (1988), a negative exeution ost on average to buying! Turning to institutional sales (panel (B) of Table 3), there is a prinipal-weighted average drop in pries of 0.14 perent from the open to the trade. Many of the same fators as in the ase for buys an aount for this hange. In marked ontrast to the behavior of pries after buys, however, the initial prie deline is almost fully reversed. As a result, there is only a small permanent hange of perent. The post-trade behavior of pries in the ase of sells is thus more supportive of effets due to short-term liquidity osts, rather than imperfet substitution or information. The results in panel (B) are reminisent of the findings in Kraus and Stoll (1972) and Holthausen et al. (1990). However, we find muh smaller prie impats than reported in these earlier studies. Overall, the evidene suggests that institutional sales are assoiated with some downward prie pressure, although the market impat is generally small and temporary. It might be argued that the differenes between the findings in Table 3 and the findings of earlier researh are due to differenes in ommission rates. If the speialist or blok trader on the other side of the trade is ompensated by a ommission as well as a prie onession, then a lower onession might be exhanged for a higher ommission. Similarly, differenes between the ommission rates for purhases and sales might aount for the differential prie impat. In Table 3, however, the prinipal-weighted

22 10 ommission rate is the same for buys and sells, at 0.17 perent of trade value (six ents per share on a stok with the average prie of $36.50). Moreover, the simple average ommission rate, 0.23 perent, is muh smaller than the mean ommission rate of 1.01 perent for the largest stoks over the period 1960 to 1979, reported by Stoll and Whaley (1983) The Asymmetri Response of Pries to Purhases and Sales A key puzzle emerges from Table 3: there is a marked asymmetry between the effet of institutional buying and selling ativity on stok pries. 2 Purhases of a stok are aompanied by an inrease in its prie, whih ontinues to rise after the trade; sales of a stok are aompanied by a drop in its prie, but there is subsequently an almost omplete reovery in the prie. Several fators, not mutually exlusive, might aount for the differenes between the effets of buying and selling ativity. "Street wisdom" suggests that brokers are willing to aommodate ustomers' sales by purhasing shares and holding them in inventory in exhange for a short-term prie onession. On the other hand, brokers are more relutant to aommodate ustomers' purhases by undertaking short positions. Sine an intermediary is less likely to be involved on the other side of an institutional purhase, it is less likely that the transation prie in the ase of a buy inorporates a fee to the intermediary in the form of a temporary prie onession. Information effets might also be stronger for purhases than for sales. Sine an institutional investor typially does not hold the market portfolio, the hoie of a partiular issue to sell, out of the limited alternatives in a portfolio, does not neessarily onvey negative information. Rather, the stoks whih are sold may already have met the portfolio's objetives, or there may be other mehanial rules, unrelated to expetations about future performane, for reduing a position. As a result, there are many liquiditymotivated reasons to dispose of a stok. In ontrast, the hoie of one speifi issue to buy, out of the numerous possibilities on the market, is likely to onvey favorable firm-speifi news.'* The information ontent of

23 11 purhases might be diluted insofar as the portfolio reeives net ash inflows. However, Table 1 suggests that purhases and sales by our sample of money managers are roughly equal. Moreover, net ash inflows to the typial money manager are a very small perentage relative to the manager's turnover. The larger, positive impat of institutional purhases ould also arise if institutions are positive feedbak traders for buys but not for sells, i.e., they intensify their buying behavior on days when the market rises. This explanation, however, is not ompatible with the data. For every day in the sample period, we measure the rate of return from the open to the lose on the S&P 500 index. Moreover, for every day, we know the dollar value of buying and selling ativity by our sample of money managers. We then alulate the dollar-weighted average return for buys and sells separately. This produes a prinipal-weighted average return on the index of 0.05 perent for buys, and 0.08 perent for sells. If anything, this finding suggests that money managers might stabilize markets through negative feedbak strategies. In summary, the prie impat of sales is not merely the reverse of the impat of purhases. While the behavior of the stok prie after buys reflets new information or inelasti exess demand urves, the prie behavior after sells is more indiative of a liquidity-related reversal. In any ase, the average and median prie effets are not large, and exeution pries for institutional trades do not differ very muh from average pries over the ourse of the day Firm Size, Trade Diffiulty and Prie Impat Prior theoretial and empirial researh suggests that the prie impat of a trade is affeted by firm size (Loeb (1983), Stoll and Whaley (1983), Keim and Madhavan (1991)), and by the size of the transation (Easley and O'Hara (1987), Glosten (1989)). Table 4 examines the behavior of the prie impat of trades as both firm size and trade omplexity (trade size relative to normal daily volume) vary. Within eah of the four ategories of firm size (desribed in Table 1), trades are divided into four groups by trade omplexity, using the quart iles of the distribution of trade omplexity (as reported in Table 2, panel ). In addition, the bottom panel of the table

24 12 aggregates aross omplexity groups within eah size group, and the last olumn in the table aggregates aross size groups. The table reports the prinipal-weighted averages. In the bottom panel of Table 4, the return from the open for buys rises monotonially as firm size delines, exept for the smallest firms. The prie ontinuation is also stronger after purhases of smaller firms. Taken together, the average prie hange from open to lose for institutional purhases is positive and tends to be higher for smaller firms, ranging from 0.29 perent for the largest firms to 0.49 perent for the smallest. For sell orders, the drop from the opening prie to the exeution prie is also stronger for smaller firms. However, the subsequent reovery is also stronger for smaller firms. As a result, there is no lear pattern aross the four size groups with respet to the permanent prie hange the prie remains roughly unhanged or delines slightly from the open to the lose for sells. The larger permanent prie hange assoiated with purhases of smaller firms ould be due to several reasons. Even a minor institutional stake in a small stok might involve several suessive trades, so that the market impat of a purhase might be spread out over several days before a reversal ours. Further, the market might interpret institutional purhases of smaller stoks as more reliable indiators of favorable private information. Unless an investment manager speializes in lower-apitalization stoks, the deision to purhase a small stok is generally risky for the manager. If the stok's performane is disappointing, the manager may be asked to aount for his deision to depart from the norm and invest in small stoks (Lakonishok, Shleifer, Thaler and Vishny (1991)). Hene, a manager must have strong favorable beliefs about a small stok to justify its purhase. Sales by institutional money managers, on the other hand, need not onvey muh new information to the market, even for the smallest stoks. Suh sales might represent "window dressing, " attempts by managers to avoid potentially embarrassing questions from their lients by removing poorly performing small stoks from their portfolios. Investment poliies regarding minimum levels of market apitalization, dividend yield, or the number of analysts following a

25 13 stok may prompt a manager to sell stoks even in the absene of unfavorable information. When transations are divided into four ategories by omplexity (the last olumn of Table 4), the results are generally similar to those obtained for trades ordered by firm size. In partiular, the prinipal-weighted average permanent prie hange for purhases inreases monotonially with trade omplexity, rising from 0.17 perent for the easiest trades to 0.39 perent for the hardest trades. The permanent prie hange for sales is generally small, even in the ategory of the hardest trades. The two polar ases in the body of Table 4 provide further detail on the assoiation between prie impat, firm size and trade omplexity. In the ase of the easiest trades in the largest firms, the prie hanges are small: the permanent impat for purhases (sales) is 0.11 perent (0.05 perent). At the other end of the sale, the permanent prie hange for the hardest purhases of the smallest stoks is 0.72 perent, omprising a return of 0.23 perent from the open and a prie ontinuation of 0.49 perent after the trade. Sell transations in this ategory are assoiated with a drop of 0.57 perent from the opening prie, but there is a subsequent reversal of 0.71 perent to the lose. Nonetheless, the prie hanges assoiated with even the hardest trades in the smallest stoks are not partiularly large, ompared to other researhers' estimates of the osts of trading small stoks in general. In partiular, sine the average stok prie of trades in this group is only about $10, even a hange of 0.72 perent is substantially less than one tik. If the volume-weighted prie is used as the benhmark, the prie impat provides little basis for disriminating between trades with different harateristis: the average impat of the easiest purhases in the largest stoks is perent, while the average impat of the most diffiult purhases of the smallest stoks is, surprisingly, even more favorable at perent. For the smallest firms, however, the size of the prie impat of both buys and sells is sensitive to whether the trade is inluded in, or exluded from, the volume-weighted average. In the ategory of the most diffiult trades in the smallest stoks, for example, exluding the trade from

26 14 the alulation of the volume-weighted average prie yields a prie impat of 0.01 perent for buys and perent for sells. The results in Table 4 onfirm the asymmetry between buys and sells aross every ategory of firm size and trade omplexity. The positive permanent impat of buys is onsistent with information effets or downward sloping demand urves due to imperfet substitution. In ontrast, sell transations are assoiated with only minor permanent prie hanges. Any initial downward pressure on pries is generally reversed by the end of the trading day, suggesting the existene of short-term liquidity osts Differenes in Prie Impat Aross Money Managers The market impat of a transation an vary with the style of the money manager and the performane of the trading desk responsible for the trade. A entral determinant of exeution performane is the portfolio manager's instrutions to the trading desk as to how an order is to be filled. For example, a value-oriented manager with low turnover will typially give muh latitude to the trading desk, sine urgeny is not onsidered ritial. On the other hand, a manager pursuing a short-term tehnial trading strategy will insist on speedy exeution, thereby onstraining the trading desk. Given the onstraints imposed by the money manager, the trading desk still has onsiderable flexibility as to how a trade is arried out (Wagner (1989)). Its hoies inlude: whether or not to employ a broker who is willing to ommit apital to failitate trades (a apital broker); how many brokers to employ; how muh of an order to expose to eah broker; the time frame within whih the trade is to be exeuted; as well as the leeway given to the broker as to how to omplete the trade (a market order, limit order or market-notheld order, for example) or how muh information about an order is displayed to the publi (as in a "sunshine trade"). In suh a ompliated proess, different managers with varying styles and levels of expertise are likely to turn in different levels of exeution performane. An extended haraterization of the various styles and trading strategies adopted by different money managers, together with their resulting impat on stok pries, is beyond the sope of this paper (see Lakonishok,

27 15 Shleifer and Vishny (1991, 1992)). Here we adopt the less ambitious tak of only doumenting the existene of dispersion aross money management firms with respet to the prie impat of their trades. In Table 5, summary statistis are presented for the distribution aross management firms of three of our measures of prie impat. For eah of the 37 money management firms, the different returns are alulated and then averaged (using trade prinipal as weights) aross all the firm's trades. The summary statistis in Table 5 are based on these 37 observations for eah prie impat measure. Considerable variation exists aross managers for both buys and sells under eah measure of prie impat. The variation annot be attributed simply to noise the average prie impat of eah manager is based on tens of thousands of trades, so that the preision of eah estimate is high. For example, the exeution performane for buys relative to the opening prie varies from perent in the tenth perentile to 0.54 perent in the ninetieth perentile, yielding a differene of a full perentage point per transation. The orresponding differene for sells is very similar, at 0.98 perent per transation. Insofar as the opening prie is known if and when a manager hooses to trade, the differenes aross managers in their exeution performane relative to the open might reflet several soures: their differential skill in seeking out liquidity; ability in trading before the release of information; as well as differenes in their responses to prie movements subsequent to the opening. The dispersion aross managers, in terms of the post-trade return till the lose, is also notable but substantially lower. For buys (sells), the tenth perentile is perent (0.01 perent) and the ninetieth perentile is 0.25 perent (0.26 perent), giving rise to a differene of 0.26 perent (0.27 perent) per transation. Given that the manager has already traded, and given that a trading strategy annot be based on the as yet unknown losing prie on the trade date, the dispersion in managers' post-trade returns should be expeted to be smaller than the dispersion in their pre-trade exeution performane. Our onfidene that the differenes aross managers an be asribed to differenes in styles and trading strategy, rather than noise, would be heightened if a manager who obtains favorable exeution for buys also fares

28 16 well for sells. This is indeed the ase: the rank orrelation aross managers between performane for buys and sells relative to the opening prie is -0.84, for performane relative to the losing prie and for the permanent prie hange. In other words, a manager who buys low relative to the opening prie (or relative to the losing prie) also tends to sell high relative to the opening prie (or relative to the losing prie). As another step in traing the soures of the ross-setional differenes in prie impat, we also obtained data from SEI on a subset of sixteen of the management firms in our sample. In partiular, data are available on eah of these managers' average turnover rate, and investment style (eah manager is lassified as pursuing either a value-oriented or growth-oriented style). Other things equal, a portfolio manager with low turnover would tend to be a more patient investor and would thus tend to have low prie impat. In addition, an investor for whom timing is more ritial (suh as a growth-oriented manager) would be expeted to have a larger impat. Based on the data for sixteen managers, a ross-setional regression onfirms that the prinipal-weighted average prie impat relative to the open for buys inreases with the turnover rate and is higher for a growth-oriented manager: the estimated interept is -0.32, while the oeffiient for turnover rate is 0.37 and the oeffiient for the dummy variable representing the manager's style (zero for a value-oriented style and one for a growth-oriented style) is The prinipal-weighted average prie drop from the open for sells also tends to be larger for a manager with high turnover and with a growth-oriented style: the estimated interept is 0.35, and the oeffiients for turnover and style are and -0.27, respetively. 6 Similar results are obtained if the prinipal-weighted return from the open to the lose is used as the dependent variable. While the results from these regressions are only suggestive (given the small number of managers), they are onsistent with the notion that the degree of urgeny to trade, as refleted in different investment style or trading strategies, is assoiated with the level of prie impat

29 Regression Results Following the lead of prior researh, the previous setions onfirm the influene of firm size and trade diffiulty on the prie impat of a trade. The unique features of our dataset enable us to suggest another potential influene, namely the identity of the manager behind eah trade. It is thus natural to ask whether, after ontrolling for firm size and trade diffiulty, the manager's identity is an important determinant of a trade's prie impat. There may also be a trade-off between the ommission ost and the market impat of the trade. These various influenes are aommodated in the following regression model: a + Pi + E 8 J s ij + E YjDij + E <Pi M ii ~1 j-1 J-l + e i For eah trade i, rj is one of the three measures of prie impat that we fous on: the perentage return from the open to the trade, from the trade to the lose, and from the open to the lose. The ommission ost for the ith trade is denoted by -, and following the ommon pratie in the investment industry, is measured in ents per share (Marshall, 1988). It is likely that the manager's trading desk pereives the trade-off (if any) in terms of the dollar ommission ost, rather than in terms of the ommission rate. In the U.S., unlike other ountries, the ommission ost for institutional investors is on a ents per share basis, irrespetive of the stok prie level, rather than in terms of the total value of the trade. Thus, for the same trade, a broker harging four ents per share will be heaper than a broker harging eight ents per share. However, the heaper broker, if assigned trades in lower-pried stoks, will appear to have a high perentage ommission rate. In evaluating the relation between ommission ost and prie impat aross trades with different pries, therefore, it is neessary to express the ommission ost on a dollar basis rather than on a perentage basis. Expressing the ommission ost relative to the trade prie would also onfound the effet of ommissions with the effet of market apitalization (sine smaller stoks tend to have lower pries). The effets of market

30 18 apitalization, trade diffiulty and managerial strategy are aptured by the dummy variables, S--, D-. and M-, respetively. For example, M- takes the value of one if the ith trade is exeuted by the jth manager and is zero otherwise. To permit identifiation, the oeffiients for the dummy variables for managers are normalized relative to the first manager in the data set. Similarly, the oeffiients for the trade diffiulty variables are expressed relative to the impat of trades in the first ategory (the easiest trades), while the oeffiients for firm size are expressed relative to the impat of trades in the largest firms. Separate regressions are fit for buy transations and sell transations. In addition, the marginal explanatory power of eah set of dummy variables is assessed by exluding eah set, one at a time, from the full model (1). Panel A of Table 6 reports the adjusted R 2 for eah speifiation of the regression model. Most of the explanatory power of the model omes from the identity of the money manager behind the trade. In ontrast, exluding the dummy variables for firm size and trade omplexity has little or no effet on the R. In light of the importane of the manager dummies, it is perhaps not surprising that the model provides the best fit in the equation for the return from the open to the trade. This measure of prie impat, to a larger extent than the others, reflets the effets of managerial trading strategy. In panel B, the oeffiients of the full model are reported for eah of the three measures of prie impat. Given the very large sample size, nearly all of the estimated oeffiients are large relative to their standard errors. Therefore, the fous of the disussion will be on the eonomi signifiane of the oeffiients. One presumption is that favorable exeution (lower prie impat) is purhased from a broker in exhange for a higher ommission fee. However, the oeffiient for the ommission ost variable for both buys and for sells (in parentheses) is very small. The most favorable evidene on substitution between the prie impat of a trade and its ommission ost emerges in the equation for the prie impat of sells relative to the losing prie. Even in this ase, however, an inrease in the ommission of one ent per share (in itself a large jump in ommissions) lowers the post-trade prie reversal by

31 perent, yielding a dollar savings of only 0.3 ents per share on a stok with the average prie of $ We also estimated the regression with the ommission ost measured relative to the trade prie as in the results reported in Table 5, no relation an be deteted between prie impat and ommission rates. As Beebower (1989) points out, however, the ommission inludes payment for researh servies and other plan expenses. The presene of suh servies, not related to trade exeution, would blur any assoiation between prie impat and the total ommission ost. In addition, some brokers may be willing to ommit their own apital to aommodate managers' trades, while others may simply proess transations. With respet to the influene of firm size and omplexity, the results in panel B onfirm the findings of the previous setions. What is partiularly noteworthy, however, is that the oeffiients of the dummy variables for money managers still display onsiderable dispersion for example, the spread between the tenth and ninetieth perentiles is 0.72 (0.85) when returns are measured from the open to the trade. While somewhat attenuated relative to the findings of Table 5, these spreads are still onsiderable. 3. The Exeution Cost of Institutional Trades The temporary and total prie impat of institutional trades, and the impat relative to various intra-day averages, reported in the previous setion, an also be interpreted as average exeution osts for purhases and sales. In partiular, the differene between the prie at whih an order is exeuted and the underlying true value of the stok amounts to a prie onession whih is a ost of trading, in addition to brokerage ommissions. While onsiderable resoures are expended within the investment ommunity on monitoring and ontrolling suh trading osts, there is little onsensus as to the magnitude of exeution osts. In pratie, part of the disagreement stems from the different hoies of a benhmark prie; the losing prie of the stok on the trade date, the opening prie and the volume-weighted average prie are all used. In Table 3, average round-trip osts inlude ommissions (whih are 0.34 perent of trade value), and market impat osts: these range

32 20 from 0.09 perent relative to the volume-weighted prie to 0.36 perent relative to the opening prie. If the losing prie is used as the benhmark, the ost of sells is roughly offset, on average, by a benefit for buys, sine there is a post-purhase average prie ontinuation. Further, if the estimates of trading ost are disaggregated by market apitalization and trade omplexity, the average market impat osts are smaller than the orresponding figures in Loeb (1983) or Stoll and Whaley (1983). In addition, the osts relative to the open tend to move with market apitalization and trade omplexity in the expeted diretion. Assuming that the deision to trade is made before the open, and thus using the opening prie as the benhmark, the round-trip ost, inluding ommissions, for the hardest trades in the smallest stoks is 1.90 perent (from Table 4); the orresponding ost for the easiest trades in the largest stoks is 0.29 perent. Costs relative to the volumeweighted prie, however, display very little variation aross trades in large and small stoks, or aross diffiult and easy trades. The various measures of exeution ost are not without shortomings. The opening prie may not be a relevant benhmark prie if the order is not submitted to the trader before trading begins. To one degree or another, eah ost measure an be gamed by traders who are being evaluated. A trader an postpone trading until lose to the end of the trading day and then hoose to exeute only those transations whose pries are better than the open or the intra-day average prie; the remaining orders are deferred. Similarly, a trader who arries out a large transation will have a major influene on the volume-weighted prie, distorting the ost alulation. None of these ost measures addresses the issue of opportunity ost (inluding the ost of unexeuted orders), or the potential adverse seletion problem (the possibility that the trader may be "bagged" by buying heaply a stok that subsequently experienes negative performane). It would thus seem advisable, in evaluating exeution performane, to onsider a broad range of ost measures, rather than a single number.

33 21 4. Summary and Conlusion Analysis of the prie impat of institutional trades sheds light on the elastiity of the exess demand urve for stoks, and on the magnitude of the ost of exeuting transations. Previous studies on the prie impat of trades, however, have foussed on the effets of blok trades and in some ases, have onsidered only the largest bloks. In these studies, moreover, the hange in the transation prie itself is used to infer whether a trade is initiated by the buyer or by the seller. In ontrast, our sample overs a more reent period and ontains more than one million trades, both large and small, by 37 large institutional money managers. Eah trade is expliitly identified as a purhase or sale by the money manager, who is also identified. The distintive features of our data set enable us to generalize and extend previous studies on the prie impat of blok trades. Overall, the evidene suggests that institutional purhases and sales of a stok are assoiated with some pressure on pries. Relative to the opening prie on the trade date, for example, buy transations are assoiated with a prinipalweighted average prie inrease of 0.22 perent while sell transations are assoiated with a prinipal-weighted average prie deline of 0.14 perent. The behavior of pries from the open to the trade an be attributable to short-run liquidity osts, prior release of information or positive feedbak trading behavior by managers. The post-trade behavior of pries is more perplexing, and displays a sharp differene between buys and sells. Speifially, the prie ontinues to rise after purhases the prinipal-weighted average return from the trade to the losing prie is 0.12 perent while the prie tends to orret itself after sales the reversal is 0.10 perent. The post-trade reversal for sells is onsistent with the existene of short-run liquidity osts, while the postpurhase behavior of pries is onsistent with information effets, or imperfetly elasti demand urves. We find that institutional purhases are assoiated with a prinipalweighted permanent prie hange from the open to the lose on the trade date of 0.34 perent, while there is only a very small permanent impat (-0.04 perent) from institutional sales. The asymmetry is also noted in